| File: | src/sentinel.c |
| Warning: | line 1992, column 14 Although the value stored to 'slave' is used in the enclosing expression, the value is never actually read from 'slave' |
Press '?' to see keyboard shortcuts
Keyboard shortcuts:
| 1 | /* Redis Sentinel implementation |
| 2 | * |
| 3 | * Copyright (c) 2009-2012, Salvatore Sanfilippo <antirez at gmail dot com> |
| 4 | * All rights reserved. |
| 5 | * |
| 6 | * Redistribution and use in source and binary forms, with or without |
| 7 | * modification, are permitted provided that the following conditions are met: |
| 8 | * |
| 9 | * * Redistributions of source code must retain the above copyright notice, |
| 10 | * this list of conditions and the following disclaimer. |
| 11 | * * Redistributions in binary form must reproduce the above copyright |
| 12 | * notice, this list of conditions and the following disclaimer in the |
| 13 | * documentation and/or other materials provided with the distribution. |
| 14 | * * Neither the name of Redis nor the names of its contributors may be used |
| 15 | * to endorse or promote products derived from this software without |
| 16 | * specific prior written permission. |
| 17 | * |
| 18 | * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" |
| 19 | * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE |
| 20 | * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE |
| 21 | * ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE |
| 22 | * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR |
| 23 | * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF |
| 24 | * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS |
| 25 | * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN |
| 26 | * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) |
| 27 | * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE |
| 28 | * POSSIBILITY OF SUCH DAMAGE. |
| 29 | */ |
| 30 | |
| 31 | #include "server.h" |
| 32 | #include "hiredis.h" |
| 33 | #ifdef USE_OPENSSL |
| 34 | #include "openssl/ssl.h" |
| 35 | #include "hiredis_ssl.h" |
| 36 | #endif |
| 37 | #include "async.h" |
| 38 | |
| 39 | #include <ctype.h> |
| 40 | #include <arpa/inet.h> |
| 41 | #include <sys/socket.h> |
| 42 | #include <sys/wait.h> |
| 43 | #include <fcntl.h> |
| 44 | |
| 45 | extern char **environ; |
| 46 | |
| 47 | #ifdef USE_OPENSSL |
| 48 | extern SSL_CTX *redis_tls_ctx; |
| 49 | extern SSL_CTX *redis_tls_client_ctx; |
| 50 | #endif |
| 51 | |
| 52 | #define REDIS_SENTINEL_PORT26379 26379 |
| 53 | |
| 54 | /* ======================== Sentinel global state =========================== */ |
| 55 | |
| 56 | /* Address object, used to describe an ip:port pair. */ |
| 57 | typedef struct sentinelAddr { |
| 58 | char *hostname; /* Hostname OR address, as specified */ |
| 59 | char *ip; /* Always a resolved address */ |
| 60 | int port; |
| 61 | } sentinelAddr; |
| 62 | |
| 63 | /* A Sentinel Redis Instance object is monitoring. */ |
| 64 | #define SRI_MASTER(1<<0) (1<<0) |
| 65 | #define SRI_SLAVE(1<<1) (1<<1) |
| 66 | #define SRI_SENTINEL(1<<2) (1<<2) |
| 67 | #define SRI_S_DOWN(1<<3) (1<<3) /* Subjectively down (no quorum). */ |
| 68 | #define SRI_O_DOWN(1<<4) (1<<4) /* Objectively down (confirmed by others). */ |
| 69 | #define SRI_MASTER_DOWN(1<<5) (1<<5) /* A Sentinel with this flag set thinks that |
| 70 | its master is down. */ |
| 71 | #define SRI_FAILOVER_IN_PROGRESS(1<<6) (1<<6) /* Failover is in progress for |
| 72 | this master. */ |
| 73 | #define SRI_PROMOTED(1<<7) (1<<7) /* Slave selected for promotion. */ |
| 74 | #define SRI_RECONF_SENT(1<<8) (1<<8) /* SLAVEOF <newmaster> sent. */ |
| 75 | #define SRI_RECONF_INPROG(1<<9) (1<<9) /* Slave synchronization in progress. */ |
| 76 | #define SRI_RECONF_DONE(1<<10) (1<<10) /* Slave synchronized with new master. */ |
| 77 | #define SRI_FORCE_FAILOVER(1<<11) (1<<11) /* Force failover with master up. */ |
| 78 | #define SRI_SCRIPT_KILL_SENT(1<<12) (1<<12) /* SCRIPT KILL already sent on -BUSY */ |
| 79 | |
| 80 | /* Note: times are in milliseconds. */ |
| 81 | #define SENTINEL_INFO_PERIOD10000 10000 |
| 82 | #define SENTINEL_PING_PERIOD1000 1000 |
| 83 | #define SENTINEL_ASK_PERIOD1000 1000 |
| 84 | #define SENTINEL_PUBLISH_PERIOD2000 2000 |
| 85 | #define SENTINEL_DEFAULT_DOWN_AFTER30000 30000 |
| 86 | #define SENTINEL_HELLO_CHANNEL"__sentinel__:hello" "__sentinel__:hello" |
| 87 | #define SENTINEL_TILT_TRIGGER2000 2000 |
| 88 | #define SENTINEL_TILT_PERIOD(1000*30) (SENTINEL_PING_PERIOD1000*30) |
| 89 | #define SENTINEL_DEFAULT_SLAVE_PRIORITY100 100 |
| 90 | #define SENTINEL_SLAVE_RECONF_TIMEOUT10000 10000 |
| 91 | #define SENTINEL_DEFAULT_PARALLEL_SYNCS1 1 |
| 92 | #define SENTINEL_MIN_LINK_RECONNECT_PERIOD15000 15000 |
| 93 | #define SENTINEL_DEFAULT_FAILOVER_TIMEOUT(60*3*1000) (60*3*1000) |
| 94 | #define SENTINEL_MAX_PENDING_COMMANDS100 100 |
| 95 | #define SENTINEL_ELECTION_TIMEOUT10000 10000 |
| 96 | #define SENTINEL_MAX_DESYNC1000 1000 |
| 97 | #define SENTINEL_DEFAULT_DENY_SCRIPTS_RECONFIG1 1 |
| 98 | #define SENTINEL_DEFAULT_RESOLVE_HOSTNAMES0 0 |
| 99 | #define SENTINEL_DEFAULT_ANNOUNCE_HOSTNAMES0 0 |
| 100 | |
| 101 | /* Failover machine different states. */ |
| 102 | #define SENTINEL_FAILOVER_STATE_NONE0 0 /* No failover in progress. */ |
| 103 | #define SENTINEL_FAILOVER_STATE_WAIT_START1 1 /* Wait for failover_start_time*/ |
| 104 | #define SENTINEL_FAILOVER_STATE_SELECT_SLAVE2 2 /* Select slave to promote */ |
| 105 | #define SENTINEL_FAILOVER_STATE_SEND_SLAVEOF_NOONE3 3 /* Slave -> Master */ |
| 106 | #define SENTINEL_FAILOVER_STATE_WAIT_PROMOTION4 4 /* Wait slave to change role */ |
| 107 | #define SENTINEL_FAILOVER_STATE_RECONF_SLAVES5 5 /* SLAVEOF newmaster */ |
| 108 | #define SENTINEL_FAILOVER_STATE_UPDATE_CONFIG6 6 /* Monitor promoted slave. */ |
| 109 | |
| 110 | #define SENTINEL_MASTER_LINK_STATUS_UP0 0 |
| 111 | #define SENTINEL_MASTER_LINK_STATUS_DOWN1 1 |
| 112 | |
| 113 | /* Generic flags that can be used with different functions. |
| 114 | * They use higher bits to avoid colliding with the function specific |
| 115 | * flags. */ |
| 116 | #define SENTINEL_NO_FLAGS0 0 |
| 117 | #define SENTINEL_GENERATE_EVENT(1<<16) (1<<16) |
| 118 | #define SENTINEL_LEADER(1<<17) (1<<17) |
| 119 | #define SENTINEL_OBSERVER(1<<18) (1<<18) |
| 120 | |
| 121 | /* Script execution flags and limits. */ |
| 122 | #define SENTINEL_SCRIPT_NONE0 0 |
| 123 | #define SENTINEL_SCRIPT_RUNNING1 1 |
| 124 | #define SENTINEL_SCRIPT_MAX_QUEUE256 256 |
| 125 | #define SENTINEL_SCRIPT_MAX_RUNNING16 16 |
| 126 | #define SENTINEL_SCRIPT_MAX_RUNTIME60000 60000 /* 60 seconds max exec time. */ |
| 127 | #define SENTINEL_SCRIPT_MAX_RETRY10 10 |
| 128 | #define SENTINEL_SCRIPT_RETRY_DELAY30000 30000 /* 30 seconds between retries. */ |
| 129 | |
| 130 | /* SENTINEL SIMULATE-FAILURE command flags. */ |
| 131 | #define SENTINEL_SIMFAILURE_NONE0 0 |
| 132 | #define SENTINEL_SIMFAILURE_CRASH_AFTER_ELECTION(1<<0) (1<<0) |
| 133 | #define SENTINEL_SIMFAILURE_CRASH_AFTER_PROMOTION(1<<1) (1<<1) |
| 134 | |
| 135 | /* The link to a sentinelRedisInstance. When we have the same set of Sentinels |
| 136 | * monitoring many masters, we have different instances representing the |
| 137 | * same Sentinels, one per master, and we need to share the hiredis connections |
| 138 | * among them. Otherwise if 5 Sentinels are monitoring 100 masters we create |
| 139 | * 500 outgoing connections instead of 5. |
| 140 | * |
| 141 | * So this structure represents a reference counted link in terms of the two |
| 142 | * hiredis connections for commands and Pub/Sub, and the fields needed for |
| 143 | * failure detection, since the ping/pong time are now local to the link: if |
| 144 | * the link is available, the instance is available. This way we don't just |
| 145 | * have 5 connections instead of 500, we also send 5 pings instead of 500. |
| 146 | * |
| 147 | * Links are shared only for Sentinels: master and slave instances have |
| 148 | * a link with refcount = 1, always. */ |
| 149 | typedef struct instanceLink { |
| 150 | int refcount; /* Number of sentinelRedisInstance owners. */ |
| 151 | int disconnected; /* Non-zero if we need to reconnect cc or pc. */ |
| 152 | int pending_commands; /* Number of commands sent waiting for a reply. */ |
| 153 | redisAsyncContext *cc; /* Hiredis context for commands. */ |
| 154 | redisAsyncContext *pc; /* Hiredis context for Pub / Sub. */ |
| 155 | mstime_t cc_conn_time; /* cc connection time. */ |
| 156 | mstime_t pc_conn_time; /* pc connection time. */ |
| 157 | mstime_t pc_last_activity; /* Last time we received any message. */ |
| 158 | mstime_t last_avail_time; /* Last time the instance replied to ping with |
| 159 | a reply we consider valid. */ |
| 160 | mstime_t act_ping_time; /* Time at which the last pending ping (no pong |
| 161 | received after it) was sent. This field is |
| 162 | set to 0 when a pong is received, and set again |
| 163 | to the current time if the value is 0 and a new |
| 164 | ping is sent. */ |
| 165 | mstime_t last_ping_time; /* Time at which we sent the last ping. This is |
| 166 | only used to avoid sending too many pings |
| 167 | during failure. Idle time is computed using |
| 168 | the act_ping_time field. */ |
| 169 | mstime_t last_pong_time; /* Last time the instance replied to ping, |
| 170 | whatever the reply was. That's used to check |
| 171 | if the link is idle and must be reconnected. */ |
| 172 | mstime_t last_reconn_time; /* Last reconnection attempt performed when |
| 173 | the link was down. */ |
| 174 | } instanceLink; |
| 175 | |
| 176 | typedef struct sentinelRedisInstance { |
| 177 | int flags; /* See SRI_... defines */ |
| 178 | char *name; /* Master name from the point of view of this sentinel. */ |
| 179 | char *runid; /* Run ID of this instance, or unique ID if is a Sentinel.*/ |
| 180 | uint64_t config_epoch; /* Configuration epoch. */ |
| 181 | sentinelAddr *addr; /* Master host. */ |
| 182 | instanceLink *link; /* Link to the instance, may be shared for Sentinels. */ |
| 183 | mstime_t last_pub_time; /* Last time we sent hello via Pub/Sub. */ |
| 184 | mstime_t last_hello_time; /* Only used if SRI_SENTINEL is set. Last time |
| 185 | we received a hello from this Sentinel |
| 186 | via Pub/Sub. */ |
| 187 | mstime_t last_master_down_reply_time; /* Time of last reply to |
| 188 | SENTINEL is-master-down command. */ |
| 189 | mstime_t s_down_since_time; /* Subjectively down since time. */ |
| 190 | mstime_t o_down_since_time; /* Objectively down since time. */ |
| 191 | mstime_t down_after_period; /* Consider it down after that period. */ |
| 192 | mstime_t info_refresh; /* Time at which we received INFO output from it. */ |
| 193 | dict *renamed_commands; /* Commands renamed in this instance: |
| 194 | Sentinel will use the alternative commands |
| 195 | mapped on this table to send things like |
| 196 | SLAVEOF, CONFING, INFO, ... */ |
| 197 | |
| 198 | /* Role and the first time we observed it. |
| 199 | * This is useful in order to delay replacing what the instance reports |
| 200 | * with our own configuration. We need to always wait some time in order |
| 201 | * to give a chance to the leader to report the new configuration before |
| 202 | * we do silly things. */ |
| 203 | int role_reported; |
| 204 | mstime_t role_reported_time; |
| 205 | mstime_t slave_conf_change_time; /* Last time slave master addr changed. */ |
| 206 | |
| 207 | /* Master specific. */ |
| 208 | dict *sentinels; /* Other sentinels monitoring the same master. */ |
| 209 | dict *slaves; /* Slaves for this master instance. */ |
| 210 | unsigned int quorum;/* Number of sentinels that need to agree on failure. */ |
| 211 | int parallel_syncs; /* How many slaves to reconfigure at same time. */ |
| 212 | char *auth_pass; /* Password to use for AUTH against master & replica. */ |
| 213 | char *auth_user; /* Username for ACLs AUTH against master & replica. */ |
| 214 | |
| 215 | /* Slave specific. */ |
| 216 | mstime_t master_link_down_time; /* Slave replication link down time. */ |
| 217 | int slave_priority; /* Slave priority according to its INFO output. */ |
| 218 | mstime_t slave_reconf_sent_time; /* Time at which we sent SLAVE OF <new> */ |
| 219 | struct sentinelRedisInstance *master; /* Master instance if it's slave. */ |
| 220 | char *slave_master_host; /* Master host as reported by INFO */ |
| 221 | int slave_master_port; /* Master port as reported by INFO */ |
| 222 | int slave_master_link_status; /* Master link status as reported by INFO */ |
| 223 | unsigned long long slave_repl_offset; /* Slave replication offset. */ |
| 224 | /* Failover */ |
| 225 | char *leader; /* If this is a master instance, this is the runid of |
| 226 | the Sentinel that should perform the failover. If |
| 227 | this is a Sentinel, this is the runid of the Sentinel |
| 228 | that this Sentinel voted as leader. */ |
| 229 | uint64_t leader_epoch; /* Epoch of the 'leader' field. */ |
| 230 | uint64_t failover_epoch; /* Epoch of the currently started failover. */ |
| 231 | int failover_state; /* See SENTINEL_FAILOVER_STATE_* defines. */ |
| 232 | mstime_t failover_state_change_time; |
| 233 | mstime_t failover_start_time; /* Last failover attempt start time. */ |
| 234 | mstime_t failover_timeout; /* Max time to refresh failover state. */ |
| 235 | mstime_t failover_delay_logged; /* For what failover_start_time value we |
| 236 | logged the failover delay. */ |
| 237 | struct sentinelRedisInstance *promoted_slave; /* Promoted slave instance. */ |
| 238 | /* Scripts executed to notify admin or reconfigure clients: when they |
| 239 | * are set to NULL no script is executed. */ |
| 240 | char *notification_script; |
| 241 | char *client_reconfig_script; |
| 242 | sds info; /* cached INFO output */ |
| 243 | } sentinelRedisInstance; |
| 244 | |
| 245 | /* Main state. */ |
| 246 | struct sentinelState { |
| 247 | char myid[CONFIG_RUN_ID_SIZE40+1]; /* This sentinel ID. */ |
| 248 | uint64_t current_epoch; /* Current epoch. */ |
| 249 | dict *masters; /* Dictionary of master sentinelRedisInstances. |
| 250 | Key is the instance name, value is the |
| 251 | sentinelRedisInstance structure pointer. */ |
| 252 | int tilt; /* Are we in TILT mode? */ |
| 253 | int running_scripts; /* Number of scripts in execution right now. */ |
| 254 | mstime_t tilt_start_time; /* When TITL started. */ |
| 255 | mstime_t previous_time; /* Last time we ran the time handler. */ |
| 256 | list *scripts_queue; /* Queue of user scripts to execute. */ |
| 257 | char *announce_ip; /* IP addr that is gossiped to other sentinels if |
| 258 | not NULL. */ |
| 259 | int announce_port; /* Port that is gossiped to other sentinels if |
| 260 | non zero. */ |
| 261 | unsigned long simfailure_flags; /* Failures simulation. */ |
| 262 | int deny_scripts_reconfig; /* Allow SENTINEL SET ... to change script |
| 263 | paths at runtime? */ |
| 264 | char *sentinel_auth_pass; /* Password to use for AUTH against other sentinel */ |
| 265 | char *sentinel_auth_user; /* Username for ACLs AUTH against other sentinel. */ |
| 266 | int resolve_hostnames; /* Support use of hostnames, assuming DNS is well configured. */ |
| 267 | int announce_hostnames; /* Announce hostnames instead of IPs when we have them. */ |
| 268 | } sentinel; |
| 269 | |
| 270 | /* A script execution job. */ |
| 271 | typedef struct sentinelScriptJob { |
| 272 | int flags; /* Script job flags: SENTINEL_SCRIPT_* */ |
| 273 | int retry_num; /* Number of times we tried to execute it. */ |
| 274 | char **argv; /* Arguments to call the script. */ |
| 275 | mstime_t start_time; /* Script execution time if the script is running, |
| 276 | otherwise 0 if we are allowed to retry the |
| 277 | execution at any time. If the script is not |
| 278 | running and it's not 0, it means: do not run |
| 279 | before the specified time. */ |
| 280 | pid_t pid; /* Script execution pid. */ |
| 281 | } sentinelScriptJob; |
| 282 | |
| 283 | /* ======================= hiredis ae.c adapters ============================= |
| 284 | * Note: this implementation is taken from hiredis/adapters/ae.h, however |
| 285 | * we have our modified copy for Sentinel in order to use our allocator |
| 286 | * and to have full control over how the adapter works. */ |
| 287 | |
| 288 | typedef struct redisAeEvents { |
| 289 | redisAsyncContext *context; |
| 290 | aeEventLoop *loop; |
| 291 | int fd; |
| 292 | int reading, writing; |
| 293 | } redisAeEvents; |
| 294 | |
| 295 | static void redisAeReadEvent(aeEventLoop *el, int fd, void *privdata, int mask) { |
| 296 | ((void)el); ((void)fd); ((void)mask); |
| 297 | |
| 298 | redisAeEvents *e = (redisAeEvents*)privdata; |
| 299 | redisAsyncHandleRead(e->context); |
| 300 | } |
| 301 | |
| 302 | static void redisAeWriteEvent(aeEventLoop *el, int fd, void *privdata, int mask) { |
| 303 | ((void)el); ((void)fd); ((void)mask); |
| 304 | |
| 305 | redisAeEvents *e = (redisAeEvents*)privdata; |
| 306 | redisAsyncHandleWrite(e->context); |
| 307 | } |
| 308 | |
| 309 | static void redisAeAddRead(void *privdata) { |
| 310 | redisAeEvents *e = (redisAeEvents*)privdata; |
| 311 | aeEventLoop *loop = e->loop; |
| 312 | if (!e->reading) { |
| 313 | e->reading = 1; |
| 314 | aeCreateFileEvent(loop,e->fd,AE_READABLE1,redisAeReadEvent,e); |
| 315 | } |
| 316 | } |
| 317 | |
| 318 | static void redisAeDelRead(void *privdata) { |
| 319 | redisAeEvents *e = (redisAeEvents*)privdata; |
| 320 | aeEventLoop *loop = e->loop; |
| 321 | if (e->reading) { |
| 322 | e->reading = 0; |
| 323 | aeDeleteFileEvent(loop,e->fd,AE_READABLE1); |
| 324 | } |
| 325 | } |
| 326 | |
| 327 | static void redisAeAddWrite(void *privdata) { |
| 328 | redisAeEvents *e = (redisAeEvents*)privdata; |
| 329 | aeEventLoop *loop = e->loop; |
| 330 | if (!e->writing) { |
| 331 | e->writing = 1; |
| 332 | aeCreateFileEvent(loop,e->fd,AE_WRITABLE2,redisAeWriteEvent,e); |
| 333 | } |
| 334 | } |
| 335 | |
| 336 | static void redisAeDelWrite(void *privdata) { |
| 337 | redisAeEvents *e = (redisAeEvents*)privdata; |
| 338 | aeEventLoop *loop = e->loop; |
| 339 | if (e->writing) { |
| 340 | e->writing = 0; |
| 341 | aeDeleteFileEvent(loop,e->fd,AE_WRITABLE2); |
| 342 | } |
| 343 | } |
| 344 | |
| 345 | static void redisAeCleanup(void *privdata) { |
| 346 | redisAeEvents *e = (redisAeEvents*)privdata; |
| 347 | redisAeDelRead(privdata); |
| 348 | redisAeDelWrite(privdata); |
| 349 | zfree(e); |
| 350 | } |
| 351 | |
| 352 | static int redisAeAttach(aeEventLoop *loop, redisAsyncContext *ac) { |
| 353 | redisContext *c = &(ac->c); |
| 354 | redisAeEvents *e; |
| 355 | |
| 356 | /* Nothing should be attached when something is already attached */ |
| 357 | if (ac->ev.data != NULL((void*)0)) |
| 358 | return C_ERR-1; |
| 359 | |
| 360 | /* Create container for context and r/w events */ |
| 361 | e = (redisAeEvents*)zmalloc(sizeof(*e)); |
| 362 | e->context = ac; |
| 363 | e->loop = loop; |
| 364 | e->fd = c->fd; |
| 365 | e->reading = e->writing = 0; |
| 366 | |
| 367 | /* Register functions to start/stop listening for events */ |
| 368 | ac->ev.addRead = redisAeAddRead; |
| 369 | ac->ev.delRead = redisAeDelRead; |
| 370 | ac->ev.addWrite = redisAeAddWrite; |
| 371 | ac->ev.delWrite = redisAeDelWrite; |
| 372 | ac->ev.cleanup = redisAeCleanup; |
| 373 | ac->ev.data = e; |
| 374 | |
| 375 | return C_OK0; |
| 376 | } |
| 377 | |
| 378 | /* ============================= Prototypes ================================= */ |
| 379 | |
| 380 | void sentinelLinkEstablishedCallback(const redisAsyncContext *c, int status); |
| 381 | void sentinelDisconnectCallback(const redisAsyncContext *c, int status); |
| 382 | void sentinelReceiveHelloMessages(redisAsyncContext *c, void *reply, void *privdata); |
| 383 | sentinelRedisInstance *sentinelGetMasterByName(char *name); |
| 384 | char *sentinelGetSubjectiveLeader(sentinelRedisInstance *master); |
| 385 | char *sentinelGetObjectiveLeader(sentinelRedisInstance *master); |
| 386 | int yesnotoi(char *s); |
| 387 | void instanceLinkConnectionError(const redisAsyncContext *c); |
| 388 | const char *sentinelRedisInstanceTypeStr(sentinelRedisInstance *ri); |
| 389 | void sentinelAbortFailover(sentinelRedisInstance *ri); |
| 390 | void sentinelEvent(int level, char *type, sentinelRedisInstance *ri, const char *fmt, ...); |
| 391 | sentinelRedisInstance *sentinelSelectSlave(sentinelRedisInstance *master); |
| 392 | void sentinelScheduleScriptExecution(char *path, ...); |
| 393 | void sentinelStartFailover(sentinelRedisInstance *master); |
| 394 | void sentinelDiscardReplyCallback(redisAsyncContext *c, void *reply, void *privdata); |
| 395 | int sentinelSendSlaveOf(sentinelRedisInstance *ri, const sentinelAddr *addr); |
| 396 | char *sentinelVoteLeader(sentinelRedisInstance *master, uint64_t req_epoch, char *req_runid, uint64_t *leader_epoch); |
| 397 | void sentinelFlushConfig(void); |
| 398 | void sentinelGenerateInitialMonitorEvents(void); |
| 399 | int sentinelSendPing(sentinelRedisInstance *ri); |
| 400 | int sentinelForceHelloUpdateForMaster(sentinelRedisInstance *master); |
| 401 | sentinelRedisInstance *getSentinelRedisInstanceByAddrAndRunID(dict *instances, char *ip, int port, char *runid); |
| 402 | void sentinelSimFailureCrash(void); |
| 403 | |
| 404 | /* ========================= Dictionary types =============================== */ |
| 405 | |
| 406 | uint64_t dictSdsHash(const void *key); |
| 407 | uint64_t dictSdsCaseHash(const void *key); |
| 408 | int dictSdsKeyCompare(void *privdata, const void *key1, const void *key2); |
| 409 | int dictSdsKeyCaseCompare(void *privdata, const void *key1, const void *key2); |
| 410 | void releaseSentinelRedisInstance(sentinelRedisInstance *ri); |
| 411 | |
| 412 | void dictInstancesValDestructor (void *privdata, void *obj) { |
| 413 | UNUSED(privdata)((void) privdata); |
| 414 | releaseSentinelRedisInstance(obj); |
| 415 | } |
| 416 | |
| 417 | /* Instance name (sds) -> instance (sentinelRedisInstance pointer) |
| 418 | * |
| 419 | * also used for: sentinelRedisInstance->sentinels dictionary that maps |
| 420 | * sentinels ip:port to last seen time in Pub/Sub hello message. */ |
| 421 | dictType instancesDictType = { |
| 422 | dictSdsHash, /* hash function */ |
| 423 | NULL((void*)0), /* key dup */ |
| 424 | NULL((void*)0), /* val dup */ |
| 425 | dictSdsKeyCompare, /* key compare */ |
| 426 | NULL((void*)0), /* key destructor */ |
| 427 | dictInstancesValDestructor,/* val destructor */ |
| 428 | NULL((void*)0) /* allow to expand */ |
| 429 | }; |
| 430 | |
| 431 | /* Instance runid (sds) -> votes (long casted to void*) |
| 432 | * |
| 433 | * This is useful into sentinelGetObjectiveLeader() function in order to |
| 434 | * count the votes and understand who is the leader. */ |
| 435 | dictType leaderVotesDictType = { |
| 436 | dictSdsHash, /* hash function */ |
| 437 | NULL((void*)0), /* key dup */ |
| 438 | NULL((void*)0), /* val dup */ |
| 439 | dictSdsKeyCompare, /* key compare */ |
| 440 | NULL((void*)0), /* key destructor */ |
| 441 | NULL((void*)0), /* val destructor */ |
| 442 | NULL((void*)0) /* allow to expand */ |
| 443 | }; |
| 444 | |
| 445 | /* Instance renamed commands table. */ |
| 446 | dictType renamedCommandsDictType = { |
| 447 | dictSdsCaseHash, /* hash function */ |
| 448 | NULL((void*)0), /* key dup */ |
| 449 | NULL((void*)0), /* val dup */ |
| 450 | dictSdsKeyCaseCompare, /* key compare */ |
| 451 | dictSdsDestructor, /* key destructor */ |
| 452 | dictSdsDestructor, /* val destructor */ |
| 453 | NULL((void*)0) /* allow to expand */ |
| 454 | }; |
| 455 | |
| 456 | /* =========================== Initialization =============================== */ |
| 457 | |
| 458 | void sentinelCommand(client *c); |
| 459 | void sentinelInfoCommand(client *c); |
| 460 | void sentinelSetCommand(client *c); |
| 461 | void sentinelPublishCommand(client *c); |
| 462 | void sentinelRoleCommand(client *c); |
| 463 | void sentinelConfigGetCommand(client *c); |
| 464 | void sentinelConfigSetCommand(client *c); |
| 465 | |
| 466 | struct redisCommand sentinelcmds[] = { |
| 467 | {"ping",pingCommand,1,"fast @connection",0,NULL((void*)0),0,0,0,0,0}, |
| 468 | {"sentinel",sentinelCommand,-2,"admin",0,NULL((void*)0),0,0,0,0,0}, |
| 469 | {"subscribe",subscribeCommand,-2,"pub-sub",0,NULL((void*)0),0,0,0,0,0}, |
| 470 | {"unsubscribe",unsubscribeCommand,-1,"pub-sub",0,NULL((void*)0),0,0,0,0,0}, |
| 471 | {"psubscribe",psubscribeCommand,-2,"pub-sub",0,NULL((void*)0),0,0,0,0,0}, |
| 472 | {"punsubscribe",punsubscribeCommand,-1,"pub-sub",0,NULL((void*)0),0,0,0,0,0}, |
| 473 | {"publish",sentinelPublishCommand,3,"pub-sub fast",0,NULL((void*)0),0,0,0,0,0}, |
| 474 | {"info",sentinelInfoCommand,-1,"random @dangerous",0,NULL((void*)0),0,0,0,0,0}, |
| 475 | {"role",sentinelRoleCommand,1,"fast read-only @dangerous",0,NULL((void*)0),0,0,0,0,0}, |
| 476 | {"client",clientCommand,-2,"admin random @connection",0,NULL((void*)0),0,0,0,0,0}, |
| 477 | {"shutdown",shutdownCommand,-1,"admin",0,NULL((void*)0),0,0,0,0,0}, |
| 478 | {"auth",authCommand,-2,"no-auth fast @connection",0,NULL((void*)0),0,0,0,0,0}, |
| 479 | {"hello",helloCommand,-1,"no-auth fast @connection",0,NULL((void*)0),0,0,0,0,0}, |
| 480 | {"acl",aclCommand,-2,"admin",0,NULL((void*)0),0,0,0,0,0,0}, |
| 481 | {"command",commandCommand,-1, "random @connection", 0,NULL((void*)0),0,0,0,0,0,0} |
| 482 | }; |
| 483 | |
| 484 | /* this array is used for sentinel config lookup, which need to be loaded |
| 485 | * before monitoring masters config to avoid dependency issues */ |
| 486 | const char *preMonitorCfgName[] = { |
| 487 | "announce-ip", |
| 488 | "announce-port", |
| 489 | "deny-scripts-reconfig", |
| 490 | "sentinel-user", |
| 491 | "sentinel-pass", |
| 492 | "current-epoch", |
| 493 | "myid", |
| 494 | "resolve-hostnames", |
| 495 | "announce-hostnames" |
| 496 | }; |
| 497 | |
| 498 | /* This function overwrites a few normal Redis config default with Sentinel |
| 499 | * specific defaults. */ |
| 500 | void initSentinelConfig(void) { |
| 501 | server.port = REDIS_SENTINEL_PORT26379; |
| 502 | server.protected_mode = 0; /* Sentinel must be exposed. */ |
| 503 | } |
| 504 | |
| 505 | void freeSentinelLoadQueueEntry(void *item); |
| 506 | |
| 507 | /* Perform the Sentinel mode initialization. */ |
| 508 | void initSentinel(void) { |
| 509 | unsigned int j; |
| 510 | |
| 511 | /* Remove usual Redis commands from the command table, then just add |
| 512 | * the SENTINEL command. */ |
| 513 | dictEmpty(server.commands,NULL((void*)0)); |
| 514 | dictEmpty(server.orig_commands,NULL((void*)0)); |
| 515 | ACLClearCommandID(); |
| 516 | for (j = 0; j < sizeof(sentinelcmds)/sizeof(sentinelcmds[0]); j++) { |
| 517 | int retval; |
| 518 | struct redisCommand *cmd = sentinelcmds+j; |
| 519 | cmd->id = ACLGetCommandID(cmd->name); /* Assign the ID used for ACL. */ |
| 520 | retval = dictAdd(server.commands, sdsnew(cmd->name), cmd); |
| 521 | serverAssert(retval == DICT_OK)((retval == 0)?(void)0 : (_serverAssert("retval == DICT_OK","sentinel.c" ,521),__builtin_unreachable())); |
| 522 | retval = dictAdd(server.orig_commands, sdsnew(cmd->name), cmd); |
| 523 | serverAssert(retval == DICT_OK)((retval == 0)?(void)0 : (_serverAssert("retval == DICT_OK","sentinel.c" ,523),__builtin_unreachable())); |
| 524 | |
| 525 | /* Translate the command string flags description into an actual |
| 526 | * set of flags. */ |
| 527 | if (populateCommandTableParseFlags(cmd,cmd->sflags) == C_ERR-1) |
| 528 | serverPanic("Unsupported command flag")_serverPanic("sentinel.c",528,"Unsupported command flag"),__builtin_unreachable (); |
| 529 | } |
| 530 | |
| 531 | /* Initialize various data structures. */ |
| 532 | sentinel.current_epoch = 0; |
| 533 | sentinel.masters = dictCreate(&instancesDictType,NULL((void*)0)); |
| 534 | sentinel.tilt = 0; |
| 535 | sentinel.tilt_start_time = 0; |
| 536 | sentinel.previous_time = mstime(); |
| 537 | sentinel.running_scripts = 0; |
| 538 | sentinel.scripts_queue = listCreate(); |
| 539 | sentinel.announce_ip = NULL((void*)0); |
| 540 | sentinel.announce_port = 0; |
| 541 | sentinel.simfailure_flags = SENTINEL_SIMFAILURE_NONE0; |
| 542 | sentinel.deny_scripts_reconfig = SENTINEL_DEFAULT_DENY_SCRIPTS_RECONFIG1; |
| 543 | sentinel.sentinel_auth_pass = NULL((void*)0); |
| 544 | sentinel.sentinel_auth_user = NULL((void*)0); |
| 545 | sentinel.resolve_hostnames = SENTINEL_DEFAULT_RESOLVE_HOSTNAMES0; |
| 546 | sentinel.announce_hostnames = SENTINEL_DEFAULT_ANNOUNCE_HOSTNAMES0; |
| 547 | memset(sentinel.myid,0,sizeof(sentinel.myid)); |
| 548 | server.sentinel_config = NULL((void*)0); |
| 549 | } |
| 550 | |
| 551 | /* This function gets called when the server is in Sentinel mode, started, |
| 552 | * loaded the configuration, and is ready for normal operations. */ |
| 553 | void sentinelIsRunning(void) { |
| 554 | int j; |
| 555 | |
| 556 | if (server.configfile == NULL((void*)0)) { |
| 557 | serverLog(LL_WARNING3, |
| 558 | "Sentinel started without a config file. Exiting..."); |
| 559 | exit(1); |
| 560 | } else if (access(server.configfile,W_OK2) == -1) { |
| 561 | serverLog(LL_WARNING3, |
| 562 | "Sentinel config file %s is not writable: %s. Exiting...", |
| 563 | server.configfile,strerror(errno(*__errno_location ()))); |
| 564 | exit(1); |
| 565 | } |
| 566 | |
| 567 | /* If this Sentinel has yet no ID set in the configuration file, we |
| 568 | * pick a random one and persist the config on disk. From now on this |
| 569 | * will be this Sentinel ID across restarts. */ |
| 570 | for (j = 0; j < CONFIG_RUN_ID_SIZE40; j++) |
| 571 | if (sentinel.myid[j] != 0) break; |
| 572 | |
| 573 | if (j == CONFIG_RUN_ID_SIZE40) { |
| 574 | /* Pick ID and persist the config. */ |
| 575 | getRandomHexChars(sentinel.myid,CONFIG_RUN_ID_SIZE40); |
| 576 | sentinelFlushConfig(); |
| 577 | } |
| 578 | |
| 579 | /* Log its ID to make debugging of issues simpler. */ |
| 580 | serverLog(LL_WARNING3,"Sentinel ID is %s", sentinel.myid); |
| 581 | |
| 582 | /* We want to generate a +monitor event for every configured master |
| 583 | * at startup. */ |
| 584 | sentinelGenerateInitialMonitorEvents(); |
| 585 | } |
| 586 | |
| 587 | /* ============================== sentinelAddr ============================== */ |
| 588 | |
| 589 | /* Create a sentinelAddr object and return it on success. |
| 590 | * On error NULL is returned and errno is set to: |
| 591 | * ENOENT: Can't resolve the hostname. |
| 592 | * EINVAL: Invalid port number. |
| 593 | */ |
| 594 | sentinelAddr *createSentinelAddr(char *hostname, int port) { |
| 595 | char ip[NET_IP_STR_LEN46]; |
| 596 | sentinelAddr *sa; |
| 597 | |
| 598 | if (port < 0 || port > 65535) { |
| 599 | errno(*__errno_location ()) = EINVAL22; |
| 600 | return NULL((void*)0); |
| 601 | } |
| 602 | if (anetResolve(NULL((void*)0),hostname,ip,sizeof(ip), |
| 603 | sentinel.resolve_hostnames ? ANET_NONE0 : ANET_IP_ONLY(1<<0)) == ANET_ERR-1) { |
| 604 | errno(*__errno_location ()) = ENOENT2; |
| 605 | return NULL((void*)0); |
| 606 | } |
| 607 | sa = zmalloc(sizeof(*sa)); |
| 608 | sa->hostname = sdsnew(hostname); |
| 609 | sa->ip = sdsnew(ip); |
| 610 | sa->port = port; |
| 611 | return sa; |
| 612 | } |
| 613 | |
| 614 | /* Return a duplicate of the source address. */ |
| 615 | sentinelAddr *dupSentinelAddr(sentinelAddr *src) { |
| 616 | sentinelAddr *sa; |
| 617 | |
| 618 | sa = zmalloc(sizeof(*sa)); |
| 619 | sa->hostname = sdsnew(src->hostname); |
| 620 | sa->ip = sdsnew(src->ip); |
| 621 | sa->port = src->port; |
| 622 | return sa; |
| 623 | } |
| 624 | |
| 625 | /* Free a Sentinel address. Can't fail. */ |
| 626 | void releaseSentinelAddr(sentinelAddr *sa) { |
| 627 | sdsfree(sa->hostname); |
| 628 | sdsfree(sa->ip); |
| 629 | zfree(sa); |
| 630 | } |
| 631 | |
| 632 | /* Return non-zero if two addresses are equal. */ |
| 633 | int sentinelAddrIsEqual(sentinelAddr *a, sentinelAddr *b) { |
| 634 | return a->port == b->port && !strcasecmp(a->ip,b->ip); |
| 635 | } |
| 636 | |
| 637 | /* Return non-zero if a hostname matches an address. */ |
| 638 | int sentinelAddrEqualsHostname(sentinelAddr *a, char *hostname) { |
| 639 | char ip[NET_IP_STR_LEN46]; |
| 640 | |
| 641 | /* We always resolve the hostname and compare it to the address */ |
| 642 | if (anetResolve(NULL((void*)0), hostname, ip, sizeof(ip), |
| 643 | sentinel.resolve_hostnames ? ANET_NONE0 : ANET_IP_ONLY(1<<0)) == ANET_ERR-1) |
| 644 | return 0; |
| 645 | return !strcasecmp(a->ip, ip); |
| 646 | } |
| 647 | |
| 648 | const char *announceSentinelAddr(const sentinelAddr *a) { |
| 649 | return sentinel.announce_hostnames ? a->hostname : a->ip; |
| 650 | } |
| 651 | |
| 652 | /* Return an allocated sds with hostname/address:port. IPv6 |
| 653 | * addresses are bracketed the same way anetFormatAddr() does. |
| 654 | */ |
| 655 | sds announceSentinelAddrAndPort(const sentinelAddr *a) { |
| 656 | const char *addr = announceSentinelAddr(a); |
| 657 | if (strchr(addr, ':') != NULL((void*)0)) |
| 658 | return sdscatprintf(sdsempty(), "[%s]:%d", addr, a->port); |
| 659 | else |
| 660 | return sdscatprintf(sdsempty(), "%s:%d", addr, a->port); |
| 661 | } |
| 662 | |
| 663 | /* =========================== Events notification ========================== */ |
| 664 | |
| 665 | /* Send an event to log, pub/sub, user notification script. |
| 666 | * |
| 667 | * 'level' is the log level for logging. Only LL_WARNING events will trigger |
| 668 | * the execution of the user notification script. |
| 669 | * |
| 670 | * 'type' is the message type, also used as a pub/sub channel name. |
| 671 | * |
| 672 | * 'ri', is the redis instance target of this event if applicable, and is |
| 673 | * used to obtain the path of the notification script to execute. |
| 674 | * |
| 675 | * The remaining arguments are printf-alike. |
| 676 | * If the format specifier starts with the two characters "%@" then ri is |
| 677 | * not NULL, and the message is prefixed with an instance identifier in the |
| 678 | * following format: |
| 679 | * |
| 680 | * <instance type> <instance name> <ip> <port> |
| 681 | * |
| 682 | * If the instance type is not master, than the additional string is |
| 683 | * added to specify the originating master: |
| 684 | * |
| 685 | * @ <master name> <master ip> <master port> |
| 686 | * |
| 687 | * Any other specifier after "%@" is processed by printf itself. |
| 688 | */ |
| 689 | void sentinelEvent(int level, char *type, sentinelRedisInstance *ri, |
| 690 | const char *fmt, ...) { |
| 691 | va_list ap; |
| 692 | char msg[LOG_MAX_LEN1024]; |
| 693 | robj *channel, *payload; |
| 694 | |
| 695 | /* Handle %@ */ |
| 696 | if (fmt[0] == '%' && fmt[1] == '@') { |
| 697 | sentinelRedisInstance *master = (ri->flags & SRI_MASTER(1<<0)) ? |
| 698 | NULL((void*)0) : ri->master; |
| 699 | |
| 700 | if (master) { |
| 701 | snprintf(msg, sizeof(msg), "%s %s %s %d @ %s %s %d", |
| 702 | sentinelRedisInstanceTypeStr(ri), |
| 703 | ri->name, announceSentinelAddr(ri->addr), ri->addr->port, |
| 704 | master->name, announceSentinelAddr(master->addr), master->addr->port); |
| 705 | } else { |
| 706 | snprintf(msg, sizeof(msg), "%s %s %s %d", |
| 707 | sentinelRedisInstanceTypeStr(ri), |
| 708 | ri->name, announceSentinelAddr(ri->addr), ri->addr->port); |
| 709 | } |
| 710 | fmt += 2; |
| 711 | } else { |
| 712 | msg[0] = '\0'; |
| 713 | } |
| 714 | |
| 715 | /* Use vsprintf for the rest of the formatting if any. */ |
| 716 | if (fmt[0] != '\0') { |
| 717 | va_start(ap, fmt)__builtin_va_start(ap, fmt); |
| 718 | vsnprintf(msg+strlen(msg), sizeof(msg)-strlen(msg), fmt, ap); |
| 719 | va_end(ap)__builtin_va_end(ap); |
| 720 | } |
| 721 | |
| 722 | /* Log the message if the log level allows it to be logged. */ |
| 723 | if (level >= server.verbosity) |
| 724 | serverLog(level,"%s %s",type,msg); |
| 725 | |
| 726 | /* Publish the message via Pub/Sub if it's not a debugging one. */ |
| 727 | if (level != LL_DEBUG0) { |
| 728 | channel = createStringObject(type,strlen(type)); |
| 729 | payload = createStringObject(msg,strlen(msg)); |
| 730 | pubsubPublishMessage(channel,payload); |
| 731 | decrRefCount(channel); |
| 732 | decrRefCount(payload); |
| 733 | } |
| 734 | |
| 735 | /* Call the notification script if applicable. */ |
| 736 | if (level == LL_WARNING3 && ri != NULL((void*)0)) { |
| 737 | sentinelRedisInstance *master = (ri->flags & SRI_MASTER(1<<0)) ? |
| 738 | ri : ri->master; |
| 739 | if (master && master->notification_script) { |
| 740 | sentinelScheduleScriptExecution(master->notification_script, |
| 741 | type,msg,NULL((void*)0)); |
| 742 | } |
| 743 | } |
| 744 | } |
| 745 | |
| 746 | /* This function is called only at startup and is used to generate a |
| 747 | * +monitor event for every configured master. The same events are also |
| 748 | * generated when a master to monitor is added at runtime via the |
| 749 | * SENTINEL MONITOR command. */ |
| 750 | void sentinelGenerateInitialMonitorEvents(void) { |
| 751 | dictIterator *di; |
| 752 | dictEntry *de; |
| 753 | |
| 754 | di = dictGetIterator(sentinel.masters); |
| 755 | while((de = dictNext(di)) != NULL((void*)0)) { |
| 756 | sentinelRedisInstance *ri = dictGetVal(de)((de)->v.val); |
| 757 | sentinelEvent(LL_WARNING3,"+monitor",ri,"%@ quorum %d",ri->quorum); |
| 758 | } |
| 759 | dictReleaseIterator(di); |
| 760 | } |
| 761 | |
| 762 | /* ============================ script execution ============================ */ |
| 763 | |
| 764 | /* Release a script job structure and all the associated data. */ |
| 765 | void sentinelReleaseScriptJob(sentinelScriptJob *sj) { |
| 766 | int j = 0; |
| 767 | |
| 768 | while(sj->argv[j]) sdsfree(sj->argv[j++]); |
| 769 | zfree(sj->argv); |
| 770 | zfree(sj); |
| 771 | } |
| 772 | |
| 773 | #define SENTINEL_SCRIPT_MAX_ARGS16 16 |
| 774 | void sentinelScheduleScriptExecution(char *path, ...) { |
| 775 | va_list ap; |
| 776 | char *argv[SENTINEL_SCRIPT_MAX_ARGS16+1]; |
| 777 | int argc = 1; |
| 778 | sentinelScriptJob *sj; |
| 779 | |
| 780 | va_start(ap, path)__builtin_va_start(ap, path); |
| 781 | while(argc < SENTINEL_SCRIPT_MAX_ARGS16) { |
| 782 | argv[argc] = va_arg(ap,char*)__builtin_va_arg(ap, char*); |
| 783 | if (!argv[argc]) break; |
| 784 | argv[argc] = sdsnew(argv[argc]); /* Copy the string. */ |
| 785 | argc++; |
| 786 | } |
| 787 | va_end(ap)__builtin_va_end(ap); |
| 788 | argv[0] = sdsnew(path); |
| 789 | |
| 790 | sj = zmalloc(sizeof(*sj)); |
| 791 | sj->flags = SENTINEL_SCRIPT_NONE0; |
| 792 | sj->retry_num = 0; |
| 793 | sj->argv = zmalloc(sizeof(char*)*(argc+1)); |
| 794 | sj->start_time = 0; |
| 795 | sj->pid = 0; |
| 796 | memcpy(sj->argv,argv,sizeof(char*)*(argc+1)); |
| 797 | |
| 798 | listAddNodeTail(sentinel.scripts_queue,sj); |
| 799 | |
| 800 | /* Remove the oldest non running script if we already hit the limit. */ |
| 801 | if (listLength(sentinel.scripts_queue)((sentinel.scripts_queue)->len) > SENTINEL_SCRIPT_MAX_QUEUE256) { |
| 802 | listNode *ln; |
| 803 | listIter li; |
| 804 | |
| 805 | listRewind(sentinel.scripts_queue,&li); |
| 806 | while ((ln = listNext(&li)) != NULL((void*)0)) { |
| 807 | sj = ln->value; |
| 808 | |
| 809 | if (sj->flags & SENTINEL_SCRIPT_RUNNING1) continue; |
| 810 | /* The first node is the oldest as we add on tail. */ |
| 811 | listDelNode(sentinel.scripts_queue,ln); |
| 812 | sentinelReleaseScriptJob(sj); |
| 813 | break; |
| 814 | } |
| 815 | serverAssert(listLength(sentinel.scripts_queue) <=((((sentinel.scripts_queue)->len) <= 256)?(void)0 : (_serverAssert ("listLength(sentinel.scripts_queue) <= SENTINEL_SCRIPT_MAX_QUEUE" ,"sentinel.c",816),__builtin_unreachable())) |
| 816 | SENTINEL_SCRIPT_MAX_QUEUE)((((sentinel.scripts_queue)->len) <= 256)?(void)0 : (_serverAssert ("listLength(sentinel.scripts_queue) <= SENTINEL_SCRIPT_MAX_QUEUE" ,"sentinel.c",816),__builtin_unreachable())); |
| 817 | } |
| 818 | } |
| 819 | |
| 820 | /* Lookup a script in the scripts queue via pid, and returns the list node |
| 821 | * (so that we can easily remove it from the queue if needed). */ |
| 822 | listNode *sentinelGetScriptListNodeByPid(pid_t pid) { |
| 823 | listNode *ln; |
| 824 | listIter li; |
| 825 | |
| 826 | listRewind(sentinel.scripts_queue,&li); |
| 827 | while ((ln = listNext(&li)) != NULL((void*)0)) { |
| 828 | sentinelScriptJob *sj = ln->value; |
| 829 | |
| 830 | if ((sj->flags & SENTINEL_SCRIPT_RUNNING1) && sj->pid == pid) |
| 831 | return ln; |
| 832 | } |
| 833 | return NULL((void*)0); |
| 834 | } |
| 835 | |
| 836 | /* Run pending scripts if we are not already at max number of running |
| 837 | * scripts. */ |
| 838 | void sentinelRunPendingScripts(void) { |
| 839 | listNode *ln; |
| 840 | listIter li; |
| 841 | mstime_t now = mstime(); |
| 842 | |
| 843 | /* Find jobs that are not running and run them, from the top to the |
| 844 | * tail of the queue, so we run older jobs first. */ |
| 845 | listRewind(sentinel.scripts_queue,&li); |
| 846 | while (sentinel.running_scripts < SENTINEL_SCRIPT_MAX_RUNNING16 && |
| 847 | (ln = listNext(&li)) != NULL((void*)0)) |
| 848 | { |
| 849 | sentinelScriptJob *sj = ln->value; |
| 850 | pid_t pid; |
| 851 | |
| 852 | /* Skip if already running. */ |
| 853 | if (sj->flags & SENTINEL_SCRIPT_RUNNING1) continue; |
| 854 | |
| 855 | /* Skip if it's a retry, but not enough time has elapsed. */ |
| 856 | if (sj->start_time && sj->start_time > now) continue; |
| 857 | |
| 858 | sj->flags |= SENTINEL_SCRIPT_RUNNING1; |
| 859 | sj->start_time = mstime(); |
| 860 | sj->retry_num++; |
| 861 | pid = fork(); |
| 862 | |
| 863 | if (pid == -1) { |
| 864 | /* Parent (fork error). |
| 865 | * We report fork errors as signal 99, in order to unify the |
| 866 | * reporting with other kind of errors. */ |
| 867 | sentinelEvent(LL_WARNING3,"-script-error",NULL((void*)0), |
| 868 | "%s %d %d", sj->argv[0], 99, 0); |
| 869 | sj->flags &= ~SENTINEL_SCRIPT_RUNNING1; |
| 870 | sj->pid = 0; |
| 871 | } else if (pid == 0) { |
| 872 | /* Child */ |
| 873 | execve(sj->argv[0],sj->argv,environ); |
| 874 | /* If we are here an error occurred. */ |
| 875 | _exit(2); /* Don't retry execution. */ |
| 876 | } else { |
| 877 | sentinel.running_scripts++; |
| 878 | sj->pid = pid; |
| 879 | sentinelEvent(LL_DEBUG0,"+script-child",NULL((void*)0),"%ld",(long)pid); |
| 880 | } |
| 881 | } |
| 882 | } |
| 883 | |
| 884 | /* How much to delay the execution of a script that we need to retry after |
| 885 | * an error? |
| 886 | * |
| 887 | * We double the retry delay for every further retry we do. So for instance |
| 888 | * if RETRY_DELAY is set to 30 seconds and the max number of retries is 10 |
| 889 | * starting from the second attempt to execute the script the delays are: |
| 890 | * 30 sec, 60 sec, 2 min, 4 min, 8 min, 16 min, 32 min, 64 min, 128 min. */ |
| 891 | mstime_t sentinelScriptRetryDelay(int retry_num) { |
| 892 | mstime_t delay = SENTINEL_SCRIPT_RETRY_DELAY30000; |
| 893 | |
| 894 | while (retry_num-- > 1) delay *= 2; |
| 895 | return delay; |
| 896 | } |
| 897 | |
| 898 | /* Check for scripts that terminated, and remove them from the queue if the |
| 899 | * script terminated successfully. If instead the script was terminated by |
| 900 | * a signal, or returned exit code "1", it is scheduled to run again if |
| 901 | * the max number of retries did not already elapsed. */ |
| 902 | void sentinelCollectTerminatedScripts(void) { |
| 903 | int statloc; |
| 904 | pid_t pid; |
| 905 | |
| 906 | while ((pid = wait3(&statloc,WNOHANG1,NULL((void*)0))) > 0) { |
| 907 | int exitcode = WEXITSTATUS(statloc)(((statloc) & 0xff00) >> 8); |
| 908 | int bysignal = 0; |
| 909 | listNode *ln; |
| 910 | sentinelScriptJob *sj; |
| 911 | |
| 912 | if (WIFSIGNALED(statloc)(((signed char) (((statloc) & 0x7f) + 1) >> 1) > 0)) bysignal = WTERMSIG(statloc)((statloc) & 0x7f); |
| 913 | sentinelEvent(LL_DEBUG0,"-script-child",NULL((void*)0),"%ld %d %d", |
| 914 | (long)pid, exitcode, bysignal); |
| 915 | |
| 916 | ln = sentinelGetScriptListNodeByPid(pid); |
| 917 | if (ln == NULL((void*)0)) { |
| 918 | serverLog(LL_WARNING3,"wait3() returned a pid (%ld) we can't find in our scripts execution queue!", (long)pid); |
| 919 | continue; |
| 920 | } |
| 921 | sj = ln->value; |
| 922 | |
| 923 | /* If the script was terminated by a signal or returns an |
| 924 | * exit code of "1" (that means: please retry), we reschedule it |
| 925 | * if the max number of retries is not already reached. */ |
| 926 | if ((bysignal || exitcode == 1) && |
| 927 | sj->retry_num != SENTINEL_SCRIPT_MAX_RETRY10) |
| 928 | { |
| 929 | sj->flags &= ~SENTINEL_SCRIPT_RUNNING1; |
| 930 | sj->pid = 0; |
| 931 | sj->start_time = mstime() + |
| 932 | sentinelScriptRetryDelay(sj->retry_num); |
| 933 | } else { |
| 934 | /* Otherwise let's remove the script, but log the event if the |
| 935 | * execution did not terminated in the best of the ways. */ |
| 936 | if (bysignal || exitcode != 0) { |
| 937 | sentinelEvent(LL_WARNING3,"-script-error",NULL((void*)0), |
| 938 | "%s %d %d", sj->argv[0], bysignal, exitcode); |
| 939 | } |
| 940 | listDelNode(sentinel.scripts_queue,ln); |
| 941 | sentinelReleaseScriptJob(sj); |
| 942 | } |
| 943 | sentinel.running_scripts--; |
| 944 | } |
| 945 | } |
| 946 | |
| 947 | /* Kill scripts in timeout, they'll be collected by the |
| 948 | * sentinelCollectTerminatedScripts() function. */ |
| 949 | void sentinelKillTimedoutScripts(void) { |
| 950 | listNode *ln; |
| 951 | listIter li; |
| 952 | mstime_t now = mstime(); |
| 953 | |
| 954 | listRewind(sentinel.scripts_queue,&li); |
| 955 | while ((ln = listNext(&li)) != NULL((void*)0)) { |
| 956 | sentinelScriptJob *sj = ln->value; |
| 957 | |
| 958 | if (sj->flags & SENTINEL_SCRIPT_RUNNING1 && |
| 959 | (now - sj->start_time) > SENTINEL_SCRIPT_MAX_RUNTIME60000) |
| 960 | { |
| 961 | sentinelEvent(LL_WARNING3,"-script-timeout",NULL((void*)0),"%s %ld", |
| 962 | sj->argv[0], (long)sj->pid); |
| 963 | kill(sj->pid,SIGKILL9); |
| 964 | } |
| 965 | } |
| 966 | } |
| 967 | |
| 968 | /* Implements SENTINEL PENDING-SCRIPTS command. */ |
| 969 | void sentinelPendingScriptsCommand(client *c) { |
| 970 | listNode *ln; |
| 971 | listIter li; |
| 972 | |
| 973 | addReplyArrayLen(c,listLength(sentinel.scripts_queue)((sentinel.scripts_queue)->len)); |
| 974 | listRewind(sentinel.scripts_queue,&li); |
| 975 | while ((ln = listNext(&li)) != NULL((void*)0)) { |
| 976 | sentinelScriptJob *sj = ln->value; |
| 977 | int j = 0; |
| 978 | |
| 979 | addReplyMapLen(c,5); |
| 980 | |
| 981 | addReplyBulkCString(c,"argv"); |
| 982 | while (sj->argv[j]) j++; |
| 983 | addReplyArrayLen(c,j); |
| 984 | j = 0; |
| 985 | while (sj->argv[j]) addReplyBulkCString(c,sj->argv[j++]); |
| 986 | |
| 987 | addReplyBulkCString(c,"flags"); |
| 988 | addReplyBulkCString(c, |
| 989 | (sj->flags & SENTINEL_SCRIPT_RUNNING1) ? "running" : "scheduled"); |
| 990 | |
| 991 | addReplyBulkCString(c,"pid"); |
| 992 | addReplyBulkLongLong(c,sj->pid); |
| 993 | |
| 994 | if (sj->flags & SENTINEL_SCRIPT_RUNNING1) { |
| 995 | addReplyBulkCString(c,"run-time"); |
| 996 | addReplyBulkLongLong(c,mstime() - sj->start_time); |
| 997 | } else { |
| 998 | mstime_t delay = sj->start_time ? (sj->start_time-mstime()) : 0; |
| 999 | if (delay < 0) delay = 0; |
| 1000 | addReplyBulkCString(c,"run-delay"); |
| 1001 | addReplyBulkLongLong(c,delay); |
| 1002 | } |
| 1003 | |
| 1004 | addReplyBulkCString(c,"retry-num"); |
| 1005 | addReplyBulkLongLong(c,sj->retry_num); |
| 1006 | } |
| 1007 | } |
| 1008 | |
| 1009 | /* This function calls, if any, the client reconfiguration script with the |
| 1010 | * following parameters: |
| 1011 | * |
| 1012 | * <master-name> <role> <state> <from-ip> <from-port> <to-ip> <to-port> |
| 1013 | * |
| 1014 | * It is called every time a failover is performed. |
| 1015 | * |
| 1016 | * <state> is currently always "failover". |
| 1017 | * <role> is either "leader" or "observer". |
| 1018 | * |
| 1019 | * from/to fields are respectively master -> promoted slave addresses for |
| 1020 | * "start" and "end". */ |
| 1021 | void sentinelCallClientReconfScript(sentinelRedisInstance *master, int role, char *state, sentinelAddr *from, sentinelAddr *to) { |
| 1022 | char fromport[32], toport[32]; |
| 1023 | |
| 1024 | if (master->client_reconfig_script == NULL((void*)0)) return; |
| 1025 | ll2string(fromport,sizeof(fromport),from->port); |
| 1026 | ll2string(toport,sizeof(toport),to->port); |
| 1027 | sentinelScheduleScriptExecution(master->client_reconfig_script, |
| 1028 | master->name, |
| 1029 | (role == SENTINEL_LEADER(1<<17)) ? "leader" : "observer", |
| 1030 | state, announceSentinelAddr(from), fromport, |
| 1031 | announceSentinelAddr(to), toport, NULL((void*)0)); |
| 1032 | } |
| 1033 | |
| 1034 | /* =============================== instanceLink ============================= */ |
| 1035 | |
| 1036 | /* Create a not yet connected link object. */ |
| 1037 | instanceLink *createInstanceLink(void) { |
| 1038 | instanceLink *link = zmalloc(sizeof(*link)); |
| 1039 | |
| 1040 | link->refcount = 1; |
| 1041 | link->disconnected = 1; |
| 1042 | link->pending_commands = 0; |
| 1043 | link->cc = NULL((void*)0); |
| 1044 | link->pc = NULL((void*)0); |
| 1045 | link->cc_conn_time = 0; |
| 1046 | link->pc_conn_time = 0; |
| 1047 | link->last_reconn_time = 0; |
| 1048 | link->pc_last_activity = 0; |
| 1049 | /* We set the act_ping_time to "now" even if we actually don't have yet |
| 1050 | * a connection with the node, nor we sent a ping. |
| 1051 | * This is useful to detect a timeout in case we'll not be able to connect |
| 1052 | * with the node at all. */ |
| 1053 | link->act_ping_time = mstime(); |
| 1054 | link->last_ping_time = 0; |
| 1055 | link->last_avail_time = mstime(); |
| 1056 | link->last_pong_time = mstime(); |
| 1057 | return link; |
| 1058 | } |
| 1059 | |
| 1060 | /* Disconnect a hiredis connection in the context of an instance link. */ |
| 1061 | void instanceLinkCloseConnection(instanceLink *link, redisAsyncContext *c) { |
| 1062 | if (c == NULL((void*)0)) return; |
| 1063 | |
| 1064 | if (link->cc == c) { |
| 1065 | link->cc = NULL((void*)0); |
| 1066 | link->pending_commands = 0; |
| 1067 | } |
| 1068 | if (link->pc == c) link->pc = NULL((void*)0); |
| 1069 | c->data = NULL((void*)0); |
| 1070 | link->disconnected = 1; |
| 1071 | redisAsyncFree(c); |
| 1072 | } |
| 1073 | |
| 1074 | /* Decrement the refcount of a link object, if it drops to zero, actually |
| 1075 | * free it and return NULL. Otherwise don't do anything and return the pointer |
| 1076 | * to the object. |
| 1077 | * |
| 1078 | * If we are not going to free the link and ri is not NULL, we rebind all the |
| 1079 | * pending requests in link->cc (hiredis connection for commands) to a |
| 1080 | * callback that will just ignore them. This is useful to avoid processing |
| 1081 | * replies for an instance that no longer exists. */ |
| 1082 | instanceLink *releaseInstanceLink(instanceLink *link, sentinelRedisInstance *ri) |
| 1083 | { |
| 1084 | serverAssert(link->refcount > 0)((link->refcount > 0)?(void)0 : (_serverAssert("link->refcount > 0" ,"sentinel.c",1084),__builtin_unreachable())); |
| 1085 | link->refcount--; |
| 1086 | if (link->refcount != 0) { |
| 1087 | if (ri && ri->link->cc) { |
| 1088 | /* This instance may have pending callbacks in the hiredis async |
| 1089 | * context, having as 'privdata' the instance that we are going to |
| 1090 | * free. Let's rewrite the callback list, directly exploiting |
| 1091 | * hiredis internal data structures, in order to bind them with |
| 1092 | * a callback that will ignore the reply at all. */ |
| 1093 | redisCallback *cb; |
| 1094 | redisCallbackList *callbacks = &link->cc->replies; |
| 1095 | |
| 1096 | cb = callbacks->head; |
| 1097 | while(cb) { |
| 1098 | if (cb->privdata == ri) { |
| 1099 | cb->fn = sentinelDiscardReplyCallback; |
| 1100 | cb->privdata = NULL((void*)0); /* Not strictly needed. */ |
| 1101 | } |
| 1102 | cb = cb->next; |
| 1103 | } |
| 1104 | } |
| 1105 | return link; /* Other active users. */ |
| 1106 | } |
| 1107 | |
| 1108 | instanceLinkCloseConnection(link,link->cc); |
| 1109 | instanceLinkCloseConnection(link,link->pc); |
| 1110 | zfree(link); |
| 1111 | return NULL((void*)0); |
| 1112 | } |
| 1113 | |
| 1114 | /* This function will attempt to share the instance link we already have |
| 1115 | * for the same Sentinel in the context of a different master, with the |
| 1116 | * instance we are passing as argument. |
| 1117 | * |
| 1118 | * This way multiple Sentinel objects that refer all to the same physical |
| 1119 | * Sentinel instance but in the context of different masters will use |
| 1120 | * a single connection, will send a single PING per second for failure |
| 1121 | * detection and so forth. |
| 1122 | * |
| 1123 | * Return C_OK if a matching Sentinel was found in the context of a |
| 1124 | * different master and sharing was performed. Otherwise C_ERR |
| 1125 | * is returned. */ |
| 1126 | int sentinelTryConnectionSharing(sentinelRedisInstance *ri) { |
| 1127 | serverAssert(ri->flags & SRI_SENTINEL)((ri->flags & (1<<2))?(void)0 : (_serverAssert("ri->flags & SRI_SENTINEL" ,"sentinel.c",1127),__builtin_unreachable())); |
| 1128 | dictIterator *di; |
| 1129 | dictEntry *de; |
| 1130 | |
| 1131 | if (ri->runid == NULL((void*)0)) return C_ERR-1; /* No way to identify it. */ |
| 1132 | if (ri->link->refcount > 1) return C_ERR-1; /* Already shared. */ |
| 1133 | |
| 1134 | di = dictGetIterator(sentinel.masters); |
| 1135 | while((de = dictNext(di)) != NULL((void*)0)) { |
| 1136 | sentinelRedisInstance *master = dictGetVal(de)((de)->v.val), *match; |
| 1137 | /* We want to share with the same physical Sentinel referenced |
| 1138 | * in other masters, so skip our master. */ |
| 1139 | if (master == ri->master) continue; |
| 1140 | match = getSentinelRedisInstanceByAddrAndRunID(master->sentinels, |
| 1141 | NULL((void*)0),0,ri->runid); |
| 1142 | if (match == NULL((void*)0)) continue; /* No match. */ |
| 1143 | if (match == ri) continue; /* Should never happen but... safer. */ |
| 1144 | |
| 1145 | /* We identified a matching Sentinel, great! Let's free our link |
| 1146 | * and use the one of the matching Sentinel. */ |
| 1147 | releaseInstanceLink(ri->link,NULL((void*)0)); |
| 1148 | ri->link = match->link; |
| 1149 | match->link->refcount++; |
| 1150 | dictReleaseIterator(di); |
| 1151 | return C_OK0; |
| 1152 | } |
| 1153 | dictReleaseIterator(di); |
| 1154 | return C_ERR-1; |
| 1155 | } |
| 1156 | |
| 1157 | /* Drop all connections to other sentinels. Returns the number of connections |
| 1158 | * dropped.*/ |
| 1159 | int sentinelDropConnections(void) { |
| 1160 | dictIterator *di; |
| 1161 | dictEntry *de; |
| 1162 | int dropped = 0; |
| 1163 | |
| 1164 | di = dictGetIterator(sentinel.masters); |
| 1165 | while ((de = dictNext(di)) != NULL((void*)0)) { |
| 1166 | dictIterator *sdi; |
| 1167 | dictEntry *sde; |
| 1168 | |
| 1169 | sentinelRedisInstance *ri = dictGetVal(de)((de)->v.val); |
| 1170 | sdi = dictGetIterator(ri->sentinels); |
| 1171 | while ((sde = dictNext(sdi)) != NULL((void*)0)) { |
| 1172 | sentinelRedisInstance *si = dictGetVal(sde)((sde)->v.val); |
| 1173 | if (!si->link->disconnected) { |
| 1174 | instanceLinkCloseConnection(si->link, si->link->pc); |
| 1175 | instanceLinkCloseConnection(si->link, si->link->cc); |
| 1176 | dropped++; |
| 1177 | } |
| 1178 | } |
| 1179 | dictReleaseIterator(sdi); |
| 1180 | } |
| 1181 | dictReleaseIterator(di); |
| 1182 | |
| 1183 | return dropped; |
| 1184 | } |
| 1185 | |
| 1186 | /* When we detect a Sentinel to switch address (reporting a different IP/port |
| 1187 | * pair in Hello messages), let's update all the matching Sentinels in the |
| 1188 | * context of other masters as well and disconnect the links, so that everybody |
| 1189 | * will be updated. |
| 1190 | * |
| 1191 | * Return the number of updated Sentinel addresses. */ |
| 1192 | int sentinelUpdateSentinelAddressInAllMasters(sentinelRedisInstance *ri) { |
| 1193 | serverAssert(ri->flags & SRI_SENTINEL)((ri->flags & (1<<2))?(void)0 : (_serverAssert("ri->flags & SRI_SENTINEL" ,"sentinel.c",1193),__builtin_unreachable())); |
| 1194 | dictIterator *di; |
| 1195 | dictEntry *de; |
| 1196 | int reconfigured = 0; |
| 1197 | |
| 1198 | di = dictGetIterator(sentinel.masters); |
| 1199 | while((de = dictNext(di)) != NULL((void*)0)) { |
| 1200 | sentinelRedisInstance *master = dictGetVal(de)((de)->v.val), *match; |
| 1201 | match = getSentinelRedisInstanceByAddrAndRunID(master->sentinels, |
| 1202 | NULL((void*)0),0,ri->runid); |
| 1203 | /* If there is no match, this master does not know about this |
| 1204 | * Sentinel, try with the next one. */ |
| 1205 | if (match == NULL((void*)0)) continue; |
| 1206 | |
| 1207 | /* Disconnect the old links if connected. */ |
| 1208 | if (match->link->cc != NULL((void*)0)) |
| 1209 | instanceLinkCloseConnection(match->link,match->link->cc); |
| 1210 | if (match->link->pc != NULL((void*)0)) |
| 1211 | instanceLinkCloseConnection(match->link,match->link->pc); |
| 1212 | |
| 1213 | if (match == ri) continue; /* Address already updated for it. */ |
| 1214 | |
| 1215 | /* Update the address of the matching Sentinel by copying the address |
| 1216 | * of the Sentinel object that received the address update. */ |
| 1217 | releaseSentinelAddr(match->addr); |
| 1218 | match->addr = dupSentinelAddr(ri->addr); |
| 1219 | reconfigured++; |
| 1220 | } |
| 1221 | dictReleaseIterator(di); |
| 1222 | if (reconfigured) |
| 1223 | sentinelEvent(LL_NOTICE2,"+sentinel-address-update", ri, |
| 1224 | "%@ %d additional matching instances", reconfigured); |
| 1225 | return reconfigured; |
| 1226 | } |
| 1227 | |
| 1228 | /* This function is called when a hiredis connection reported an error. |
| 1229 | * We set it to NULL and mark the link as disconnected so that it will be |
| 1230 | * reconnected again. |
| 1231 | * |
| 1232 | * Note: we don't free the hiredis context as hiredis will do it for us |
| 1233 | * for async connections. */ |
| 1234 | void instanceLinkConnectionError(const redisAsyncContext *c) { |
| 1235 | instanceLink *link = c->data; |
| 1236 | int pubsub; |
| 1237 | |
| 1238 | if (!link) return; |
| 1239 | |
| 1240 | pubsub = (link->pc == c); |
| 1241 | if (pubsub) |
| 1242 | link->pc = NULL((void*)0); |
| 1243 | else |
| 1244 | link->cc = NULL((void*)0); |
| 1245 | link->disconnected = 1; |
| 1246 | } |
| 1247 | |
| 1248 | /* Hiredis connection established / disconnected callbacks. We need them |
| 1249 | * just to cleanup our link state. */ |
| 1250 | void sentinelLinkEstablishedCallback(const redisAsyncContext *c, int status) { |
| 1251 | if (status != C_OK0) instanceLinkConnectionError(c); |
| 1252 | } |
| 1253 | |
| 1254 | void sentinelDisconnectCallback(const redisAsyncContext *c, int status) { |
| 1255 | UNUSED(status)((void) status); |
| 1256 | instanceLinkConnectionError(c); |
| 1257 | } |
| 1258 | |
| 1259 | /* ========================== sentinelRedisInstance ========================= */ |
| 1260 | |
| 1261 | /* Create a redis instance, the following fields must be populated by the |
| 1262 | * caller if needed: |
| 1263 | * runid: set to NULL but will be populated once INFO output is received. |
| 1264 | * info_refresh: is set to 0 to mean that we never received INFO so far. |
| 1265 | * |
| 1266 | * If SRI_MASTER is set into initial flags the instance is added to |
| 1267 | * sentinel.masters table. |
| 1268 | * |
| 1269 | * if SRI_SLAVE or SRI_SENTINEL is set then 'master' must be not NULL and the |
| 1270 | * instance is added into master->slaves or master->sentinels table. |
| 1271 | * |
| 1272 | * If the instance is a slave or sentinel, the name parameter is ignored and |
| 1273 | * is created automatically as hostname:port. |
| 1274 | * |
| 1275 | * The function fails if hostname can't be resolved or port is out of range. |
| 1276 | * When this happens NULL is returned and errno is set accordingly to the |
| 1277 | * createSentinelAddr() function. |
| 1278 | * |
| 1279 | * The function may also fail and return NULL with errno set to EBUSY if |
| 1280 | * a master with the same name, a slave with the same address, or a sentinel |
| 1281 | * with the same ID already exists. */ |
| 1282 | |
| 1283 | sentinelRedisInstance *createSentinelRedisInstance(char *name, int flags, char *hostname, int port, int quorum, sentinelRedisInstance *master) { |
| 1284 | sentinelRedisInstance *ri; |
| 1285 | sentinelAddr *addr; |
| 1286 | dict *table = NULL((void*)0); |
| 1287 | sds sdsname; |
| 1288 | |
| 1289 | serverAssert(flags & (SRI_MASTER|SRI_SLAVE|SRI_SENTINEL))((flags & ((1<<0)|(1<<1)|(1<<2)))?(void )0 : (_serverAssert("flags & (SRI_MASTER|SRI_SLAVE|SRI_SENTINEL)" ,"sentinel.c",1289),__builtin_unreachable())); |
| 1290 | serverAssert((flags & SRI_MASTER) || master != NULL)(((flags & (1<<0)) || master != ((void*)0))?(void)0 : (_serverAssert("(flags & SRI_MASTER) || master != NULL" ,"sentinel.c",1290),__builtin_unreachable())); |
| 1291 | |
| 1292 | /* Check address validity. */ |
| 1293 | addr = createSentinelAddr(hostname,port); |
| 1294 | if (addr == NULL((void*)0)) return NULL((void*)0); |
| 1295 | |
| 1296 | /* For slaves use ip/host:port as name. */ |
| 1297 | if (flags & SRI_SLAVE(1<<1)) |
| 1298 | sdsname = announceSentinelAddrAndPort(addr); |
| 1299 | else |
| 1300 | sdsname = sdsnew(name); |
| 1301 | |
| 1302 | /* Make sure the entry is not duplicated. This may happen when the same |
| 1303 | * name for a master is used multiple times inside the configuration or |
| 1304 | * if we try to add multiple times a slave or sentinel with same ip/port |
| 1305 | * to a master. */ |
| 1306 | if (flags & SRI_MASTER(1<<0)) table = sentinel.masters; |
| 1307 | else if (flags & SRI_SLAVE(1<<1)) table = master->slaves; |
| 1308 | else if (flags & SRI_SENTINEL(1<<2)) table = master->sentinels; |
| 1309 | if (dictFind(table,sdsname)) { |
| 1310 | releaseSentinelAddr(addr); |
| 1311 | sdsfree(sdsname); |
| 1312 | errno(*__errno_location ()) = EBUSY16; |
| 1313 | return NULL((void*)0); |
| 1314 | } |
| 1315 | |
| 1316 | /* Create the instance object. */ |
| 1317 | ri = zmalloc(sizeof(*ri)); |
| 1318 | /* Note that all the instances are started in the disconnected state, |
| 1319 | * the event loop will take care of connecting them. */ |
| 1320 | ri->flags = flags; |
| 1321 | ri->name = sdsname; |
| 1322 | ri->runid = NULL((void*)0); |
| 1323 | ri->config_epoch = 0; |
| 1324 | ri->addr = addr; |
| 1325 | ri->link = createInstanceLink(); |
| 1326 | ri->last_pub_time = mstime(); |
| 1327 | ri->last_hello_time = mstime(); |
| 1328 | ri->last_master_down_reply_time = mstime(); |
| 1329 | ri->s_down_since_time = 0; |
| 1330 | ri->o_down_since_time = 0; |
| 1331 | ri->down_after_period = master ? master->down_after_period : |
| 1332 | SENTINEL_DEFAULT_DOWN_AFTER30000; |
| 1333 | ri->master_link_down_time = 0; |
| 1334 | ri->auth_pass = NULL((void*)0); |
| 1335 | ri->auth_user = NULL((void*)0); |
| 1336 | ri->slave_priority = SENTINEL_DEFAULT_SLAVE_PRIORITY100; |
| 1337 | ri->slave_reconf_sent_time = 0; |
| 1338 | ri->slave_master_host = NULL((void*)0); |
| 1339 | ri->slave_master_port = 0; |
| 1340 | ri->slave_master_link_status = SENTINEL_MASTER_LINK_STATUS_DOWN1; |
| 1341 | ri->slave_repl_offset = 0; |
| 1342 | ri->sentinels = dictCreate(&instancesDictType,NULL((void*)0)); |
| 1343 | ri->quorum = quorum; |
| 1344 | ri->parallel_syncs = SENTINEL_DEFAULT_PARALLEL_SYNCS1; |
| 1345 | ri->master = master; |
| 1346 | ri->slaves = dictCreate(&instancesDictType,NULL((void*)0)); |
| 1347 | ri->info_refresh = 0; |
| 1348 | ri->renamed_commands = dictCreate(&renamedCommandsDictType,NULL((void*)0)); |
| 1349 | |
| 1350 | /* Failover state. */ |
| 1351 | ri->leader = NULL((void*)0); |
| 1352 | ri->leader_epoch = 0; |
| 1353 | ri->failover_epoch = 0; |
| 1354 | ri->failover_state = SENTINEL_FAILOVER_STATE_NONE0; |
| 1355 | ri->failover_state_change_time = 0; |
| 1356 | ri->failover_start_time = 0; |
| 1357 | ri->failover_timeout = SENTINEL_DEFAULT_FAILOVER_TIMEOUT(60*3*1000); |
| 1358 | ri->failover_delay_logged = 0; |
| 1359 | ri->promoted_slave = NULL((void*)0); |
| 1360 | ri->notification_script = NULL((void*)0); |
| 1361 | ri->client_reconfig_script = NULL((void*)0); |
| 1362 | ri->info = NULL((void*)0); |
| 1363 | |
| 1364 | /* Role */ |
| 1365 | ri->role_reported = ri->flags & (SRI_MASTER(1<<0)|SRI_SLAVE(1<<1)); |
| 1366 | ri->role_reported_time = mstime(); |
| 1367 | ri->slave_conf_change_time = mstime(); |
| 1368 | |
| 1369 | /* Add into the right table. */ |
| 1370 | dictAdd(table, ri->name, ri); |
| 1371 | return ri; |
| 1372 | } |
| 1373 | |
| 1374 | /* Release this instance and all its slaves, sentinels, hiredis connections. |
| 1375 | * This function does not take care of unlinking the instance from the main |
| 1376 | * masters table (if it is a master) or from its master sentinels/slaves table |
| 1377 | * if it is a slave or sentinel. */ |
| 1378 | void releaseSentinelRedisInstance(sentinelRedisInstance *ri) { |
| 1379 | /* Release all its slaves or sentinels if any. */ |
| 1380 | dictRelease(ri->sentinels); |
| 1381 | dictRelease(ri->slaves); |
| 1382 | |
| 1383 | /* Disconnect the instance. */ |
| 1384 | releaseInstanceLink(ri->link,ri); |
| 1385 | |
| 1386 | /* Free other resources. */ |
| 1387 | sdsfree(ri->name); |
| 1388 | sdsfree(ri->runid); |
| 1389 | sdsfree(ri->notification_script); |
| 1390 | sdsfree(ri->client_reconfig_script); |
| 1391 | sdsfree(ri->slave_master_host); |
| 1392 | sdsfree(ri->leader); |
| 1393 | sdsfree(ri->auth_pass); |
| 1394 | sdsfree(ri->auth_user); |
| 1395 | sdsfree(ri->info); |
| 1396 | releaseSentinelAddr(ri->addr); |
| 1397 | dictRelease(ri->renamed_commands); |
| 1398 | |
| 1399 | /* Clear state into the master if needed. */ |
| 1400 | if ((ri->flags & SRI_SLAVE(1<<1)) && (ri->flags & SRI_PROMOTED(1<<7)) && ri->master) |
| 1401 | ri->master->promoted_slave = NULL((void*)0); |
| 1402 | |
| 1403 | zfree(ri); |
| 1404 | } |
| 1405 | |
| 1406 | /* Lookup a slave in a master Redis instance, by ip and port. */ |
| 1407 | sentinelRedisInstance *sentinelRedisInstanceLookupSlave( |
| 1408 | sentinelRedisInstance *ri, char *slave_addr, int port) |
| 1409 | { |
| 1410 | sds key; |
| 1411 | sentinelRedisInstance *slave; |
| 1412 | sentinelAddr *addr; |
| 1413 | |
| 1414 | serverAssert(ri->flags & SRI_MASTER)((ri->flags & (1<<0))?(void)0 : (_serverAssert("ri->flags & SRI_MASTER" ,"sentinel.c",1414),__builtin_unreachable())); |
| 1415 | |
| 1416 | /* We need to handle a slave_addr that is potentially a hostname. |
| 1417 | * If that is the case, depending on configuration we either resolve |
| 1418 | * it and use the IP addres or fail. |
| 1419 | */ |
| 1420 | addr = createSentinelAddr(slave_addr, port); |
| 1421 | if (!addr) return NULL((void*)0); |
| 1422 | key = announceSentinelAddrAndPort(addr); |
| 1423 | releaseSentinelAddr(addr); |
| 1424 | |
| 1425 | slave = dictFetchValue(ri->slaves,key); |
| 1426 | sdsfree(key); |
| 1427 | return slave; |
| 1428 | } |
| 1429 | |
| 1430 | /* Return the name of the type of the instance as a string. */ |
| 1431 | const char *sentinelRedisInstanceTypeStr(sentinelRedisInstance *ri) { |
| 1432 | if (ri->flags & SRI_MASTER(1<<0)) return "master"; |
| 1433 | else if (ri->flags & SRI_SLAVE(1<<1)) return "slave"; |
| 1434 | else if (ri->flags & SRI_SENTINEL(1<<2)) return "sentinel"; |
| 1435 | else return "unknown"; |
| 1436 | } |
| 1437 | |
| 1438 | /* This function remove the Sentinel with the specified ID from the |
| 1439 | * specified master. |
| 1440 | * |
| 1441 | * If "runid" is NULL the function returns ASAP. |
| 1442 | * |
| 1443 | * This function is useful because on Sentinels address switch, we want to |
| 1444 | * remove our old entry and add a new one for the same ID but with the new |
| 1445 | * address. |
| 1446 | * |
| 1447 | * The function returns 1 if the matching Sentinel was removed, otherwise |
| 1448 | * 0 if there was no Sentinel with this ID. */ |
| 1449 | int removeMatchingSentinelFromMaster(sentinelRedisInstance *master, char *runid) { |
| 1450 | dictIterator *di; |
| 1451 | dictEntry *de; |
| 1452 | int removed = 0; |
| 1453 | |
| 1454 | if (runid == NULL((void*)0)) return 0; |
| 1455 | |
| 1456 | di = dictGetSafeIterator(master->sentinels); |
| 1457 | while((de = dictNext(di)) != NULL((void*)0)) { |
| 1458 | sentinelRedisInstance *ri = dictGetVal(de)((de)->v.val); |
| 1459 | |
| 1460 | if (ri->runid && strcmp(ri->runid,runid) == 0) { |
| 1461 | dictDelete(master->sentinels,ri->name); |
| 1462 | removed++; |
| 1463 | } |
| 1464 | } |
| 1465 | dictReleaseIterator(di); |
| 1466 | return removed; |
| 1467 | } |
| 1468 | |
| 1469 | /* Search an instance with the same runid, ip and port into a dictionary |
| 1470 | * of instances. Return NULL if not found, otherwise return the instance |
| 1471 | * pointer. |
| 1472 | * |
| 1473 | * runid or addr can be NULL. In such a case the search is performed only |
| 1474 | * by the non-NULL field. */ |
| 1475 | sentinelRedisInstance *getSentinelRedisInstanceByAddrAndRunID(dict *instances, char *addr, int port, char *runid) { |
| 1476 | dictIterator *di; |
| 1477 | dictEntry *de; |
| 1478 | sentinelRedisInstance *instance = NULL((void*)0); |
| 1479 | sentinelAddr *ri_addr = NULL((void*)0); |
| 1480 | |
| 1481 | serverAssert(addr || runid)((addr || runid)?(void)0 : (_serverAssert("addr || runid","sentinel.c" ,1481),__builtin_unreachable())); /* User must pass at least one search param. */ |
| 1482 | if (addr != NULL((void*)0)) { |
| 1483 | /* Resolve addr, we use the IP as a key even if a hostname is used */ |
| 1484 | ri_addr = createSentinelAddr(addr, port); |
| 1485 | if (!ri_addr) return NULL((void*)0); |
| 1486 | } |
| 1487 | di = dictGetIterator(instances); |
| 1488 | while((de = dictNext(di)) != NULL((void*)0)) { |
| 1489 | sentinelRedisInstance *ri = dictGetVal(de)((de)->v.val); |
| 1490 | |
| 1491 | if (runid && !ri->runid) continue; |
| 1492 | if ((runid == NULL((void*)0) || strcmp(ri->runid, runid) == 0) && |
| 1493 | (addr == NULL((void*)0) || (strcmp(ri->addr->ip, ri_addr->ip) == 0 && |
| 1494 | ri->addr->port == port))) |
| 1495 | { |
| 1496 | instance = ri; |
| 1497 | break; |
| 1498 | } |
| 1499 | } |
| 1500 | dictReleaseIterator(di); |
| 1501 | if (ri_addr != NULL((void*)0)) |
| 1502 | releaseSentinelAddr(ri_addr); |
| 1503 | |
| 1504 | return instance; |
| 1505 | } |
| 1506 | |
| 1507 | /* Master lookup by name */ |
| 1508 | sentinelRedisInstance *sentinelGetMasterByName(char *name) { |
| 1509 | sentinelRedisInstance *ri; |
| 1510 | sds sdsname = sdsnew(name); |
| 1511 | |
| 1512 | ri = dictFetchValue(sentinel.masters,sdsname); |
| 1513 | sdsfree(sdsname); |
| 1514 | return ri; |
| 1515 | } |
| 1516 | |
| 1517 | /* Add the specified flags to all the instances in the specified dictionary. */ |
| 1518 | void sentinelAddFlagsToDictOfRedisInstances(dict *instances, int flags) { |
| 1519 | dictIterator *di; |
| 1520 | dictEntry *de; |
| 1521 | |
| 1522 | di = dictGetIterator(instances); |
| 1523 | while((de = dictNext(di)) != NULL((void*)0)) { |
| 1524 | sentinelRedisInstance *ri = dictGetVal(de)((de)->v.val); |
| 1525 | ri->flags |= flags; |
| 1526 | } |
| 1527 | dictReleaseIterator(di); |
| 1528 | } |
| 1529 | |
| 1530 | /* Remove the specified flags to all the instances in the specified |
| 1531 | * dictionary. */ |
| 1532 | void sentinelDelFlagsToDictOfRedisInstances(dict *instances, int flags) { |
| 1533 | dictIterator *di; |
| 1534 | dictEntry *de; |
| 1535 | |
| 1536 | di = dictGetIterator(instances); |
| 1537 | while((de = dictNext(di)) != NULL((void*)0)) { |
| 1538 | sentinelRedisInstance *ri = dictGetVal(de)((de)->v.val); |
| 1539 | ri->flags &= ~flags; |
| 1540 | } |
| 1541 | dictReleaseIterator(di); |
| 1542 | } |
| 1543 | |
| 1544 | /* Reset the state of a monitored master: |
| 1545 | * 1) Remove all slaves. |
| 1546 | * 2) Remove all sentinels. |
| 1547 | * 3) Remove most of the flags resulting from runtime operations. |
| 1548 | * 4) Reset timers to their default value. For example after a reset it will be |
| 1549 | * possible to failover again the same master ASAP, without waiting the |
| 1550 | * failover timeout delay. |
| 1551 | * 5) In the process of doing this undo the failover if in progress. |
| 1552 | * 6) Disconnect the connections with the master (will reconnect automatically). |
| 1553 | */ |
| 1554 | |
| 1555 | #define SENTINEL_RESET_NO_SENTINELS(1<<0) (1<<0) |
| 1556 | void sentinelResetMaster(sentinelRedisInstance *ri, int flags) { |
| 1557 | serverAssert(ri->flags & SRI_MASTER)((ri->flags & (1<<0))?(void)0 : (_serverAssert("ri->flags & SRI_MASTER" ,"sentinel.c",1557),__builtin_unreachable())); |
| 1558 | dictRelease(ri->slaves); |
| 1559 | ri->slaves = dictCreate(&instancesDictType,NULL((void*)0)); |
| 1560 | if (!(flags & SENTINEL_RESET_NO_SENTINELS(1<<0))) { |
| 1561 | dictRelease(ri->sentinels); |
| 1562 | ri->sentinels = dictCreate(&instancesDictType,NULL((void*)0)); |
| 1563 | } |
| 1564 | instanceLinkCloseConnection(ri->link,ri->link->cc); |
| 1565 | instanceLinkCloseConnection(ri->link,ri->link->pc); |
| 1566 | ri->flags &= SRI_MASTER(1<<0); |
| 1567 | if (ri->leader) { |
| 1568 | sdsfree(ri->leader); |
| 1569 | ri->leader = NULL((void*)0); |
| 1570 | } |
| 1571 | ri->failover_state = SENTINEL_FAILOVER_STATE_NONE0; |
| 1572 | ri->failover_state_change_time = 0; |
| 1573 | ri->failover_start_time = 0; /* We can failover again ASAP. */ |
| 1574 | ri->promoted_slave = NULL((void*)0); |
| 1575 | sdsfree(ri->runid); |
| 1576 | sdsfree(ri->slave_master_host); |
| 1577 | ri->runid = NULL((void*)0); |
| 1578 | ri->slave_master_host = NULL((void*)0); |
| 1579 | ri->link->act_ping_time = mstime(); |
| 1580 | ri->link->last_ping_time = 0; |
| 1581 | ri->link->last_avail_time = mstime(); |
| 1582 | ri->link->last_pong_time = mstime(); |
| 1583 | ri->role_reported_time = mstime(); |
| 1584 | ri->role_reported = SRI_MASTER(1<<0); |
| 1585 | if (flags & SENTINEL_GENERATE_EVENT(1<<16)) |
| 1586 | sentinelEvent(LL_WARNING3,"+reset-master",ri,"%@"); |
| 1587 | } |
| 1588 | |
| 1589 | /* Call sentinelResetMaster() on every master with a name matching the specified |
| 1590 | * pattern. */ |
| 1591 | int sentinelResetMastersByPattern(char *pattern, int flags) { |
| 1592 | dictIterator *di; |
| 1593 | dictEntry *de; |
| 1594 | int reset = 0; |
| 1595 | |
| 1596 | di = dictGetIterator(sentinel.masters); |
| 1597 | while((de = dictNext(di)) != NULL((void*)0)) { |
| 1598 | sentinelRedisInstance *ri = dictGetVal(de)((de)->v.val); |
| 1599 | |
| 1600 | if (ri->name) { |
| 1601 | if (stringmatch(pattern,ri->name,0)) { |
| 1602 | sentinelResetMaster(ri,flags); |
| 1603 | reset++; |
| 1604 | } |
| 1605 | } |
| 1606 | } |
| 1607 | dictReleaseIterator(di); |
| 1608 | return reset; |
| 1609 | } |
| 1610 | |
| 1611 | /* Reset the specified master with sentinelResetMaster(), and also change |
| 1612 | * the ip:port address, but take the name of the instance unmodified. |
| 1613 | * |
| 1614 | * This is used to handle the +switch-master event. |
| 1615 | * |
| 1616 | * The function returns C_ERR if the address can't be resolved for some |
| 1617 | * reason. Otherwise C_OK is returned. */ |
| 1618 | int sentinelResetMasterAndChangeAddress(sentinelRedisInstance *master, char *hostname, int port) { |
| 1619 | sentinelAddr *oldaddr, *newaddr; |
| 1620 | sentinelAddr **slaves = NULL((void*)0); |
| 1621 | int numslaves = 0, j; |
| 1622 | dictIterator *di; |
| 1623 | dictEntry *de; |
| 1624 | |
| 1625 | newaddr = createSentinelAddr(hostname,port); |
| 1626 | if (newaddr == NULL((void*)0)) return C_ERR-1; |
| 1627 | |
| 1628 | /* There can be only 0 or 1 slave that has the newaddr. |
| 1629 | * and It can add old master 1 more slave. |
| 1630 | * so It allocates dictSize(master->slaves) + 1 */ |
| 1631 | slaves = zmalloc(sizeof(sentinelAddr*)*(dictSize(master->slaves)((master->slaves)->ht[0].used+(master->slaves)->ht [1].used) + 1)); |
| 1632 | |
| 1633 | /* Don't include the one having the address we are switching to. */ |
| 1634 | di = dictGetIterator(master->slaves); |
| 1635 | while((de = dictNext(di)) != NULL((void*)0)) { |
| 1636 | sentinelRedisInstance *slave = dictGetVal(de)((de)->v.val); |
| 1637 | |
| 1638 | if (sentinelAddrIsEqual(slave->addr,newaddr)) continue; |
| 1639 | slaves[numslaves++] = dupSentinelAddr(slave->addr); |
| 1640 | } |
| 1641 | dictReleaseIterator(di); |
| 1642 | |
| 1643 | /* If we are switching to a different address, include the old address |
| 1644 | * as a slave as well, so that we'll be able to sense / reconfigure |
| 1645 | * the old master. */ |
| 1646 | if (!sentinelAddrIsEqual(newaddr,master->addr)) { |
| 1647 | slaves[numslaves++] = dupSentinelAddr(master->addr); |
| 1648 | } |
| 1649 | |
| 1650 | /* Reset and switch address. */ |
| 1651 | sentinelResetMaster(master,SENTINEL_RESET_NO_SENTINELS(1<<0)); |
| 1652 | oldaddr = master->addr; |
| 1653 | master->addr = newaddr; |
| 1654 | master->o_down_since_time = 0; |
| 1655 | master->s_down_since_time = 0; |
| 1656 | |
| 1657 | /* Add slaves back. */ |
| 1658 | for (j = 0; j < numslaves; j++) { |
| 1659 | sentinelRedisInstance *slave; |
| 1660 | |
| 1661 | slave = createSentinelRedisInstance(NULL((void*)0),SRI_SLAVE(1<<1),slaves[j]->hostname, |
| 1662 | slaves[j]->port, master->quorum, master); |
| 1663 | releaseSentinelAddr(slaves[j]); |
| 1664 | if (slave) sentinelEvent(LL_NOTICE2,"+slave",slave,"%@"); |
| 1665 | } |
| 1666 | zfree(slaves); |
| 1667 | |
| 1668 | /* Release the old address at the end so we are safe even if the function |
| 1669 | * gets the master->addr->ip and master->addr->port as arguments. */ |
| 1670 | releaseSentinelAddr(oldaddr); |
| 1671 | sentinelFlushConfig(); |
| 1672 | return C_OK0; |
| 1673 | } |
| 1674 | |
| 1675 | /* Return non-zero if there was no SDOWN or ODOWN error associated to this |
| 1676 | * instance in the latest 'ms' milliseconds. */ |
| 1677 | int sentinelRedisInstanceNoDownFor(sentinelRedisInstance *ri, mstime_t ms) { |
| 1678 | mstime_t most_recent; |
| 1679 | |
| 1680 | most_recent = ri->s_down_since_time; |
| 1681 | if (ri->o_down_since_time > most_recent) |
| 1682 | most_recent = ri->o_down_since_time; |
| 1683 | return most_recent == 0 || (mstime() - most_recent) > ms; |
| 1684 | } |
| 1685 | |
| 1686 | /* Return the current master address, that is, its address or the address |
| 1687 | * of the promoted slave if already operational. */ |
| 1688 | sentinelAddr *sentinelGetCurrentMasterAddress(sentinelRedisInstance *master) { |
| 1689 | /* If we are failing over the master, and the state is already |
| 1690 | * SENTINEL_FAILOVER_STATE_RECONF_SLAVES or greater, it means that we |
| 1691 | * already have the new configuration epoch in the master, and the |
| 1692 | * slave acknowledged the configuration switch. Advertise the new |
| 1693 | * address. */ |
| 1694 | if ((master->flags & SRI_FAILOVER_IN_PROGRESS(1<<6)) && |
| 1695 | master->promoted_slave && |
| 1696 | master->failover_state >= SENTINEL_FAILOVER_STATE_RECONF_SLAVES5) |
| 1697 | { |
| 1698 | return master->promoted_slave->addr; |
| 1699 | } else { |
| 1700 | return master->addr; |
| 1701 | } |
| 1702 | } |
| 1703 | |
| 1704 | /* This function sets the down_after_period field value in 'master' to all |
| 1705 | * the slaves and sentinel instances connected to this master. */ |
| 1706 | void sentinelPropagateDownAfterPeriod(sentinelRedisInstance *master) { |
| 1707 | dictIterator *di; |
| 1708 | dictEntry *de; |
| 1709 | int j; |
| 1710 | dict *d[] = {master->slaves, master->sentinels, NULL((void*)0)}; |
| 1711 | |
| 1712 | for (j = 0; d[j]; j++) { |
| 1713 | di = dictGetIterator(d[j]); |
| 1714 | while((de = dictNext(di)) != NULL((void*)0)) { |
| 1715 | sentinelRedisInstance *ri = dictGetVal(de)((de)->v.val); |
| 1716 | ri->down_after_period = master->down_after_period; |
| 1717 | } |
| 1718 | dictReleaseIterator(di); |
| 1719 | } |
| 1720 | } |
| 1721 | |
| 1722 | char *sentinelGetInstanceTypeString(sentinelRedisInstance *ri) { |
| 1723 | if (ri->flags & SRI_MASTER(1<<0)) return "master"; |
| 1724 | else if (ri->flags & SRI_SLAVE(1<<1)) return "slave"; |
| 1725 | else if (ri->flags & SRI_SENTINEL(1<<2)) return "sentinel"; |
| 1726 | else return "unknown"; |
| 1727 | } |
| 1728 | |
| 1729 | /* This function is used in order to send commands to Redis instances: the |
| 1730 | * commands we send from Sentinel may be renamed, a common case is a master |
| 1731 | * with CONFIG and SLAVEOF commands renamed for security concerns. In that |
| 1732 | * case we check the ri->renamed_command table (or if the instance is a slave, |
| 1733 | * we check the one of the master), and map the command that we should send |
| 1734 | * to the set of renamed commads. However, if the command was not renamed, |
| 1735 | * we just return "command" itself. */ |
| 1736 | char *sentinelInstanceMapCommand(sentinelRedisInstance *ri, char *command) { |
| 1737 | sds sc = sdsnew(command); |
| 1738 | if (ri->master) ri = ri->master; |
| 1739 | char *retval = dictFetchValue(ri->renamed_commands, sc); |
| 1740 | sdsfree(sc); |
| 1741 | return retval ? retval : command; |
| 1742 | } |
| 1743 | |
| 1744 | /* ============================ Config handling ============================= */ |
| 1745 | |
| 1746 | /* Generalise handling create instance error. Use SRI_MASTER, SRI_SLAVE or |
| 1747 | * SRI_SENTINEL as a role value. */ |
| 1748 | const char *sentinelCheckCreateInstanceErrors(int role) { |
| 1749 | switch(errno(*__errno_location ())) { |
| 1750 | case EBUSY16: |
| 1751 | switch (role) { |
| 1752 | case SRI_MASTER(1<<0): |
| 1753 | return "Duplicate master name."; |
| 1754 | case SRI_SLAVE(1<<1): |
| 1755 | return "Duplicate hostname and port for replica."; |
| 1756 | case SRI_SENTINEL(1<<2): |
| 1757 | return "Duplicate runid for sentinel."; |
| 1758 | default: |
| 1759 | serverAssert(0)((0)?(void)0 : (_serverAssert("0","sentinel.c",1759),__builtin_unreachable ())); |
| 1760 | break; |
| 1761 | } |
| 1762 | break; |
| 1763 | case ENOENT2: |
| 1764 | return "Can't resolve instance hostname."; |
| 1765 | case EINVAL22: |
| 1766 | return "Invalid port number."; |
| 1767 | default: |
| 1768 | return "Unknown Error for creating instances."; |
| 1769 | } |
| 1770 | } |
| 1771 | |
| 1772 | /* init function for server.sentinel_config */ |
| 1773 | void initializeSentinelConfig() { |
| 1774 | server.sentinel_config = zmalloc(sizeof(struct sentinelConfig)); |
| 1775 | server.sentinel_config->monitor_cfg = listCreate(); |
| 1776 | server.sentinel_config->pre_monitor_cfg = listCreate(); |
| 1777 | server.sentinel_config->post_monitor_cfg = listCreate(); |
| 1778 | listSetFreeMethod(server.sentinel_config->monitor_cfg,freeSentinelLoadQueueEntry)((server.sentinel_config->monitor_cfg)->free = (freeSentinelLoadQueueEntry )); |
| 1779 | listSetFreeMethod(server.sentinel_config->pre_monitor_cfg,freeSentinelLoadQueueEntry)((server.sentinel_config->pre_monitor_cfg)->free = (freeSentinelLoadQueueEntry )); |
| 1780 | listSetFreeMethod(server.sentinel_config->post_monitor_cfg,freeSentinelLoadQueueEntry)((server.sentinel_config->post_monitor_cfg)->free = (freeSentinelLoadQueueEntry )); |
| 1781 | } |
| 1782 | |
| 1783 | /* destroy function for server.sentinel_config */ |
| 1784 | void freeSentinelConfig() { |
| 1785 | /* release these three config queues since we will not use it anymore */ |
| 1786 | listRelease(server.sentinel_config->pre_monitor_cfg); |
| 1787 | listRelease(server.sentinel_config->monitor_cfg); |
| 1788 | listRelease(server.sentinel_config->post_monitor_cfg); |
| 1789 | zfree(server.sentinel_config); |
| 1790 | server.sentinel_config = NULL((void*)0); |
| 1791 | } |
| 1792 | |
| 1793 | /* Search config name in pre monitor config name array, return 1 if found, |
| 1794 | * 0 if not found. */ |
| 1795 | int searchPreMonitorCfgName(const char *name) { |
| 1796 | for (unsigned int i = 0; i < sizeof(preMonitorCfgName)/sizeof(preMonitorCfgName[0]); i++) { |
| 1797 | if (!strcasecmp(preMonitorCfgName[i],name)) return 1; |
| 1798 | } |
| 1799 | return 0; |
| 1800 | } |
| 1801 | |
| 1802 | /* free method for sentinelLoadQueueEntry when release the list */ |
| 1803 | void freeSentinelLoadQueueEntry(void *item) { |
| 1804 | struct sentinelLoadQueueEntry *entry = item; |
| 1805 | sdsfreesplitres(entry->argv,entry->argc); |
| 1806 | sdsfree(entry->line); |
| 1807 | zfree(entry); |
| 1808 | } |
| 1809 | |
| 1810 | /* This function is used for queuing sentinel configuration, the main |
| 1811 | * purpose of this function is to delay parsing the sentinel config option |
| 1812 | * in order to avoid the order dependent issue from the config. */ |
| 1813 | void queueSentinelConfig(sds *argv, int argc, int linenum, sds line) { |
| 1814 | int i; |
| 1815 | struct sentinelLoadQueueEntry *entry; |
| 1816 | |
| 1817 | /* initialize sentinel_config for the first call */ |
| 1818 | if (server.sentinel_config == NULL((void*)0)) initializeSentinelConfig(); |
| 1819 | |
| 1820 | entry = zmalloc(sizeof(struct sentinelLoadQueueEntry)); |
| 1821 | entry->argv = zmalloc(sizeof(char*)*argc); |
| 1822 | entry->argc = argc; |
| 1823 | entry->linenum = linenum; |
| 1824 | entry->line = sdsdup(line); |
| 1825 | for (i = 0; i < argc; i++) { |
| 1826 | entry->argv[i] = sdsdup(argv[i]); |
| 1827 | } |
| 1828 | /* Separate config lines with pre monitor config, monitor config and |
| 1829 | * post monitor config, in order to parsing config dependencies |
| 1830 | * correctly. */ |
| 1831 | if (!strcasecmp(argv[0],"monitor")) { |
| 1832 | listAddNodeTail(server.sentinel_config->monitor_cfg,entry); |
| 1833 | } else if (searchPreMonitorCfgName(argv[0])) { |
| 1834 | listAddNodeTail(server.sentinel_config->pre_monitor_cfg,entry); |
| 1835 | } else{ |
| 1836 | listAddNodeTail(server.sentinel_config->post_monitor_cfg,entry); |
| 1837 | } |
| 1838 | } |
| 1839 | |
| 1840 | /* This function is used for loading the sentinel configuration from |
| 1841 | * pre_monitor_cfg, monitor_cfg and post_monitor_cfg list */ |
| 1842 | void loadSentinelConfigFromQueue(void) { |
| 1843 | const char *err = NULL((void*)0); |
| 1844 | listIter li; |
| 1845 | listNode *ln; |
| 1846 | int linenum = 0; |
| 1847 | sds line = NULL((void*)0); |
| 1848 | |
| 1849 | /* if there is no sentinel_config entry, we can return immediately */ |
| 1850 | if (server.sentinel_config == NULL((void*)0)) return; |
| 1851 | |
| 1852 | /* loading from pre monitor config queue first to avoid dependency issues */ |
| 1853 | listRewind(server.sentinel_config->pre_monitor_cfg,&li); |
| 1854 | while((ln = listNext(&li))) { |
| 1855 | struct sentinelLoadQueueEntry *entry = ln->value; |
| 1856 | err = sentinelHandleConfiguration(entry->argv,entry->argc); |
| 1857 | if (err) { |
| 1858 | linenum = entry->linenum; |
| 1859 | line = entry->line; |
| 1860 | goto loaderr; |
| 1861 | } |
| 1862 | } |
| 1863 | |
| 1864 | /* loading from monitor config queue */ |
| 1865 | listRewind(server.sentinel_config->monitor_cfg,&li); |
| 1866 | while((ln = listNext(&li))) { |
| 1867 | struct sentinelLoadQueueEntry *entry = ln->value; |
| 1868 | err = sentinelHandleConfiguration(entry->argv,entry->argc); |
| 1869 | if (err) { |
| 1870 | linenum = entry->linenum; |
| 1871 | line = entry->line; |
| 1872 | goto loaderr; |
| 1873 | } |
| 1874 | } |
| 1875 | |
| 1876 | /* loading from the post monitor config queue */ |
| 1877 | listRewind(server.sentinel_config->post_monitor_cfg,&li); |
| 1878 | while((ln = listNext(&li))) { |
| 1879 | struct sentinelLoadQueueEntry *entry = ln->value; |
| 1880 | err = sentinelHandleConfiguration(entry->argv,entry->argc); |
| 1881 | if (err) { |
| 1882 | linenum = entry->linenum; |
| 1883 | line = entry->line; |
| 1884 | goto loaderr; |
| 1885 | } |
| 1886 | } |
| 1887 | |
| 1888 | /* free sentinel_config when config loading is finished */ |
| 1889 | freeSentinelConfig(); |
| 1890 | return; |
| 1891 | |
| 1892 | loaderr: |
| 1893 | fprintf(stderrstderr, "\n*** FATAL CONFIG FILE ERROR (Redis %s) ***\n", |
| 1894 | REDIS_VERSION"6.2.1"); |
| 1895 | fprintf(stderrstderr, "Reading the configuration file, at line %d\n", linenum); |
| 1896 | fprintf(stderrstderr, ">>> '%s'\n", line); |
| 1897 | fprintf(stderrstderr, "%s\n", err); |
| 1898 | exit(1); |
| 1899 | } |
| 1900 | |
| 1901 | const char *sentinelHandleConfiguration(char **argv, int argc) { |
| 1902 | |
| 1903 | sentinelRedisInstance *ri; |
| 1904 | |
| 1905 | if (!strcasecmp(argv[0],"monitor") && argc == 5) { |
| 1906 | /* monitor <name> <host> <port> <quorum> */ |
| 1907 | int quorum = atoi(argv[4]); |
| 1908 | |
| 1909 | if (quorum <= 0) return "Quorum must be 1 or greater."; |
| 1910 | if (createSentinelRedisInstance(argv[1],SRI_MASTER(1<<0),argv[2], |
| 1911 | atoi(argv[3]),quorum,NULL((void*)0)) == NULL((void*)0)) |
| 1912 | { |
| 1913 | return sentinelCheckCreateInstanceErrors(SRI_MASTER(1<<0)); |
| 1914 | } |
| 1915 | } else if (!strcasecmp(argv[0],"down-after-milliseconds") && argc == 3) { |
| 1916 | /* down-after-milliseconds <name> <milliseconds> */ |
| 1917 | ri = sentinelGetMasterByName(argv[1]); |
| 1918 | if (!ri) return "No such master with specified name."; |
| 1919 | ri->down_after_period = atoi(argv[2]); |
| 1920 | if (ri->down_after_period <= 0) |
| 1921 | return "negative or zero time parameter."; |
| 1922 | sentinelPropagateDownAfterPeriod(ri); |
| 1923 | } else if (!strcasecmp(argv[0],"failover-timeout") && argc == 3) { |
| 1924 | /* failover-timeout <name> <milliseconds> */ |
| 1925 | ri = sentinelGetMasterByName(argv[1]); |
| 1926 | if (!ri) return "No such master with specified name."; |
| 1927 | ri->failover_timeout = atoi(argv[2]); |
| 1928 | if (ri->failover_timeout <= 0) |
| 1929 | return "negative or zero time parameter."; |
| 1930 | } else if (!strcasecmp(argv[0],"parallel-syncs") && argc == 3) { |
| 1931 | /* parallel-syncs <name> <milliseconds> */ |
| 1932 | ri = sentinelGetMasterByName(argv[1]); |
| 1933 | if (!ri) return "No such master with specified name."; |
| 1934 | ri->parallel_syncs = atoi(argv[2]); |
| 1935 | } else if (!strcasecmp(argv[0],"notification-script") && argc == 3) { |
| 1936 | /* notification-script <name> <path> */ |
| 1937 | ri = sentinelGetMasterByName(argv[1]); |
| 1938 | if (!ri) return "No such master with specified name."; |
| 1939 | if (access(argv[2],X_OK1) == -1) |
| 1940 | return "Notification script seems non existing or non executable."; |
| 1941 | ri->notification_script = sdsnew(argv[2]); |
| 1942 | } else if (!strcasecmp(argv[0],"client-reconfig-script") && argc == 3) { |
| 1943 | /* client-reconfig-script <name> <path> */ |
| 1944 | ri = sentinelGetMasterByName(argv[1]); |
| 1945 | if (!ri) return "No such master with specified name."; |
| 1946 | if (access(argv[2],X_OK1) == -1) |
| 1947 | return "Client reconfiguration script seems non existing or " |
| 1948 | "non executable."; |
| 1949 | ri->client_reconfig_script = sdsnew(argv[2]); |
| 1950 | } else if (!strcasecmp(argv[0],"auth-pass") && argc == 3) { |
| 1951 | /* auth-pass <name> <password> */ |
| 1952 | ri = sentinelGetMasterByName(argv[1]); |
| 1953 | if (!ri) return "No such master with specified name."; |
| 1954 | ri->auth_pass = sdsnew(argv[2]); |
| 1955 | } else if (!strcasecmp(argv[0],"auth-user") && argc == 3) { |
| 1956 | /* auth-user <name> <username> */ |
| 1957 | ri = sentinelGetMasterByName(argv[1]); |
| 1958 | if (!ri) return "No such master with specified name."; |
| 1959 | ri->auth_user = sdsnew(argv[2]); |
| 1960 | } else if (!strcasecmp(argv[0],"current-epoch") && argc == 2) { |
| 1961 | /* current-epoch <epoch> */ |
| 1962 | unsigned long long current_epoch = strtoull(argv[1],NULL((void*)0),10); |
| 1963 | if (current_epoch > sentinel.current_epoch) |
| 1964 | sentinel.current_epoch = current_epoch; |
| 1965 | } else if (!strcasecmp(argv[0],"myid") && argc == 2) { |
| 1966 | if (strlen(argv[1]) != CONFIG_RUN_ID_SIZE40) |
| 1967 | return "Malformed Sentinel id in myid option."; |
| 1968 | memcpy(sentinel.myid,argv[1],CONFIG_RUN_ID_SIZE40); |
| 1969 | } else if (!strcasecmp(argv[0],"config-epoch") && argc == 3) { |
| 1970 | /* config-epoch <name> <epoch> */ |
| 1971 | ri = sentinelGetMasterByName(argv[1]); |
| 1972 | if (!ri) return "No such master with specified name."; |
| 1973 | ri->config_epoch = strtoull(argv[2],NULL((void*)0),10); |
| 1974 | /* The following update of current_epoch is not really useful as |
| 1975 | * now the current epoch is persisted on the config file, but |
| 1976 | * we leave this check here for redundancy. */ |
| 1977 | if (ri->config_epoch > sentinel.current_epoch) |
| 1978 | sentinel.current_epoch = ri->config_epoch; |
| 1979 | } else if (!strcasecmp(argv[0],"leader-epoch") && argc == 3) { |
| 1980 | /* leader-epoch <name> <epoch> */ |
| 1981 | ri = sentinelGetMasterByName(argv[1]); |
| 1982 | if (!ri) return "No such master with specified name."; |
| 1983 | ri->leader_epoch = strtoull(argv[2],NULL((void*)0),10); |
| 1984 | } else if ((!strcasecmp(argv[0],"known-slave") || |
| 1985 | !strcasecmp(argv[0],"known-replica")) && argc == 4) |
| 1986 | { |
| 1987 | sentinelRedisInstance *slave; |
| 1988 | |
| 1989 | /* known-replica <name> <ip> <port> */ |
| 1990 | ri = sentinelGetMasterByName(argv[1]); |
| 1991 | if (!ri) return "No such master with specified name."; |
| 1992 | if ((slave = createSentinelRedisInstance(NULL((void*)0),SRI_SLAVE(1<<1),argv[2], |
Although the value stored to 'slave' is used in the enclosing expression, the value is never actually read from 'slave' | |
| 1993 | atoi(argv[3]), ri->quorum, ri)) == NULL((void*)0)) |
| 1994 | { |
| 1995 | return sentinelCheckCreateInstanceErrors(SRI_SLAVE(1<<1)); |
| 1996 | } |
| 1997 | } else if (!strcasecmp(argv[0],"known-sentinel") && |
| 1998 | (argc == 4 || argc == 5)) { |
| 1999 | sentinelRedisInstance *si; |
| 2000 | |
| 2001 | if (argc == 5) { /* Ignore the old form without runid. */ |
| 2002 | /* known-sentinel <name> <ip> <port> [runid] */ |
| 2003 | ri = sentinelGetMasterByName(argv[1]); |
| 2004 | if (!ri) return "No such master with specified name."; |
| 2005 | if ((si = createSentinelRedisInstance(argv[4],SRI_SENTINEL(1<<2),argv[2], |
| 2006 | atoi(argv[3]), ri->quorum, ri)) == NULL((void*)0)) |
| 2007 | { |
| 2008 | return sentinelCheckCreateInstanceErrors(SRI_SENTINEL(1<<2)); |
| 2009 | } |
| 2010 | si->runid = sdsnew(argv[4]); |
| 2011 | sentinelTryConnectionSharing(si); |
| 2012 | } |
| 2013 | } else if (!strcasecmp(argv[0],"rename-command") && argc == 4) { |
| 2014 | /* rename-command <name> <command> <renamed-command> */ |
| 2015 | ri = sentinelGetMasterByName(argv[1]); |
| 2016 | if (!ri) return "No such master with specified name."; |
| 2017 | sds oldcmd = sdsnew(argv[2]); |
| 2018 | sds newcmd = sdsnew(argv[3]); |
| 2019 | if (dictAdd(ri->renamed_commands,oldcmd,newcmd) != DICT_OK0) { |
| 2020 | sdsfree(oldcmd); |
| 2021 | sdsfree(newcmd); |
| 2022 | return "Same command renamed multiple times with rename-command."; |
| 2023 | } |
| 2024 | } else if (!strcasecmp(argv[0],"announce-ip") && argc == 2) { |
| 2025 | /* announce-ip <ip-address> */ |
| 2026 | if (strlen(argv[1])) |
| 2027 | sentinel.announce_ip = sdsnew(argv[1]); |
| 2028 | } else if (!strcasecmp(argv[0],"announce-port") && argc == 2) { |
| 2029 | /* announce-port <port> */ |
| 2030 | sentinel.announce_port = atoi(argv[1]); |
| 2031 | } else if (!strcasecmp(argv[0],"deny-scripts-reconfig") && argc == 2) { |
| 2032 | /* deny-scripts-reconfig <yes|no> */ |
| 2033 | if ((sentinel.deny_scripts_reconfig = yesnotoi(argv[1])) == -1) { |
| 2034 | return "Please specify yes or no for the " |
| 2035 | "deny-scripts-reconfig options."; |
| 2036 | } |
| 2037 | } else if (!strcasecmp(argv[0],"sentinel-user") && argc == 2) { |
| 2038 | /* sentinel-user <user-name> */ |
| 2039 | if (strlen(argv[1])) |
| 2040 | sentinel.sentinel_auth_user = sdsnew(argv[1]); |
| 2041 | } else if (!strcasecmp(argv[0],"sentinel-pass") && argc == 2) { |
| 2042 | /* sentinel-pass <password> */ |
| 2043 | if (strlen(argv[1])) |
| 2044 | sentinel.sentinel_auth_pass = sdsnew(argv[1]); |
| 2045 | } else if (!strcasecmp(argv[0],"resolve-hostnames") && argc == 2) { |
| 2046 | /* resolve-hostnames <yes|no> */ |
| 2047 | if ((sentinel.resolve_hostnames = yesnotoi(argv[1])) == -1) { |
| 2048 | return "Please specify yes or not for the resolve-hostnames option."; |
| 2049 | } |
| 2050 | } else if (!strcasecmp(argv[0],"announce-hostnames") && argc == 2) { |
| 2051 | /* announce-hostnames <yes|no> */ |
| 2052 | if ((sentinel.announce_hostnames = yesnotoi(argv[1])) == -1) { |
| 2053 | return "Please specify yes or not for the announce-hostnames option."; |
| 2054 | } |
| 2055 | } else { |
| 2056 | return "Unrecognized sentinel configuration statement."; |
| 2057 | } |
| 2058 | return NULL((void*)0); |
| 2059 | } |
| 2060 | |
| 2061 | /* Implements CONFIG REWRITE for "sentinel" option. |
| 2062 | * This is used not just to rewrite the configuration given by the user |
| 2063 | * (the configured masters) but also in order to retain the state of |
| 2064 | * Sentinel across restarts: config epoch of masters, associated slaves |
| 2065 | * and sentinel instances, and so forth. */ |
| 2066 | void rewriteConfigSentinelOption(struct rewriteConfigState *state) { |
| 2067 | dictIterator *di, *di2; |
| 2068 | dictEntry *de; |
| 2069 | sds line; |
| 2070 | |
| 2071 | /* sentinel unique ID. */ |
| 2072 | line = sdscatprintf(sdsempty(), "sentinel myid %s", sentinel.myid); |
| 2073 | rewriteConfigRewriteLine(state,"sentinel myid",line,1); |
| 2074 | |
| 2075 | /* sentinel deny-scripts-reconfig. */ |
| 2076 | line = sdscatprintf(sdsempty(), "sentinel deny-scripts-reconfig %s", |
| 2077 | sentinel.deny_scripts_reconfig ? "yes" : "no"); |
| 2078 | rewriteConfigRewriteLine(state,"sentinel deny-scripts-reconfig",line, |
| 2079 | sentinel.deny_scripts_reconfig != SENTINEL_DEFAULT_DENY_SCRIPTS_RECONFIG1); |
| 2080 | |
| 2081 | /* sentinel resolve-hostnames. |
| 2082 | * This must be included early in the file so it is already in effect |
| 2083 | * when reading the file. |
| 2084 | */ |
| 2085 | line = sdscatprintf(sdsempty(), "sentinel resolve-hostnames %s", |
| 2086 | sentinel.resolve_hostnames ? "yes" : "no"); |
| 2087 | rewriteConfigRewriteLine(state,"sentinel resolve-hostnames",line, |
| 2088 | sentinel.resolve_hostnames != SENTINEL_DEFAULT_RESOLVE_HOSTNAMES0); |
| 2089 | |
| 2090 | /* sentinel announce-hostnames. */ |
| 2091 | line = sdscatprintf(sdsempty(), "sentinel announce-hostnames %s", |
| 2092 | sentinel.announce_hostnames ? "yes" : "no"); |
| 2093 | rewriteConfigRewriteLine(state,"sentinel announce-hostnames",line, |
| 2094 | sentinel.announce_hostnames != SENTINEL_DEFAULT_ANNOUNCE_HOSTNAMES0); |
| 2095 | |
| 2096 | /* For every master emit a "sentinel monitor" config entry. */ |
| 2097 | di = dictGetIterator(sentinel.masters); |
| 2098 | while((de = dictNext(di)) != NULL((void*)0)) { |
| 2099 | sentinelRedisInstance *master, *ri; |
| 2100 | sentinelAddr *master_addr; |
| 2101 | |
| 2102 | /* sentinel monitor */ |
| 2103 | master = dictGetVal(de)((de)->v.val); |
| 2104 | master_addr = sentinelGetCurrentMasterAddress(master); |
| 2105 | line = sdscatprintf(sdsempty(),"sentinel monitor %s %s %d %d", |
| 2106 | master->name, announceSentinelAddr(master_addr), master_addr->port, |
| 2107 | master->quorum); |
| 2108 | rewriteConfigRewriteLine(state,"sentinel monitor",line,1); |
| 2109 | /* rewriteConfigMarkAsProcessed is handled after the loop */ |
| 2110 | |
| 2111 | /* sentinel down-after-milliseconds */ |
| 2112 | if (master->down_after_period != SENTINEL_DEFAULT_DOWN_AFTER30000) { |
| 2113 | line = sdscatprintf(sdsempty(), |
| 2114 | "sentinel down-after-milliseconds %s %ld", |
| 2115 | master->name, (long) master->down_after_period); |
| 2116 | rewriteConfigRewriteLine(state,"sentinel down-after-milliseconds",line,1); |
| 2117 | /* rewriteConfigMarkAsProcessed is handled after the loop */ |
| 2118 | } |
| 2119 | |
| 2120 | /* sentinel failover-timeout */ |
| 2121 | if (master->failover_timeout != SENTINEL_DEFAULT_FAILOVER_TIMEOUT(60*3*1000)) { |
| 2122 | line = sdscatprintf(sdsempty(), |
| 2123 | "sentinel failover-timeout %s %ld", |
| 2124 | master->name, (long) master->failover_timeout); |
| 2125 | rewriteConfigRewriteLine(state,"sentinel failover-timeout",line,1); |
| 2126 | /* rewriteConfigMarkAsProcessed is handled after the loop */ |
| 2127 | |
| 2128 | } |
| 2129 | |
| 2130 | /* sentinel parallel-syncs */ |
| 2131 | if (master->parallel_syncs != SENTINEL_DEFAULT_PARALLEL_SYNCS1) { |
| 2132 | line = sdscatprintf(sdsempty(), |
| 2133 | "sentinel parallel-syncs %s %d", |
| 2134 | master->name, master->parallel_syncs); |
| 2135 | rewriteConfigRewriteLine(state,"sentinel parallel-syncs",line,1); |
| 2136 | /* rewriteConfigMarkAsProcessed is handled after the loop */ |
| 2137 | } |
| 2138 | |
| 2139 | /* sentinel notification-script */ |
| 2140 | if (master->notification_script) { |
| 2141 | line = sdscatprintf(sdsempty(), |
| 2142 | "sentinel notification-script %s %s", |
| 2143 | master->name, master->notification_script); |
| 2144 | rewriteConfigRewriteLine(state,"sentinel notification-script",line,1); |
| 2145 | /* rewriteConfigMarkAsProcessed is handled after the loop */ |
| 2146 | } |
| 2147 | |
| 2148 | /* sentinel client-reconfig-script */ |
| 2149 | if (master->client_reconfig_script) { |
| 2150 | line = sdscatprintf(sdsempty(), |
| 2151 | "sentinel client-reconfig-script %s %s", |
| 2152 | master->name, master->client_reconfig_script); |
| 2153 | rewriteConfigRewriteLine(state,"sentinel client-reconfig-script",line,1); |
| 2154 | /* rewriteConfigMarkAsProcessed is handled after the loop */ |
| 2155 | } |
| 2156 | |
| 2157 | /* sentinel auth-pass & auth-user */ |
| 2158 | if (master->auth_pass) { |
| 2159 | line = sdscatprintf(sdsempty(), |
| 2160 | "sentinel auth-pass %s %s", |
| 2161 | master->name, master->auth_pass); |
| 2162 | rewriteConfigRewriteLine(state,"sentinel auth-pass",line,1); |
| 2163 | /* rewriteConfigMarkAsProcessed is handled after the loop */ |
| 2164 | } |
| 2165 | |
| 2166 | if (master->auth_user) { |
| 2167 | line = sdscatprintf(sdsempty(), |
| 2168 | "sentinel auth-user %s %s", |
| 2169 | master->name, master->auth_user); |
| 2170 | rewriteConfigRewriteLine(state,"sentinel auth-user",line,1); |
| 2171 | /* rewriteConfigMarkAsProcessed is handled after the loop */ |
| 2172 | } |
| 2173 | |
| 2174 | /* sentinel config-epoch */ |
| 2175 | line = sdscatprintf(sdsempty(), |
| 2176 | "sentinel config-epoch %s %llu", |
| 2177 | master->name, (unsigned long long) master->config_epoch); |
| 2178 | rewriteConfigRewriteLine(state,"sentinel config-epoch",line,1); |
| 2179 | /* rewriteConfigMarkAsProcessed is handled after the loop */ |
| 2180 | |
| 2181 | |
| 2182 | /* sentinel leader-epoch */ |
| 2183 | line = sdscatprintf(sdsempty(), |
| 2184 | "sentinel leader-epoch %s %llu", |
| 2185 | master->name, (unsigned long long) master->leader_epoch); |
| 2186 | rewriteConfigRewriteLine(state,"sentinel leader-epoch",line,1); |
| 2187 | /* rewriteConfigMarkAsProcessed is handled after the loop */ |
| 2188 | |
| 2189 | /* sentinel known-slave */ |
| 2190 | di2 = dictGetIterator(master->slaves); |
| 2191 | while((de = dictNext(di2)) != NULL((void*)0)) { |
| 2192 | sentinelAddr *slave_addr; |
| 2193 | |
| 2194 | ri = dictGetVal(de)((de)->v.val); |
| 2195 | slave_addr = ri->addr; |
| 2196 | |
| 2197 | /* If master_addr (obtained using sentinelGetCurrentMasterAddress() |
| 2198 | * so it may be the address of the promoted slave) is equal to this |
| 2199 | * slave's address, a failover is in progress and the slave was |
| 2200 | * already successfully promoted. So as the address of this slave |
| 2201 | * we use the old master address instead. */ |
| 2202 | if (sentinelAddrIsEqual(slave_addr,master_addr)) |
| 2203 | slave_addr = master->addr; |
| 2204 | line = sdscatprintf(sdsempty(), |
| 2205 | "sentinel known-replica %s %s %d", |
| 2206 | master->name, announceSentinelAddr(slave_addr), slave_addr->port); |
| 2207 | rewriteConfigRewriteLine(state,"sentinel known-replica",line,1); |
| 2208 | /* rewriteConfigMarkAsProcessed is handled after the loop */ |
| 2209 | } |
| 2210 | dictReleaseIterator(di2); |
| 2211 | |
| 2212 | /* sentinel known-sentinel */ |
| 2213 | di2 = dictGetIterator(master->sentinels); |
| 2214 | while((de = dictNext(di2)) != NULL((void*)0)) { |
| 2215 | ri = dictGetVal(de)((de)->v.val); |
| 2216 | if (ri->runid == NULL((void*)0)) continue; |
| 2217 | line = sdscatprintf(sdsempty(), |
| 2218 | "sentinel known-sentinel %s %s %d %s", |
| 2219 | master->name, announceSentinelAddr(ri->addr), ri->addr->port, ri->runid); |
| 2220 | rewriteConfigRewriteLine(state,"sentinel known-sentinel",line,1); |
| 2221 | /* rewriteConfigMarkAsProcessed is handled after the loop */ |
| 2222 | } |
| 2223 | dictReleaseIterator(di2); |
| 2224 | |
| 2225 | /* sentinel rename-command */ |
| 2226 | di2 = dictGetIterator(master->renamed_commands); |
| 2227 | while((de = dictNext(di2)) != NULL((void*)0)) { |
| 2228 | sds oldname = dictGetKey(de)((de)->key); |
| 2229 | sds newname = dictGetVal(de)((de)->v.val); |
| 2230 | line = sdscatprintf(sdsempty(), |
| 2231 | "sentinel rename-command %s %s %s", |
| 2232 | master->name, oldname, newname); |
| 2233 | rewriteConfigRewriteLine(state,"sentinel rename-command",line,1); |
| 2234 | /* rewriteConfigMarkAsProcessed is handled after the loop */ |
| 2235 | } |
| 2236 | dictReleaseIterator(di2); |
| 2237 | } |
| 2238 | |
| 2239 | /* sentinel current-epoch is a global state valid for all the masters. */ |
| 2240 | line = sdscatprintf(sdsempty(), |
| 2241 | "sentinel current-epoch %llu", (unsigned long long) sentinel.current_epoch); |
| 2242 | rewriteConfigRewriteLine(state,"sentinel current-epoch",line,1); |
| 2243 | |
| 2244 | /* sentinel announce-ip. */ |
| 2245 | if (sentinel.announce_ip) { |
| 2246 | line = sdsnew("sentinel announce-ip "); |
| 2247 | line = sdscatrepr(line, sentinel.announce_ip, sdslen(sentinel.announce_ip)); |
| 2248 | rewriteConfigRewriteLine(state,"sentinel announce-ip",line,1); |
| 2249 | } else { |
| 2250 | rewriteConfigMarkAsProcessed(state,"sentinel announce-ip"); |
| 2251 | } |
| 2252 | |
| 2253 | /* sentinel announce-port. */ |
| 2254 | if (sentinel.announce_port) { |
| 2255 | line = sdscatprintf(sdsempty(),"sentinel announce-port %d", |
| 2256 | sentinel.announce_port); |
| 2257 | rewriteConfigRewriteLine(state,"sentinel announce-port",line,1); |
| 2258 | } else { |
| 2259 | rewriteConfigMarkAsProcessed(state,"sentinel announce-port"); |
| 2260 | } |
| 2261 | |
| 2262 | /* sentinel sentinel-user. */ |
| 2263 | if (sentinel.sentinel_auth_user) { |
| 2264 | line = sdscatprintf(sdsempty(), "sentinel sentinel-user %s", sentinel.sentinel_auth_user); |
| 2265 | rewriteConfigRewriteLine(state,"sentinel sentinel-user",line,1); |
| 2266 | } else { |
| 2267 | rewriteConfigMarkAsProcessed(state,"sentinel sentinel-user"); |
| 2268 | } |
| 2269 | |
| 2270 | /* sentinel sentinel-pass. */ |
| 2271 | if (sentinel.sentinel_auth_pass) { |
| 2272 | line = sdscatprintf(sdsempty(), "sentinel sentinel-pass %s", sentinel.sentinel_auth_pass); |
| 2273 | rewriteConfigRewriteLine(state,"sentinel sentinel-pass",line,1); |
| 2274 | } else { |
| 2275 | rewriteConfigMarkAsProcessed(state,"sentinel sentinel-pass"); |
| 2276 | } |
| 2277 | |
| 2278 | dictReleaseIterator(di); |
| 2279 | |
| 2280 | /* NOTE: the purpose here is in case due to the state change, the config rewrite |
| 2281 | does not handle the configs, however, previously the config was set in the config file, |
| 2282 | rewriteConfigMarkAsProcessed should be put here to mark it as processed in order to |
| 2283 | delete the old config entry. |
| 2284 | */ |
| 2285 | rewriteConfigMarkAsProcessed(state,"sentinel monitor"); |
| 2286 | rewriteConfigMarkAsProcessed(state,"sentinel down-after-milliseconds"); |
| 2287 | rewriteConfigMarkAsProcessed(state,"sentinel failover-timeout"); |
| 2288 | rewriteConfigMarkAsProcessed(state,"sentinel parallel-syncs"); |
| 2289 | rewriteConfigMarkAsProcessed(state,"sentinel notification-script"); |
| 2290 | rewriteConfigMarkAsProcessed(state,"sentinel client-reconfig-script"); |
| 2291 | rewriteConfigMarkAsProcessed(state,"sentinel auth-pass"); |
| 2292 | rewriteConfigMarkAsProcessed(state,"sentinel auth-user"); |
| 2293 | rewriteConfigMarkAsProcessed(state,"sentinel config-epoch"); |
| 2294 | rewriteConfigMarkAsProcessed(state,"sentinel leader-epoch"); |
| 2295 | rewriteConfigMarkAsProcessed(state,"sentinel known-replica"); |
| 2296 | rewriteConfigMarkAsProcessed(state,"sentinel known-sentinel"); |
| 2297 | rewriteConfigMarkAsProcessed(state,"sentinel rename-command"); |
| 2298 | } |
| 2299 | |
| 2300 | /* This function uses the config rewriting Redis engine in order to persist |
| 2301 | * the state of the Sentinel in the current configuration file. |
| 2302 | * |
| 2303 | * Before returning the function calls fsync() against the generated |
| 2304 | * configuration file to make sure changes are committed to disk. |
| 2305 | * |
| 2306 | * On failure the function logs a warning on the Redis log. */ |
| 2307 | void sentinelFlushConfig(void) { |
| 2308 | int fd = -1; |
| 2309 | int saved_hz = server.hz; |
| 2310 | int rewrite_status; |
| 2311 | |
| 2312 | server.hz = CONFIG_DEFAULT_HZ10; |
| 2313 | rewrite_status = rewriteConfig(server.configfile, 0); |
| 2314 | server.hz = saved_hz; |
| 2315 | |
| 2316 | if (rewrite_status == -1) goto werr; |
| 2317 | if ((fd = open(server.configfile,O_RDONLY00)) == -1) goto werr; |
| 2318 | if (fsync(fd) == -1) goto werr; |
| 2319 | if (close(fd) == EOF(-1)) goto werr; |
| 2320 | return; |
| 2321 | |
| 2322 | werr: |
| 2323 | if (fd != -1) close(fd); |
| 2324 | serverLog(LL_WARNING3,"WARNING: Sentinel was not able to save the new configuration on disk!!!: %s", strerror(errno(*__errno_location ()))); |
| 2325 | } |
| 2326 | |
| 2327 | /* ====================== hiredis connection handling ======================= */ |
| 2328 | |
| 2329 | /* Send the AUTH command with the specified master password if needed. |
| 2330 | * Note that for slaves the password set for the master is used. |
| 2331 | * |
| 2332 | * In case this Sentinel requires a password as well, via the "requirepass" |
| 2333 | * configuration directive, we assume we should use the local password in |
| 2334 | * order to authenticate when connecting with the other Sentinels as well. |
| 2335 | * So basically all the Sentinels share the same password and use it to |
| 2336 | * authenticate reciprocally. |
| 2337 | * |
| 2338 | * We don't check at all if the command was successfully transmitted |
| 2339 | * to the instance as if it fails Sentinel will detect the instance down, |
| 2340 | * will disconnect and reconnect the link and so forth. */ |
| 2341 | void sentinelSendAuthIfNeeded(sentinelRedisInstance *ri, redisAsyncContext *c) { |
| 2342 | char *auth_pass = NULL((void*)0); |
| 2343 | char *auth_user = NULL((void*)0); |
| 2344 | |
| 2345 | if (ri->flags & SRI_MASTER(1<<0)) { |
| 2346 | auth_pass = ri->auth_pass; |
| 2347 | auth_user = ri->auth_user; |
| 2348 | } else if (ri->flags & SRI_SLAVE(1<<1)) { |
| 2349 | auth_pass = ri->master->auth_pass; |
| 2350 | auth_user = ri->master->auth_user; |
| 2351 | } else if (ri->flags & SRI_SENTINEL(1<<2)) { |
| 2352 | /* If sentinel_auth_user is NULL, AUTH will use default user |
| 2353 | with sentinel_auth_pass to authenticate */ |
| 2354 | if (sentinel.sentinel_auth_pass) { |
| 2355 | auth_pass = sentinel.sentinel_auth_pass; |
| 2356 | auth_user = sentinel.sentinel_auth_user; |
| 2357 | } else { |
| 2358 | /* Compatibility with old configs. requirepass is used |
| 2359 | * for both incoming and outgoing authentication. */ |
| 2360 | auth_pass = server.requirepass; |
| 2361 | auth_user = NULL((void*)0); |
| 2362 | } |
| 2363 | } |
| 2364 | |
| 2365 | if (auth_pass && auth_user == NULL((void*)0)) { |
| 2366 | if (redisAsyncCommand(c, sentinelDiscardReplyCallback, ri, "%s %s", |
| 2367 | sentinelInstanceMapCommand(ri,"AUTH"), |
| 2368 | auth_pass) == C_OK0) ri->link->pending_commands++; |
| 2369 | } else if (auth_pass && auth_user) { |
| 2370 | /* If we also have an username, use the ACL-style AUTH command |
| 2371 | * with two arguments, username and password. */ |
| 2372 | if (redisAsyncCommand(c, sentinelDiscardReplyCallback, ri, "%s %s %s", |
| 2373 | sentinelInstanceMapCommand(ri,"AUTH"), |
| 2374 | auth_user, auth_pass) == C_OK0) ri->link->pending_commands++; |
| 2375 | } |
| 2376 | } |
| 2377 | |
| 2378 | /* Use CLIENT SETNAME to name the connection in the Redis instance as |
| 2379 | * sentinel-<first_8_chars_of_runid>-<connection_type> |
| 2380 | * The connection type is "cmd" or "pubsub" as specified by 'type'. |
| 2381 | * |
| 2382 | * This makes it possible to list all the sentinel instances connected |
| 2383 | * to a Redis server with CLIENT LIST, grepping for a specific name format. */ |
| 2384 | void sentinelSetClientName(sentinelRedisInstance *ri, redisAsyncContext *c, char *type) { |
| 2385 | char name[64]; |
| 2386 | |
| 2387 | snprintf(name,sizeof(name),"sentinel-%.8s-%s",sentinel.myid,type); |
| 2388 | if (redisAsyncCommand(c, sentinelDiscardReplyCallback, ri, |
| 2389 | "%s SETNAME %s", |
| 2390 | sentinelInstanceMapCommand(ri,"CLIENT"), |
| 2391 | name) == C_OK0) |
| 2392 | { |
| 2393 | ri->link->pending_commands++; |
| 2394 | } |
| 2395 | } |
| 2396 | |
| 2397 | static int instanceLinkNegotiateTLS(redisAsyncContext *context) { |
| 2398 | #ifndef USE_OPENSSL |
| 2399 | (void) context; |
| 2400 | #else |
| 2401 | if (!redis_tls_ctx) return C_ERR-1; |
| 2402 | SSL *ssl = SSL_new(redis_tls_client_ctx ? redis_tls_client_ctx : redis_tls_ctx); |
| 2403 | if (!ssl) return C_ERR-1; |
| 2404 | |
| 2405 | if (redisInitiateSSL(&context->c, ssl) == REDIS_ERR-1) return C_ERR-1; |
| 2406 | #endif |
| 2407 | return C_OK0; |
| 2408 | } |
| 2409 | |
| 2410 | /* Create the async connections for the instance link if the link |
| 2411 | * is disconnected. Note that link->disconnected is true even if just |
| 2412 | * one of the two links (commands and pub/sub) is missing. */ |
| 2413 | void sentinelReconnectInstance(sentinelRedisInstance *ri) { |
| 2414 | if (ri->link->disconnected == 0) return; |
| 2415 | if (ri->addr->port == 0) return; /* port == 0 means invalid address. */ |
| 2416 | instanceLink *link = ri->link; |
| 2417 | mstime_t now = mstime(); |
| 2418 | |
| 2419 | if (now - ri->link->last_reconn_time < SENTINEL_PING_PERIOD1000) return; |
| 2420 | ri->link->last_reconn_time = now; |
| 2421 | |
| 2422 | /* Commands connection. */ |
| 2423 | if (link->cc == NULL((void*)0)) { |
| 2424 | link->cc = redisAsyncConnectBind(ri->addr->ip,ri->addr->port,NET_FIRST_BIND_ADDR(server.bindaddr_count ? server.bindaddr[0] : ((void*)0))); |
| 2425 | if (!link->cc->err) anetCloexec(link->cc->c.fd); |
| 2426 | if (!link->cc->err && server.tls_replication && |
| 2427 | (instanceLinkNegotiateTLS(link->cc) == C_ERR-1)) { |
| 2428 | sentinelEvent(LL_DEBUG0,"-cmd-link-reconnection",ri,"%@ #Failed to initialize TLS"); |
| 2429 | instanceLinkCloseConnection(link,link->cc); |
| 2430 | } else if (link->cc->err) { |
| 2431 | sentinelEvent(LL_DEBUG0,"-cmd-link-reconnection",ri,"%@ #%s", |
| 2432 | link->cc->errstr); |
| 2433 | instanceLinkCloseConnection(link,link->cc); |
| 2434 | } else { |
| 2435 | link->pending_commands = 0; |
| 2436 | link->cc_conn_time = mstime(); |
| 2437 | link->cc->data = link; |
| 2438 | redisAeAttach(server.el,link->cc); |
| 2439 | redisAsyncSetConnectCallback(link->cc, |
| 2440 | sentinelLinkEstablishedCallback); |
| 2441 | redisAsyncSetDisconnectCallback(link->cc, |
| 2442 | sentinelDisconnectCallback); |
| 2443 | sentinelSendAuthIfNeeded(ri,link->cc); |
| 2444 | sentinelSetClientName(ri,link->cc,"cmd"); |
| 2445 | |
| 2446 | /* Send a PING ASAP when reconnecting. */ |
| 2447 | sentinelSendPing(ri); |
| 2448 | } |
| 2449 | } |
| 2450 | /* Pub / Sub */ |
| 2451 | if ((ri->flags & (SRI_MASTER(1<<0)|SRI_SLAVE(1<<1))) && link->pc == NULL((void*)0)) { |
| 2452 | link->pc = redisAsyncConnectBind(ri->addr->ip,ri->addr->port,NET_FIRST_BIND_ADDR(server.bindaddr_count ? server.bindaddr[0] : ((void*)0))); |
| 2453 | if (!link->pc->err) anetCloexec(link->pc->c.fd); |
| 2454 | if (!link->pc->err && server.tls_replication && |
| 2455 | (instanceLinkNegotiateTLS(link->pc) == C_ERR-1)) { |
| 2456 | sentinelEvent(LL_DEBUG0,"-pubsub-link-reconnection",ri,"%@ #Failed to initialize TLS"); |
| 2457 | } else if (link->pc->err) { |
| 2458 | sentinelEvent(LL_DEBUG0,"-pubsub-link-reconnection",ri,"%@ #%s", |
| 2459 | link->pc->errstr); |
| 2460 | instanceLinkCloseConnection(link,link->pc); |
| 2461 | } else { |
| 2462 | int retval; |
| 2463 | link->pc_conn_time = mstime(); |
| 2464 | link->pc->data = link; |
| 2465 | redisAeAttach(server.el,link->pc); |
| 2466 | redisAsyncSetConnectCallback(link->pc, |
| 2467 | sentinelLinkEstablishedCallback); |
| 2468 | redisAsyncSetDisconnectCallback(link->pc, |
| 2469 | sentinelDisconnectCallback); |
| 2470 | sentinelSendAuthIfNeeded(ri,link->pc); |
| 2471 | sentinelSetClientName(ri,link->pc,"pubsub"); |
| 2472 | /* Now we subscribe to the Sentinels "Hello" channel. */ |
| 2473 | retval = redisAsyncCommand(link->pc, |
| 2474 | sentinelReceiveHelloMessages, ri, "%s %s", |
| 2475 | sentinelInstanceMapCommand(ri,"SUBSCRIBE"), |
| 2476 | SENTINEL_HELLO_CHANNEL"__sentinel__:hello"); |
| 2477 | if (retval != C_OK0) { |
| 2478 | /* If we can't subscribe, the Pub/Sub connection is useless |
| 2479 | * and we can simply disconnect it and try again. */ |
| 2480 | instanceLinkCloseConnection(link,link->pc); |
| 2481 | return; |
| 2482 | } |
| 2483 | } |
| 2484 | } |
| 2485 | /* Clear the disconnected status only if we have both the connections |
| 2486 | * (or just the commands connection if this is a sentinel instance). */ |
| 2487 | if (link->cc && (ri->flags & SRI_SENTINEL(1<<2) || link->pc)) |
| 2488 | link->disconnected = 0; |
| 2489 | } |
| 2490 | |
| 2491 | /* ======================== Redis instances pinging ======================== */ |
| 2492 | |
| 2493 | /* Return true if master looks "sane", that is: |
| 2494 | * 1) It is actually a master in the current configuration. |
| 2495 | * 2) It reports itself as a master. |
| 2496 | * 3) It is not SDOWN or ODOWN. |
| 2497 | * 4) We obtained last INFO no more than two times the INFO period time ago. */ |
| 2498 | int sentinelMasterLooksSane(sentinelRedisInstance *master) { |
| 2499 | return |
| 2500 | master->flags & SRI_MASTER(1<<0) && |
| 2501 | master->role_reported == SRI_MASTER(1<<0) && |
| 2502 | (master->flags & (SRI_S_DOWN(1<<3)|SRI_O_DOWN(1<<4))) == 0 && |
| 2503 | (mstime() - master->info_refresh) < SENTINEL_INFO_PERIOD10000*2; |
| 2504 | } |
| 2505 | |
| 2506 | /* Process the INFO output from masters. */ |
| 2507 | void sentinelRefreshInstanceInfo(sentinelRedisInstance *ri, const char *info) { |
| 2508 | sds *lines; |
| 2509 | int numlines, j; |
| 2510 | int role = 0; |
| 2511 | |
| 2512 | /* cache full INFO output for instance */ |
| 2513 | sdsfree(ri->info); |
| 2514 | ri->info = sdsnew(info); |
| 2515 | |
| 2516 | /* The following fields must be reset to a given value in the case they |
| 2517 | * are not found at all in the INFO output. */ |
| 2518 | ri->master_link_down_time = 0; |
| 2519 | |
| 2520 | /* Process line by line. */ |
| 2521 | lines = sdssplitlen(info,strlen(info),"\r\n",2,&numlines); |
| 2522 | for (j = 0; j < numlines; j++) { |
| 2523 | sentinelRedisInstance *slave; |
| 2524 | sds l = lines[j]; |
| 2525 | |
| 2526 | /* run_id:<40 hex chars>*/ |
| 2527 | if (sdslen(l) >= 47 && !memcmp(l,"run_id:",7)) { |
| 2528 | if (ri->runid == NULL((void*)0)) { |
| 2529 | ri->runid = sdsnewlen(l+7,40); |
| 2530 | } else { |
| 2531 | if (strncmp(ri->runid,l+7,40) != 0) { |
| 2532 | sentinelEvent(LL_NOTICE2,"+reboot",ri,"%@"); |
| 2533 | sdsfree(ri->runid); |
| 2534 | ri->runid = sdsnewlen(l+7,40); |
| 2535 | } |
| 2536 | } |
| 2537 | } |
| 2538 | |
| 2539 | /* old versions: slave0:<ip>,<port>,<state> |
| 2540 | * new versions: slave0:ip=127.0.0.1,port=9999,... */ |
| 2541 | if ((ri->flags & SRI_MASTER(1<<0)) && |
| 2542 | sdslen(l) >= 7 && |
| 2543 | !memcmp(l,"slave",5) && isdigit(l[5])((*__ctype_b_loc ())[(int) ((l[5]))] & (unsigned short int ) _ISdigit)) |
| 2544 | { |
| 2545 | char *ip, *port, *end; |
| 2546 | |
| 2547 | if (strstr(l,"ip=") == NULL((void*)0)) { |
| 2548 | /* Old format. */ |
| 2549 | ip = strchr(l,':'); if (!ip) continue; |
| 2550 | ip++; /* Now ip points to start of ip address. */ |
| 2551 | port = strchr(ip,','); if (!port) continue; |
| 2552 | *port = '\0'; /* nul term for easy access. */ |
| 2553 | port++; /* Now port points to start of port number. */ |
| 2554 | end = strchr(port,','); if (!end) continue; |
| 2555 | *end = '\0'; /* nul term for easy access. */ |
| 2556 | } else { |
| 2557 | /* New format. */ |
| 2558 | ip = strstr(l,"ip="); if (!ip) continue; |
| 2559 | ip += 3; /* Now ip points to start of ip address. */ |
| 2560 | port = strstr(l,"port="); if (!port) continue; |
| 2561 | port += 5; /* Now port points to start of port number. */ |
| 2562 | /* Nul term both fields for easy access. */ |
| 2563 | end = strchr(ip,','); if (end) *end = '\0'; |
| 2564 | end = strchr(port,','); if (end) *end = '\0'; |
| 2565 | } |
| 2566 | |
| 2567 | /* Check if we already have this slave into our table, |
| 2568 | * otherwise add it. */ |
| 2569 | if (sentinelRedisInstanceLookupSlave(ri,ip,atoi(port)) == NULL((void*)0)) { |
| 2570 | if ((slave = createSentinelRedisInstance(NULL((void*)0),SRI_SLAVE(1<<1),ip, |
| 2571 | atoi(port), ri->quorum, ri)) != NULL((void*)0)) |
| 2572 | { |
| 2573 | sentinelEvent(LL_NOTICE2,"+slave",slave,"%@"); |
| 2574 | sentinelFlushConfig(); |
| 2575 | } |
| 2576 | } |
| 2577 | } |
| 2578 | |
| 2579 | /* master_link_down_since_seconds:<seconds> */ |
| 2580 | if (sdslen(l) >= 32 && |
| 2581 | !memcmp(l,"master_link_down_since_seconds",30)) |
| 2582 | { |
| 2583 | ri->master_link_down_time = strtoll(l+31,NULL((void*)0),10)*1000; |
| 2584 | } |
| 2585 | |
| 2586 | /* role:<role> */ |
| 2587 | if (sdslen(l) >= 11 && !memcmp(l,"role:master",11)) role = SRI_MASTER(1<<0); |
| 2588 | else if (sdslen(l) >= 10 && !memcmp(l,"role:slave",10)) role = SRI_SLAVE(1<<1); |
| 2589 | |
| 2590 | if (role == SRI_SLAVE(1<<1)) { |
| 2591 | /* master_host:<host> */ |
| 2592 | if (sdslen(l) >= 12 && !memcmp(l,"master_host:",12)) { |
| 2593 | if (ri->slave_master_host == NULL((void*)0) || |
| 2594 | strcasecmp(l+12,ri->slave_master_host)) |
| 2595 | { |
| 2596 | sdsfree(ri->slave_master_host); |
| 2597 | ri->slave_master_host = sdsnew(l+12); |
| 2598 | ri->slave_conf_change_time = mstime(); |
| 2599 | } |
| 2600 | } |
| 2601 | |
| 2602 | /* master_port:<port> */ |
| 2603 | if (sdslen(l) >= 12 && !memcmp(l,"master_port:",12)) { |
| 2604 | int slave_master_port = atoi(l+12); |
| 2605 | |
| 2606 | if (ri->slave_master_port != slave_master_port) { |
| 2607 | ri->slave_master_port = slave_master_port; |
| 2608 | ri->slave_conf_change_time = mstime(); |
| 2609 | } |
| 2610 | } |
| 2611 | |
| 2612 | /* master_link_status:<status> */ |
| 2613 | if (sdslen(l) >= 19 && !memcmp(l,"master_link_status:",19)) { |
| 2614 | ri->slave_master_link_status = |
| 2615 | (strcasecmp(l+19,"up") == 0) ? |
| 2616 | SENTINEL_MASTER_LINK_STATUS_UP0 : |
| 2617 | SENTINEL_MASTER_LINK_STATUS_DOWN1; |
| 2618 | } |
| 2619 | |
| 2620 | /* slave_priority:<priority> */ |
| 2621 | if (sdslen(l) >= 15 && !memcmp(l,"slave_priority:",15)) |
| 2622 | ri->slave_priority = atoi(l+15); |
| 2623 | |
| 2624 | /* slave_repl_offset:<offset> */ |
| 2625 | if (sdslen(l) >= 18 && !memcmp(l,"slave_repl_offset:",18)) |
| 2626 | ri->slave_repl_offset = strtoull(l+18,NULL((void*)0),10); |
| 2627 | } |
| 2628 | } |
| 2629 | ri->info_refresh = mstime(); |
| 2630 | sdsfreesplitres(lines,numlines); |
| 2631 | |
| 2632 | /* ---------------------------- Acting half ----------------------------- |
| 2633 | * Some things will not happen if sentinel.tilt is true, but some will |
| 2634 | * still be processed. */ |
| 2635 | |
| 2636 | /* Remember when the role changed. */ |
| 2637 | if (role != ri->role_reported) { |
| 2638 | ri->role_reported_time = mstime(); |
| 2639 | ri->role_reported = role; |
| 2640 | if (role == SRI_SLAVE(1<<1)) ri->slave_conf_change_time = mstime(); |
| 2641 | /* Log the event with +role-change if the new role is coherent or |
| 2642 | * with -role-change if there is a mismatch with the current config. */ |
| 2643 | sentinelEvent(LL_VERBOSE1, |
| 2644 | ((ri->flags & (SRI_MASTER(1<<0)|SRI_SLAVE(1<<1))) == role) ? |
| 2645 | "+role-change" : "-role-change", |
| 2646 | ri, "%@ new reported role is %s", |
| 2647 | role == SRI_MASTER(1<<0) ? "master" : "slave", |
| 2648 | ri->flags & SRI_MASTER(1<<0) ? "master" : "slave"); |
| 2649 | } |
| 2650 | |
| 2651 | /* None of the following conditions are processed when in tilt mode, so |
| 2652 | * return asap. */ |
| 2653 | if (sentinel.tilt) return; |
| 2654 | |
| 2655 | /* Handle master -> slave role switch. */ |
| 2656 | if ((ri->flags & SRI_MASTER(1<<0)) && role == SRI_SLAVE(1<<1)) { |
| 2657 | /* Nothing to do, but masters claiming to be slaves are |
| 2658 | * considered to be unreachable by Sentinel, so eventually |
| 2659 | * a failover will be triggered. */ |
| 2660 | } |
| 2661 | |
| 2662 | /* Handle slave -> master role switch. */ |
| 2663 | if ((ri->flags & SRI_SLAVE(1<<1)) && role == SRI_MASTER(1<<0)) { |
| 2664 | /* If this is a promoted slave we can change state to the |
| 2665 | * failover state machine. */ |
| 2666 | if ((ri->flags & SRI_PROMOTED(1<<7)) && |
| 2667 | (ri->master->flags & SRI_FAILOVER_IN_PROGRESS(1<<6)) && |
| 2668 | (ri->master->failover_state == |
| 2669 | SENTINEL_FAILOVER_STATE_WAIT_PROMOTION4)) |
| 2670 | { |
| 2671 | /* Now that we are sure the slave was reconfigured as a master |
| 2672 | * set the master configuration epoch to the epoch we won the |
| 2673 | * election to perform this failover. This will force the other |
| 2674 | * Sentinels to update their config (assuming there is not |
| 2675 | * a newer one already available). */ |
| 2676 | ri->master->config_epoch = ri->master->failover_epoch; |
| 2677 | ri->master->failover_state = SENTINEL_FAILOVER_STATE_RECONF_SLAVES5; |
| 2678 | ri->master->failover_state_change_time = mstime(); |
| 2679 | sentinelFlushConfig(); |
| 2680 | sentinelEvent(LL_WARNING3,"+promoted-slave",ri,"%@"); |
| 2681 | if (sentinel.simfailure_flags & |
| 2682 | SENTINEL_SIMFAILURE_CRASH_AFTER_PROMOTION(1<<1)) |
| 2683 | sentinelSimFailureCrash(); |
| 2684 | sentinelEvent(LL_WARNING3,"+failover-state-reconf-slaves", |
| 2685 | ri->master,"%@"); |
| 2686 | sentinelCallClientReconfScript(ri->master,SENTINEL_LEADER(1<<17), |
| 2687 | "start",ri->master->addr,ri->addr); |
| 2688 | sentinelForceHelloUpdateForMaster(ri->master); |
| 2689 | } else { |
| 2690 | /* A slave turned into a master. We want to force our view and |
| 2691 | * reconfigure as slave. Wait some time after the change before |
| 2692 | * going forward, to receive new configs if any. */ |
| 2693 | mstime_t wait_time = SENTINEL_PUBLISH_PERIOD2000*4; |
| 2694 | |
| 2695 | if (!(ri->flags & SRI_PROMOTED(1<<7)) && |
| 2696 | sentinelMasterLooksSane(ri->master) && |
| 2697 | sentinelRedisInstanceNoDownFor(ri,wait_time) && |
| 2698 | mstime() - ri->role_reported_time > wait_time) |
| 2699 | { |
| 2700 | int retval = sentinelSendSlaveOf(ri,ri->master->addr); |
| 2701 | if (retval == C_OK0) |
| 2702 | sentinelEvent(LL_NOTICE2,"+convert-to-slave",ri,"%@"); |
| 2703 | } |
| 2704 | } |
| 2705 | } |
| 2706 | |
| 2707 | /* Handle slaves replicating to a different master address. */ |
| 2708 | if ((ri->flags & SRI_SLAVE(1<<1)) && |
| 2709 | role == SRI_SLAVE(1<<1) && |
| 2710 | (ri->slave_master_port != ri->master->addr->port || |
| 2711 | !sentinelAddrEqualsHostname(ri->master->addr, ri->slave_master_host))) |
| 2712 | { |
| 2713 | mstime_t wait_time = ri->master->failover_timeout; |
| 2714 | |
| 2715 | /* Make sure the master is sane before reconfiguring this instance |
| 2716 | * into a slave. */ |
| 2717 | if (sentinelMasterLooksSane(ri->master) && |
| 2718 | sentinelRedisInstanceNoDownFor(ri,wait_time) && |
| 2719 | mstime() - ri->slave_conf_change_time > wait_time) |
| 2720 | { |
| 2721 | int retval = sentinelSendSlaveOf(ri,ri->master->addr); |
| 2722 | if (retval == C_OK0) |
| 2723 | sentinelEvent(LL_NOTICE2,"+fix-slave-config",ri,"%@"); |
| 2724 | } |
| 2725 | } |
| 2726 | |
| 2727 | /* Detect if the slave that is in the process of being reconfigured |
| 2728 | * changed state. */ |
| 2729 | if ((ri->flags & SRI_SLAVE(1<<1)) && role == SRI_SLAVE(1<<1) && |
| 2730 | (ri->flags & (SRI_RECONF_SENT(1<<8)|SRI_RECONF_INPROG(1<<9)))) |
| 2731 | { |
| 2732 | /* SRI_RECONF_SENT -> SRI_RECONF_INPROG. */ |
| 2733 | if ((ri->flags & SRI_RECONF_SENT(1<<8)) && |
| 2734 | ri->slave_master_host && |
| 2735 | sentinelAddrEqualsHostname(ri->master->promoted_slave->addr, |
| 2736 | ri->slave_master_host) && |
| 2737 | ri->slave_master_port == ri->master->promoted_slave->addr->port) |
| 2738 | { |
| 2739 | ri->flags &= ~SRI_RECONF_SENT(1<<8); |
| 2740 | ri->flags |= SRI_RECONF_INPROG(1<<9); |
| 2741 | sentinelEvent(LL_NOTICE2,"+slave-reconf-inprog",ri,"%@"); |
| 2742 | } |
| 2743 | |
| 2744 | /* SRI_RECONF_INPROG -> SRI_RECONF_DONE */ |
| 2745 | if ((ri->flags & SRI_RECONF_INPROG(1<<9)) && |
| 2746 | ri->slave_master_link_status == SENTINEL_MASTER_LINK_STATUS_UP0) |
| 2747 | { |
| 2748 | ri->flags &= ~SRI_RECONF_INPROG(1<<9); |
| 2749 | ri->flags |= SRI_RECONF_DONE(1<<10); |
| 2750 | sentinelEvent(LL_NOTICE2,"+slave-reconf-done",ri,"%@"); |
| 2751 | } |
| 2752 | } |
| 2753 | } |
| 2754 | |
| 2755 | void sentinelInfoReplyCallback(redisAsyncContext *c, void *reply, void *privdata) { |
| 2756 | sentinelRedisInstance *ri = privdata; |
| 2757 | instanceLink *link = c->data; |
| 2758 | redisReply *r; |
| 2759 | |
| 2760 | if (!reply || !link) return; |
| 2761 | link->pending_commands--; |
| 2762 | r = reply; |
| 2763 | |
| 2764 | if (r->type == REDIS_REPLY_STRING1) |
| 2765 | sentinelRefreshInstanceInfo(ri,r->str); |
| 2766 | } |
| 2767 | |
| 2768 | /* Just discard the reply. We use this when we are not monitoring the return |
| 2769 | * value of the command but its effects directly. */ |
| 2770 | void sentinelDiscardReplyCallback(redisAsyncContext *c, void *reply, void *privdata) { |
| 2771 | instanceLink *link = c->data; |
| 2772 | UNUSED(reply)((void) reply); |
| 2773 | UNUSED(privdata)((void) privdata); |
| 2774 | |
| 2775 | if (link) link->pending_commands--; |
| 2776 | } |
| 2777 | |
| 2778 | void sentinelPingReplyCallback(redisAsyncContext *c, void *reply, void *privdata) { |
| 2779 | sentinelRedisInstance *ri = privdata; |
| 2780 | instanceLink *link = c->data; |
| 2781 | redisReply *r; |
| 2782 | |
| 2783 | if (!reply || !link) return; |
| 2784 | link->pending_commands--; |
| 2785 | r = reply; |
| 2786 | |
| 2787 | if (r->type == REDIS_REPLY_STATUS5 || |
| 2788 | r->type == REDIS_REPLY_ERROR6) { |
| 2789 | /* Update the "instance available" field only if this is an |
| 2790 | * acceptable reply. */ |
| 2791 | if (strncmp(r->str,"PONG",4) == 0 || |
| 2792 | strncmp(r->str,"LOADING",7) == 0 || |
| 2793 | strncmp(r->str,"MASTERDOWN",10) == 0) |
| 2794 | { |
| 2795 | link->last_avail_time = mstime(); |
| 2796 | link->act_ping_time = 0; /* Flag the pong as received. */ |
| 2797 | } else { |
| 2798 | /* Send a SCRIPT KILL command if the instance appears to be |
| 2799 | * down because of a busy script. */ |
| 2800 | if (strncmp(r->str,"BUSY",4) == 0 && |
| 2801 | (ri->flags & SRI_S_DOWN(1<<3)) && |
| 2802 | !(ri->flags & SRI_SCRIPT_KILL_SENT(1<<12))) |
| 2803 | { |
| 2804 | if (redisAsyncCommand(ri->link->cc, |
| 2805 | sentinelDiscardReplyCallback, ri, |
| 2806 | "%s KILL", |
| 2807 | sentinelInstanceMapCommand(ri,"SCRIPT")) == C_OK0) |
| 2808 | { |
| 2809 | ri->link->pending_commands++; |
| 2810 | } |
| 2811 | ri->flags |= SRI_SCRIPT_KILL_SENT(1<<12); |
| 2812 | } |
| 2813 | } |
| 2814 | } |
| 2815 | link->last_pong_time = mstime(); |
| 2816 | } |
| 2817 | |
| 2818 | /* This is called when we get the reply about the PUBLISH command we send |
| 2819 | * to the master to advertise this sentinel. */ |
| 2820 | void sentinelPublishReplyCallback(redisAsyncContext *c, void *reply, void *privdata) { |
| 2821 | sentinelRedisInstance *ri = privdata; |
| 2822 | instanceLink *link = c->data; |
| 2823 | redisReply *r; |
| 2824 | |
| 2825 | if (!reply || !link) return; |
| 2826 | link->pending_commands--; |
| 2827 | r = reply; |
| 2828 | |
| 2829 | /* Only update pub_time if we actually published our message. Otherwise |
| 2830 | * we'll retry again in 100 milliseconds. */ |
| 2831 | if (r->type != REDIS_REPLY_ERROR6) |
| 2832 | ri->last_pub_time = mstime(); |
| 2833 | } |
| 2834 | |
| 2835 | /* Process a hello message received via Pub/Sub in master or slave instance, |
| 2836 | * or sent directly to this sentinel via the (fake) PUBLISH command of Sentinel. |
| 2837 | * |
| 2838 | * If the master name specified in the message is not known, the message is |
| 2839 | * discarded. */ |
| 2840 | void sentinelProcessHelloMessage(char *hello, int hello_len) { |
| 2841 | /* Format is composed of 8 tokens: |
| 2842 | * 0=ip,1=port,2=runid,3=current_epoch,4=master_name, |
| 2843 | * 5=master_ip,6=master_port,7=master_config_epoch. */ |
| 2844 | int numtokens, port, removed, master_port; |
| 2845 | uint64_t current_epoch, master_config_epoch; |
| 2846 | char **token = sdssplitlen(hello, hello_len, ",", 1, &numtokens); |
| 2847 | sentinelRedisInstance *si, *master; |
| 2848 | |
| 2849 | if (numtokens == 8) { |
| 2850 | /* Obtain a reference to the master this hello message is about */ |
| 2851 | master = sentinelGetMasterByName(token[4]); |
| 2852 | if (!master) goto cleanup; /* Unknown master, skip the message. */ |
| 2853 | |
| 2854 | /* First, try to see if we already have this sentinel. */ |
| 2855 | port = atoi(token[1]); |
| 2856 | master_port = atoi(token[6]); |
| 2857 | si = getSentinelRedisInstanceByAddrAndRunID( |
| 2858 | master->sentinels,token[0],port,token[2]); |
| 2859 | current_epoch = strtoull(token[3],NULL((void*)0),10); |
| 2860 | master_config_epoch = strtoull(token[7],NULL((void*)0),10); |
| 2861 | |
| 2862 | if (!si) { |
| 2863 | /* If not, remove all the sentinels that have the same runid |
| 2864 | * because there was an address change, and add the same Sentinel |
| 2865 | * with the new address back. */ |
| 2866 | removed = removeMatchingSentinelFromMaster(master,token[2]); |
| 2867 | if (removed) { |
| 2868 | sentinelEvent(LL_NOTICE2,"+sentinel-address-switch",master, |
| 2869 | "%@ ip %s port %d for %s", token[0],port,token[2]); |
| 2870 | } else { |
| 2871 | /* Check if there is another Sentinel with the same address this |
| 2872 | * new one is reporting. What we do if this happens is to set its |
| 2873 | * port to 0, to signal the address is invalid. We'll update it |
| 2874 | * later if we get an HELLO message. */ |
| 2875 | sentinelRedisInstance *other = |
| 2876 | getSentinelRedisInstanceByAddrAndRunID( |
| 2877 | master->sentinels, token[0],port,NULL((void*)0)); |
| 2878 | if (other) { |
| 2879 | sentinelEvent(LL_NOTICE2,"+sentinel-invalid-addr",other,"%@"); |
| 2880 | other->addr->port = 0; /* It means: invalid address. */ |
| 2881 | sentinelUpdateSentinelAddressInAllMasters(other); |
| 2882 | } |
| 2883 | } |
| 2884 | |
| 2885 | /* Add the new sentinel. */ |
| 2886 | si = createSentinelRedisInstance(token[2],SRI_SENTINEL(1<<2), |
| 2887 | token[0],port,master->quorum,master); |
| 2888 | |
| 2889 | if (si) { |
| 2890 | if (!removed) sentinelEvent(LL_NOTICE2,"+sentinel",si,"%@"); |
| 2891 | /* The runid is NULL after a new instance creation and |
| 2892 | * for Sentinels we don't have a later chance to fill it, |
| 2893 | * so do it now. */ |
| 2894 | si->runid = sdsnew(token[2]); |
| 2895 | sentinelTryConnectionSharing(si); |
| 2896 | if (removed) sentinelUpdateSentinelAddressInAllMasters(si); |
| 2897 | sentinelFlushConfig(); |
| 2898 | } |
| 2899 | } |
| 2900 | |
| 2901 | /* Update local current_epoch if received current_epoch is greater.*/ |
| 2902 | if (current_epoch > sentinel.current_epoch) { |
| 2903 | sentinel.current_epoch = current_epoch; |
| 2904 | sentinelFlushConfig(); |
| 2905 | sentinelEvent(LL_WARNING3,"+new-epoch",master,"%llu", |
| 2906 | (unsigned long long) sentinel.current_epoch); |
| 2907 | } |
| 2908 | |
| 2909 | /* Update master info if received configuration is newer. */ |
| 2910 | if (si && master->config_epoch < master_config_epoch) { |
| 2911 | master->config_epoch = master_config_epoch; |
| 2912 | if (master_port != master->addr->port || |
| 2913 | !sentinelAddrEqualsHostname(master->addr, token[5])) |
| 2914 | { |
| 2915 | sentinelAddr *old_addr; |
| 2916 | |
| 2917 | sentinelEvent(LL_WARNING3,"+config-update-from",si,"%@"); |
| 2918 | sentinelEvent(LL_WARNING3,"+switch-master", |
| 2919 | master,"%s %s %d %s %d", |
| 2920 | master->name, |
| 2921 | announceSentinelAddr(master->addr), master->addr->port, |
| 2922 | token[5], master_port); |
| 2923 | |
| 2924 | old_addr = dupSentinelAddr(master->addr); |
| 2925 | sentinelResetMasterAndChangeAddress(master, token[5], master_port); |
| 2926 | sentinelCallClientReconfScript(master, |
| 2927 | SENTINEL_OBSERVER(1<<18),"start", |
| 2928 | old_addr,master->addr); |
| 2929 | releaseSentinelAddr(old_addr); |
| 2930 | } |
| 2931 | } |
| 2932 | |
| 2933 | /* Update the state of the Sentinel. */ |
| 2934 | if (si) si->last_hello_time = mstime(); |
| 2935 | } |
| 2936 | |
| 2937 | cleanup: |
| 2938 | sdsfreesplitres(token,numtokens); |
| 2939 | } |
| 2940 | |
| 2941 | |
| 2942 | /* This is our Pub/Sub callback for the Hello channel. It's useful in order |
| 2943 | * to discover other sentinels attached at the same master. */ |
| 2944 | void sentinelReceiveHelloMessages(redisAsyncContext *c, void *reply, void *privdata) { |
| 2945 | sentinelRedisInstance *ri = privdata; |
| 2946 | redisReply *r; |
| 2947 | UNUSED(c)((void) c); |
| 2948 | |
| 2949 | if (!reply || !ri) return; |
| 2950 | r = reply; |
| 2951 | |
| 2952 | /* Update the last activity in the pubsub channel. Note that since we |
| 2953 | * receive our messages as well this timestamp can be used to detect |
| 2954 | * if the link is probably disconnected even if it seems otherwise. */ |
| 2955 | ri->link->pc_last_activity = mstime(); |
| 2956 | |
| 2957 | /* Sanity check in the reply we expect, so that the code that follows |
| 2958 | * can avoid to check for details. */ |
| 2959 | if (r->type != REDIS_REPLY_ARRAY2 || |
| 2960 | r->elements != 3 || |
| 2961 | r->element[0]->type != REDIS_REPLY_STRING1 || |
| 2962 | r->element[1]->type != REDIS_REPLY_STRING1 || |
| 2963 | r->element[2]->type != REDIS_REPLY_STRING1 || |
| 2964 | strcmp(r->element[0]->str,"message") != 0) return; |
| 2965 | |
| 2966 | /* We are not interested in meeting ourselves */ |
| 2967 | if (strstr(r->element[2]->str,sentinel.myid) != NULL((void*)0)) return; |
| 2968 | |
| 2969 | sentinelProcessHelloMessage(r->element[2]->str, r->element[2]->len); |
| 2970 | } |
| 2971 | |
| 2972 | /* Send a "Hello" message via Pub/Sub to the specified 'ri' Redis |
| 2973 | * instance in order to broadcast the current configuration for this |
| 2974 | * master, and to advertise the existence of this Sentinel at the same time. |
| 2975 | * |
| 2976 | * The message has the following format: |
| 2977 | * |
| 2978 | * sentinel_ip,sentinel_port,sentinel_runid,current_epoch, |
| 2979 | * master_name,master_ip,master_port,master_config_epoch. |
| 2980 | * |
| 2981 | * Returns C_OK if the PUBLISH was queued correctly, otherwise |
| 2982 | * C_ERR is returned. */ |
| 2983 | int sentinelSendHello(sentinelRedisInstance *ri) { |
| 2984 | char ip[NET_IP_STR_LEN46]; |
| 2985 | char payload[NET_IP_STR_LEN46+1024]; |
| 2986 | int retval; |
| 2987 | char *announce_ip; |
| 2988 | int announce_port; |
| 2989 | sentinelRedisInstance *master = (ri->flags & SRI_MASTER(1<<0)) ? ri : ri->master; |
| 2990 | sentinelAddr *master_addr = sentinelGetCurrentMasterAddress(master); |
| 2991 | |
| 2992 | if (ri->link->disconnected) return C_ERR-1; |
| 2993 | |
| 2994 | /* Use the specified announce address if specified, otherwise try to |
| 2995 | * obtain our own IP address. */ |
| 2996 | if (sentinel.announce_ip) { |
| 2997 | announce_ip = sentinel.announce_ip; |
| 2998 | } else { |
| 2999 | if (anetFdToString(ri->link->cc->c.fd,ip,sizeof(ip),NULL((void*)0),FD_TO_SOCK_NAME1) == -1) |
| 3000 | return C_ERR-1; |
| 3001 | announce_ip = ip; |
| 3002 | } |
| 3003 | if (sentinel.announce_port) announce_port = sentinel.announce_port; |
| 3004 | else if (server.tls_replication && server.tls_port) announce_port = server.tls_port; |
| 3005 | else announce_port = server.port; |
| 3006 | |
| 3007 | /* Format and send the Hello message. */ |
| 3008 | snprintf(payload,sizeof(payload), |
| 3009 | "%s,%d,%s,%llu," /* Info about this sentinel. */ |
| 3010 | "%s,%s,%d,%llu", /* Info about current master. */ |
| 3011 | announce_ip, announce_port, sentinel.myid, |
| 3012 | (unsigned long long) sentinel.current_epoch, |
| 3013 | /* --- */ |
| 3014 | master->name,announceSentinelAddr(master_addr),master_addr->port, |
| 3015 | (unsigned long long) master->config_epoch); |
| 3016 | retval = redisAsyncCommand(ri->link->cc, |
| 3017 | sentinelPublishReplyCallback, ri, "%s %s %s", |
| 3018 | sentinelInstanceMapCommand(ri,"PUBLISH"), |
| 3019 | SENTINEL_HELLO_CHANNEL"__sentinel__:hello",payload); |
| 3020 | if (retval != C_OK0) return C_ERR-1; |
| 3021 | ri->link->pending_commands++; |
| 3022 | return C_OK0; |
| 3023 | } |
| 3024 | |
| 3025 | /* Reset last_pub_time in all the instances in the specified dictionary |
| 3026 | * in order to force the delivery of a Hello update ASAP. */ |
| 3027 | void sentinelForceHelloUpdateDictOfRedisInstances(dict *instances) { |
| 3028 | dictIterator *di; |
| 3029 | dictEntry *de; |
| 3030 | |
| 3031 | di = dictGetSafeIterator(instances); |
| 3032 | while((de = dictNext(di)) != NULL((void*)0)) { |
| 3033 | sentinelRedisInstance *ri = dictGetVal(de)((de)->v.val); |
| 3034 | if (ri->last_pub_time >= (SENTINEL_PUBLISH_PERIOD2000+1)) |
| 3035 | ri->last_pub_time -= (SENTINEL_PUBLISH_PERIOD2000+1); |
| 3036 | } |
| 3037 | dictReleaseIterator(di); |
| 3038 | } |
| 3039 | |
| 3040 | /* This function forces the delivery of a "Hello" message (see |
| 3041 | * sentinelSendHello() top comment for further information) to all the Redis |
| 3042 | * and Sentinel instances related to the specified 'master'. |
| 3043 | * |
| 3044 | * It is technically not needed since we send an update to every instance |
| 3045 | * with a period of SENTINEL_PUBLISH_PERIOD milliseconds, however when a |
| 3046 | * Sentinel upgrades a configuration it is a good idea to deliver an update |
| 3047 | * to the other Sentinels ASAP. */ |
| 3048 | int sentinelForceHelloUpdateForMaster(sentinelRedisInstance *master) { |
| 3049 | if (!(master->flags & SRI_MASTER(1<<0))) return C_ERR-1; |
| 3050 | if (master->last_pub_time >= (SENTINEL_PUBLISH_PERIOD2000+1)) |
| 3051 | master->last_pub_time -= (SENTINEL_PUBLISH_PERIOD2000+1); |
| 3052 | sentinelForceHelloUpdateDictOfRedisInstances(master->sentinels); |
| 3053 | sentinelForceHelloUpdateDictOfRedisInstances(master->slaves); |
| 3054 | return C_OK0; |
| 3055 | } |
| 3056 | |
| 3057 | /* Send a PING to the specified instance and refresh the act_ping_time |
| 3058 | * if it is zero (that is, if we received a pong for the previous ping). |
| 3059 | * |
| 3060 | * On error zero is returned, and we can't consider the PING command |
| 3061 | * queued in the connection. */ |
| 3062 | int sentinelSendPing(sentinelRedisInstance *ri) { |
| 3063 | int retval = redisAsyncCommand(ri->link->cc, |
| 3064 | sentinelPingReplyCallback, ri, "%s", |
| 3065 | sentinelInstanceMapCommand(ri,"PING")); |
| 3066 | if (retval == C_OK0) { |
| 3067 | ri->link->pending_commands++; |
| 3068 | ri->link->last_ping_time = mstime(); |
| 3069 | /* We update the active ping time only if we received the pong for |
| 3070 | * the previous ping, otherwise we are technically waiting since the |
| 3071 | * first ping that did not receive a reply. */ |
| 3072 | if (ri->link->act_ping_time == 0) |
| 3073 | ri->link->act_ping_time = ri->link->last_ping_time; |
| 3074 | return 1; |
| 3075 | } else { |
| 3076 | return 0; |
| 3077 | } |
| 3078 | } |
| 3079 | |
| 3080 | /* Send periodic PING, INFO, and PUBLISH to the Hello channel to |
| 3081 | * the specified master or slave instance. */ |
| 3082 | void sentinelSendPeriodicCommands(sentinelRedisInstance *ri) { |
| 3083 | mstime_t now = mstime(); |
| 3084 | mstime_t info_period, ping_period; |
| 3085 | int retval; |
| 3086 | |
| 3087 | /* Return ASAP if we have already a PING or INFO already pending, or |
| 3088 | * in the case the instance is not properly connected. */ |
| 3089 | if (ri->link->disconnected) return; |
| 3090 | |
| 3091 | /* For INFO, PING, PUBLISH that are not critical commands to send we |
| 3092 | * also have a limit of SENTINEL_MAX_PENDING_COMMANDS. We don't |
| 3093 | * want to use a lot of memory just because a link is not working |
| 3094 | * properly (note that anyway there is a redundant protection about this, |
| 3095 | * that is, the link will be disconnected and reconnected if a long |
| 3096 | * timeout condition is detected. */ |
| 3097 | if (ri->link->pending_commands >= |
| 3098 | SENTINEL_MAX_PENDING_COMMANDS100 * ri->link->refcount) return; |
| 3099 | |
| 3100 | /* If this is a slave of a master in O_DOWN condition we start sending |
| 3101 | * it INFO every second, instead of the usual SENTINEL_INFO_PERIOD |
| 3102 | * period. In this state we want to closely monitor slaves in case they |
| 3103 | * are turned into masters by another Sentinel, or by the sysadmin. |
| 3104 | * |
| 3105 | * Similarly we monitor the INFO output more often if the slave reports |
| 3106 | * to be disconnected from the master, so that we can have a fresh |
| 3107 | * disconnection time figure. */ |
| 3108 | if ((ri->flags & SRI_SLAVE(1<<1)) && |
| 3109 | ((ri->master->flags & (SRI_O_DOWN(1<<4)|SRI_FAILOVER_IN_PROGRESS(1<<6))) || |
| 3110 | (ri->master_link_down_time != 0))) |
| 3111 | { |
| 3112 | info_period = 1000; |
| 3113 | } else { |
| 3114 | info_period = SENTINEL_INFO_PERIOD10000; |
| 3115 | } |
| 3116 | |
| 3117 | /* We ping instances every time the last received pong is older than |
| 3118 | * the configured 'down-after-milliseconds' time, but every second |
| 3119 | * anyway if 'down-after-milliseconds' is greater than 1 second. */ |
| 3120 | ping_period = ri->down_after_period; |
| 3121 | if (ping_period > SENTINEL_PING_PERIOD1000) ping_period = SENTINEL_PING_PERIOD1000; |
| 3122 | |
| 3123 | /* Send INFO to masters and slaves, not sentinels. */ |
| 3124 | if ((ri->flags & SRI_SENTINEL(1<<2)) == 0 && |
| 3125 | (ri->info_refresh == 0 || |
| 3126 | (now - ri->info_refresh) > info_period)) |
| 3127 | { |
| 3128 | retval = redisAsyncCommand(ri->link->cc, |
| 3129 | sentinelInfoReplyCallback, ri, "%s", |
| 3130 | sentinelInstanceMapCommand(ri,"INFO")); |
| 3131 | if (retval == C_OK0) ri->link->pending_commands++; |
| 3132 | } |
| 3133 | |
| 3134 | /* Send PING to all the three kinds of instances. */ |
| 3135 | if ((now - ri->link->last_pong_time) > ping_period && |
| 3136 | (now - ri->link->last_ping_time) > ping_period/2) { |
| 3137 | sentinelSendPing(ri); |
| 3138 | } |
| 3139 | |
| 3140 | /* PUBLISH hello messages to all the three kinds of instances. */ |
| 3141 | if ((now - ri->last_pub_time) > SENTINEL_PUBLISH_PERIOD2000) { |
| 3142 | sentinelSendHello(ri); |
| 3143 | } |
| 3144 | } |
| 3145 | |
| 3146 | /* =========================== SENTINEL command ============================= */ |
| 3147 | |
| 3148 | /* SENTINEL CONFIG SET <option> */ |
| 3149 | void sentinelConfigSetCommand(client *c) { |
| 3150 | robj *o = c->argv[3]; |
| 3151 | robj *val = c->argv[4]; |
| 3152 | long long numval; |
| 3153 | int drop_conns = 0; |
| 3154 | |
| 3155 | if (!strcasecmp(o->ptr, "resolve-hostnames")) { |
| 3156 | if ((numval = yesnotoi(val->ptr)) == -1) goto badfmt; |
| 3157 | sentinel.resolve_hostnames = numval; |
| 3158 | } else if (!strcasecmp(o->ptr, "announce-hostnames")) { |
| 3159 | if ((numval = yesnotoi(val->ptr)) == -1) goto badfmt; |
| 3160 | sentinel.announce_hostnames = numval; |
| 3161 | } else if (!strcasecmp(o->ptr, "announce-ip")) { |
| 3162 | if (sentinel.announce_ip) sdsfree(sentinel.announce_ip); |
| 3163 | sentinel.announce_ip = sdsnew(val->ptr); |
| 3164 | } else if (!strcasecmp(o->ptr, "announce-port")) { |
| 3165 | if (getLongLongFromObject(val, &numval) == C_ERR-1 || |
| 3166 | numval < 0 || numval > 65535) |
| 3167 | goto badfmt; |
| 3168 | sentinel.announce_port = numval; |
| 3169 | } else if (!strcasecmp(o->ptr, "sentinel-user")) { |
| 3170 | sdsfree(sentinel.sentinel_auth_user); |
| 3171 | sentinel.sentinel_auth_user = sdsnew(val->ptr); |
| 3172 | drop_conns = 1; |
| 3173 | } else if (!strcasecmp(o->ptr, "sentinel-pass")) { |
| 3174 | sdsfree(sentinel.sentinel_auth_pass); |
| 3175 | sentinel.sentinel_auth_pass = sdsnew(val->ptr); |
| 3176 | drop_conns = 1; |
| 3177 | } else { |
| 3178 | addReplyErrorFormat(c, "Invalid argument '%s' to SENTINEL CONFIG SET", |
| 3179 | (char *) o->ptr); |
| 3180 | return; |
| 3181 | } |
| 3182 | |
| 3183 | sentinelFlushConfig(); |
| 3184 | addReply(c, shared.ok); |
| 3185 | |
| 3186 | /* Drop Sentinel connections to initiate a reconnect if needed. */ |
| 3187 | if (drop_conns) |
| 3188 | sentinelDropConnections(); |
| 3189 | |
| 3190 | return; |
| 3191 | |
| 3192 | badfmt: |
| 3193 | addReplyErrorFormat(c, "Invalid value '%s' to SENTINEL CONFIG SET '%s'", |
| 3194 | (char *) val->ptr, (char *) o->ptr); |
| 3195 | } |
| 3196 | |
| 3197 | /* SENTINEL CONFIG GET <option> */ |
| 3198 | void sentinelConfigGetCommand(client *c) { |
| 3199 | robj *o = c->argv[3]; |
| 3200 | const char *pattern = o->ptr; |
| 3201 | void *replylen = addReplyDeferredLen(c); |
| 3202 | int matches = 0; |
| 3203 | |
| 3204 | if (stringmatch(pattern,"resolve-hostnames",1)) { |
| 3205 | addReplyBulkCString(c,"resolve-hostnames"); |
| 3206 | addReplyBulkCString(c,sentinel.resolve_hostnames ? "yes" : "no"); |
| 3207 | matches++; |
| 3208 | } |
| 3209 | |
| 3210 | if (stringmatch(pattern, "announce-hostnames", 1)) { |
| 3211 | addReplyBulkCString(c,"announce-hostnames"); |
| 3212 | addReplyBulkCString(c,sentinel.announce_hostnames ? "yes" : "no"); |
| 3213 | matches++; |
| 3214 | } |
| 3215 | |
| 3216 | if (stringmatch(pattern, "announce-ip", 1)) { |
| 3217 | addReplyBulkCString(c,"announce-ip"); |
| 3218 | addReplyBulkCString(c,sentinel.announce_ip ? sentinel.announce_ip : ""); |
| 3219 | matches++; |
| 3220 | } |
| 3221 | |
| 3222 | if (stringmatch(pattern, "announce-port", 1)) { |
| 3223 | addReplyBulkCString(c, "announce-port"); |
| 3224 | addReplyBulkLongLong(c, sentinel.announce_port); |
| 3225 | matches++; |
| 3226 | } |
| 3227 | |
| 3228 | if (stringmatch(pattern, "sentinel-user", 1)) { |
| 3229 | addReplyBulkCString(c, "sentinel-user"); |
| 3230 | addReplyBulkCString(c, sentinel.sentinel_auth_user ? sentinel.sentinel_auth_user : ""); |
| 3231 | matches++; |
| 3232 | } |
| 3233 | |
| 3234 | if (stringmatch(pattern, "sentinel-pass", 1)) { |
| 3235 | addReplyBulkCString(c, "sentinel-pass"); |
| 3236 | addReplyBulkCString(c, sentinel.sentinel_auth_pass ? sentinel.sentinel_auth_pass : ""); |
| 3237 | matches++; |
| 3238 | } |
| 3239 | |
| 3240 | setDeferredMapLen(c, replylen, matches); |
| 3241 | } |
| 3242 | |
| 3243 | const char *sentinelFailoverStateStr(int state) { |
| 3244 | switch(state) { |
| 3245 | case SENTINEL_FAILOVER_STATE_NONE0: return "none"; |
| 3246 | case SENTINEL_FAILOVER_STATE_WAIT_START1: return "wait_start"; |
| 3247 | case SENTINEL_FAILOVER_STATE_SELECT_SLAVE2: return "select_slave"; |
| 3248 | case SENTINEL_FAILOVER_STATE_SEND_SLAVEOF_NOONE3: return "send_slaveof_noone"; |
| 3249 | case SENTINEL_FAILOVER_STATE_WAIT_PROMOTION4: return "wait_promotion"; |
| 3250 | case SENTINEL_FAILOVER_STATE_RECONF_SLAVES5: return "reconf_slaves"; |
| 3251 | case SENTINEL_FAILOVER_STATE_UPDATE_CONFIG6: return "update_config"; |
| 3252 | default: return "unknown"; |
| 3253 | } |
| 3254 | } |
| 3255 | |
| 3256 | /* Redis instance to Redis protocol representation. */ |
| 3257 | void addReplySentinelRedisInstance(client *c, sentinelRedisInstance *ri) { |
| 3258 | char *flags = sdsempty(); |
| 3259 | void *mbl; |
| 3260 | int fields = 0; |
| 3261 | |
| 3262 | mbl = addReplyDeferredLen(c); |
| 3263 | |
| 3264 | addReplyBulkCString(c,"name"); |
| 3265 | addReplyBulkCString(c,ri->name); |
| 3266 | fields++; |
| 3267 | |
| 3268 | addReplyBulkCString(c,"ip"); |
| 3269 | addReplyBulkCString(c,announceSentinelAddr(ri->addr)); |
| 3270 | fields++; |
| 3271 | |
| 3272 | addReplyBulkCString(c,"port"); |
| 3273 | addReplyBulkLongLong(c,ri->addr->port); |
| 3274 | fields++; |
| 3275 | |
| 3276 | addReplyBulkCString(c,"runid"); |
| 3277 | addReplyBulkCString(c,ri->runid ? ri->runid : ""); |
| 3278 | fields++; |
| 3279 | |
| 3280 | addReplyBulkCString(c,"flags"); |
| 3281 | if (ri->flags & SRI_S_DOWN(1<<3)) flags = sdscat(flags,"s_down,"); |
| 3282 | if (ri->flags & SRI_O_DOWN(1<<4)) flags = sdscat(flags,"o_down,"); |
| 3283 | if (ri->flags & SRI_MASTER(1<<0)) flags = sdscat(flags,"master,"); |
| 3284 | if (ri->flags & SRI_SLAVE(1<<1)) flags = sdscat(flags,"slave,"); |
| 3285 | if (ri->flags & SRI_SENTINEL(1<<2)) flags = sdscat(flags,"sentinel,"); |
| 3286 | if (ri->link->disconnected) flags = sdscat(flags,"disconnected,"); |
| 3287 | if (ri->flags & SRI_MASTER_DOWN(1<<5)) flags = sdscat(flags,"master_down,"); |
| 3288 | if (ri->flags & SRI_FAILOVER_IN_PROGRESS(1<<6)) |
| 3289 | flags = sdscat(flags,"failover_in_progress,"); |
| 3290 | if (ri->flags & SRI_PROMOTED(1<<7)) flags = sdscat(flags,"promoted,"); |
| 3291 | if (ri->flags & SRI_RECONF_SENT(1<<8)) flags = sdscat(flags,"reconf_sent,"); |
| 3292 | if (ri->flags & SRI_RECONF_INPROG(1<<9)) flags = sdscat(flags,"reconf_inprog,"); |
| 3293 | if (ri->flags & SRI_RECONF_DONE(1<<10)) flags = sdscat(flags,"reconf_done,"); |
| 3294 | |
| 3295 | if (sdslen(flags) != 0) sdsrange(flags,0,-2); /* remove last "," */ |
| 3296 | addReplyBulkCString(c,flags); |
| 3297 | sdsfree(flags); |
| 3298 | fields++; |
| 3299 | |
| 3300 | addReplyBulkCString(c,"link-pending-commands"); |
| 3301 | addReplyBulkLongLong(c,ri->link->pending_commands); |
| 3302 | fields++; |
| 3303 | |
| 3304 | addReplyBulkCString(c,"link-refcount"); |
| 3305 | addReplyBulkLongLong(c,ri->link->refcount); |
| 3306 | fields++; |
| 3307 | |
| 3308 | if (ri->flags & SRI_FAILOVER_IN_PROGRESS(1<<6)) { |
| 3309 | addReplyBulkCString(c,"failover-state"); |
| 3310 | addReplyBulkCString(c,(char*)sentinelFailoverStateStr(ri->failover_state)); |
| 3311 | fields++; |
| 3312 | } |
| 3313 | |
| 3314 | addReplyBulkCString(c,"last-ping-sent"); |
| 3315 | addReplyBulkLongLong(c, |
| 3316 | ri->link->act_ping_time ? (mstime() - ri->link->act_ping_time) : 0); |
| 3317 | fields++; |
| 3318 | |
| 3319 | addReplyBulkCString(c,"last-ok-ping-reply"); |
| 3320 | addReplyBulkLongLong(c,mstime() - ri->link->last_avail_time); |
| 3321 | fields++; |
| 3322 | |
| 3323 | addReplyBulkCString(c,"last-ping-reply"); |
| 3324 | addReplyBulkLongLong(c,mstime() - ri->link->last_pong_time); |
| 3325 | fields++; |
| 3326 | |
| 3327 | if (ri->flags & SRI_S_DOWN(1<<3)) { |
| 3328 | addReplyBulkCString(c,"s-down-time"); |
| 3329 | addReplyBulkLongLong(c,mstime()-ri->s_down_since_time); |
| 3330 | fields++; |
| 3331 | } |
| 3332 | |
| 3333 | if (ri->flags & SRI_O_DOWN(1<<4)) { |
| 3334 | addReplyBulkCString(c,"o-down-time"); |
| 3335 | addReplyBulkLongLong(c,mstime()-ri->o_down_since_time); |
| 3336 | fields++; |
| 3337 | } |
| 3338 | |
| 3339 | addReplyBulkCString(c,"down-after-milliseconds"); |
| 3340 | addReplyBulkLongLong(c,ri->down_after_period); |
| 3341 | fields++; |
| 3342 | |
| 3343 | /* Masters and Slaves */ |
| 3344 | if (ri->flags & (SRI_MASTER(1<<0)|SRI_SLAVE(1<<1))) { |
| 3345 | addReplyBulkCString(c,"info-refresh"); |
| 3346 | addReplyBulkLongLong(c,mstime() - ri->info_refresh); |
| 3347 | fields++; |
| 3348 | |
| 3349 | addReplyBulkCString(c,"role-reported"); |
| 3350 | addReplyBulkCString(c, (ri->role_reported == SRI_MASTER(1<<0)) ? "master" : |
| 3351 | "slave"); |
| 3352 | fields++; |
| 3353 | |
| 3354 | addReplyBulkCString(c,"role-reported-time"); |
| 3355 | addReplyBulkLongLong(c,mstime() - ri->role_reported_time); |
| 3356 | fields++; |
| 3357 | } |
| 3358 | |
| 3359 | /* Only masters */ |
| 3360 | if (ri->flags & SRI_MASTER(1<<0)) { |
| 3361 | addReplyBulkCString(c,"config-epoch"); |
| 3362 | addReplyBulkLongLong(c,ri->config_epoch); |
| 3363 | fields++; |
| 3364 | |
| 3365 | addReplyBulkCString(c,"num-slaves"); |
| 3366 | addReplyBulkLongLong(c,dictSize(ri->slaves)((ri->slaves)->ht[0].used+(ri->slaves)->ht[1].used )); |
| 3367 | fields++; |
| 3368 | |
| 3369 | addReplyBulkCString(c,"num-other-sentinels"); |
| 3370 | addReplyBulkLongLong(c,dictSize(ri->sentinels)((ri->sentinels)->ht[0].used+(ri->sentinels)->ht[ 1].used)); |
| 3371 | fields++; |
| 3372 | |
| 3373 | addReplyBulkCString(c,"quorum"); |
| 3374 | addReplyBulkLongLong(c,ri->quorum); |
| 3375 | fields++; |
| 3376 | |
| 3377 | addReplyBulkCString(c,"failover-timeout"); |
| 3378 | addReplyBulkLongLong(c,ri->failover_timeout); |
| 3379 | fields++; |
| 3380 | |
| 3381 | addReplyBulkCString(c,"parallel-syncs"); |
| 3382 | addReplyBulkLongLong(c,ri->parallel_syncs); |
| 3383 | fields++; |
| 3384 | |
| 3385 | if (ri->notification_script) { |
| 3386 | addReplyBulkCString(c,"notification-script"); |
| 3387 | addReplyBulkCString(c,ri->notification_script); |
| 3388 | fields++; |
| 3389 | } |
| 3390 | |
| 3391 | if (ri->client_reconfig_script) { |
| 3392 | addReplyBulkCString(c,"client-reconfig-script"); |
| 3393 | addReplyBulkCString(c,ri->client_reconfig_script); |
| 3394 | fields++; |
| 3395 | } |
| 3396 | } |
| 3397 | |
| 3398 | /* Only slaves */ |
| 3399 | if (ri->flags & SRI_SLAVE(1<<1)) { |
| 3400 | addReplyBulkCString(c,"master-link-down-time"); |
| 3401 | addReplyBulkLongLong(c,ri->master_link_down_time); |
| 3402 | fields++; |
| 3403 | |
| 3404 | addReplyBulkCString(c,"master-link-status"); |
| 3405 | addReplyBulkCString(c, |
| 3406 | (ri->slave_master_link_status == SENTINEL_MASTER_LINK_STATUS_UP0) ? |
| 3407 | "ok" : "err"); |
| 3408 | fields++; |
| 3409 | |
| 3410 | addReplyBulkCString(c,"master-host"); |
| 3411 | addReplyBulkCString(c, |
| 3412 | ri->slave_master_host ? ri->slave_master_host : "?"); |
| 3413 | fields++; |
| 3414 | |
| 3415 | addReplyBulkCString(c,"master-port"); |
| 3416 | addReplyBulkLongLong(c,ri->slave_master_port); |
| 3417 | fields++; |
| 3418 | |
| 3419 | addReplyBulkCString(c,"slave-priority"); |
| 3420 | addReplyBulkLongLong(c,ri->slave_priority); |
| 3421 | fields++; |
| 3422 | |
| 3423 | addReplyBulkCString(c,"slave-repl-offset"); |
| 3424 | addReplyBulkLongLong(c,ri->slave_repl_offset); |
| 3425 | fields++; |
| 3426 | } |
| 3427 | |
| 3428 | /* Only sentinels */ |
| 3429 | if (ri->flags & SRI_SENTINEL(1<<2)) { |
| 3430 | addReplyBulkCString(c,"last-hello-message"); |
| 3431 | addReplyBulkLongLong(c,mstime() - ri->last_hello_time); |
| 3432 | fields++; |
| 3433 | |
| 3434 | addReplyBulkCString(c,"voted-leader"); |
| 3435 | addReplyBulkCString(c,ri->leader ? ri->leader : "?"); |
| 3436 | fields++; |
| 3437 | |
| 3438 | addReplyBulkCString(c,"voted-leader-epoch"); |
| 3439 | addReplyBulkLongLong(c,ri->leader_epoch); |
| 3440 | fields++; |
| 3441 | } |
| 3442 | |
| 3443 | setDeferredMapLen(c,mbl,fields); |
| 3444 | } |
| 3445 | |
| 3446 | /* Output a number of instances contained inside a dictionary as |
| 3447 | * Redis protocol. */ |
| 3448 | void addReplyDictOfRedisInstances(client *c, dict *instances) { |
| 3449 | dictIterator *di; |
| 3450 | dictEntry *de; |
| 3451 | |
| 3452 | di = dictGetIterator(instances); |
| 3453 | addReplyArrayLen(c,dictSize(instances)((instances)->ht[0].used+(instances)->ht[1].used)); |
| 3454 | while((de = dictNext(di)) != NULL((void*)0)) { |
| 3455 | sentinelRedisInstance *ri = dictGetVal(de)((de)->v.val); |
| 3456 | |
| 3457 | addReplySentinelRedisInstance(c,ri); |
| 3458 | } |
| 3459 | dictReleaseIterator(di); |
| 3460 | } |
| 3461 | |
| 3462 | /* Lookup the named master into sentinel.masters. |
| 3463 | * If the master is not found reply to the client with an error and returns |
| 3464 | * NULL. */ |
| 3465 | sentinelRedisInstance *sentinelGetMasterByNameOrReplyError(client *c, |
| 3466 | robj *name) |
| 3467 | { |
| 3468 | sentinelRedisInstance *ri; |
| 3469 | |
| 3470 | ri = dictFetchValue(sentinel.masters,name->ptr); |
| 3471 | if (!ri) { |
| 3472 | addReplyError(c,"No such master with that name"); |
| 3473 | return NULL((void*)0); |
| 3474 | } |
| 3475 | return ri; |
| 3476 | } |
| 3477 | |
| 3478 | #define SENTINEL_ISQR_OK0 0 |
| 3479 | #define SENTINEL_ISQR_NOQUORUM(1<<0) (1<<0) |
| 3480 | #define SENTINEL_ISQR_NOAUTH(1<<1) (1<<1) |
| 3481 | int sentinelIsQuorumReachable(sentinelRedisInstance *master, int *usableptr) { |
| 3482 | dictIterator *di; |
| 3483 | dictEntry *de; |
| 3484 | int usable = 1; /* Number of usable Sentinels. Init to 1 to count myself. */ |
| 3485 | int result = SENTINEL_ISQR_OK0; |
| 3486 | int voters = dictSize(master->sentinels)((master->sentinels)->ht[0].used+(master->sentinels) ->ht[1].used)+1; /* Known Sentinels + myself. */ |
| 3487 | |
| 3488 | di = dictGetIterator(master->sentinels); |
| 3489 | while((de = dictNext(di)) != NULL((void*)0)) { |
| 3490 | sentinelRedisInstance *ri = dictGetVal(de)((de)->v.val); |
| 3491 | |
| 3492 | if (ri->flags & (SRI_S_DOWN(1<<3)|SRI_O_DOWN(1<<4))) continue; |
| 3493 | usable++; |
| 3494 | } |
| 3495 | dictReleaseIterator(di); |
| 3496 | |
| 3497 | if (usable < (int)master->quorum) result |= SENTINEL_ISQR_NOQUORUM(1<<0); |
| 3498 | if (usable < voters/2+1) result |= SENTINEL_ISQR_NOAUTH(1<<1); |
| 3499 | if (usableptr) *usableptr = usable; |
| 3500 | return result; |
| 3501 | } |
| 3502 | |
| 3503 | void sentinelCommand(client *c) { |
| 3504 | if (c->argc == 2 && !strcasecmp(c->argv[1]->ptr,"help")) { |
| 3505 | const char *help[] = { |
| 3506 | "CKQUORUM <master-name>", |
| 3507 | " Check if the current Sentinel configuration is able to reach the quorum", |
| 3508 | " needed to failover a master and the majority needed to authorize the", |
| 3509 | " failover.", |
| 3510 | "CONFIG SET <param> <value>", |
| 3511 | " Set a global Sentinel configuration parameter.", |
| 3512 | "CONFIG GET <param>", |
| 3513 | " Get global Sentinel configuration parameter.", |
| 3514 | "GET-MASTER-ADDR-BY-NAME <master-name>", |
| 3515 | " Return the ip and port number of the master with that name.", |
| 3516 | "FAILOVER <master-name>", |
| 3517 | " Manually failover a master node without asking for agreement from other", |
| 3518 | " Sentinels", |
| 3519 | "FLUSHCONFIG", |
| 3520 | " Force Sentinel to rewrite its configuration on disk, including the current", |
| 3521 | " Sentinel state.", |
| 3522 | "INFO-CACHE <master-name>", |
| 3523 | " Return last cached INFO output from masters and all its replicas.", |
| 3524 | "IS-MASTER-DOWN-BY-ADDR <ip> <port> <current-epoch> <runid>", |
| 3525 | " Check if the master specified by ip:port is down from current Sentinel's", |
| 3526 | " point of view.", |
| 3527 | "MASTER <master-name>", |
| 3528 | " Show the state and info of the specified master.", |
| 3529 | "MASTERS", |
| 3530 | " Show a list of monitored masters and their state.", |
| 3531 | "MONITOR <name> <ip> <port> <quorum>", |
| 3532 | " Start monitoring a new master with the specified name, ip, port and quorum.", |
| 3533 | "MYID", |
| 3534 | " Return the ID of the Sentinel instance.", |
| 3535 | "PENDING-SCRIPTS", |
| 3536 | " Get pending scripts information.", |
| 3537 | "REMOVE <master-name>", |
| 3538 | " Remove master from Sentinel's monitor list.", |
| 3539 | "REPLICAS <master-name>", |
| 3540 | " Show a list of replicas for this master and their state.", |
| 3541 | "RESET <pattern>", |
| 3542 | " Reset masters for specific master name matching this pattern.", |
| 3543 | "SENTINELS <master-name>", |
| 3544 | " Show a list of Sentinel instances for this master and their state.", |
| 3545 | "SET <master-name> <option> <value>", |
| 3546 | " Set configuration paramters for certain masters.", |
| 3547 | "SIMULATE-FAILURE (CRASH-AFTER-ELECTION|CRASH-AFTER-PROMOTION|HELP)", |
| 3548 | " Simulate a Sentinel crash.", |
| 3549 | NULL((void*)0) |
| 3550 | }; |
| 3551 | addReplyHelp(c, help); |
| 3552 | } else if (!strcasecmp(c->argv[1]->ptr,"masters")) { |
| 3553 | /* SENTINEL MASTERS */ |
| 3554 | if (c->argc != 2) goto numargserr; |
| 3555 | addReplyDictOfRedisInstances(c,sentinel.masters); |
| 3556 | } else if (!strcasecmp(c->argv[1]->ptr,"master")) { |
| 3557 | /* SENTINEL MASTER <name> */ |
| 3558 | sentinelRedisInstance *ri; |
| 3559 | |
| 3560 | if (c->argc != 3) goto numargserr; |
| 3561 | if ((ri = sentinelGetMasterByNameOrReplyError(c,c->argv[2])) |
| 3562 | == NULL((void*)0)) return; |
| 3563 | addReplySentinelRedisInstance(c,ri); |
| 3564 | } else if (!strcasecmp(c->argv[1]->ptr,"slaves") || |
| 3565 | !strcasecmp(c->argv[1]->ptr,"replicas")) |
| 3566 | { |
| 3567 | /* SENTINEL REPLICAS <master-name> */ |
| 3568 | sentinelRedisInstance *ri; |
| 3569 | |
| 3570 | if (c->argc != 3) goto numargserr; |
| 3571 | if ((ri = sentinelGetMasterByNameOrReplyError(c,c->argv[2])) == NULL((void*)0)) |
| 3572 | return; |
| 3573 | addReplyDictOfRedisInstances(c,ri->slaves); |
| 3574 | } else if (!strcasecmp(c->argv[1]->ptr,"sentinels")) { |
| 3575 | /* SENTINEL SENTINELS <master-name> */ |
| 3576 | sentinelRedisInstance *ri; |
| 3577 | |
| 3578 | if (c->argc != 3) goto numargserr; |
| 3579 | if ((ri = sentinelGetMasterByNameOrReplyError(c,c->argv[2])) == NULL((void*)0)) |
| 3580 | return; |
| 3581 | addReplyDictOfRedisInstances(c,ri->sentinels); |
| 3582 | } else if (!strcasecmp(c->argv[1]->ptr,"myid") && c->argc == 2) { |
| 3583 | /* SENTINEL MYID */ |
| 3584 | addReplyBulkCBuffer(c,sentinel.myid,CONFIG_RUN_ID_SIZE40); |
| 3585 | } else if (!strcasecmp(c->argv[1]->ptr,"is-master-down-by-addr")) { |
| 3586 | /* SENTINEL IS-MASTER-DOWN-BY-ADDR <ip> <port> <current-epoch> <runid> |
| 3587 | * |
| 3588 | * Arguments: |
| 3589 | * |
| 3590 | * ip and port are the ip and port of the master we want to be |
| 3591 | * checked by Sentinel. Note that the command will not check by |
| 3592 | * name but just by master, in theory different Sentinels may monitor |
| 3593 | * different masters with the same name. |
| 3594 | * |
| 3595 | * current-epoch is needed in order to understand if we are allowed |
| 3596 | * to vote for a failover leader or not. Each Sentinel can vote just |
| 3597 | * one time per epoch. |
| 3598 | * |
| 3599 | * runid is "*" if we are not seeking for a vote from the Sentinel |
| 3600 | * in order to elect the failover leader. Otherwise it is set to the |
| 3601 | * runid we want the Sentinel to vote if it did not already voted. |
| 3602 | */ |
| 3603 | sentinelRedisInstance *ri; |
| 3604 | long long req_epoch; |
| 3605 | uint64_t leader_epoch = 0; |
| 3606 | char *leader = NULL((void*)0); |
| 3607 | long port; |
| 3608 | int isdown = 0; |
| 3609 | |
| 3610 | if (c->argc != 6) goto numargserr; |
| 3611 | if (getLongFromObjectOrReply(c,c->argv[3],&port,NULL((void*)0)) != C_OK0 || |
| 3612 | getLongLongFromObjectOrReply(c,c->argv[4],&req_epoch,NULL((void*)0)) |
| 3613 | != C_OK0) |
| 3614 | return; |
| 3615 | ri = getSentinelRedisInstanceByAddrAndRunID(sentinel.masters, |
| 3616 | c->argv[2]->ptr,port,NULL((void*)0)); |
| 3617 | |
| 3618 | /* It exists? Is actually a master? Is subjectively down? It's down. |
| 3619 | * Note: if we are in tilt mode we always reply with "0". */ |
| 3620 | if (!sentinel.tilt && ri && (ri->flags & SRI_S_DOWN(1<<3)) && |
| 3621 | (ri->flags & SRI_MASTER(1<<0))) |
| 3622 | isdown = 1; |
| 3623 | |
| 3624 | /* Vote for the master (or fetch the previous vote) if the request |
| 3625 | * includes a runid, otherwise the sender is not seeking for a vote. */ |
| 3626 | if (ri && ri->flags & SRI_MASTER(1<<0) && strcasecmp(c->argv[5]->ptr,"*")) { |
| 3627 | leader = sentinelVoteLeader(ri,(uint64_t)req_epoch, |
| 3628 | c->argv[5]->ptr, |
| 3629 | &leader_epoch); |
| 3630 | } |
| 3631 | |
| 3632 | /* Reply with a three-elements multi-bulk reply: |
| 3633 | * down state, leader, vote epoch. */ |
| 3634 | addReplyArrayLen(c,3); |
| 3635 | addReply(c, isdown ? shared.cone : shared.czero); |
| 3636 | addReplyBulkCString(c, leader ? leader : "*"); |
| 3637 | addReplyLongLong(c, (long long)leader_epoch); |
| 3638 | if (leader) sdsfree(leader); |
| 3639 | } else if (!strcasecmp(c->argv[1]->ptr,"reset")) { |
| 3640 | /* SENTINEL RESET <pattern> */ |
| 3641 | if (c->argc != 3) goto numargserr; |
| 3642 | addReplyLongLong(c,sentinelResetMastersByPattern(c->argv[2]->ptr,SENTINEL_GENERATE_EVENT(1<<16))); |
| 3643 | } else if (!strcasecmp(c->argv[1]->ptr,"get-master-addr-by-name")) { |
| 3644 | /* SENTINEL GET-MASTER-ADDR-BY-NAME <master-name> */ |
| 3645 | sentinelRedisInstance *ri; |
| 3646 | |
| 3647 | if (c->argc != 3) goto numargserr; |
| 3648 | ri = sentinelGetMasterByName(c->argv[2]->ptr); |
| 3649 | if (ri == NULL((void*)0)) { |
| 3650 | addReplyNullArray(c); |
| 3651 | } else { |
| 3652 | sentinelAddr *addr = sentinelGetCurrentMasterAddress(ri); |
| 3653 | |
| 3654 | addReplyArrayLen(c,2); |
| 3655 | addReplyBulkCString(c,announceSentinelAddr(addr)); |
| 3656 | addReplyBulkLongLong(c,addr->port); |
| 3657 | } |
| 3658 | } else if (!strcasecmp(c->argv[1]->ptr,"failover")) { |
| 3659 | /* SENTINEL FAILOVER <master-name> */ |
| 3660 | sentinelRedisInstance *ri; |
| 3661 | |
| 3662 | if (c->argc != 3) goto numargserr; |
| 3663 | if ((ri = sentinelGetMasterByNameOrReplyError(c,c->argv[2])) == NULL((void*)0)) |
| 3664 | return; |
| 3665 | if (ri->flags & SRI_FAILOVER_IN_PROGRESS(1<<6)) { |
| 3666 | addReplySds(c,sdsnew("-INPROG Failover already in progress\r\n")); |
| 3667 | return; |
| 3668 | } |
| 3669 | if (sentinelSelectSlave(ri) == NULL((void*)0)) { |
| 3670 | addReplySds(c,sdsnew("-NOGOODSLAVE No suitable replica to promote\r\n")); |
| 3671 | return; |
| 3672 | } |
| 3673 | serverLog(LL_WARNING3,"Executing user requested FAILOVER of '%s'", |
| 3674 | ri->name); |
| 3675 | sentinelStartFailover(ri); |
| 3676 | ri->flags |= SRI_FORCE_FAILOVER(1<<11); |
| 3677 | addReply(c,shared.ok); |
| 3678 | } else if (!strcasecmp(c->argv[1]->ptr,"pending-scripts")) { |
| 3679 | /* SENTINEL PENDING-SCRIPTS */ |
| 3680 | |
| 3681 | if (c->argc != 2) goto numargserr; |
| 3682 | sentinelPendingScriptsCommand(c); |
| 3683 | } else if (!strcasecmp(c->argv[1]->ptr,"monitor")) { |
| 3684 | /* SENTINEL MONITOR <name> <ip> <port> <quorum> */ |
| 3685 | sentinelRedisInstance *ri; |
| 3686 | long quorum, port; |
| 3687 | char ip[NET_IP_STR_LEN46]; |
| 3688 | |
| 3689 | if (c->argc != 6) goto numargserr; |
| 3690 | if (getLongFromObjectOrReply(c,c->argv[5],&quorum,"Invalid quorum") |
| 3691 | != C_OK0) return; |
| 3692 | if (getLongFromObjectOrReply(c,c->argv[4],&port,"Invalid port") |
| 3693 | != C_OK0) return; |
| 3694 | |
| 3695 | if (quorum <= 0) { |
| 3696 | addReplyError(c, "Quorum must be 1 or greater."); |
| 3697 | return; |
| 3698 | } |
| 3699 | |
| 3700 | /* If resolve-hostnames is used, actual DNS resolution may take place. |
| 3701 | * Otherwise just validate address. |
| 3702 | */ |
| 3703 | if (anetResolve(NULL((void*)0),c->argv[3]->ptr,ip,sizeof(ip), |
| 3704 | sentinel.resolve_hostnames ? ANET_NONE0 : ANET_IP_ONLY(1<<0)) == ANET_ERR-1) { |
| 3705 | addReplyError(c, "Invalid IP address or hostname specified"); |
| 3706 | return; |
| 3707 | } |
| 3708 | |
| 3709 | /* Parameters are valid. Try to create the master instance. */ |
| 3710 | ri = createSentinelRedisInstance(c->argv[2]->ptr,SRI_MASTER(1<<0), |
| 3711 | c->argv[3]->ptr,port,quorum,NULL((void*)0)); |
| 3712 | if (ri == NULL((void*)0)) { |
| 3713 | switch(errno(*__errno_location ())) { |
| 3714 | case EBUSY16: |
| 3715 | addReplyError(c,"Duplicated master name"); |
| 3716 | break; |
| 3717 | case EINVAL22: |
| 3718 | addReplyError(c,"Invalid port number"); |
| 3719 | break; |
| 3720 | default: |
| 3721 | addReplyError(c,"Unspecified error adding the instance"); |
| 3722 | break; |
| 3723 | } |
| 3724 | } else { |
| 3725 | sentinelFlushConfig(); |
| 3726 | sentinelEvent(LL_WARNING3,"+monitor",ri,"%@ quorum %d",ri->quorum); |
| 3727 | addReply(c,shared.ok); |
| 3728 | } |
| 3729 | } else if (!strcasecmp(c->argv[1]->ptr,"flushconfig")) { |
| 3730 | if (c->argc != 2) goto numargserr; |
| 3731 | sentinelFlushConfig(); |
| 3732 | addReply(c,shared.ok); |
| 3733 | return; |
| 3734 | } else if (!strcasecmp(c->argv[1]->ptr,"remove")) { |
| 3735 | /* SENTINEL REMOVE <name> */ |
| 3736 | sentinelRedisInstance *ri; |
| 3737 | |
| 3738 | if (c->argc != 3) goto numargserr; |
| 3739 | if ((ri = sentinelGetMasterByNameOrReplyError(c,c->argv[2])) |
| 3740 | == NULL((void*)0)) return; |
| 3741 | sentinelEvent(LL_WARNING3,"-monitor",ri,"%@"); |
| 3742 | dictDelete(sentinel.masters,c->argv[2]->ptr); |
| 3743 | sentinelFlushConfig(); |
| 3744 | addReply(c,shared.ok); |
| 3745 | } else if (!strcasecmp(c->argv[1]->ptr,"ckquorum")) { |
| 3746 | /* SENTINEL CKQUORUM <name> */ |
| 3747 | sentinelRedisInstance *ri; |
| 3748 | int usable; |
| 3749 | |
| 3750 | if (c->argc != 3) goto numargserr; |
| 3751 | if ((ri = sentinelGetMasterByNameOrReplyError(c,c->argv[2])) |
| 3752 | == NULL((void*)0)) return; |
| 3753 | int result = sentinelIsQuorumReachable(ri,&usable); |
| 3754 | if (result == SENTINEL_ISQR_OK0) { |
| 3755 | addReplySds(c, sdscatfmt(sdsempty(), |
| 3756 | "+OK %i usable Sentinels. Quorum and failover authorization " |
| 3757 | "can be reached\r\n",usable)); |
| 3758 | } else { |
| 3759 | sds e = sdscatfmt(sdsempty(), |
| 3760 | "-NOQUORUM %i usable Sentinels. ",usable); |
| 3761 | if (result & SENTINEL_ISQR_NOQUORUM(1<<0)) |
| 3762 | e = sdscat(e,"Not enough available Sentinels to reach the" |
| 3763 | " specified quorum for this master"); |
| 3764 | if (result & SENTINEL_ISQR_NOAUTH(1<<1)) { |
| 3765 | if (result & SENTINEL_ISQR_NOQUORUM(1<<0)) e = sdscat(e,". "); |
| 3766 | e = sdscat(e, "Not enough available Sentinels to reach the" |
| 3767 | " majority and authorize a failover"); |
| 3768 | } |
| 3769 | e = sdscat(e,"\r\n"); |
| 3770 | addReplySds(c,e); |
| 3771 | } |
| 3772 | } else if (!strcasecmp(c->argv[1]->ptr,"set")) { |
| 3773 | if (c->argc < 3) goto numargserr; |
| 3774 | sentinelSetCommand(c); |
| 3775 | } else if (!strcasecmp(c->argv[1]->ptr,"config")) { |
| 3776 | if (c->argc < 3) goto numargserr; |
| 3777 | if (!strcasecmp(c->argv[2]->ptr,"set") && c->argc == 5) |
| 3778 | sentinelConfigSetCommand(c); |
| 3779 | else if (!strcasecmp(c->argv[2]->ptr,"get") && c->argc == 4) |
| 3780 | sentinelConfigGetCommand(c); |
| 3781 | else |
| 3782 | addReplyError(c, "Only SENTINEL CONFIG GET <option> / SET <option> <value> are supported."); |
| 3783 | } else if (!strcasecmp(c->argv[1]->ptr,"info-cache")) { |
| 3784 | /* SENTINEL INFO-CACHE <name> */ |
| 3785 | if (c->argc < 2) goto numargserr; |
| 3786 | mstime_t now = mstime(); |
| 3787 | |
| 3788 | /* Create an ad-hoc dictionary type so that we can iterate |
| 3789 | * a dictionary composed of just the master groups the user |
| 3790 | * requested. */ |
| 3791 | dictType copy_keeper = instancesDictType; |
| 3792 | copy_keeper.valDestructor = NULL((void*)0); |
| 3793 | dict *masters_local = sentinel.masters; |
| 3794 | if (c->argc > 2) { |
| 3795 | masters_local = dictCreate(©_keeper, NULL((void*)0)); |
| 3796 | |
| 3797 | for (int i = 2; i < c->argc; i++) { |
| 3798 | sentinelRedisInstance *ri; |
| 3799 | ri = sentinelGetMasterByName(c->argv[i]->ptr); |
| 3800 | if (!ri) continue; /* ignore non-existing names */ |
| 3801 | dictAdd(masters_local, ri->name, ri); |
| 3802 | } |
| 3803 | } |
| 3804 | |
| 3805 | /* Reply format: |
| 3806 | * 1.) master name |
| 3807 | * 2.) 1.) info from master |
| 3808 | * 2.) info from replica |
| 3809 | * ... |
| 3810 | * 3.) other master name |
| 3811 | * ... |
| 3812 | */ |
| 3813 | addReplyArrayLen(c,dictSize(masters_local)((masters_local)->ht[0].used+(masters_local)->ht[1].used ) * 2); |
| 3814 | |
| 3815 | dictIterator *di; |
| 3816 | dictEntry *de; |
| 3817 | di = dictGetIterator(masters_local); |
| 3818 | while ((de = dictNext(di)) != NULL((void*)0)) { |
| 3819 | sentinelRedisInstance *ri = dictGetVal(de)((de)->v.val); |
| 3820 | addReplyBulkCBuffer(c,ri->name,strlen(ri->name)); |
| 3821 | addReplyArrayLen(c,dictSize(ri->slaves)((ri->slaves)->ht[0].used+(ri->slaves)->ht[1].used ) + 1); /* +1 for self */ |
| 3822 | addReplyArrayLen(c,2); |
| 3823 | addReplyLongLong(c, now - ri->info_refresh); |
| 3824 | if (ri->info) |
| 3825 | addReplyBulkCBuffer(c,ri->info,sdslen(ri->info)); |
| 3826 | else |
| 3827 | addReplyNull(c); |
| 3828 | |
| 3829 | dictIterator *sdi; |
| 3830 | dictEntry *sde; |
| 3831 | sdi = dictGetIterator(ri->slaves); |
| 3832 | while ((sde = dictNext(sdi)) != NULL((void*)0)) { |
| 3833 | sentinelRedisInstance *sri = dictGetVal(sde)((sde)->v.val); |
| 3834 | addReplyArrayLen(c,2); |
| 3835 | addReplyLongLong(c, now - sri->info_refresh); |
| 3836 | if (sri->info) |
| 3837 | addReplyBulkCBuffer(c,sri->info,sdslen(sri->info)); |
| 3838 | else |
| 3839 | addReplyNull(c); |
| 3840 | } |
| 3841 | dictReleaseIterator(sdi); |
| 3842 | } |
| 3843 | dictReleaseIterator(di); |
| 3844 | if (masters_local != sentinel.masters) dictRelease(masters_local); |
| 3845 | } else if (!strcasecmp(c->argv[1]->ptr,"simulate-failure")) { |
| 3846 | /* SENTINEL SIMULATE-FAILURE <flag> <flag> ... <flag> */ |
| 3847 | int j; |
| 3848 | |
| 3849 | sentinel.simfailure_flags = SENTINEL_SIMFAILURE_NONE0; |
| 3850 | for (j = 2; j < c->argc; j++) { |
| 3851 | if (!strcasecmp(c->argv[j]->ptr,"crash-after-election")) { |
| 3852 | sentinel.simfailure_flags |= |
| 3853 | SENTINEL_SIMFAILURE_CRASH_AFTER_ELECTION(1<<0); |
| 3854 | serverLog(LL_WARNING3,"Failure simulation: this Sentinel " |
| 3855 | "will crash after being successfully elected as failover " |
| 3856 | "leader"); |
| 3857 | } else if (!strcasecmp(c->argv[j]->ptr,"crash-after-promotion")) { |
| 3858 | sentinel.simfailure_flags |= |
| 3859 | SENTINEL_SIMFAILURE_CRASH_AFTER_PROMOTION(1<<1); |
| 3860 | serverLog(LL_WARNING3,"Failure simulation: this Sentinel " |
| 3861 | "will crash after promoting the selected replica to master"); |
| 3862 | } else if (!strcasecmp(c->argv[j]->ptr,"help")) { |
| 3863 | addReplyArrayLen(c,2); |
| 3864 | addReplyBulkCString(c,"crash-after-election"); |
| 3865 | addReplyBulkCString(c,"crash-after-promotion"); |
| 3866 | } else { |
| 3867 | addReplyError(c,"Unknown failure simulation specified"); |
| 3868 | return; |
| 3869 | } |
| 3870 | } |
| 3871 | addReply(c,shared.ok); |
| 3872 | } else { |
| 3873 | addReplySubcommandSyntaxError(c); |
| 3874 | } |
| 3875 | return; |
| 3876 | |
| 3877 | numargserr: |
| 3878 | addReplyErrorFormat(c,"Wrong number of arguments for 'sentinel %s'", |
| 3879 | (char*)c->argv[1]->ptr); |
| 3880 | } |
| 3881 | |
| 3882 | #define info_section_from_redis(section_name)do { if (defsections || allsections || !strcasecmp(section,section_name )) { sds redissection; if (sections++) info = sdscat(info,"\r\n" ); redissection = genRedisInfoString(section_name); info = sdscatlen (info,redissection,sdslen(redissection)); sdsfree(redissection ); } } while(0) do { \ |
| 3883 | if (defsections || allsections || !strcasecmp(section,section_name)) { \ |
| 3884 | sds redissection; \ |
| 3885 | if (sections++) info = sdscat(info,"\r\n"); \ |
| 3886 | redissection = genRedisInfoString(section_name); \ |
| 3887 | info = sdscatlen(info,redissection,sdslen(redissection)); \ |
| 3888 | sdsfree(redissection); \ |
| 3889 | } \ |
| 3890 | } while(0) |
| 3891 | |
| 3892 | /* SENTINEL INFO [section] */ |
| 3893 | void sentinelInfoCommand(client *c) { |
| 3894 | if (c->argc > 2) { |
| 3895 | addReplyErrorObject(c,shared.syntaxerr); |
| 3896 | return; |
| 3897 | } |
| 3898 | |
| 3899 | int defsections = 0, allsections = 0; |
| 3900 | char *section = c->argc == 2 ? c->argv[1]->ptr : NULL((void*)0); |
| 3901 | if (section) { |
| 3902 | allsections = !strcasecmp(section,"all"); |
| 3903 | defsections = !strcasecmp(section,"default"); |
| 3904 | } else { |
| 3905 | defsections = 1; |
| 3906 | } |
| 3907 | |
| 3908 | int sections = 0; |
| 3909 | sds info = sdsempty(); |
| 3910 | |
| 3911 | info_section_from_redis("server")do { if (defsections || allsections || !strcasecmp(section,"server" )) { sds redissection; if (sections++) info = sdscat(info,"\r\n" ); redissection = genRedisInfoString("server"); info = sdscatlen (info,redissection,sdslen(redissection)); sdsfree(redissection ); } } while(0); |
| 3912 | info_section_from_redis("clients")do { if (defsections || allsections || !strcasecmp(section,"clients" )) { sds redissection; if (sections++) info = sdscat(info,"\r\n" ); redissection = genRedisInfoString("clients"); info = sdscatlen (info,redissection,sdslen(redissection)); sdsfree(redissection ); } } while(0); |
| 3913 | info_section_from_redis("cpu")do { if (defsections || allsections || !strcasecmp(section,"cpu" )) { sds redissection; if (sections++) info = sdscat(info,"\r\n" ); redissection = genRedisInfoString("cpu"); info = sdscatlen (info,redissection,sdslen(redissection)); sdsfree(redissection ); } } while(0); |
| 3914 | info_section_from_redis("stats")do { if (defsections || allsections || !strcasecmp(section,"stats" )) { sds redissection; if (sections++) info = sdscat(info,"\r\n" ); redissection = genRedisInfoString("stats"); info = sdscatlen (info,redissection,sdslen(redissection)); sdsfree(redissection ); } } while(0); |
| 3915 | |
| 3916 | if (defsections || allsections || !strcasecmp(section,"sentinel")) { |
| 3917 | dictIterator *di; |
| 3918 | dictEntry *de; |
| 3919 | int master_id = 0; |
| 3920 | |
| 3921 | if (sections++) info = sdscat(info,"\r\n"); |
| 3922 | info = sdscatprintf(info, |
| 3923 | "# Sentinel\r\n" |
| 3924 | "sentinel_masters:%lu\r\n" |
| 3925 | "sentinel_tilt:%d\r\n" |
| 3926 | "sentinel_running_scripts:%d\r\n" |
| 3927 | "sentinel_scripts_queue_length:%ld\r\n" |
| 3928 | "sentinel_simulate_failure_flags:%lu\r\n", |
| 3929 | dictSize(sentinel.masters)((sentinel.masters)->ht[0].used+(sentinel.masters)->ht[ 1].used), |
| 3930 | sentinel.tilt, |
| 3931 | sentinel.running_scripts, |
| 3932 | listLength(sentinel.scripts_queue)((sentinel.scripts_queue)->len), |
| 3933 | sentinel.simfailure_flags); |
| 3934 | |
| 3935 | di = dictGetIterator(sentinel.masters); |
| 3936 | while((de = dictNext(di)) != NULL((void*)0)) { |
| 3937 | sentinelRedisInstance *ri = dictGetVal(de)((de)->v.val); |
| 3938 | char *status = "ok"; |
| 3939 | |
| 3940 | if (ri->flags & SRI_O_DOWN(1<<4)) status = "odown"; |
| 3941 | else if (ri->flags & SRI_S_DOWN(1<<3)) status = "sdown"; |
| 3942 | info = sdscatprintf(info, |
| 3943 | "master%d:name=%s,status=%s,address=%s:%d," |
| 3944 | "slaves=%lu,sentinels=%lu\r\n", |
| 3945 | master_id++, ri->name, status, |
| 3946 | announceSentinelAddr(ri->addr), ri->addr->port, |
| 3947 | dictSize(ri->slaves)((ri->slaves)->ht[0].used+(ri->slaves)->ht[1].used ), |
| 3948 | dictSize(ri->sentinels)((ri->sentinels)->ht[0].used+(ri->sentinels)->ht[ 1].used)+1); |
| 3949 | } |
| 3950 | dictReleaseIterator(di); |
| 3951 | } |
| 3952 | |
| 3953 | addReplyBulkSds(c, info); |
| 3954 | } |
| 3955 | |
| 3956 | /* Implements Sentinel version of the ROLE command. The output is |
| 3957 | * "sentinel" and the list of currently monitored master names. */ |
| 3958 | void sentinelRoleCommand(client *c) { |
| 3959 | dictIterator *di; |
| 3960 | dictEntry *de; |
| 3961 | |
| 3962 | addReplyArrayLen(c,2); |
| 3963 | addReplyBulkCBuffer(c,"sentinel",8); |
| 3964 | addReplyArrayLen(c,dictSize(sentinel.masters)((sentinel.masters)->ht[0].used+(sentinel.masters)->ht[ 1].used)); |
| 3965 | |
| 3966 | di = dictGetIterator(sentinel.masters); |
| 3967 | while((de = dictNext(di)) != NULL((void*)0)) { |
| 3968 | sentinelRedisInstance *ri = dictGetVal(de)((de)->v.val); |
| 3969 | |
| 3970 | addReplyBulkCString(c,ri->name); |
| 3971 | } |
| 3972 | dictReleaseIterator(di); |
| 3973 | } |
| 3974 | |
| 3975 | /* SENTINEL SET <mastername> [<option> <value> ...] */ |
| 3976 | void sentinelSetCommand(client *c) { |
| 3977 | sentinelRedisInstance *ri; |
| 3978 | int j, changes = 0; |
| 3979 | int badarg = 0; /* Bad argument position for error reporting. */ |
| 3980 | char *option; |
| 3981 | |
| 3982 | if ((ri = sentinelGetMasterByNameOrReplyError(c,c->argv[2])) |
| 3983 | == NULL((void*)0)) return; |
| 3984 | |
| 3985 | /* Process option - value pairs. */ |
| 3986 | for (j = 3; j < c->argc; j++) { |
| 3987 | int moreargs = (c->argc-1) - j; |
| 3988 | option = c->argv[j]->ptr; |
| 3989 | long long ll; |
| 3990 | int old_j = j; /* Used to know what to log as an event. */ |
| 3991 | |
| 3992 | if (!strcasecmp(option,"down-after-milliseconds") && moreargs > 0) { |
| 3993 | /* down-after-millisecodns <milliseconds> */ |
| 3994 | robj *o = c->argv[++j]; |
| 3995 | if (getLongLongFromObject(o,&ll) == C_ERR-1 || ll <= 0) { |
| 3996 | badarg = j; |
| 3997 | goto badfmt; |
| 3998 | } |
| 3999 | ri->down_after_period = ll; |
| 4000 | sentinelPropagateDownAfterPeriod(ri); |
| 4001 | changes++; |
| 4002 | } else if (!strcasecmp(option,"failover-timeout") && moreargs > 0) { |
| 4003 | /* failover-timeout <milliseconds> */ |
| 4004 | robj *o = c->argv[++j]; |
| 4005 | if (getLongLongFromObject(o,&ll) == C_ERR-1 || ll <= 0) { |
| 4006 | badarg = j; |
| 4007 | goto badfmt; |
| 4008 | } |
| 4009 | ri->failover_timeout = ll; |
| 4010 | changes++; |
| 4011 | } else if (!strcasecmp(option,"parallel-syncs") && moreargs > 0) { |
| 4012 | /* parallel-syncs <milliseconds> */ |
| 4013 | robj *o = c->argv[++j]; |
| 4014 | if (getLongLongFromObject(o,&ll) == C_ERR-1 || ll <= 0) { |
| 4015 | badarg = j; |
| 4016 | goto badfmt; |
| 4017 | } |
| 4018 | ri->parallel_syncs = ll; |
| 4019 | changes++; |
| 4020 | } else if (!strcasecmp(option,"notification-script") && moreargs > 0) { |
| 4021 | /* notification-script <path> */ |
| 4022 | char *value = c->argv[++j]->ptr; |
| 4023 | if (sentinel.deny_scripts_reconfig) { |
| 4024 | addReplyError(c, |
| 4025 | "Reconfiguration of scripts path is denied for " |
| 4026 | "security reasons. Check the deny-scripts-reconfig " |
| 4027 | "configuration directive in your Sentinel configuration"); |
| 4028 | goto seterr; |
| 4029 | } |
| 4030 | |
| 4031 | if (strlen(value) && access(value,X_OK1) == -1) { |
| 4032 | addReplyError(c, |
| 4033 | "Notification script seems non existing or non executable"); |
| 4034 | goto seterr; |
| 4035 | } |
| 4036 | sdsfree(ri->notification_script); |
| 4037 | ri->notification_script = strlen(value) ? sdsnew(value) : NULL((void*)0); |
| 4038 | changes++; |
| 4039 | } else if (!strcasecmp(option,"client-reconfig-script") && moreargs > 0) { |
| 4040 | /* client-reconfig-script <path> */ |
| 4041 | char *value = c->argv[++j]->ptr; |
| 4042 | if (sentinel.deny_scripts_reconfig) { |
| 4043 | addReplyError(c, |
| 4044 | "Reconfiguration of scripts path is denied for " |
| 4045 | "security reasons. Check the deny-scripts-reconfig " |
| 4046 | "configuration directive in your Sentinel configuration"); |
| 4047 | goto seterr; |
| 4048 | } |
| 4049 | |
| 4050 | if (strlen(value) && access(value,X_OK1) == -1) { |
| 4051 | addReplyError(c, |
| 4052 | "Client reconfiguration script seems non existing or " |
| 4053 | "non executable"); |
| 4054 | goto seterr; |
| 4055 | } |
| 4056 | sdsfree(ri->client_reconfig_script); |
| 4057 | ri->client_reconfig_script = strlen(value) ? sdsnew(value) : NULL((void*)0); |
| 4058 | changes++; |
| 4059 | } else if (!strcasecmp(option,"auth-pass") && moreargs > 0) { |
| 4060 | /* auth-pass <password> */ |
| 4061 | char *value = c->argv[++j]->ptr; |
| 4062 | sdsfree(ri->auth_pass); |
| 4063 | ri->auth_pass = strlen(value) ? sdsnew(value) : NULL((void*)0); |
| 4064 | changes++; |
| 4065 | } else if (!strcasecmp(option,"auth-user") && moreargs > 0) { |
| 4066 | /* auth-user <username> */ |
| 4067 | char *value = c->argv[++j]->ptr; |
| 4068 | sdsfree(ri->auth_user); |
| 4069 | ri->auth_user = strlen(value) ? sdsnew(value) : NULL((void*)0); |
| 4070 | changes++; |
| 4071 | } else if (!strcasecmp(option,"quorum") && moreargs > 0) { |
| 4072 | /* quorum <count> */ |
| 4073 | robj *o = c->argv[++j]; |
| 4074 | if (getLongLongFromObject(o,&ll) == C_ERR-1 || ll <= 0) { |
| 4075 | badarg = j; |
| 4076 | goto badfmt; |
| 4077 | } |
| 4078 | ri->quorum = ll; |
| 4079 | changes++; |
| 4080 | } else if (!strcasecmp(option,"rename-command") && moreargs > 1) { |
| 4081 | /* rename-command <oldname> <newname> */ |
| 4082 | sds oldname = c->argv[++j]->ptr; |
| 4083 | sds newname = c->argv[++j]->ptr; |
| 4084 | |
| 4085 | if ((sdslen(oldname) == 0) || (sdslen(newname) == 0)) { |
| 4086 | badarg = sdslen(newname) ? j-1 : j; |
| 4087 | goto badfmt; |
| 4088 | } |
| 4089 | |
| 4090 | /* Remove any older renaming for this command. */ |
| 4091 | dictDelete(ri->renamed_commands,oldname); |
| 4092 | |
| 4093 | /* If the target name is the same as the source name there |
| 4094 | * is no need to add an entry mapping to itself. */ |
| 4095 | if (!dictSdsKeyCaseCompare(NULL((void*)0),oldname,newname)) { |
| 4096 | oldname = sdsdup(oldname); |
| 4097 | newname = sdsdup(newname); |
| 4098 | dictAdd(ri->renamed_commands,oldname,newname); |
| 4099 | } |
| 4100 | changes++; |
| 4101 | } else { |
| 4102 | addReplyErrorFormat(c,"Unknown option or number of arguments for " |
| 4103 | "SENTINEL SET '%s'", option); |
| 4104 | goto seterr; |
| 4105 | } |
| 4106 | |
| 4107 | /* Log the event. */ |
| 4108 | int numargs = j-old_j+1; |
| 4109 | switch(numargs) { |
| 4110 | case 2: |
| 4111 | sentinelEvent(LL_WARNING3,"+set",ri,"%@ %s %s",c->argv[old_j]->ptr, |
| 4112 | c->argv[old_j+1]->ptr); |
| 4113 | break; |
| 4114 | case 3: |
| 4115 | sentinelEvent(LL_WARNING3,"+set",ri,"%@ %s %s %s",c->argv[old_j]->ptr, |
| 4116 | c->argv[old_j+1]->ptr, |
| 4117 | c->argv[old_j+2]->ptr); |
| 4118 | break; |
| 4119 | default: |
| 4120 | sentinelEvent(LL_WARNING3,"+set",ri,"%@ %s",c->argv[old_j]->ptr); |
| 4121 | break; |
| 4122 | } |
| 4123 | } |
| 4124 | |
| 4125 | if (changes) sentinelFlushConfig(); |
| 4126 | addReply(c,shared.ok); |
| 4127 | return; |
| 4128 | |
| 4129 | badfmt: /* Bad format errors */ |
| 4130 | addReplyErrorFormat(c,"Invalid argument '%s' for SENTINEL SET '%s'", |
| 4131 | (char*)c->argv[badarg]->ptr,option); |
| 4132 | seterr: |
| 4133 | if (changes) sentinelFlushConfig(); |
| 4134 | return; |
| 4135 | } |
| 4136 | |
| 4137 | /* Our fake PUBLISH command: it is actually useful only to receive hello messages |
| 4138 | * from the other sentinel instances, and publishing to a channel other than |
| 4139 | * SENTINEL_HELLO_CHANNEL is forbidden. |
| 4140 | * |
| 4141 | * Because we have a Sentinel PUBLISH, the code to send hello messages is the same |
| 4142 | * for all the three kind of instances: masters, slaves, sentinels. */ |
| 4143 | void sentinelPublishCommand(client *c) { |
| 4144 | if (strcmp(c->argv[1]->ptr,SENTINEL_HELLO_CHANNEL"__sentinel__:hello")) { |
| 4145 | addReplyError(c, "Only HELLO messages are accepted by Sentinel instances."); |
| 4146 | return; |
| 4147 | } |
| 4148 | sentinelProcessHelloMessage(c->argv[2]->ptr,sdslen(c->argv[2]->ptr)); |
| 4149 | addReplyLongLong(c,1); |
| 4150 | } |
| 4151 | |
| 4152 | /* ===================== SENTINEL availability checks ======================= */ |
| 4153 | |
| 4154 | /* Is this instance down from our point of view? */ |
| 4155 | void sentinelCheckSubjectivelyDown(sentinelRedisInstance *ri) { |
| 4156 | mstime_t elapsed = 0; |
| 4157 | |
| 4158 | if (ri->link->act_ping_time) |
| 4159 | elapsed = mstime() - ri->link->act_ping_time; |
| 4160 | else if (ri->link->disconnected) |
| 4161 | elapsed = mstime() - ri->link->last_avail_time; |
| 4162 | |
| 4163 | /* Check if we are in need for a reconnection of one of the |
| 4164 | * links, because we are detecting low activity. |
| 4165 | * |
| 4166 | * 1) Check if the command link seems connected, was connected not less |
| 4167 | * than SENTINEL_MIN_LINK_RECONNECT_PERIOD, but still we have a |
| 4168 | * pending ping for more than half the timeout. */ |
| 4169 | if (ri->link->cc && |
| 4170 | (mstime() - ri->link->cc_conn_time) > |
| 4171 | SENTINEL_MIN_LINK_RECONNECT_PERIOD15000 && |
| 4172 | ri->link->act_ping_time != 0 && /* There is a pending ping... */ |
| 4173 | /* The pending ping is delayed, and we did not receive |
| 4174 | * error replies as well. */ |
| 4175 | (mstime() - ri->link->act_ping_time) > (ri->down_after_period/2) && |
| 4176 | (mstime() - ri->link->last_pong_time) > (ri->down_after_period/2)) |
| 4177 | { |
| 4178 | instanceLinkCloseConnection(ri->link,ri->link->cc); |
| 4179 | } |
| 4180 | |
| 4181 | /* 2) Check if the pubsub link seems connected, was connected not less |
| 4182 | * than SENTINEL_MIN_LINK_RECONNECT_PERIOD, but still we have no |
| 4183 | * activity in the Pub/Sub channel for more than |
| 4184 | * SENTINEL_PUBLISH_PERIOD * 3. |
| 4185 | */ |
| 4186 | if (ri->link->pc && |
| 4187 | (mstime() - ri->link->pc_conn_time) > |
| 4188 | SENTINEL_MIN_LINK_RECONNECT_PERIOD15000 && |
| 4189 | (mstime() - ri->link->pc_last_activity) > (SENTINEL_PUBLISH_PERIOD2000*3)) |
| 4190 | { |
| 4191 | instanceLinkCloseConnection(ri->link,ri->link->pc); |
| 4192 | } |
| 4193 | |
| 4194 | /* Update the SDOWN flag. We believe the instance is SDOWN if: |
| 4195 | * |
| 4196 | * 1) It is not replying. |
| 4197 | * 2) We believe it is a master, it reports to be a slave for enough time |
| 4198 | * to meet the down_after_period, plus enough time to get two times |
| 4199 | * INFO report from the instance. */ |
| 4200 | if (elapsed > ri->down_after_period || |
| 4201 | (ri->flags & SRI_MASTER(1<<0) && |
| 4202 | ri->role_reported == SRI_SLAVE(1<<1) && |
| 4203 | mstime() - ri->role_reported_time > |
| 4204 | (ri->down_after_period+SENTINEL_INFO_PERIOD10000*2))) |
| 4205 | { |
| 4206 | /* Is subjectively down */ |
| 4207 | if ((ri->flags & SRI_S_DOWN(1<<3)) == 0) { |
| 4208 | sentinelEvent(LL_WARNING3,"+sdown",ri,"%@"); |
| 4209 | ri->s_down_since_time = mstime(); |
| 4210 | ri->flags |= SRI_S_DOWN(1<<3); |
| 4211 | } |
| 4212 | } else { |
| 4213 | /* Is subjectively up */ |
| 4214 | if (ri->flags & SRI_S_DOWN(1<<3)) { |
| 4215 | sentinelEvent(LL_WARNING3,"-sdown",ri,"%@"); |
| 4216 | ri->flags &= ~(SRI_S_DOWN(1<<3)|SRI_SCRIPT_KILL_SENT(1<<12)); |
| 4217 | } |
| 4218 | } |
| 4219 | } |
| 4220 | |
| 4221 | /* Is this instance down according to the configured quorum? |
| 4222 | * |
| 4223 | * Note that ODOWN is a weak quorum, it only means that enough Sentinels |
| 4224 | * reported in a given time range that the instance was not reachable. |
| 4225 | * However messages can be delayed so there are no strong guarantees about |
| 4226 | * N instances agreeing at the same time about the down state. */ |
| 4227 | void sentinelCheckObjectivelyDown(sentinelRedisInstance *master) { |
| 4228 | dictIterator *di; |
| 4229 | dictEntry *de; |
| 4230 | unsigned int quorum = 0, odown = 0; |
| 4231 | |
| 4232 | if (master->flags & SRI_S_DOWN(1<<3)) { |
| 4233 | /* Is down for enough sentinels? */ |
| 4234 | quorum = 1; /* the current sentinel. */ |
| 4235 | /* Count all the other sentinels. */ |
| 4236 | di = dictGetIterator(master->sentinels); |
| 4237 | while((de = dictNext(di)) != NULL((void*)0)) { |
| 4238 | sentinelRedisInstance *ri = dictGetVal(de)((de)->v.val); |
| 4239 | |
| 4240 | if (ri->flags & SRI_MASTER_DOWN(1<<5)) quorum++; |
| 4241 | } |
| 4242 | dictReleaseIterator(di); |
| 4243 | if (quorum >= master->quorum) odown = 1; |
| 4244 | } |
| 4245 | |
| 4246 | /* Set the flag accordingly to the outcome. */ |
| 4247 | if (odown) { |
| 4248 | if ((master->flags & SRI_O_DOWN(1<<4)) == 0) { |
| 4249 | sentinelEvent(LL_WARNING3,"+odown",master,"%@ #quorum %d/%d", |
| 4250 | quorum, master->quorum); |
| 4251 | master->flags |= SRI_O_DOWN(1<<4); |
| 4252 | master->o_down_since_time = mstime(); |
| 4253 | } |
| 4254 | } else { |
| 4255 | if (master->flags & SRI_O_DOWN(1<<4)) { |
| 4256 | sentinelEvent(LL_WARNING3,"-odown",master,"%@"); |
| 4257 | master->flags &= ~SRI_O_DOWN(1<<4); |
| 4258 | } |
| 4259 | } |
| 4260 | } |
| 4261 | |
| 4262 | /* Receive the SENTINEL is-master-down-by-addr reply, see the |
| 4263 | * sentinelAskMasterStateToOtherSentinels() function for more information. */ |
| 4264 | void sentinelReceiveIsMasterDownReply(redisAsyncContext *c, void *reply, void *privdata) { |
| 4265 | sentinelRedisInstance *ri = privdata; |
| 4266 | instanceLink *link = c->data; |
| 4267 | redisReply *r; |
| 4268 | |
| 4269 | if (!reply || !link) return; |
| 4270 | link->pending_commands--; |
| 4271 | r = reply; |
| 4272 | |
| 4273 | /* Ignore every error or unexpected reply. |
| 4274 | * Note that if the command returns an error for any reason we'll |
| 4275 | * end clearing the SRI_MASTER_DOWN flag for timeout anyway. */ |
| 4276 | if (r->type == REDIS_REPLY_ARRAY2 && r->elements == 3 && |
| 4277 | r->element[0]->type == REDIS_REPLY_INTEGER3 && |
| 4278 | r->element[1]->type == REDIS_REPLY_STRING1 && |
| 4279 | r->element[2]->type == REDIS_REPLY_INTEGER3) |
| 4280 | { |
| 4281 | ri->last_master_down_reply_time = mstime(); |
| 4282 | if (r->element[0]->integer == 1) { |
| 4283 | ri->flags |= SRI_MASTER_DOWN(1<<5); |
| 4284 | } else { |
| 4285 | ri->flags &= ~SRI_MASTER_DOWN(1<<5); |
| 4286 | } |
| 4287 | if (strcmp(r->element[1]->str,"*")) { |
| 4288 | /* If the runid in the reply is not "*" the Sentinel actually |
| 4289 | * replied with a vote. */ |
| 4290 | sdsfree(ri->leader); |
| 4291 | if ((long long)ri->leader_epoch != r->element[2]->integer) |
| 4292 | serverLog(LL_WARNING3, |
| 4293 | "%s voted for %s %llu", ri->name, |
| 4294 | r->element[1]->str, |
| 4295 | (unsigned long long) r->element[2]->integer); |
| 4296 | ri->leader = sdsnew(r->element[1]->str); |
| 4297 | ri->leader_epoch = r->element[2]->integer; |
| 4298 | } |
| 4299 | } |
| 4300 | } |
| 4301 | |
| 4302 | /* If we think the master is down, we start sending |
| 4303 | * SENTINEL IS-MASTER-DOWN-BY-ADDR requests to other sentinels |
| 4304 | * in order to get the replies that allow to reach the quorum |
| 4305 | * needed to mark the master in ODOWN state and trigger a failover. */ |
| 4306 | #define SENTINEL_ASK_FORCED(1<<0) (1<<0) |
| 4307 | void sentinelAskMasterStateToOtherSentinels(sentinelRedisInstance *master, int flags) { |
| 4308 | dictIterator *di; |
| 4309 | dictEntry *de; |
| 4310 | |
| 4311 | di = dictGetIterator(master->sentinels); |
| 4312 | while((de = dictNext(di)) != NULL((void*)0)) { |
| 4313 | sentinelRedisInstance *ri = dictGetVal(de)((de)->v.val); |
| 4314 | mstime_t elapsed = mstime() - ri->last_master_down_reply_time; |
| 4315 | char port[32]; |
| 4316 | int retval; |
| 4317 | |
| 4318 | /* If the master state from other sentinel is too old, we clear it. */ |
| 4319 | if (elapsed > SENTINEL_ASK_PERIOD1000*5) { |
| 4320 | ri->flags &= ~SRI_MASTER_DOWN(1<<5); |
| 4321 | sdsfree(ri->leader); |
| 4322 | ri->leader = NULL((void*)0); |
| 4323 | } |
| 4324 | |
| 4325 | /* Only ask if master is down to other sentinels if: |
| 4326 | * |
| 4327 | * 1) We believe it is down, or there is a failover in progress. |
| 4328 | * 2) Sentinel is connected. |
| 4329 | * 3) We did not receive the info within SENTINEL_ASK_PERIOD ms. */ |
| 4330 | if ((master->flags & SRI_S_DOWN(1<<3)) == 0) continue; |
| 4331 | if (ri->link->disconnected) continue; |
| 4332 | if (!(flags & SENTINEL_ASK_FORCED(1<<0)) && |
| 4333 | mstime() - ri->last_master_down_reply_time < SENTINEL_ASK_PERIOD1000) |
| 4334 | continue; |
| 4335 | |
| 4336 | /* Ask */ |
| 4337 | ll2string(port,sizeof(port),master->addr->port); |
| 4338 | retval = redisAsyncCommand(ri->link->cc, |
| 4339 | sentinelReceiveIsMasterDownReply, ri, |
| 4340 | "%s is-master-down-by-addr %s %s %llu %s", |
| 4341 | sentinelInstanceMapCommand(ri,"SENTINEL"), |
| 4342 | announceSentinelAddr(master->addr), port, |
| 4343 | sentinel.current_epoch, |
| 4344 | (master->failover_state > SENTINEL_FAILOVER_STATE_NONE0) ? |
| 4345 | sentinel.myid : "*"); |
| 4346 | if (retval == C_OK0) ri->link->pending_commands++; |
| 4347 | } |
| 4348 | dictReleaseIterator(di); |
| 4349 | } |
| 4350 | |
| 4351 | /* =============================== FAILOVER ================================= */ |
| 4352 | |
| 4353 | /* Crash because of user request via SENTINEL simulate-failure command. */ |
| 4354 | void sentinelSimFailureCrash(void) { |
| 4355 | serverLog(LL_WARNING3, |
| 4356 | "Sentinel CRASH because of SENTINEL simulate-failure"); |
| 4357 | exit(99); |
| 4358 | } |
| 4359 | |
| 4360 | /* Vote for the sentinel with 'req_runid' or return the old vote if already |
| 4361 | * voted for the specified 'req_epoch' or one greater. |
| 4362 | * |
| 4363 | * If a vote is not available returns NULL, otherwise return the Sentinel |
| 4364 | * runid and populate the leader_epoch with the epoch of the vote. */ |
| 4365 | char *sentinelVoteLeader(sentinelRedisInstance *master, uint64_t req_epoch, char *req_runid, uint64_t *leader_epoch) { |
| 4366 | if (req_epoch > sentinel.current_epoch) { |
| 4367 | sentinel.current_epoch = req_epoch; |
| 4368 | sentinelFlushConfig(); |
| 4369 | sentinelEvent(LL_WARNING3,"+new-epoch",master,"%llu", |
| 4370 | (unsigned long long) sentinel.current_epoch); |
| 4371 | } |
| 4372 | |
| 4373 | if (master->leader_epoch < req_epoch && sentinel.current_epoch <= req_epoch) |
| 4374 | { |
| 4375 | sdsfree(master->leader); |
| 4376 | master->leader = sdsnew(req_runid); |
| 4377 | master->leader_epoch = sentinel.current_epoch; |
| 4378 | sentinelFlushConfig(); |
| 4379 | sentinelEvent(LL_WARNING3,"+vote-for-leader",master,"%s %llu", |
| 4380 | master->leader, (unsigned long long) master->leader_epoch); |
| 4381 | /* If we did not voted for ourselves, set the master failover start |
| 4382 | * time to now, in order to force a delay before we can start a |
| 4383 | * failover for the same master. */ |
| 4384 | if (strcasecmp(master->leader,sentinel.myid)) |
| 4385 | master->failover_start_time = mstime()+rand()%SENTINEL_MAX_DESYNC1000; |
| 4386 | } |
| 4387 | |
| 4388 | *leader_epoch = master->leader_epoch; |
| 4389 | return master->leader ? sdsnew(master->leader) : NULL((void*)0); |
| 4390 | } |
| 4391 | |
| 4392 | struct sentinelLeader { |
| 4393 | char *runid; |
| 4394 | unsigned long votes; |
| 4395 | }; |
| 4396 | |
| 4397 | /* Helper function for sentinelGetLeader, increment the counter |
| 4398 | * relative to the specified runid. */ |
| 4399 | int sentinelLeaderIncr(dict *counters, char *runid) { |
| 4400 | dictEntry *existing, *de; |
| 4401 | uint64_t oldval; |
| 4402 | |
| 4403 | de = dictAddRaw(counters,runid,&existing); |
| 4404 | if (existing) { |
| 4405 | oldval = dictGetUnsignedIntegerVal(existing)((existing)->v.u64); |
| 4406 | dictSetUnsignedIntegerVal(existing,oldval+1)do { (existing)->v.u64 = oldval+1; } while(0); |
| 4407 | return oldval+1; |
| 4408 | } else { |
| 4409 | serverAssert(de != NULL)((de != ((void*)0))?(void)0 : (_serverAssert("de != NULL","sentinel.c" ,4409),__builtin_unreachable())); |
| 4410 | dictSetUnsignedIntegerVal(de,1)do { (de)->v.u64 = 1; } while(0); |
| 4411 | return 1; |
| 4412 | } |
| 4413 | } |
| 4414 | |
| 4415 | /* Scan all the Sentinels attached to this master to check if there |
| 4416 | * is a leader for the specified epoch. |
| 4417 | * |
| 4418 | * To be a leader for a given epoch, we should have the majority of |
| 4419 | * the Sentinels we know (ever seen since the last SENTINEL RESET) that |
| 4420 | * reported the same instance as leader for the same epoch. */ |
| 4421 | char *sentinelGetLeader(sentinelRedisInstance *master, uint64_t epoch) { |
| 4422 | dict *counters; |
| 4423 | dictIterator *di; |
| 4424 | dictEntry *de; |
| 4425 | unsigned int voters = 0, voters_quorum; |
| 4426 | char *myvote; |
| 4427 | char *winner = NULL((void*)0); |
| 4428 | uint64_t leader_epoch; |
| 4429 | uint64_t max_votes = 0; |
| 4430 | |
| 4431 | serverAssert(master->flags & (SRI_O_DOWN|SRI_FAILOVER_IN_PROGRESS))((master->flags & ((1<<4)|(1<<6)))?(void)0 : (_serverAssert("master->flags & (SRI_O_DOWN|SRI_FAILOVER_IN_PROGRESS)" ,"sentinel.c",4431),__builtin_unreachable())); |
| 4432 | counters = dictCreate(&leaderVotesDictType,NULL((void*)0)); |
| 4433 | |
| 4434 | voters = dictSize(master->sentinels)((master->sentinels)->ht[0].used+(master->sentinels) ->ht[1].used)+1; /* All the other sentinels and me.*/ |
| 4435 | |
| 4436 | /* Count other sentinels votes */ |
| 4437 | di = dictGetIterator(master->sentinels); |
| 4438 | while((de = dictNext(di)) != NULL((void*)0)) { |
| 4439 | sentinelRedisInstance *ri = dictGetVal(de)((de)->v.val); |
| 4440 | if (ri->leader != NULL((void*)0) && ri->leader_epoch == sentinel.current_epoch) |
| 4441 | sentinelLeaderIncr(counters,ri->leader); |
| 4442 | } |
| 4443 | dictReleaseIterator(di); |
| 4444 | |
| 4445 | /* Check what's the winner. For the winner to win, it needs two conditions: |
| 4446 | * 1) Absolute majority between voters (50% + 1). |
| 4447 | * 2) And anyway at least master->quorum votes. */ |
| 4448 | di = dictGetIterator(counters); |
| 4449 | while((de = dictNext(di)) != NULL((void*)0)) { |
| 4450 | uint64_t votes = dictGetUnsignedIntegerVal(de)((de)->v.u64); |
| 4451 | |
| 4452 | if (votes > max_votes) { |
| 4453 | max_votes = votes; |
| 4454 | winner = dictGetKey(de)((de)->key); |
| 4455 | } |
| 4456 | } |
| 4457 | dictReleaseIterator(di); |
| 4458 | |
| 4459 | /* Count this Sentinel vote: |
| 4460 | * if this Sentinel did not voted yet, either vote for the most |
| 4461 | * common voted sentinel, or for itself if no vote exists at all. */ |
| 4462 | if (winner) |
| 4463 | myvote = sentinelVoteLeader(master,epoch,winner,&leader_epoch); |
| 4464 | else |
| 4465 | myvote = sentinelVoteLeader(master,epoch,sentinel.myid,&leader_epoch); |
| 4466 | |
| 4467 | if (myvote && leader_epoch == epoch) { |
| 4468 | uint64_t votes = sentinelLeaderIncr(counters,myvote); |
| 4469 | |
| 4470 | if (votes > max_votes) { |
| 4471 | max_votes = votes; |
| 4472 | winner = myvote; |
| 4473 | } |
| 4474 | } |
| 4475 | |
| 4476 | voters_quorum = voters/2+1; |
| 4477 | if (winner && (max_votes < voters_quorum || max_votes < master->quorum)) |
| 4478 | winner = NULL((void*)0); |
| 4479 | |
| 4480 | winner = winner ? sdsnew(winner) : NULL((void*)0); |
| 4481 | sdsfree(myvote); |
| 4482 | dictRelease(counters); |
| 4483 | return winner; |
| 4484 | } |
| 4485 | |
| 4486 | /* Send SLAVEOF to the specified instance, always followed by a |
| 4487 | * CONFIG REWRITE command in order to store the new configuration on disk |
| 4488 | * when possible (that is, if the Redis instance is recent enough to support |
| 4489 | * config rewriting, and if the server was started with a configuration file). |
| 4490 | * |
| 4491 | * If Host is NULL the function sends "SLAVEOF NO ONE". |
| 4492 | * |
| 4493 | * The command returns C_OK if the SLAVEOF command was accepted for |
| 4494 | * (later) delivery otherwise C_ERR. The command replies are just |
| 4495 | * discarded. */ |
| 4496 | int sentinelSendSlaveOf(sentinelRedisInstance *ri, const sentinelAddr *addr) { |
| 4497 | char portstr[32]; |
| 4498 | const char *host; |
| 4499 | int retval; |
| 4500 | |
| 4501 | /* If host is NULL we send SLAVEOF NO ONE that will turn the instance |
| 4502 | * into a master. */ |
| 4503 | if (!addr) { |
| 4504 | host = "NO"; |
| 4505 | memcpy(portstr,"ONE",4); |
| 4506 | } else { |
| 4507 | host = announceSentinelAddr(addr); |
| 4508 | ll2string(portstr,sizeof(portstr),addr->port); |
| 4509 | } |
| 4510 | |
| 4511 | /* In order to send SLAVEOF in a safe way, we send a transaction performing |
| 4512 | * the following tasks: |
| 4513 | * 1) Reconfigure the instance according to the specified host/port params. |
| 4514 | * 2) Rewrite the configuration. |
| 4515 | * 3) Disconnect all clients (but this one sending the command) in order |
| 4516 | * to trigger the ask-master-on-reconnection protocol for connected |
| 4517 | * clients. |
| 4518 | * |
| 4519 | * Note that we don't check the replies returned by commands, since we |
| 4520 | * will observe instead the effects in the next INFO output. */ |
| 4521 | retval = redisAsyncCommand(ri->link->cc, |
| 4522 | sentinelDiscardReplyCallback, ri, "%s", |
| 4523 | sentinelInstanceMapCommand(ri,"MULTI")); |
| 4524 | if (retval == C_ERR-1) return retval; |
| 4525 | ri->link->pending_commands++; |
| 4526 | |
| 4527 | retval = redisAsyncCommand(ri->link->cc, |
| 4528 | sentinelDiscardReplyCallback, ri, "%s %s %s", |
| 4529 | sentinelInstanceMapCommand(ri,"SLAVEOF"), |
| 4530 | host, portstr); |
| 4531 | if (retval == C_ERR-1) return retval; |
| 4532 | ri->link->pending_commands++; |
| 4533 | |
| 4534 | retval = redisAsyncCommand(ri->link->cc, |
| 4535 | sentinelDiscardReplyCallback, ri, "%s REWRITE", |
| 4536 | sentinelInstanceMapCommand(ri,"CONFIG")); |
| 4537 | if (retval == C_ERR-1) return retval; |
| 4538 | ri->link->pending_commands++; |
| 4539 | |
| 4540 | /* CLIENT KILL TYPE <type> is only supported starting from Redis 2.8.12, |
| 4541 | * however sending it to an instance not understanding this command is not |
| 4542 | * an issue because CLIENT is variadic command, so Redis will not |
| 4543 | * recognized as a syntax error, and the transaction will not fail (but |
| 4544 | * only the unsupported command will fail). */ |
| 4545 | for (int type = 0; type < 2; type++) { |
| 4546 | retval = redisAsyncCommand(ri->link->cc, |
| 4547 | sentinelDiscardReplyCallback, ri, "%s KILL TYPE %s", |
| 4548 | sentinelInstanceMapCommand(ri,"CLIENT"), |
| 4549 | type == 0 ? "normal" : "pubsub"); |
| 4550 | if (retval == C_ERR-1) return retval; |
| 4551 | ri->link->pending_commands++; |
| 4552 | } |
| 4553 | |
| 4554 | retval = redisAsyncCommand(ri->link->cc, |
| 4555 | sentinelDiscardReplyCallback, ri, "%s", |
| 4556 | sentinelInstanceMapCommand(ri,"EXEC")); |
| 4557 | if (retval == C_ERR-1) return retval; |
| 4558 | ri->link->pending_commands++; |
| 4559 | |
| 4560 | return C_OK0; |
| 4561 | } |
| 4562 | |
| 4563 | /* Setup the master state to start a failover. */ |
| 4564 | void sentinelStartFailover(sentinelRedisInstance *master) { |
| 4565 | serverAssert(master->flags & SRI_MASTER)((master->flags & (1<<0))?(void)0 : (_serverAssert ("master->flags & SRI_MASTER","sentinel.c",4565),__builtin_unreachable ())); |
| 4566 | |
| 4567 | master->failover_state = SENTINEL_FAILOVER_STATE_WAIT_START1; |
| 4568 | master->flags |= SRI_FAILOVER_IN_PROGRESS(1<<6); |
| 4569 | master->failover_epoch = ++sentinel.current_epoch; |
| 4570 | sentinelEvent(LL_WARNING3,"+new-epoch",master,"%llu", |
| 4571 | (unsigned long long) sentinel.current_epoch); |
| 4572 | sentinelEvent(LL_WARNING3,"+try-failover",master,"%@"); |
| 4573 | master->failover_start_time = mstime()+rand()%SENTINEL_MAX_DESYNC1000; |
| 4574 | master->failover_state_change_time = mstime(); |
| 4575 | } |
| 4576 | |
| 4577 | /* This function checks if there are the conditions to start the failover, |
| 4578 | * that is: |
| 4579 | * |
| 4580 | * 1) Master must be in ODOWN condition. |
| 4581 | * 2) No failover already in progress. |
| 4582 | * 3) No failover already attempted recently. |
| 4583 | * |
| 4584 | * We still don't know if we'll win the election so it is possible that we |
| 4585 | * start the failover but that we'll not be able to act. |
| 4586 | * |
| 4587 | * Return non-zero if a failover was started. */ |
| 4588 | int sentinelStartFailoverIfNeeded(sentinelRedisInstance *master) { |
| 4589 | /* We can't failover if the master is not in O_DOWN state. */ |
| 4590 | if (!(master->flags & SRI_O_DOWN(1<<4))) return 0; |
| 4591 | |
| 4592 | /* Failover already in progress? */ |
| 4593 | if (master->flags & SRI_FAILOVER_IN_PROGRESS(1<<6)) return 0; |
| 4594 | |
| 4595 | /* Last failover attempt started too little time ago? */ |
| 4596 | if (mstime() - master->failover_start_time < |
| 4597 | master->failover_timeout*2) |
| 4598 | { |
| 4599 | if (master->failover_delay_logged != master->failover_start_time) { |
| 4600 | time_t clock = (master->failover_start_time + |
| 4601 | master->failover_timeout*2) / 1000; |
| 4602 | char ctimebuf[26]; |
| 4603 | |
| 4604 | ctime_r(&clock,ctimebuf); |
| 4605 | ctimebuf[24] = '\0'; /* Remove newline. */ |
| 4606 | master->failover_delay_logged = master->failover_start_time; |
| 4607 | serverLog(LL_WARNING3, |
| 4608 | "Next failover delay: I will not start a failover before %s", |
| 4609 | ctimebuf); |
| 4610 | } |
| 4611 | return 0; |
| 4612 | } |
| 4613 | |
| 4614 | sentinelStartFailover(master); |
| 4615 | return 1; |
| 4616 | } |
| 4617 | |
| 4618 | /* Select a suitable slave to promote. The current algorithm only uses |
| 4619 | * the following parameters: |
| 4620 | * |
| 4621 | * 1) None of the following conditions: S_DOWN, O_DOWN, DISCONNECTED. |
| 4622 | * 2) Last time the slave replied to ping no more than 5 times the PING period. |
| 4623 | * 3) info_refresh not older than 3 times the INFO refresh period. |
| 4624 | * 4) master_link_down_time no more than: |
| 4625 | * (now - master->s_down_since_time) + (master->down_after_period * 10). |
| 4626 | * Basically since the master is down from our POV, the slave reports |
| 4627 | * to be disconnected no more than 10 times the configured down-after-period. |
| 4628 | * This is pretty much black magic but the idea is, the master was not |
| 4629 | * available so the slave may be lagging, but not over a certain time. |
| 4630 | * Anyway we'll select the best slave according to replication offset. |
| 4631 | * 5) Slave priority can't be zero, otherwise the slave is discarded. |
| 4632 | * |
| 4633 | * Among all the slaves matching the above conditions we select the slave |
| 4634 | * with, in order of sorting key: |
| 4635 | * |
| 4636 | * - lower slave_priority. |
| 4637 | * - bigger processed replication offset. |
| 4638 | * - lexicographically smaller runid. |
| 4639 | * |
| 4640 | * Basically if runid is the same, the slave that processed more commands |
| 4641 | * from the master is selected. |
| 4642 | * |
| 4643 | * The function returns the pointer to the selected slave, otherwise |
| 4644 | * NULL if no suitable slave was found. |
| 4645 | */ |
| 4646 | |
| 4647 | /* Helper for sentinelSelectSlave(). This is used by qsort() in order to |
| 4648 | * sort suitable slaves in a "better first" order, to take the first of |
| 4649 | * the list. */ |
| 4650 | int compareSlavesForPromotion(const void *a, const void *b) { |
| 4651 | sentinelRedisInstance **sa = (sentinelRedisInstance **)a, |
| 4652 | **sb = (sentinelRedisInstance **)b; |
| 4653 | char *sa_runid, *sb_runid; |
| 4654 | |
| 4655 | if ((*sa)->slave_priority != (*sb)->slave_priority) |
| 4656 | return (*sa)->slave_priority - (*sb)->slave_priority; |
| 4657 | |
| 4658 | /* If priority is the same, select the slave with greater replication |
| 4659 | * offset (processed more data from the master). */ |
| 4660 | if ((*sa)->slave_repl_offset > (*sb)->slave_repl_offset) { |
| 4661 | return -1; /* a < b */ |
| 4662 | } else if ((*sa)->slave_repl_offset < (*sb)->slave_repl_offset) { |
| 4663 | return 1; /* a > b */ |
| 4664 | } |
| 4665 | |
| 4666 | /* If the replication offset is the same select the slave with that has |
| 4667 | * the lexicographically smaller runid. Note that we try to handle runid |
| 4668 | * == NULL as there are old Redis versions that don't publish runid in |
| 4669 | * INFO. A NULL runid is considered bigger than any other runid. */ |
| 4670 | sa_runid = (*sa)->runid; |
| 4671 | sb_runid = (*sb)->runid; |
| 4672 | if (sa_runid == NULL((void*)0) && sb_runid == NULL((void*)0)) return 0; |
| 4673 | else if (sa_runid == NULL((void*)0)) return 1; /* a > b */ |
| 4674 | else if (sb_runid == NULL((void*)0)) return -1; /* a < b */ |
| 4675 | return strcasecmp(sa_runid, sb_runid); |
| 4676 | } |
| 4677 | |
| 4678 | sentinelRedisInstance *sentinelSelectSlave(sentinelRedisInstance *master) { |
| 4679 | sentinelRedisInstance **instance = |
| 4680 | zmalloc(sizeof(instance[0])*dictSize(master->slaves)((master->slaves)->ht[0].used+(master->slaves)->ht [1].used)); |
| 4681 | sentinelRedisInstance *selected = NULL((void*)0); |
| 4682 | int instances = 0; |
| 4683 | dictIterator *di; |
| 4684 | dictEntry *de; |
| 4685 | mstime_t max_master_down_time = 0; |
| 4686 | |
| 4687 | if (master->flags & SRI_S_DOWN(1<<3)) |
| 4688 | max_master_down_time += mstime() - master->s_down_since_time; |
| 4689 | max_master_down_time += master->down_after_period * 10; |
| 4690 | |
| 4691 | di = dictGetIterator(master->slaves); |
| 4692 | while((de = dictNext(di)) != NULL((void*)0)) { |
| 4693 | sentinelRedisInstance *slave = dictGetVal(de)((de)->v.val); |
| 4694 | mstime_t info_validity_time; |
| 4695 | |
| 4696 | if (slave->flags & (SRI_S_DOWN(1<<3)|SRI_O_DOWN(1<<4))) continue; |
| 4697 | if (slave->link->disconnected) continue; |
| 4698 | if (mstime() - slave->link->last_avail_time > SENTINEL_PING_PERIOD1000*5) continue; |
| 4699 | if (slave->slave_priority == 0) continue; |
| 4700 | |
| 4701 | /* If the master is in SDOWN state we get INFO for slaves every second. |
| 4702 | * Otherwise we get it with the usual period so we need to account for |
| 4703 | * a larger delay. */ |
| 4704 | if (master->flags & SRI_S_DOWN(1<<3)) |
| 4705 | info_validity_time = SENTINEL_PING_PERIOD1000*5; |
| 4706 | else |
| 4707 | info_validity_time = SENTINEL_INFO_PERIOD10000*3; |
| 4708 | if (mstime() - slave->info_refresh > info_validity_time) continue; |
| 4709 | if (slave->master_link_down_time > max_master_down_time) continue; |
| 4710 | instance[instances++] = slave; |
| 4711 | } |
| 4712 | dictReleaseIterator(di); |
| 4713 | if (instances) { |
| 4714 | qsort(instance,instances,sizeof(sentinelRedisInstance*), |
| 4715 | compareSlavesForPromotion); |
| 4716 | selected = instance[0]; |
| 4717 | } |
| 4718 | zfree(instance); |
| 4719 | return selected; |
| 4720 | } |
| 4721 | |
| 4722 | /* ---------------- Failover state machine implementation ------------------- */ |
| 4723 | void sentinelFailoverWaitStart(sentinelRedisInstance *ri) { |
| 4724 | char *leader; |
| 4725 | int isleader; |
| 4726 | |
| 4727 | /* Check if we are the leader for the failover epoch. */ |
| 4728 | leader = sentinelGetLeader(ri, ri->failover_epoch); |
| 4729 | isleader = leader && strcasecmp(leader,sentinel.myid) == 0; |
| 4730 | sdsfree(leader); |
| 4731 | |
| 4732 | /* If I'm not the leader, and it is not a forced failover via |
| 4733 | * SENTINEL FAILOVER, then I can't continue with the failover. */ |
| 4734 | if (!isleader && !(ri->flags & SRI_FORCE_FAILOVER(1<<11))) { |
| 4735 | int election_timeout = SENTINEL_ELECTION_TIMEOUT10000; |
| 4736 | |
| 4737 | /* The election timeout is the MIN between SENTINEL_ELECTION_TIMEOUT |
| 4738 | * and the configured failover timeout. */ |
| 4739 | if (election_timeout > ri->failover_timeout) |
| 4740 | election_timeout = ri->failover_timeout; |
| 4741 | /* Abort the failover if I'm not the leader after some time. */ |
| 4742 | if (mstime() - ri->failover_start_time > election_timeout) { |
| 4743 | sentinelEvent(LL_WARNING3,"-failover-abort-not-elected",ri,"%@"); |
| 4744 | sentinelAbortFailover(ri); |
| 4745 | } |
| 4746 | return; |
| 4747 | } |
| 4748 | sentinelEvent(LL_WARNING3,"+elected-leader",ri,"%@"); |
| 4749 | if (sentinel.simfailure_flags & SENTINEL_SIMFAILURE_CRASH_AFTER_ELECTION(1<<0)) |
| 4750 | sentinelSimFailureCrash(); |
| 4751 | ri->failover_state = SENTINEL_FAILOVER_STATE_SELECT_SLAVE2; |
| 4752 | ri->failover_state_change_time = mstime(); |
| 4753 | sentinelEvent(LL_WARNING3,"+failover-state-select-slave",ri,"%@"); |
| 4754 | } |
| 4755 | |
| 4756 | void sentinelFailoverSelectSlave(sentinelRedisInstance *ri) { |
| 4757 | sentinelRedisInstance *slave = sentinelSelectSlave(ri); |
| 4758 | |
| 4759 | /* We don't handle the timeout in this state as the function aborts |
| 4760 | * the failover or go forward in the next state. */ |
| 4761 | if (slave == NULL((void*)0)) { |
| 4762 | sentinelEvent(LL_WARNING3,"-failover-abort-no-good-slave",ri,"%@"); |
| 4763 | sentinelAbortFailover(ri); |
| 4764 | } else { |
| 4765 | sentinelEvent(LL_WARNING3,"+selected-slave",slave,"%@"); |
| 4766 | slave->flags |= SRI_PROMOTED(1<<7); |
| 4767 | ri->promoted_slave = slave; |
| 4768 | ri->failover_state = SENTINEL_FAILOVER_STATE_SEND_SLAVEOF_NOONE3; |
| 4769 | ri->failover_state_change_time = mstime(); |
| 4770 | sentinelEvent(LL_NOTICE2,"+failover-state-send-slaveof-noone", |
| 4771 | slave, "%@"); |
| 4772 | } |
| 4773 | } |
| 4774 | |
| 4775 | void sentinelFailoverSendSlaveOfNoOne(sentinelRedisInstance *ri) { |
| 4776 | int retval; |
| 4777 | |
| 4778 | /* We can't send the command to the promoted slave if it is now |
| 4779 | * disconnected. Retry again and again with this state until the timeout |
| 4780 | * is reached, then abort the failover. */ |
| 4781 | if (ri->promoted_slave->link->disconnected) { |
| 4782 | if (mstime() - ri->failover_state_change_time > ri->failover_timeout) { |
| 4783 | sentinelEvent(LL_WARNING3,"-failover-abort-slave-timeout",ri,"%@"); |
| 4784 | sentinelAbortFailover(ri); |
| 4785 | } |
| 4786 | return; |
| 4787 | } |
| 4788 | |
| 4789 | /* Send SLAVEOF NO ONE command to turn the slave into a master. |
| 4790 | * We actually register a generic callback for this command as we don't |
| 4791 | * really care about the reply. We check if it worked indirectly observing |
| 4792 | * if INFO returns a different role (master instead of slave). */ |
| 4793 | retval = sentinelSendSlaveOf(ri->promoted_slave,NULL((void*)0)); |
| 4794 | if (retval != C_OK0) return; |
| 4795 | sentinelEvent(LL_NOTICE2, "+failover-state-wait-promotion", |
| 4796 | ri->promoted_slave,"%@"); |
| 4797 | ri->failover_state = SENTINEL_FAILOVER_STATE_WAIT_PROMOTION4; |
| 4798 | ri->failover_state_change_time = mstime(); |
| 4799 | } |
| 4800 | |
| 4801 | /* We actually wait for promotion indirectly checking with INFO when the |
| 4802 | * slave turns into a master. */ |
| 4803 | void sentinelFailoverWaitPromotion(sentinelRedisInstance *ri) { |
| 4804 | /* Just handle the timeout. Switching to the next state is handled |
| 4805 | * by the function parsing the INFO command of the promoted slave. */ |
| 4806 | if (mstime() - ri->failover_state_change_time > ri->failover_timeout) { |
| 4807 | sentinelEvent(LL_WARNING3,"-failover-abort-slave-timeout",ri,"%@"); |
| 4808 | sentinelAbortFailover(ri); |
| 4809 | } |
| 4810 | } |
| 4811 | |
| 4812 | void sentinelFailoverDetectEnd(sentinelRedisInstance *master) { |
| 4813 | int not_reconfigured = 0, timeout = 0; |
| 4814 | dictIterator *di; |
| 4815 | dictEntry *de; |
| 4816 | mstime_t elapsed = mstime() - master->failover_state_change_time; |
| 4817 | |
| 4818 | /* We can't consider failover finished if the promoted slave is |
| 4819 | * not reachable. */ |
| 4820 | if (master->promoted_slave == NULL((void*)0) || |
| 4821 | master->promoted_slave->flags & SRI_S_DOWN(1<<3)) return; |
| 4822 | |
| 4823 | /* The failover terminates once all the reachable slaves are properly |
| 4824 | * configured. */ |
| 4825 | di = dictGetIterator(master->slaves); |
| 4826 | while((de = dictNext(di)) != NULL((void*)0)) { |
| 4827 | sentinelRedisInstance *slave = dictGetVal(de)((de)->v.val); |
| 4828 | |
| 4829 | if (slave->flags & (SRI_PROMOTED(1<<7)|SRI_RECONF_DONE(1<<10))) continue; |
| 4830 | if (slave->flags & SRI_S_DOWN(1<<3)) continue; |
| 4831 | not_reconfigured++; |
| 4832 | } |
| 4833 | dictReleaseIterator(di); |
| 4834 | |
| 4835 | /* Force end of failover on timeout. */ |
| 4836 | if (elapsed > master->failover_timeout) { |
| 4837 | not_reconfigured = 0; |
| 4838 | timeout = 1; |
| 4839 | sentinelEvent(LL_WARNING3,"+failover-end-for-timeout",master,"%@"); |
| 4840 | } |
| 4841 | |
| 4842 | if (not_reconfigured == 0) { |
| 4843 | sentinelEvent(LL_WARNING3,"+failover-end",master,"%@"); |
| 4844 | master->failover_state = SENTINEL_FAILOVER_STATE_UPDATE_CONFIG6; |
| 4845 | master->failover_state_change_time = mstime(); |
| 4846 | } |
| 4847 | |
| 4848 | /* If I'm the leader it is a good idea to send a best effort SLAVEOF |
| 4849 | * command to all the slaves still not reconfigured to replicate with |
| 4850 | * the new master. */ |
| 4851 | if (timeout) { |
| 4852 | dictIterator *di; |
| 4853 | dictEntry *de; |
| 4854 | |
| 4855 | di = dictGetIterator(master->slaves); |
| 4856 | while((de = dictNext(di)) != NULL((void*)0)) { |
| 4857 | sentinelRedisInstance *slave = dictGetVal(de)((de)->v.val); |
| 4858 | int retval; |
| 4859 | |
| 4860 | if (slave->flags & (SRI_PROMOTED(1<<7)|SRI_RECONF_DONE(1<<10)|SRI_RECONF_SENT(1<<8))) continue; |
| 4861 | if (slave->link->disconnected) continue; |
| 4862 | |
| 4863 | retval = sentinelSendSlaveOf(slave,master->promoted_slave->addr); |
| 4864 | if (retval == C_OK0) { |
| 4865 | sentinelEvent(LL_NOTICE2,"+slave-reconf-sent-be",slave,"%@"); |
| 4866 | slave->flags |= SRI_RECONF_SENT(1<<8); |
| 4867 | } |
| 4868 | } |
| 4869 | dictReleaseIterator(di); |
| 4870 | } |
| 4871 | } |
| 4872 | |
| 4873 | /* Send SLAVE OF <new master address> to all the remaining slaves that |
| 4874 | * still don't appear to have the configuration updated. */ |
| 4875 | void sentinelFailoverReconfNextSlave(sentinelRedisInstance *master) { |
| 4876 | dictIterator *di; |
| 4877 | dictEntry *de; |
| 4878 | int in_progress = 0; |
| 4879 | |
| 4880 | di = dictGetIterator(master->slaves); |
| 4881 | while((de = dictNext(di)) != NULL((void*)0)) { |
| 4882 | sentinelRedisInstance *slave = dictGetVal(de)((de)->v.val); |
| 4883 | |
| 4884 | if (slave->flags & (SRI_RECONF_SENT(1<<8)|SRI_RECONF_INPROG(1<<9))) |
| 4885 | in_progress++; |
| 4886 | } |
| 4887 | dictReleaseIterator(di); |
| 4888 | |
| 4889 | di = dictGetIterator(master->slaves); |
| 4890 | while(in_progress < master->parallel_syncs && |
| 4891 | (de = dictNext(di)) != NULL((void*)0)) |
| 4892 | { |
| 4893 | sentinelRedisInstance *slave = dictGetVal(de)((de)->v.val); |
| 4894 | int retval; |
| 4895 | |
| 4896 | /* Skip the promoted slave, and already configured slaves. */ |
| 4897 | if (slave->flags & (SRI_PROMOTED(1<<7)|SRI_RECONF_DONE(1<<10))) continue; |
| 4898 | |
| 4899 | /* If too much time elapsed without the slave moving forward to |
| 4900 | * the next state, consider it reconfigured even if it is not. |
| 4901 | * Sentinels will detect the slave as misconfigured and fix its |
| 4902 | * configuration later. */ |
| 4903 | if ((slave->flags & SRI_RECONF_SENT(1<<8)) && |
| 4904 | (mstime() - slave->slave_reconf_sent_time) > |
| 4905 | SENTINEL_SLAVE_RECONF_TIMEOUT10000) |
| 4906 | { |
| 4907 | sentinelEvent(LL_NOTICE2,"-slave-reconf-sent-timeout",slave,"%@"); |
| 4908 | slave->flags &= ~SRI_RECONF_SENT(1<<8); |
| 4909 | slave->flags |= SRI_RECONF_DONE(1<<10); |
| 4910 | } |
| 4911 | |
| 4912 | /* Nothing to do for instances that are disconnected or already |
| 4913 | * in RECONF_SENT state. */ |
| 4914 | if (slave->flags & (SRI_RECONF_SENT(1<<8)|SRI_RECONF_INPROG(1<<9))) continue; |
| 4915 | if (slave->link->disconnected) continue; |
| 4916 | |
| 4917 | /* Send SLAVEOF <new master>. */ |
| 4918 | retval = sentinelSendSlaveOf(slave,master->promoted_slave->addr); |
| 4919 | if (retval == C_OK0) { |
| 4920 | slave->flags |= SRI_RECONF_SENT(1<<8); |
| 4921 | slave->slave_reconf_sent_time = mstime(); |
| 4922 | sentinelEvent(LL_NOTICE2,"+slave-reconf-sent",slave,"%@"); |
| 4923 | in_progress++; |
| 4924 | } |
| 4925 | } |
| 4926 | dictReleaseIterator(di); |
| 4927 | |
| 4928 | /* Check if all the slaves are reconfigured and handle timeout. */ |
| 4929 | sentinelFailoverDetectEnd(master); |
| 4930 | } |
| 4931 | |
| 4932 | /* This function is called when the slave is in |
| 4933 | * SENTINEL_FAILOVER_STATE_UPDATE_CONFIG state. In this state we need |
| 4934 | * to remove it from the master table and add the promoted slave instead. */ |
| 4935 | void sentinelFailoverSwitchToPromotedSlave(sentinelRedisInstance *master) { |
| 4936 | sentinelRedisInstance *ref = master->promoted_slave ? |
| 4937 | master->promoted_slave : master; |
| 4938 | |
| 4939 | sentinelEvent(LL_WARNING3,"+switch-master",master,"%s %s %d %s %d", |
| 4940 | master->name, announceSentinelAddr(master->addr), master->addr->port, |
| 4941 | announceSentinelAddr(ref->addr), ref->addr->port); |
| 4942 | |
| 4943 | sentinelResetMasterAndChangeAddress(master,ref->addr->hostname,ref->addr->port); |
| 4944 | } |
| 4945 | |
| 4946 | void sentinelFailoverStateMachine(sentinelRedisInstance *ri) { |
| 4947 | serverAssert(ri->flags & SRI_MASTER)((ri->flags & (1<<0))?(void)0 : (_serverAssert("ri->flags & SRI_MASTER" ,"sentinel.c",4947),__builtin_unreachable())); |
| 4948 | |
| 4949 | if (!(ri->flags & SRI_FAILOVER_IN_PROGRESS(1<<6))) return; |
| 4950 | |
| 4951 | switch(ri->failover_state) { |
| 4952 | case SENTINEL_FAILOVER_STATE_WAIT_START1: |
| 4953 | sentinelFailoverWaitStart(ri); |
| 4954 | break; |
| 4955 | case SENTINEL_FAILOVER_STATE_SELECT_SLAVE2: |
| 4956 | sentinelFailoverSelectSlave(ri); |
| 4957 | break; |
| 4958 | case SENTINEL_FAILOVER_STATE_SEND_SLAVEOF_NOONE3: |
| 4959 | sentinelFailoverSendSlaveOfNoOne(ri); |
| 4960 | break; |
| 4961 | case SENTINEL_FAILOVER_STATE_WAIT_PROMOTION4: |
| 4962 | sentinelFailoverWaitPromotion(ri); |
| 4963 | break; |
| 4964 | case SENTINEL_FAILOVER_STATE_RECONF_SLAVES5: |
| 4965 | sentinelFailoverReconfNextSlave(ri); |
| 4966 | break; |
| 4967 | } |
| 4968 | } |
| 4969 | |
| 4970 | /* Abort a failover in progress: |
| 4971 | * |
| 4972 | * This function can only be called before the promoted slave acknowledged |
| 4973 | * the slave -> master switch. Otherwise the failover can't be aborted and |
| 4974 | * will reach its end (possibly by timeout). */ |
| 4975 | void sentinelAbortFailover(sentinelRedisInstance *ri) { |
| 4976 | serverAssert(ri->flags & SRI_FAILOVER_IN_PROGRESS)((ri->flags & (1<<6))?(void)0 : (_serverAssert("ri->flags & SRI_FAILOVER_IN_PROGRESS" ,"sentinel.c",4976),__builtin_unreachable())); |
| 4977 | serverAssert(ri->failover_state <= SENTINEL_FAILOVER_STATE_WAIT_PROMOTION)((ri->failover_state <= 4)?(void)0 : (_serverAssert("ri->failover_state <= SENTINEL_FAILOVER_STATE_WAIT_PROMOTION" ,"sentinel.c",4977),__builtin_unreachable())); |
| 4978 | |
| 4979 | ri->flags &= ~(SRI_FAILOVER_IN_PROGRESS(1<<6)|SRI_FORCE_FAILOVER(1<<11)); |
| 4980 | ri->failover_state = SENTINEL_FAILOVER_STATE_NONE0; |
| 4981 | ri->failover_state_change_time = mstime(); |
| 4982 | if (ri->promoted_slave) { |
| 4983 | ri->promoted_slave->flags &= ~SRI_PROMOTED(1<<7); |
| 4984 | ri->promoted_slave = NULL((void*)0); |
| 4985 | } |
| 4986 | } |
| 4987 | |
| 4988 | /* ======================== SENTINEL timer handler ========================== |
| 4989 | * This is the "main" our Sentinel, being sentinel completely non blocking |
| 4990 | * in design. The function is called every second. |
| 4991 | * -------------------------------------------------------------------------- */ |
| 4992 | |
| 4993 | /* Perform scheduled operations for the specified Redis instance. */ |
| 4994 | void sentinelHandleRedisInstance(sentinelRedisInstance *ri) { |
| 4995 | /* ========== MONITORING HALF ============ */ |
| 4996 | /* Every kind of instance */ |
| 4997 | sentinelReconnectInstance(ri); |
| 4998 | sentinelSendPeriodicCommands(ri); |
| 4999 | |
| 5000 | /* ============== ACTING HALF ============= */ |
| 5001 | /* We don't proceed with the acting half if we are in TILT mode. |
| 5002 | * TILT happens when we find something odd with the time, like a |
| 5003 | * sudden change in the clock. */ |
| 5004 | if (sentinel.tilt) { |
| 5005 | if (mstime()-sentinel.tilt_start_time < SENTINEL_TILT_PERIOD(1000*30)) return; |
| 5006 | sentinel.tilt = 0; |
| 5007 | sentinelEvent(LL_WARNING3,"-tilt",NULL((void*)0),"#tilt mode exited"); |
| 5008 | } |
| 5009 | |
| 5010 | /* Every kind of instance */ |
| 5011 | sentinelCheckSubjectivelyDown(ri); |
| 5012 | |
| 5013 | /* Masters and slaves */ |
| 5014 | if (ri->flags & (SRI_MASTER(1<<0)|SRI_SLAVE(1<<1))) { |
| 5015 | /* Nothing so far. */ |
| 5016 | } |
| 5017 | |
| 5018 | /* Only masters */ |
| 5019 | if (ri->flags & SRI_MASTER(1<<0)) { |
| 5020 | sentinelCheckObjectivelyDown(ri); |
| 5021 | if (sentinelStartFailoverIfNeeded(ri)) |
| 5022 | sentinelAskMasterStateToOtherSentinels(ri,SENTINEL_ASK_FORCED(1<<0)); |
| 5023 | sentinelFailoverStateMachine(ri); |
| 5024 | sentinelAskMasterStateToOtherSentinels(ri,SENTINEL_NO_FLAGS0); |
| 5025 | } |
| 5026 | } |
| 5027 | |
| 5028 | /* Perform scheduled operations for all the instances in the dictionary. |
| 5029 | * Recursively call the function against dictionaries of slaves. */ |
| 5030 | void sentinelHandleDictOfRedisInstances(dict *instances) { |
| 5031 | dictIterator *di; |
| 5032 | dictEntry *de; |
| 5033 | sentinelRedisInstance *switch_to_promoted = NULL((void*)0); |
| 5034 | |
| 5035 | /* There are a number of things we need to perform against every master. */ |
| 5036 | di = dictGetIterator(instances); |
| 5037 | while((de = dictNext(di)) != NULL((void*)0)) { |
| 5038 | sentinelRedisInstance *ri = dictGetVal(de)((de)->v.val); |
| 5039 | |
| 5040 | sentinelHandleRedisInstance(ri); |
| 5041 | if (ri->flags & SRI_MASTER(1<<0)) { |
| 5042 | sentinelHandleDictOfRedisInstances(ri->slaves); |
| 5043 | sentinelHandleDictOfRedisInstances(ri->sentinels); |
| 5044 | if (ri->failover_state == SENTINEL_FAILOVER_STATE_UPDATE_CONFIG6) { |
| 5045 | switch_to_promoted = ri; |
| 5046 | } |
| 5047 | } |
| 5048 | } |
| 5049 | if (switch_to_promoted) |
| 5050 | sentinelFailoverSwitchToPromotedSlave(switch_to_promoted); |
| 5051 | dictReleaseIterator(di); |
| 5052 | } |
| 5053 | |
| 5054 | /* This function checks if we need to enter the TITL mode. |
| 5055 | * |
| 5056 | * The TILT mode is entered if we detect that between two invocations of the |
| 5057 | * timer interrupt, a negative amount of time, or too much time has passed. |
| 5058 | * Note that we expect that more or less just 100 milliseconds will pass |
| 5059 | * if everything is fine. However we'll see a negative number or a |
| 5060 | * difference bigger than SENTINEL_TILT_TRIGGER milliseconds if one of the |
| 5061 | * following conditions happen: |
| 5062 | * |
| 5063 | * 1) The Sentinel process for some time is blocked, for every kind of |
| 5064 | * random reason: the load is huge, the computer was frozen for some time |
| 5065 | * in I/O or alike, the process was stopped by a signal. Everything. |
| 5066 | * 2) The system clock was altered significantly. |
| 5067 | * |
| 5068 | * Under both this conditions we'll see everything as timed out and failing |
| 5069 | * without good reasons. Instead we enter the TILT mode and wait |
| 5070 | * for SENTINEL_TILT_PERIOD to elapse before starting to act again. |
| 5071 | * |
| 5072 | * During TILT time we still collect information, we just do not act. */ |
| 5073 | void sentinelCheckTiltCondition(void) { |
| 5074 | mstime_t now = mstime(); |
| 5075 | mstime_t delta = now - sentinel.previous_time; |
| 5076 | |
| 5077 | if (delta < 0 || delta > SENTINEL_TILT_TRIGGER2000) { |
| 5078 | sentinel.tilt = 1; |
| 5079 | sentinel.tilt_start_time = mstime(); |
| 5080 | sentinelEvent(LL_WARNING3,"+tilt",NULL((void*)0),"#tilt mode entered"); |
| 5081 | } |
| 5082 | sentinel.previous_time = mstime(); |
| 5083 | } |
| 5084 | |
| 5085 | void sentinelTimer(void) { |
| 5086 | sentinelCheckTiltCondition(); |
| 5087 | sentinelHandleDictOfRedisInstances(sentinel.masters); |
| 5088 | sentinelRunPendingScripts(); |
| 5089 | sentinelCollectTerminatedScripts(); |
| 5090 | sentinelKillTimedoutScripts(); |
| 5091 | |
| 5092 | /* We continuously change the frequency of the Redis "timer interrupt" |
| 5093 | * in order to desynchronize every Sentinel from every other. |
| 5094 | * This non-determinism avoids that Sentinels started at the same time |
| 5095 | * exactly continue to stay synchronized asking to be voted at the |
| 5096 | * same time again and again (resulting in nobody likely winning the |
| 5097 | * election because of split brain voting). */ |
| 5098 | server.hz = CONFIG_DEFAULT_HZ10 + rand() % CONFIG_DEFAULT_HZ10; |
| 5099 | } |