]>
Commit | Line | Data |
---|---|---|
120ba392 | 1 | /* Redis Sentinel implementation |
2 | * ----------------------------- | |
3 | * | |
4 | * Copyright (c) 2009-2012, Salvatore Sanfilippo <antirez at gmail dot com> | |
5 | * All rights reserved. | |
6 | * | |
7 | * Redistribution and use in source and binary forms, with or without | |
8 | * modification, are permitted provided that the following conditions are met: | |
9 | * | |
10 | * * Redistributions of source code must retain the above copyright notice, | |
11 | * this list of conditions and the following disclaimer. | |
12 | * * Redistributions in binary form must reproduce the above copyright | |
13 | * notice, this list of conditions and the following disclaimer in the | |
14 | * documentation and/or other materials provided with the distribution. | |
15 | * * Neither the name of Redis nor the names of its contributors may be used | |
16 | * to endorse or promote products derived from this software without | |
17 | * specific prior written permission. | |
18 | * | |
19 | * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" | |
20 | * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE | |
21 | * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE | |
22 | * ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE | |
23 | * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR | |
24 | * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF | |
25 | * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS | |
26 | * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN | |
27 | * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) | |
28 | * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE | |
29 | * POSSIBILITY OF SUCH DAMAGE. | |
30 | */ | |
31 | ||
32 | #include "redis.h" | |
33 | #include "hiredis.h" | |
34 | #include "async.h" | |
35 | ||
36 | #include <ctype.h> | |
37 | #include <arpa/inet.h> | |
38 | #include <sys/socket.h> | |
39 | ||
40 | #define REDIS_SENTINEL_PORT 26379 | |
41 | ||
42 | /* ======================== Sentinel global state =========================== */ | |
43 | ||
44 | typedef long long mstime_t; /* millisecond time type. */ | |
45 | ||
46 | /* Address object, used to describe an ip:port pair. */ | |
47 | typedef struct sentinelAddr { | |
48 | char *ip; | |
49 | int port; | |
50 | } sentinelAddr; | |
51 | ||
52 | /* A Sentinel Redis Instance object is monitoring. */ | |
53 | #define SRI_MASTER (1<<0) | |
54 | #define SRI_SLAVE (1<<1) | |
55 | #define SRI_SENTINEL (1<<2) | |
56 | #define SRI_DISCONNECTED (1<<3) | |
57 | #define SRI_S_DOWN (1<<4) /* Subjectively down (no quorum). */ | |
58 | #define SRI_O_DOWN (1<<5) /* Objectively down (quorum reached). */ | |
59 | #define SRI_MASTER_DOWN (1<<6) /* A Sentinel with this flag set thinks that | |
60 | its master is down. */ | |
61 | /* SRI_CAN_FAILOVER when set in an SRI_MASTER instance means that we are | |
62 | * allowed to perform the failover for this master. | |
63 | * When set in a SRI_SENTINEL instance means that sentinel is allowed to | |
64 | * perform the failover on its master. */ | |
65 | #define SRI_CAN_FAILOVER (1<<7) | |
66 | #define SRI_FAILOVER_IN_PROGRESS (1<<8) /* Failover is in progress for | |
67 | this master. */ | |
68 | #define SRI_I_AM_THE_LEADER (1<<9) /* We are the leader for this master. */ | |
69 | #define SRI_PROMOTED (1<<10) /* Slave selected for promotion. */ | |
70 | #define SRI_RECONF_SENT (1<<11) /* SLAVEOF <newmaster> sent. */ | |
71 | #define SRI_RECONF_INPROG (1<<12) /* Slave synchronization in progress. */ | |
72 | #define SRI_RECONF_DONE (1<<13) /* Slave synchronized with new master. */ | |
73 | ||
74 | #define SENTINEL_INFO_PERIOD 10000 | |
75 | #define SENTINEL_PING_PERIOD 1000 | |
76 | #define SENTINEL_ASK_PERIOD 1000 | |
77 | #define SENTINEL_PUBLISH_PERIOD 5000 | |
78 | #define SENTINEL_DOWN_AFTER_PERIOD 30000 | |
79 | #define SENTINEL_HELLO_CHANNEL "__sentinel__:hello" | |
80 | #define SENTINEL_TILT_TRIGGER 2000 | |
81 | #define SENTINEL_TILT_PERIOD (SENTINEL_PING_PERIOD*30) | |
82 | #define SENTINEL_DEFAULT_SLAVE_PRIORITY 100 | |
83 | #define SENTINEL_PROMOTION_RETRY_PERIOD 30000 | |
84 | #define SENTINEL_SLAVE_RECONF_RETRY_PERIOD 10000 | |
85 | #define SENTINEL_DEFAULT_PARALLEL_SYNCS 1 | |
86 | #define SENTINEL_MIN_LINK_RECONNECT_PERIOD 15000 | |
87 | #define SENTINEL_DEFAULT_FAILOVER_TIMEOUT (60*15*1000) | |
88 | #define SENTINEL_MAX_PENDING_COMMANDS 100 | |
89 | #define SENTINEL_EXTENDED_SDOWN_MULTIPLIER 10 | |
90 | ||
91 | /* How many milliseconds is an information valid? This applies for instance | |
92 | * to the reply to SENTINEL IS-MASTER-DOWN-BY-ADDR replies. */ | |
93 | #define SENTINEL_INFO_VALIDITY_TIME 5000 | |
94 | #define SENTINEL_FAILOVER_FIXED_DELAY 5000 | |
95 | #define SENTINEL_FAILOVER_MAX_RANDOM_DELAY 10000 | |
96 | ||
97 | /* Failover machine different states. */ | |
98 | #define SENTINEL_FAILOVER_STATE_NONE 0 /* No failover in progress. */ | |
99 | #define SENTINEL_FAILOVER_STATE_WAIT_START 1 /* Wait for failover_start_time*/ | |
100 | #define SENTINEL_FAILOVER_STATE_SELECT_SLAVE 2 /* Select slave to promote */ | |
101 | #define SENTINEL_FAILOVER_STATE_SEND_SLAVEOF_NOONE 3 /* Slave -> Master */ | |
102 | #define SENTINEL_FAILOVER_STATE_WAIT_PROMOTION 4 /* Wait slave to change role */ | |
103 | #define SENTINEL_FAILOVER_STATE_RECONF_SLAVES 5 /* SLAVEOF newmaster */ | |
104 | #define SENTINEL_FAILOVER_STATE_WAIT_NEXT_SLAVE 6 /* wait replication */ | |
105 | #define SENTINEL_FAILOVER_STATE_ALERT_CLIENTS 7 /* Run user script. */ | |
106 | #define SENTINEL_FAILOVER_STATE_WAIT_ALERT_SCRIPT 8 /* Wait script exec. */ | |
107 | #define SENTINEL_FAILOVER_STATE_DETECT_END 9 /* Check for failover end. */ | |
108 | #define SENTINEL_FAILOVER_STATE_UPDATE_CONFIG 10 /* Monitor promoted slave. */ | |
109 | ||
110 | #define SENTINEL_MASTER_LINK_STATUS_UP 0 | |
111 | #define SENTINEL_MASTER_LINK_STATUS_DOWN 1 | |
112 | ||
113 | typedef struct sentinelRedisInstance { | |
114 | int flags; /* See SRI_... defines */ | |
115 | char *name; /* Master name from the point of view of this sentinel. */ | |
116 | char *runid; /* run ID of this instance. */ | |
117 | sentinelAddr *addr; /* Master host. */ | |
118 | redisAsyncContext *cc; /* Hiredis context for commands. */ | |
119 | redisAsyncContext *pc; /* Hiredis context for Pub / Sub. */ | |
120 | int pending_commands; /* Number of commands sent waiting for a reply. */ | |
121 | mstime_t cc_conn_time; /* cc connection time. */ | |
122 | mstime_t pc_conn_time; /* pc connection time. */ | |
123 | mstime_t pc_last_activity; /* Last time we received any message. */ | |
124 | mstime_t last_avail_time; /* Last time the instance replied to ping with | |
125 | a reply we consider valid. */ | |
126 | mstime_t last_pong_time; /* Last time the instance replied to ping, | |
127 | whatever the reply was. That's used to check | |
128 | if the link is idle and must be reconnected. */ | |
129 | mstime_t last_pub_time; /* Last time we sent hello via Pub/Sub. */ | |
130 | mstime_t last_hello_time; /* Only used if SRI_SENTINEL is set. Last time | |
131 | we received an hello from this Sentinel | |
132 | via Pub/Sub. */ | |
133 | mstime_t last_master_down_reply_time; /* Time of last reply to | |
134 | SENTINEL is-master-down command. */ | |
135 | mstime_t s_down_since_time; /* Subjectively down since time. */ | |
136 | mstime_t o_down_since_time; /* Objectively down since time. */ | |
137 | mstime_t down_after_period; /* Consider it down after that period. */ | |
138 | mstime_t info_refresh; /* Time at which we received INFO output from it. */ | |
139 | ||
140 | /* Master specific. */ | |
141 | dict *sentinels; /* Other sentinels monitoring the same master. */ | |
142 | dict *slaves; /* Slaves for this master instance. */ | |
143 | int quorum; /* Number of sentinels that need to agree on failure. */ | |
144 | int parallel_syncs; /* How many slaves to reconfigure at same time. */ | |
145 | ||
146 | /* Slave specific. */ | |
147 | mstime_t master_link_down_time; /* Slave replication link down time. */ | |
148 | int slave_priority; /* Slave priority according to its INFO output. */ | |
149 | mstime_t slave_reconf_sent_time; /* Time at which we sent SLAVE OF <new> */ | |
150 | struct sentinelRedisInstance *master; /* Master instance if SRI_SLAVE is set. */ | |
151 | char *slave_master_host; /* Master host as reported by INFO */ | |
152 | int slave_master_port; /* Master port as reported by INFO */ | |
153 | int slave_master_link_status; /* Master link status as reported by INFO */ | |
154 | /* Failover */ | |
155 | char *leader; /* If this is a master instance, this is the runid of | |
156 | the Sentinel that should perform the failover. If | |
157 | this is a Sentinel, this is the runid of the Sentinel | |
158 | that this other Sentinel is voting as leader. | |
159 | This field is valid only if SRI_MASTER_DOWN is | |
160 | set on the Sentinel instance. */ | |
161 | int failover_state; /* See SENTINEL_FAILOVER_STATE_* defines. */ | |
162 | mstime_t failover_state_change_time; | |
163 | mstime_t failover_start_time; /* When to start to failover if leader. */ | |
164 | mstime_t failover_timeout; /* Max time to refresh failover state. */ | |
165 | struct sentinelRedisInstance *promoted_slave; /* Promoted slave instance. */ | |
166 | /* Scripts executed to notify admin or reconfigure clients: when they | |
167 | * are set to NULL no script is executed. */ | |
168 | char *notify_script; | |
169 | char *client_reconfig_script; | |
170 | } sentinelRedisInstance; | |
171 | ||
172 | /* Main state. */ | |
173 | struct sentinelState { | |
174 | dict *masters; /* Dictionary of master sentinelRedisInstances. | |
175 | Key is the instance name, value is the | |
176 | sentinelRedisInstance structure pointer. */ | |
177 | int tilt; /* Are we in TILT mode? */ | |
178 | mstime_t tilt_start_time; /* When TITL started. */ | |
179 | mstime_t previous_time; /* Time last time we ran the time handler. */ | |
180 | } sentinel; | |
181 | ||
182 | /* ======================= hiredis ae.c adapters ============================= | |
183 | * Note: this implementation is taken from hiredis/adapters/ae.h, however | |
184 | * we have our modified copy for Sentinel in order to use our allocator | |
185 | * and to have full control over how the adapter works. */ | |
186 | ||
187 | typedef struct redisAeEvents { | |
188 | redisAsyncContext *context; | |
189 | aeEventLoop *loop; | |
190 | int fd; | |
191 | int reading, writing; | |
192 | } redisAeEvents; | |
193 | ||
194 | static void redisAeReadEvent(aeEventLoop *el, int fd, void *privdata, int mask) { | |
195 | ((void)el); ((void)fd); ((void)mask); | |
196 | ||
197 | redisAeEvents *e = (redisAeEvents*)privdata; | |
198 | redisAsyncHandleRead(e->context); | |
199 | } | |
200 | ||
201 | static void redisAeWriteEvent(aeEventLoop *el, int fd, void *privdata, int mask) { | |
202 | ((void)el); ((void)fd); ((void)mask); | |
203 | ||
204 | redisAeEvents *e = (redisAeEvents*)privdata; | |
205 | redisAsyncHandleWrite(e->context); | |
206 | } | |
207 | ||
208 | static void redisAeAddRead(void *privdata) { | |
209 | redisAeEvents *e = (redisAeEvents*)privdata; | |
210 | aeEventLoop *loop = e->loop; | |
211 | if (!e->reading) { | |
212 | e->reading = 1; | |
213 | aeCreateFileEvent(loop,e->fd,AE_READABLE,redisAeReadEvent,e); | |
214 | } | |
215 | } | |
216 | ||
217 | static void redisAeDelRead(void *privdata) { | |
218 | redisAeEvents *e = (redisAeEvents*)privdata; | |
219 | aeEventLoop *loop = e->loop; | |
220 | if (e->reading) { | |
221 | e->reading = 0; | |
222 | aeDeleteFileEvent(loop,e->fd,AE_READABLE); | |
223 | } | |
224 | } | |
225 | ||
226 | static void redisAeAddWrite(void *privdata) { | |
227 | redisAeEvents *e = (redisAeEvents*)privdata; | |
228 | aeEventLoop *loop = e->loop; | |
229 | if (!e->writing) { | |
230 | e->writing = 1; | |
231 | aeCreateFileEvent(loop,e->fd,AE_WRITABLE,redisAeWriteEvent,e); | |
232 | } | |
233 | } | |
234 | ||
235 | static void redisAeDelWrite(void *privdata) { | |
236 | redisAeEvents *e = (redisAeEvents*)privdata; | |
237 | aeEventLoop *loop = e->loop; | |
238 | if (e->writing) { | |
239 | e->writing = 0; | |
240 | aeDeleteFileEvent(loop,e->fd,AE_WRITABLE); | |
241 | } | |
242 | } | |
243 | ||
244 | static void redisAeCleanup(void *privdata) { | |
245 | redisAeEvents *e = (redisAeEvents*)privdata; | |
246 | redisAeDelRead(privdata); | |
247 | redisAeDelWrite(privdata); | |
248 | zfree(e); | |
249 | } | |
250 | ||
251 | static int redisAeAttach(aeEventLoop *loop, redisAsyncContext *ac) { | |
252 | redisContext *c = &(ac->c); | |
253 | redisAeEvents *e; | |
254 | ||
255 | /* Nothing should be attached when something is already attached */ | |
256 | if (ac->ev.data != NULL) | |
257 | return REDIS_ERR; | |
258 | ||
259 | /* Create container for context and r/w events */ | |
260 | e = (redisAeEvents*)zmalloc(sizeof(*e)); | |
261 | e->context = ac; | |
262 | e->loop = loop; | |
263 | e->fd = c->fd; | |
264 | e->reading = e->writing = 0; | |
265 | ||
266 | /* Register functions to start/stop listening for events */ | |
267 | ac->ev.addRead = redisAeAddRead; | |
268 | ac->ev.delRead = redisAeDelRead; | |
269 | ac->ev.addWrite = redisAeAddWrite; | |
270 | ac->ev.delWrite = redisAeDelWrite; | |
271 | ac->ev.cleanup = redisAeCleanup; | |
272 | ac->ev.data = e; | |
273 | ||
274 | return REDIS_OK; | |
275 | } | |
276 | ||
277 | /* ============================= Prototypes ================================= */ | |
278 | ||
279 | void sentinelLinkEstablishedCallback(const redisAsyncContext *c, int status); | |
280 | void sentinelDisconnectCallback(const redisAsyncContext *c, int status); | |
281 | void sentinelReceiveHelloMessages(redisAsyncContext *c, void *reply, void *privdata); | |
282 | sentinelRedisInstance *sentinelGetMasterByName(char *name); | |
283 | char *sentinelGetSubjectiveLeader(sentinelRedisInstance *master); | |
284 | char *sentinelGetObjectiveLeader(sentinelRedisInstance *master); | |
285 | int yesnotoi(char *s); | |
286 | void sentinelDisconnectInstanceFromContext(const redisAsyncContext *c); | |
287 | const char *sentinelRedisInstanceTypeStr(sentinelRedisInstance *ri); | |
288 | ||
289 | /* ========================= Dictionary types =============================== */ | |
290 | ||
291 | unsigned int dictSdsHash(const void *key); | |
292 | int dictSdsKeyCompare(void *privdata, const void *key1, const void *key2); | |
293 | void releaseSentinelRedisInstance(sentinelRedisInstance *ri); | |
294 | ||
295 | void dictInstancesValDestructor (void *privdata, void *obj) { | |
296 | releaseSentinelRedisInstance(obj); | |
297 | } | |
298 | ||
299 | /* Instance name (sds) -> instance (sentinelRedisInstance pointer) | |
300 | * | |
301 | * also used for: sentinelRedisInstance->sentinels dictionary that maps | |
302 | * sentinels ip:port to last seen time in Pub/Sub hello message. */ | |
303 | dictType instancesDictType = { | |
304 | dictSdsHash, /* hash function */ | |
305 | NULL, /* key dup */ | |
306 | NULL, /* val dup */ | |
307 | dictSdsKeyCompare, /* key compare */ | |
308 | NULL, /* key destructor */ | |
309 | dictInstancesValDestructor /* val destructor */ | |
310 | }; | |
311 | ||
312 | /* Instance runid (sds) -> votes (long casted to void*) | |
313 | * | |
314 | * This is useful into sentinelGetObjectiveLeader() function in order to | |
315 | * count the votes and understand who is the leader. */ | |
316 | dictType leaderVotesDictType = { | |
317 | dictSdsHash, /* hash function */ | |
318 | NULL, /* key dup */ | |
319 | NULL, /* val dup */ | |
320 | dictSdsKeyCompare, /* key compare */ | |
321 | NULL, /* key destructor */ | |
322 | NULL /* val destructor */ | |
323 | }; | |
324 | ||
325 | /* =========================== Initialization =============================== */ | |
326 | ||
327 | void sentinelCommand(redisClient *c); | |
328 | ||
329 | struct redisCommand sentinelcmds[] = { | |
330 | {"ping",pingCommand,1,"",0,NULL,0,0,0,0,0}, | |
331 | {"sentinel",sentinelCommand,-2,"",0,NULL,0,0,0,0,0}, | |
332 | {"subscribe",subscribeCommand,-2,"",0,NULL,0,0,0,0,0}, | |
333 | {"unsubscribe",unsubscribeCommand,-1,"",0,NULL,0,0,0,0,0}, | |
334 | {"psubscribe",psubscribeCommand,-2,"",0,NULL,0,0,0,0,0}, | |
335 | {"punsubscribe",punsubscribeCommand,-1,"",0,NULL,0,0,0,0,0} | |
336 | }; | |
337 | ||
338 | /* This function overwrites a few normal Redis config default with Sentinel | |
339 | * specific defaults. */ | |
340 | void initSentinelConfig(void) { | |
341 | server.port = REDIS_SENTINEL_PORT; | |
342 | } | |
343 | ||
344 | /* Perform the Sentinel mode initialization. */ | |
345 | void initSentinel(void) { | |
346 | int j; | |
347 | ||
348 | /* Remove usual Redis commands from the command table, then just add | |
349 | * the SENTINEL command. */ | |
350 | dictEmpty(server.commands); | |
351 | for (j = 0; j < sizeof(sentinelcmds)/sizeof(sentinelcmds[0]); j++) { | |
352 | int retval; | |
353 | struct redisCommand *cmd = sentinelcmds+j; | |
354 | ||
355 | retval = dictAdd(server.commands, sdsnew(cmd->name), cmd); | |
356 | redisAssert(retval == DICT_OK); | |
357 | } | |
358 | ||
359 | /* Initialize various data structures. */ | |
360 | sentinel.masters = dictCreate(&instancesDictType,NULL); | |
361 | sentinel.tilt = 0; | |
362 | sentinel.tilt_start_time = mstime(); | |
363 | sentinel.previous_time = mstime(); | |
364 | } | |
365 | ||
366 | /* ============================== sentinelAddr ============================== */ | |
367 | ||
368 | /* Create a sentinelAddr object and return it on success. | |
369 | * On error NULL is returned and errno is set to: | |
370 | * ENOENT: Can't resolve the hostname. | |
371 | * EINVAL: Invalid port number. | |
372 | */ | |
373 | sentinelAddr *createSentinelAddr(char *hostname, int port) { | |
374 | char buf[32]; | |
375 | sentinelAddr *sa; | |
376 | ||
377 | if (port <= 0 || port > 65535) { | |
378 | errno = EINVAL; | |
379 | return NULL; | |
380 | } | |
381 | if (anetResolve(NULL,hostname,buf) == ANET_ERR) { | |
382 | errno = ENOENT; | |
383 | return NULL; | |
384 | } | |
385 | sa = zmalloc(sizeof(*sa)); | |
386 | sa->ip = sdsnew(buf); | |
387 | sa->port = port; | |
388 | return sa; | |
389 | } | |
390 | ||
391 | /* Free a Sentinel address. Can't fail. */ | |
392 | void releaseSentinelAddr(sentinelAddr *sa) { | |
393 | sdsfree(sa->ip); | |
394 | zfree(sa); | |
395 | } | |
396 | ||
397 | /* =========================== Events notification ========================== */ | |
398 | ||
399 | void sentinelCallNotificationScript(char *scriptpath, char *type, char *msg) { | |
400 | /* TODO: implement it. */ | |
401 | } | |
402 | ||
403 | /* Send an event to log, pub/sub, user notification script. | |
404 | * | |
405 | * 'level' is the log level for logging. Only REDIS_WARNING events will trigger | |
406 | * the execution of the user notification script. | |
407 | * | |
408 | * 'type' is the message type, also used as a pub/sub channel name. | |
409 | * | |
410 | * 'ri', is the redis instance target of this event if applicable, and is | |
411 | * used to obtain the path of the notification script to execute. | |
412 | * | |
413 | * The remaining arguments are printf-alike. | |
414 | * If the format specifier starts with the two characters "%@" then ri is | |
415 | * not NULL, and the message is prefixed with an instance identifier in the | |
416 | * following format: | |
417 | * | |
418 | * <instance type> <instance name> <ip> <port> | |
419 | * | |
420 | * If the instance type is not master, than the additional string is | |
421 | * added to specify the originating master: | |
422 | * | |
423 | * @ <master name> <master ip> <master port> | |
424 | * | |
425 | * Any other specifier after "%@" is processed by printf itself. | |
426 | */ | |
427 | void sentinelEvent(int level, char *type, sentinelRedisInstance *ri, | |
428 | const char *fmt, ...) { | |
429 | va_list ap; | |
430 | char msg[REDIS_MAX_LOGMSG_LEN]; | |
431 | robj *channel, *payload; | |
432 | ||
433 | /* Handle %@ */ | |
434 | if (fmt[0] == '%' && fmt[1] == '@') { | |
435 | sentinelRedisInstance *master = (ri->flags & SRI_MASTER) ? | |
436 | NULL : ri->master; | |
437 | ||
438 | if (master) { | |
439 | snprintf(msg, sizeof(msg), "%s %s %s %d @ %s %s %d", | |
440 | sentinelRedisInstanceTypeStr(ri), | |
441 | ri->name, ri->addr->ip, ri->addr->port, | |
442 | master->name, master->addr->ip, master->addr->port); | |
443 | } else { | |
444 | snprintf(msg, sizeof(msg), "%s %s %s %d", | |
445 | sentinelRedisInstanceTypeStr(ri), | |
446 | ri->name, ri->addr->ip, ri->addr->port); | |
447 | } | |
448 | fmt += 2; | |
449 | } else { | |
450 | msg[0] = '\0'; | |
451 | } | |
452 | ||
453 | /* Use vsprintf for the rest of the formatting if any. */ | |
454 | if (fmt[0] != '\0') { | |
455 | va_start(ap, fmt); | |
456 | vsnprintf(msg+strlen(msg), sizeof(msg)-strlen(msg), fmt, ap); | |
457 | va_end(ap); | |
458 | } | |
459 | ||
460 | /* Log the message if the log level allows it to be logged. */ | |
461 | if (level >= server.verbosity) | |
462 | redisLog(level,"%s %s",type,msg); | |
463 | ||
464 | /* Publish the message via Pub/Sub if it's not a debugging one. */ | |
465 | if (level != REDIS_DEBUG) { | |
466 | channel = createStringObject(type,strlen(type)); | |
467 | payload = createStringObject(msg,strlen(msg)); | |
468 | pubsubPublishMessage(channel,payload); | |
469 | decrRefCount(channel); | |
470 | decrRefCount(payload); | |
471 | } | |
472 | ||
473 | /* Call the notification script if applicable. */ | |
474 | if (level == REDIS_WARNING && ri != NULL) { | |
475 | sentinelRedisInstance *master = (ri->flags & SRI_MASTER) ? | |
476 | ri : ri->master; | |
477 | if (master->notify_script) { | |
478 | sentinelCallNotificationScript(master->notify_script,type,msg); | |
479 | } | |
480 | } | |
481 | } | |
482 | ||
483 | /* ========================== sentinelRedisInstance ========================= */ | |
484 | ||
485 | /* Create a redis instance, the following fields must be populated by the | |
486 | * caller if needed: | |
487 | * runid: set to NULL but will be populated once INFO output is received. | |
488 | * info_refresh: is set to 0 to mean that we never received INFO so far. | |
489 | * | |
490 | * If SRI_MASTER is set into initial flags the instance is added to | |
491 | * sentinel.masters table. | |
492 | * | |
493 | * if SRI_SLAVE or SRI_SENTINEL is set then 'master' must be not NULL and the | |
494 | * instance is added into master->slaves or master->sentinels table. | |
495 | * | |
496 | * If the instance is a slave or sentinel, the name parameter is ignored and | |
497 | * is created automatically as hostname:port. | |
498 | * | |
499 | * The function fails if hostname can't be resolved or port is out of range. | |
500 | * When this happens NULL is returned and errno is set accordingly to the | |
501 | * createSentinelAddr() function. | |
502 | * | |
503 | * The function may also fail and return NULL with errno set to EBUSY if | |
504 | * a master or slave with the same name already exists. */ | |
505 | sentinelRedisInstance *createSentinelRedisInstance(char *name, int flags, char *hostname, int port, int quorum, sentinelRedisInstance *master) { | |
506 | sentinelRedisInstance *ri; | |
507 | sentinelAddr *addr; | |
508 | dict *table; | |
509 | char slavename[128], *sdsname; | |
510 | ||
511 | redisAssert(flags & (SRI_MASTER|SRI_SLAVE|SRI_SENTINEL)); | |
512 | redisAssert((flags & SRI_MASTER) || master != NULL); | |
513 | ||
514 | /* Check address validity. */ | |
515 | addr = createSentinelAddr(hostname,port); | |
516 | if (addr == NULL) return NULL; | |
517 | ||
518 | /* For slaves and sentinel we use ip:port as name. */ | |
519 | if (flags & (SRI_SLAVE|SRI_SENTINEL)) { | |
520 | snprintf(slavename,sizeof(slavename),"%s:%d",hostname,port); | |
521 | name = slavename; | |
522 | } | |
523 | ||
524 | /* Make sure the entry is not duplicated. This may happen when the same | |
525 | * name for a master is used multiple times inside the configuration or | |
526 | * if we try to add multiple times a slave or sentinel with same ip/port | |
527 | * to a master. */ | |
528 | if (flags & SRI_MASTER) table = sentinel.masters; | |
529 | else if (flags & SRI_SLAVE) table = master->slaves; | |
530 | else if (flags & SRI_SENTINEL) table = master->sentinels; | |
531 | sdsname = sdsnew(name); | |
532 | if (dictFind(table,sdsname)) { | |
533 | sdsfree(sdsname); | |
534 | errno = EBUSY; | |
535 | return NULL; | |
536 | } | |
537 | ||
538 | /* Create the instance object. */ | |
539 | ri = zmalloc(sizeof(*ri)); | |
540 | /* Note that all the instances are started in the disconnected state, | |
541 | * the event loop will take care of connecting them. */ | |
542 | ri->flags = flags | SRI_DISCONNECTED; | |
543 | ri->name = sdsname; | |
544 | ri->runid = NULL; | |
545 | ri->addr = addr; | |
546 | ri->cc = NULL; | |
547 | ri->pc = NULL; | |
548 | ri->pending_commands = 0; | |
549 | ri->cc_conn_time = 0; | |
550 | ri->pc_conn_time = 0; | |
551 | ri->pc_last_activity = 0; | |
552 | ri->last_avail_time = mstime(); | |
553 | ri->last_pong_time = mstime(); | |
554 | ri->last_pub_time = mstime(); | |
555 | ri->last_hello_time = mstime(); | |
556 | ri->last_master_down_reply_time = mstime(); | |
557 | ri->s_down_since_time = 0; | |
558 | ri->o_down_since_time = 0; | |
559 | ri->down_after_period = master ? master->down_after_period : | |
560 | SENTINEL_DOWN_AFTER_PERIOD; | |
561 | ri->master_link_down_time = 0; | |
562 | ri->slave_priority = SENTINEL_DEFAULT_SLAVE_PRIORITY; | |
563 | ri->slave_reconf_sent_time = 0; | |
564 | ri->slave_master_host = NULL; | |
565 | ri->slave_master_port = 0; | |
566 | ri->slave_master_link_status = SENTINEL_MASTER_LINK_STATUS_DOWN; | |
567 | ri->sentinels = dictCreate(&instancesDictType,NULL); | |
568 | ri->quorum = quorum; | |
569 | ri->parallel_syncs = SENTINEL_DEFAULT_PARALLEL_SYNCS; | |
570 | ri->master = master; | |
571 | ri->slaves = dictCreate(&instancesDictType,NULL); | |
572 | ri->info_refresh = 0; | |
573 | ||
574 | /* Failover state. */ | |
575 | ri->leader = NULL; | |
576 | ri->failover_state = SENTINEL_FAILOVER_STATE_NONE; | |
577 | ri->failover_state_change_time = 0; | |
578 | ri->failover_start_time = 0; | |
579 | ri->failover_timeout = SENTINEL_DEFAULT_FAILOVER_TIMEOUT; | |
580 | ri->promoted_slave = NULL; | |
581 | ri->notify_script = NULL; | |
582 | ri->client_reconfig_script = NULL; | |
583 | ||
584 | /* Add into the right table. */ | |
585 | dictAdd(table, ri->name, ri); | |
586 | return ri; | |
587 | } | |
588 | ||
589 | /* Release this instance and all its slaves, sentinels, hiredis connections. | |
590 | * This function also takes care of unlinking the instance from the main | |
591 | * masters table (if it is a master) or from its master sentinels/slaves table | |
592 | * if it is a slave or sentinel. */ | |
593 | void releaseSentinelRedisInstance(sentinelRedisInstance *ri) { | |
594 | /* Release all its slaves or sentinels if any. */ | |
595 | dictRelease(ri->sentinels); | |
596 | dictRelease(ri->slaves); | |
597 | ||
598 | /* Release hiredis connections. Note that redisAsyncFree() will call | |
599 | * the disconnection callback. */ | |
600 | if (ri->cc) { | |
601 | redisAsyncFree(ri->cc); | |
602 | ri->cc = NULL; | |
603 | } | |
604 | if (ri->pc) { | |
605 | redisAsyncFree(ri->pc); | |
606 | ri->pc = NULL; | |
607 | } | |
608 | ||
609 | /* Free other resources. */ | |
610 | sdsfree(ri->name); | |
611 | sdsfree(ri->runid); | |
612 | sdsfree(ri->notify_script); | |
613 | sdsfree(ri->client_reconfig_script); | |
614 | sdsfree(ri->slave_master_host); | |
615 | sdsfree(ri->leader); | |
616 | releaseSentinelAddr(ri->addr); | |
617 | ||
618 | /* Clear state into the master if needed. */ | |
619 | if ((ri->flags & SRI_SLAVE) && (ri->flags & SRI_PROMOTED) && ri->master) | |
620 | ri->master->promoted_slave = NULL; | |
621 | ||
622 | zfree(ri); | |
623 | } | |
624 | ||
625 | /* Lookup a slave in a master Redis instance, by ip and port. */ | |
626 | sentinelRedisInstance *sentinelRedisInstanceLookupSlave( | |
627 | sentinelRedisInstance *ri, char *ip, int port) | |
628 | { | |
629 | sds key; | |
630 | sentinelRedisInstance *slave; | |
631 | ||
632 | redisAssert(ri->flags & SRI_MASTER); | |
633 | key = sdscatprintf(sdsempty(),"%s:%d",ip,port); | |
634 | slave = dictFetchValue(ri->slaves,key); | |
635 | sdsfree(key); | |
636 | return slave; | |
637 | } | |
638 | ||
639 | /* Return the name of the type of the instance as a string. */ | |
640 | const char *sentinelRedisInstanceTypeStr(sentinelRedisInstance *ri) { | |
641 | if (ri->flags & SRI_MASTER) return "master"; | |
642 | else if (ri->flags & SRI_SLAVE) return "slave"; | |
643 | else if (ri->flags & SRI_SENTINEL) return "sentinel"; | |
644 | else return "unknown"; | |
645 | } | |
646 | ||
647 | /* This function removes all the instances found in the dictionary of instances | |
648 | * 'd', having either: | |
649 | * | |
650 | * 1) The same ip/port as specified. | |
651 | * 2) The same runid. | |
652 | * | |
653 | * "1" and "2" don't need to verify at the same time, just one is enough. | |
654 | * If "runid" is NULL it is not checked. | |
655 | * Similarly if "ip" is NULL it is not checked. | |
656 | * | |
657 | * This function is useful because every time we add a new Sentinel into | |
658 | * a master's Sentinels dictionary, we want to be very sure about not | |
659 | * having duplicated instances for any reason. This is so important because | |
660 | * we use those other sentinels in order to run our quorum protocol to | |
661 | * understand if it's time to proceeed with the fail over. | |
662 | * | |
663 | * Making sure no duplication is possible we greately improve the robustness | |
664 | * of the quorum (otherwise we may end counting the same instance multiple | |
665 | * times for some reason). | |
666 | * | |
667 | * The function returns the number of Sentinels removed. */ | |
668 | int removeMatchingSentinelsFromMaster(sentinelRedisInstance *master, char *ip, int port, char *runid) { | |
669 | dictIterator *di; | |
670 | dictEntry *de; | |
671 | int removed = 0; | |
672 | ||
673 | di = dictGetSafeIterator(master->sentinels); | |
674 | while((de = dictNext(di)) != NULL) { | |
675 | sentinelRedisInstance *ri = dictGetVal(de); | |
676 | ||
677 | if ((ri->runid && runid && strcmp(ri->runid,runid) == 0) || | |
678 | (ip && strcmp(ri->addr->ip,ip) == 0 && port == ri->addr->port)) | |
679 | { | |
680 | dictDelete(master->sentinels,ri->name); | |
681 | removed++; | |
682 | } | |
683 | } | |
684 | dictReleaseIterator(di); | |
685 | return removed; | |
686 | } | |
687 | ||
688 | /* Search an instance with the same runid, ip and port into a dictionary | |
689 | * of instances. Return NULL if not found, otherwise return the instance | |
690 | * pointer. | |
691 | * | |
692 | * runid or ip can be NULL. In such a case the search is performed only | |
693 | * by the non-NULL field. */ | |
694 | sentinelRedisInstance *getSentinelRedisInstanceByAddrAndRunID(dict *instances, char *ip, int port, char *runid) { | |
695 | dictIterator *di; | |
696 | dictEntry *de; | |
697 | sentinelRedisInstance *instance = NULL; | |
698 | ||
699 | redisAssert(ip || runid); /* User must pass at least one search param. */ | |
700 | di = dictGetIterator(instances); | |
701 | while((de = dictNext(di)) != NULL) { | |
702 | sentinelRedisInstance *ri = dictGetVal(de); | |
703 | ||
704 | if (runid && !ri->runid) continue; | |
705 | if ((runid == NULL || strcmp(ri->runid, runid) == 0) && | |
706 | (ip == NULL || (strcmp(ri->addr->ip, ip) == 0 && | |
707 | ri->addr->port == port))) | |
708 | { | |
709 | instance = ri; | |
710 | break; | |
711 | } | |
712 | } | |
713 | dictReleaseIterator(di); | |
714 | return instance; | |
715 | } | |
716 | ||
717 | /* Simple master lookup by name */ | |
718 | sentinelRedisInstance *sentinelGetMasterByName(char *name) { | |
719 | sentinelRedisInstance *ri; | |
720 | sds sdsname = sdsnew(name); | |
721 | ||
722 | ri = dictFetchValue(sentinel.masters,sdsname); | |
723 | sdsfree(sdsname); | |
724 | return ri; | |
725 | } | |
726 | ||
727 | /* Add the specified flags to all the instances in the specified dictionary. */ | |
728 | void sentinelAddFlagsToDictOfRedisInstances(dict *instances, int flags) { | |
729 | dictIterator *di; | |
730 | dictEntry *de; | |
731 | ||
732 | di = dictGetIterator(instances); | |
733 | while((de = dictNext(di)) != NULL) { | |
734 | sentinelRedisInstance *ri = dictGetVal(de); | |
735 | ri->flags |= flags; | |
736 | } | |
737 | dictReleaseIterator(di); | |
738 | } | |
739 | ||
740 | /* Remove the specified flags to all the instances in the specified | |
741 | * dictionary. */ | |
742 | void sentinelDelFlagsToDictOfRedisInstances(dict *instances, int flags) { | |
743 | dictIterator *di; | |
744 | dictEntry *de; | |
745 | ||
746 | di = dictGetIterator(instances); | |
747 | while((de = dictNext(di)) != NULL) { | |
748 | sentinelRedisInstance *ri = dictGetVal(de); | |
749 | ri->flags &= ~flags; | |
750 | } | |
751 | dictReleaseIterator(di); | |
752 | } | |
753 | ||
754 | /* Reset the state of a monitored master: | |
755 | * 1) Remove all slaves. | |
756 | * 2) Remove all sentinels. | |
757 | * 3) Remove most of the flags resulting from runtime operations. | |
758 | * 4) Reset timers to their default value. | |
759 | * 5) In the process of doing this undo the failover if in progress. | |
760 | * 6) Disconnect the connections with the master (will reconnect automatically). | |
761 | */ | |
762 | void sentinelResetMaster(sentinelRedisInstance *ri) { | |
763 | redisAssert(ri->flags & SRI_MASTER); | |
764 | dictRelease(ri->slaves); | |
765 | dictRelease(ri->sentinels); | |
766 | ri->slaves = dictCreate(&instancesDictType,NULL); | |
767 | ri->sentinels = dictCreate(&instancesDictType,NULL); | |
768 | if (ri->cc) redisAsyncFree(ri->cc); | |
769 | if (ri->pc) redisAsyncFree(ri->pc); | |
770 | ri->flags &= SRI_MASTER|SRI_CAN_FAILOVER|SRI_DISCONNECTED; | |
771 | if (ri->leader) { | |
772 | sdsfree(ri->leader); | |
773 | ri->leader = NULL; | |
774 | } | |
775 | ri->failover_state = SENTINEL_FAILOVER_STATE_NONE; | |
776 | ri->failover_state_change_time = 0; | |
777 | ri->failover_start_time = 0; | |
778 | ri->promoted_slave = NULL; | |
779 | sentinelEvent(REDIS_WARNING,"+reset-master",ri,"%@"); | |
780 | } | |
781 | ||
782 | /* Call sentinelResetMaster() on every master with a name matching the specified | |
783 | * pattern. */ | |
784 | int sentinelResetMastersByPattern(char *pattern) { | |
785 | dictIterator *di; | |
786 | dictEntry *de; | |
787 | int reset = 0; | |
788 | ||
789 | di = dictGetIterator(sentinel.masters); | |
790 | while((de = dictNext(di)) != NULL) { | |
791 | sentinelRedisInstance *ri = dictGetVal(de); | |
792 | ||
793 | if (ri->name) { | |
794 | if (stringmatch(pattern,ri->name,0)) { | |
795 | sentinelResetMaster(ri); | |
796 | reset++; | |
797 | } | |
798 | } | |
799 | } | |
800 | dictReleaseIterator(di); | |
801 | return reset; | |
802 | } | |
803 | ||
804 | /* ============================ Config handling ============================= */ | |
805 | char *sentinelHandleConfiguration(char **argv, int argc) { | |
806 | sentinelRedisInstance *ri; | |
807 | ||
808 | if (!strcasecmp(argv[0],"monitor") && argc == 5) { | |
809 | /* monitor <name> <host> <port> <quorum> */ | |
810 | int quorum = atoi(argv[4]); | |
811 | ||
812 | if (quorum <= 0) return "Quorum must be 1 or greater."; | |
813 | if (createSentinelRedisInstance(argv[1],SRI_MASTER,argv[2], | |
814 | atoi(argv[3]),quorum,NULL) == NULL) | |
815 | { | |
816 | switch(errno) { | |
817 | case EBUSY: return "Duplicated master name."; | |
818 | case ENOENT: return "Can't resolve master instance hostname."; | |
819 | case EINVAL: return "Invalid port number"; | |
820 | } | |
821 | } | |
822 | } else if (!strcasecmp(argv[0],"down-after-milliseconds") && argc == 3) { | |
823 | /* down-after-milliseconds <name> <milliseconds> */ | |
824 | ri = sentinelGetMasterByName(argv[1]); | |
825 | if (!ri) return "No such master with specified name."; | |
826 | ri->down_after_period = atoi(argv[2]); | |
827 | if (ri->down_after_period <= 0) | |
828 | return "negative or zero time parameter."; | |
829 | } else if (!strcasecmp(argv[0],"failover-timeout") && argc == 3) { | |
830 | /* failover-timeout <name> <milliseconds> */ | |
831 | ri = sentinelGetMasterByName(argv[1]); | |
832 | if (!ri) return "No such master with specified name."; | |
833 | ri->failover_timeout = atoi(argv[2]); | |
834 | if (ri->failover_timeout <= 0) | |
835 | return "negative or zero time parameter."; | |
836 | } else if (!strcasecmp(argv[0],"can-failover") && argc == 3) { | |
837 | /* can-failover <name> <yes/no> */ | |
838 | int yesno = yesnotoi(argv[2]); | |
839 | ||
840 | ri = sentinelGetMasterByName(argv[1]); | |
841 | if (!ri) return "No such master with specified name."; | |
842 | if (yesno == -1) return "Argument must be either yes or no."; | |
843 | if (yesno) | |
844 | ri->flags |= SRI_CAN_FAILOVER; | |
845 | else | |
846 | ri->flags &= ~SRI_CAN_FAILOVER; | |
847 | } else if (!strcasecmp(argv[0],"parallel-syncs") && argc == 3) { | |
848 | /* parallel-syncs <name> <milliseconds> */ | |
849 | ri = sentinelGetMasterByName(argv[1]); | |
850 | if (!ri) return "No such master with specified name."; | |
851 | ri->parallel_syncs = atoi(argv[2]); | |
852 | } else { | |
853 | return "Unrecognized sentinel configuration statement."; | |
854 | } | |
855 | return NULL; | |
856 | } | |
857 | ||
858 | /* ====================== hiredis connection handling ======================= */ | |
859 | ||
860 | /* This function takes an hiredis context that is in an error condition | |
861 | * and make sure to mark the instance as disconnected performing the | |
862 | * cleanup needed. | |
863 | * | |
864 | * Note: we don't free the hiredis context as hiredis will do it for us | |
865 | * for async conenctions. */ | |
866 | void sentinelDisconnectInstanceFromContext(const redisAsyncContext *c) { | |
867 | sentinelRedisInstance *ri = c->data; | |
868 | int pubsub = (ri->pc == c); | |
869 | ||
870 | sentinelEvent(REDIS_DEBUG, pubsub ? "-pubsub-link" : "-cmd-link", ri, | |
871 | "%@ #%s", c->errstr); | |
872 | if (pubsub) | |
873 | ri->pc = NULL; | |
874 | else | |
875 | ri->cc = NULL; | |
876 | ri->flags |= SRI_DISCONNECTED; | |
877 | } | |
878 | ||
879 | void sentinelLinkEstablishedCallback(const redisAsyncContext *c, int status) { | |
880 | if (status != REDIS_OK) { | |
881 | sentinelDisconnectInstanceFromContext(c); | |
882 | } else { | |
883 | sentinelRedisInstance *ri = c->data; | |
884 | int pubsub = (ri->pc == c); | |
885 | ||
886 | sentinelEvent(REDIS_DEBUG, pubsub ? "+pubsub-link" : "+cmd-link", ri, | |
887 | "%@"); | |
888 | } | |
889 | } | |
890 | ||
891 | void sentinelDisconnectCallback(const redisAsyncContext *c, int status) { | |
892 | sentinelDisconnectInstanceFromContext(c); | |
893 | } | |
894 | ||
895 | /* Create the async connections for the specified instance if the instance | |
896 | * is disconnected. Note that the SRI_DISCONNECTED flag is set even if just | |
897 | * one of the two links (commands and pub/sub) is missing. */ | |
898 | void sentinelReconnectInstance(sentinelRedisInstance *ri) { | |
899 | if (!(ri->flags & SRI_DISCONNECTED)) return; | |
900 | ||
901 | /* Commands connection. */ | |
902 | if (ri->cc == NULL) { | |
903 | ri->cc = redisAsyncConnect(ri->addr->ip,ri->addr->port); | |
904 | if (ri->cc->err) { | |
905 | sentinelEvent(REDIS_DEBUG,"-cmd-link-reconnection",ri,"%@ #%s", | |
906 | ri->cc->errstr); | |
907 | redisAsyncFree(ri->cc); | |
908 | ri->cc = NULL; | |
909 | } else { | |
910 | ri->cc_conn_time = mstime(); | |
911 | ri->cc->data = ri; | |
912 | redisAeAttach(server.el,ri->cc); | |
913 | redisAsyncSetConnectCallback(ri->cc, | |
914 | sentinelLinkEstablishedCallback); | |
915 | redisAsyncSetDisconnectCallback(ri->cc, | |
916 | sentinelDisconnectCallback); | |
917 | } | |
918 | } | |
919 | /* Pub / Sub */ | |
920 | if ((ri->flags & SRI_MASTER) && ri->pc == NULL) { | |
921 | ri->pc = redisAsyncConnect(ri->addr->ip,ri->addr->port); | |
922 | if (ri->pc->err) { | |
923 | sentinelEvent(REDIS_DEBUG,"-pubsub-link-reconnection",ri,"%@ #%s", | |
924 | ri->pc->errstr); | |
925 | redisAsyncFree(ri->pc); | |
926 | ri->pc = NULL; | |
927 | } else { | |
928 | int retval; | |
929 | ||
930 | ri->pc_conn_time = mstime(); | |
931 | ri->pc->data = ri; | |
932 | redisAeAttach(server.el,ri->pc); | |
933 | redisAsyncSetConnectCallback(ri->pc, | |
934 | sentinelLinkEstablishedCallback); | |
935 | redisAsyncSetDisconnectCallback(ri->pc, | |
936 | sentinelDisconnectCallback); | |
937 | /* Now we subscribe to the Sentinels "Hello" channel. */ | |
938 | retval = redisAsyncCommand(ri->pc, | |
939 | sentinelReceiveHelloMessages, NULL, "SUBSCRIBE %s", | |
940 | SENTINEL_HELLO_CHANNEL); | |
941 | if (retval != REDIS_OK) { | |
942 | /* If we can't subscribe, the Pub/Sub connection is useless | |
943 | * and we can simply disconnect it and try again. */ | |
944 | redisAsyncFree(ri->pc); | |
945 | ri->pc = NULL; | |
946 | return; | |
947 | } | |
948 | } | |
949 | } | |
950 | /* Clear the DISCONNECTED flags only if we have both the connections | |
951 | * (or just the commands connection if this is a slave or a | |
952 | * sentinel instance). */ | |
953 | if (ri->cc && (ri->flags & (SRI_SLAVE|SRI_SENTINEL) || ri->pc)) | |
954 | ri->flags &= ~SRI_DISCONNECTED; | |
955 | } | |
956 | ||
957 | /* ======================== Redis instances pinging ======================== */ | |
958 | ||
959 | /* Process the INFO output from masters. */ | |
960 | void sentinelRefreshInstanceInfo(sentinelRedisInstance *ri, const char *info) { | |
961 | sds *lines; | |
962 | int numlines, j; | |
963 | int role = 0; | |
964 | ||
965 | ||
966 | /* The following fields must be reset to a given value in the case they | |
967 | * are not found at all in the INFO output. */ | |
968 | ri->master_link_down_time = 0; | |
969 | ||
970 | /* Process line by line. */ | |
971 | lines = sdssplitlen(info,strlen(info),"\r\n",2,&numlines); | |
972 | for (j = 0; j < numlines; j++) { | |
973 | sentinelRedisInstance *slave; | |
974 | sds l = lines[j]; | |
975 | ||
976 | /* run_id:<40 hex chars>*/ | |
977 | if (sdslen(l) >= 47 && !memcmp(l,"run_id:",7)) { | |
978 | if (ri->runid == NULL) { | |
979 | ri->runid = sdsnewlen(l+7,40); | |
980 | } else { | |
981 | /* TODO: check if run_id has changed. This means the | |
982 | * instance has been restarted, we want to set a flag | |
983 | * and notify this event. */ | |
984 | } | |
985 | } | |
986 | ||
987 | /* slave0:<ip>,<port>,<state> */ | |
988 | if ((ri->flags & SRI_MASTER) && | |
989 | sdslen(l) >= 7 && | |
990 | !memcmp(l,"slave",5) && isdigit(l[5])) | |
991 | { | |
992 | char *ip, *port, *end; | |
993 | ||
994 | ip = strchr(l,':'); if (!ip) continue; | |
995 | ip++; /* Now ip points to start of ip address. */ | |
996 | port = strchr(ip,','); if (!port) continue; | |
997 | *port = '\0'; /* nul term for easy access. */ | |
998 | port++; /* Now port points to start of port number. */ | |
999 | end = strchr(port,','); if (!end) continue; | |
1000 | *end = '\0'; /* nul term for easy access. */ | |
1001 | ||
1002 | /* Check if we already have this slave into our table, | |
1003 | * otherwise add it. */ | |
1004 | if (sentinelRedisInstanceLookupSlave(ri,ip,atoi(port)) == NULL) { | |
1005 | if ((slave = createSentinelRedisInstance(NULL,SRI_SLAVE,ip, | |
1006 | atoi(port), ri->quorum,ri)) != NULL) | |
1007 | { | |
1008 | sentinelEvent(REDIS_NOTICE,"+slave",slave,"%@"); | |
1009 | } | |
1010 | } | |
1011 | } | |
1012 | ||
1013 | /* master_link_down_since_seconds:<seconds> */ | |
1014 | if (sdslen(l) >= 32 && | |
1015 | !memcmp(l,"master_link_down_since_seconds",30)) | |
1016 | { | |
1017 | ri->master_link_down_time = strtoll(l+31,NULL,10)*1000; | |
1018 | } | |
1019 | ||
1020 | /* role:<role> */ | |
1021 | if (!memcmp(l,"role:master",11)) role = SRI_MASTER; | |
1022 | else if (!memcmp(l,"role:slave",10)) role = SRI_SLAVE; | |
1023 | ||
1024 | if (role == SRI_SLAVE) { | |
1025 | /* master_host:<host> */ | |
1026 | if (sdslen(l) >= 12 && !memcmp(l,"master_host:",12)) { | |
1027 | sdsfree(ri->slave_master_host); | |
1028 | ri->slave_master_host = sdsnew(l+12); | |
1029 | } | |
1030 | ||
1031 | /* master_port:<port> */ | |
1032 | if (sdslen(l) >= 12 && !memcmp(l,"master_port:",12)) | |
1033 | ri->slave_master_port = atoi(l+12); | |
1034 | ||
1035 | /* master_link_status:<status> */ | |
1036 | if (sdslen(l) >= 19 && !memcmp(l,"master_link_status:",19)) { | |
1037 | ri->slave_master_link_status = | |
1038 | (strcasecmp(l+19,"up") == 0) ? | |
1039 | SENTINEL_MASTER_LINK_STATUS_UP : | |
1040 | SENTINEL_MASTER_LINK_STATUS_DOWN; | |
1041 | } | |
1042 | } | |
1043 | } | |
1044 | ri->info_refresh = mstime(); | |
1045 | sdsfreesplitres(lines,numlines); | |
1046 | ||
1047 | if (sentinel.tilt) return; | |
1048 | ||
1049 | /* Act if a slave turned into a master. */ | |
1050 | if ((ri->flags & SRI_SLAVE) && role == SRI_MASTER) { | |
1051 | if (ri->flags & SRI_PROMOTED) { | |
1052 | /* If this is a promoted slave we can change state to the | |
1053 | * failover state machine. */ | |
1054 | if (ri->master && | |
1055 | (ri->master->flags & SRI_FAILOVER_IN_PROGRESS) && | |
1056 | (ri->master->flags & SRI_I_AM_THE_LEADER) && | |
1057 | (ri->master->failover_state == | |
1058 | SENTINEL_FAILOVER_STATE_WAIT_PROMOTION)) | |
1059 | { | |
1060 | ri->master->failover_state = SENTINEL_FAILOVER_STATE_RECONF_SLAVES; | |
1061 | ri->master->failover_state_change_time = mstime(); | |
1062 | sentinelEvent(REDIS_WARNING,"+promoted-slave",ri,"%@"); | |
1063 | sentinelEvent(REDIS_WARNING,"+failover-state-reconf-slaves", | |
1064 | ri->master,"%@"); | |
1065 | } | |
1066 | } else { | |
1067 | /* Otherwise we interpret this as the start of the failover. */ | |
1068 | if (ri->master && | |
1069 | (ri->master->flags & SRI_FAILOVER_IN_PROGRESS) == 0) | |
1070 | { | |
1071 | ri->master->flags |= SRI_FAILOVER_IN_PROGRESS; | |
1072 | sentinelEvent(REDIS_WARNING,"failover-detected",ri->master,"%@"); | |
1073 | ri->master->failover_state = SENTINEL_FAILOVER_STATE_DETECT_END; | |
1074 | ri->master->failover_state_change_time = mstime(); | |
1075 | ri->master->promoted_slave = ri; | |
1076 | ri->flags |= SRI_PROMOTED; | |
1077 | /* We are an observer, so we can only assume that the leader | |
1078 | * is reconfiguring the slave instances. For this reason we | |
1079 | * set all the instances as RECONF_SENT waiting for progresses | |
1080 | * on this side. */ | |
1081 | sentinelAddFlagsToDictOfRedisInstances(ri->master->slaves, | |
1082 | SRI_RECONF_SENT); | |
1083 | } | |
1084 | } | |
1085 | } | |
1086 | ||
1087 | /* Detect if the slave that is in the process of being reconfigured | |
1088 | * changed state. */ | |
1089 | if ((ri->flags & SRI_SLAVE) && role == SRI_SLAVE && | |
1090 | (ri->flags & (SRI_RECONF_SENT|SRI_RECONF_INPROG))) | |
1091 | { | |
1092 | /* SRI_RECONF_SENT -> SRI_RECONF_INPROG. */ | |
1093 | if ((ri->flags & SRI_RECONF_SENT) && | |
1094 | ri->slave_master_host && | |
1095 | strcmp(ri->slave_master_host, | |
1096 | ri->master->promoted_slave->addr->ip) == 0 && | |
1097 | ri->slave_master_port == ri->master->promoted_slave->addr->port) | |
1098 | { | |
1099 | ri->flags &= ~SRI_RECONF_SENT; | |
1100 | ri->flags |= SRI_RECONF_INPROG; | |
1101 | sentinelEvent(REDIS_NOTICE,"+slave-reconf-inprog",ri,"%@"); | |
1102 | } | |
1103 | ||
1104 | /* SRI_RECONF_INPROG -> SRI_RECONF_DONE */ | |
1105 | if ((ri->flags & SRI_RECONF_INPROG) && | |
1106 | ri->slave_master_link_status == SENTINEL_MASTER_LINK_STATUS_UP) | |
1107 | { | |
1108 | ri->flags &= ~SRI_RECONF_INPROG; | |
1109 | ri->flags |= SRI_RECONF_DONE; | |
1110 | sentinelEvent(REDIS_NOTICE,"+slave-reconf-done",ri,"%@"); | |
1111 | /* If we are moving forward (a new slave is now configured) | |
1112 | * we update the change_time as we are conceptually passing | |
1113 | * to the next slave. */ | |
1114 | ri->failover_state_change_time = mstime(); | |
1115 | } | |
1116 | } | |
1117 | } | |
1118 | ||
1119 | void sentinelInfoReplyCallback(redisAsyncContext *c, void *reply, void *privdata) { | |
1120 | sentinelRedisInstance *ri = c->data; | |
1121 | redisReply *r; | |
1122 | ||
1123 | ri->pending_commands--; | |
1124 | if (!reply) return; | |
1125 | r = reply; | |
1126 | ||
1127 | if (r->type == REDIS_REPLY_STRING) { | |
1128 | sentinelRefreshInstanceInfo(ri,r->str); | |
1129 | } | |
1130 | } | |
1131 | ||
1132 | /* Just discard the reply. We use this when we are not monitoring the return | |
1133 | * value of the command but its effects directly. */ | |
1134 | void sentinelDiscardReplyCallback(redisAsyncContext *c, void *reply, void *privdata) { | |
1135 | sentinelRedisInstance *ri = c->data; | |
1136 | ||
1137 | ri->pending_commands--; | |
1138 | } | |
1139 | ||
1140 | void sentinelPingReplyCallback(redisAsyncContext *c, void *reply, void *privdata) { | |
1141 | sentinelRedisInstance *ri = c->data; | |
1142 | redisReply *r; | |
1143 | ||
1144 | ri->pending_commands--; | |
1145 | if (!reply) return; | |
1146 | r = reply; | |
1147 | ||
1148 | if (r->type == REDIS_REPLY_STATUS || | |
1149 | r->type == REDIS_REPLY_ERROR) { | |
1150 | /* Update the "instance available" field only if this is an | |
1151 | * acceptable reply. */ | |
1152 | if (strncmp(r->str,"PONG",4) == 0 || | |
1153 | strncmp(r->str,"LOADING",7) == 0 || | |
1154 | strncmp(r->str,"MASTERDOWN",10) == 0) | |
1155 | { | |
1156 | ri->last_avail_time = mstime(); | |
1157 | } | |
1158 | } | |
1159 | ri->last_pong_time = mstime(); | |
1160 | } | |
1161 | ||
1162 | /* This is called when we get the reply about the PUBLISH command we send | |
1163 | * to the master to advertise this sentinel. */ | |
1164 | void sentinelPublishReplyCallback(redisAsyncContext *c, void *reply, void *privdata) { | |
1165 | sentinelRedisInstance *ri = c->data; | |
1166 | redisReply *r; | |
1167 | ||
1168 | ri->pending_commands--; | |
1169 | if (!reply) return; | |
1170 | r = reply; | |
1171 | ||
1172 | /* Only update pub_time if we actually published our message. Otherwise | |
1173 | * we'll retry against in 100 milliseconds. */ | |
1174 | if (r->type != REDIS_REPLY_ERROR) | |
1175 | ri->last_pub_time = mstime(); | |
1176 | } | |
1177 | ||
1178 | /* This is our Pub/Sub callback for the Hello channel. It's useful in order | |
1179 | * to discover other sentinels attached at the same master. */ | |
1180 | void sentinelReceiveHelloMessages(redisAsyncContext *c, void *reply, void *privdata) { | |
1181 | sentinelRedisInstance *ri = c->data; | |
1182 | redisReply *r; | |
1183 | ||
1184 | if (!reply) return; | |
1185 | r = reply; | |
1186 | ||
1187 | /* Update the last activity in the pubsub channel. Note that since we | |
1188 | * receive our messages as well this timestamp can be used to detect | |
1189 | * if the link is probably diconnected even if it seems otherwise. */ | |
1190 | ri->pc_last_activity = mstime(); | |
1191 | ||
1192 | /* Sanity check in the reply we expect, so that the code that follows | |
1193 | * can avoid to check for details. */ | |
1194 | if (r->type != REDIS_REPLY_ARRAY || | |
1195 | r->elements != 3 || | |
1196 | r->element[0]->type != REDIS_REPLY_STRING || | |
1197 | r->element[1]->type != REDIS_REPLY_STRING || | |
1198 | r->element[2]->type != REDIS_REPLY_STRING || | |
1199 | strcmp(r->element[0]->str,"message") != 0) return; | |
1200 | ||
1201 | /* We are not interested in meeting ourselves */ | |
1202 | if (strstr(r->element[2]->str,server.runid) != NULL) return; | |
1203 | ||
1204 | { | |
1205 | int numtokens, port, removed, canfailover; | |
1206 | char **token = sdssplitlen(r->element[2]->str, | |
1207 | r->element[2]->len, | |
1208 | ":",1,&numtokens); | |
1209 | sentinelRedisInstance *sentinel; | |
1210 | ||
1211 | if (numtokens == 4) { | |
1212 | /* First, try to see if we already have this sentinel. */ | |
1213 | port = atoi(token[1]); | |
1214 | canfailover = atoi(token[3]); | |
1215 | sentinel = getSentinelRedisInstanceByAddrAndRunID( | |
1216 | ri->sentinels,token[0],port,token[2]); | |
1217 | ||
1218 | if (!sentinel) { | |
1219 | /* If not, remove all the sentinels that have the same runid | |
1220 | * OR the same ip/port, because it's either a restart or a | |
1221 | * network topology change. */ | |
1222 | removed = removeMatchingSentinelsFromMaster(ri,token[0],port, | |
1223 | token[2]); | |
1224 | if (removed) { | |
1225 | sentinelEvent(REDIS_NOTICE,"-dup-sentinel",ri, | |
1226 | "%@ #duplicate of %s:%d or %s", | |
1227 | token[0],port,token[2]); | |
1228 | } | |
1229 | ||
1230 | /* Add the new sentinel. */ | |
1231 | sentinel = createSentinelRedisInstance(NULL,SRI_SENTINEL, | |
1232 | token[0],port,ri->quorum,ri); | |
1233 | if (sentinel) { | |
1234 | sentinelEvent(REDIS_NOTICE,"+sentinel",sentinel,"%@"); | |
1235 | /* The runid is NULL after a new instance creation and | |
1236 | * for Sentinels we don't have a later chance to fill it, | |
1237 | * so do it now. */ | |
1238 | sentinel->runid = sdsnew(token[2]); | |
1239 | } | |
1240 | } | |
1241 | ||
1242 | /* Update the state of the Sentinel. */ | |
1243 | if (sentinel) { | |
1244 | sentinel->last_hello_time = mstime(); | |
1245 | if (canfailover) | |
1246 | sentinel->flags |= SRI_CAN_FAILOVER; | |
1247 | else | |
1248 | sentinel->flags &= ~SRI_CAN_FAILOVER; | |
1249 | } | |
1250 | } | |
1251 | sdsfreesplitres(token,numtokens); | |
1252 | } | |
1253 | } | |
1254 | ||
1255 | void sentinelPingInstance(sentinelRedisInstance *ri) { | |
1256 | mstime_t now = mstime(); | |
1257 | mstime_t info_period; | |
1258 | int retval; | |
1259 | ||
1260 | /* Return ASAP if we have already a PING or INFO already pending, or | |
1261 | * in the case the instance is not properly connected. */ | |
1262 | if (ri->flags & SRI_DISCONNECTED) return; | |
1263 | ||
1264 | /* For INFO, PING, PUBLISH that are not critical commands to send we | |
1265 | * also have a limit of SENTINEL_MAX_PENDING_COMMANDS. We don't | |
1266 | * want to use a lot of memory just because a link is not working | |
1267 | * properly (note that anyway there is a redundant protection about this, | |
1268 | * that is, the link will be disconnected and reconnected if a long | |
1269 | * timeout condition is detected. */ | |
1270 | if (ri->pending_commands >= SENTINEL_MAX_PENDING_COMMANDS) return; | |
1271 | ||
1272 | /* If this is a slave of a master in O_DOWN condition we start sending | |
1273 | * it INFO every second, instead of the usual SENTINEL_INFO_PERIOD | |
1274 | * period. In this state we want to closely monitor slaves in case they | |
1275 | * are turned into masters by another Sentinel, or by the sysadmin. */ | |
1276 | if ((ri->flags & SRI_SLAVE) && | |
1277 | (ri->master->flags & (SRI_O_DOWN|SRI_FAILOVER_IN_PROGRESS))) { | |
1278 | info_period = 1000; | |
1279 | } else { | |
1280 | info_period = SENTINEL_INFO_PERIOD; | |
1281 | } | |
1282 | ||
1283 | if ((ri->flags & SRI_SENTINEL) == 0 && | |
1284 | (ri->info_refresh == 0 || | |
1285 | (now - ri->info_refresh) > info_period)) | |
1286 | { | |
1287 | /* Send INFO to masters and slaves, not sentinels. */ | |
1288 | retval = redisAsyncCommand(ri->cc, | |
1289 | sentinelInfoReplyCallback, NULL, "INFO"); | |
1290 | if (retval != REDIS_OK) return; | |
1291 | ri->pending_commands++; | |
1292 | } else if ((now - ri->last_pong_time) > SENTINEL_PING_PERIOD) { | |
1293 | /* Send PING to all the three kinds of instances. */ | |
1294 | retval = redisAsyncCommand(ri->cc, | |
1295 | sentinelPingReplyCallback, NULL, "PING"); | |
1296 | if (retval != REDIS_OK) return; | |
1297 | ri->pending_commands++; | |
1298 | } else if ((ri->flags & SRI_MASTER) && | |
1299 | (now - ri->last_pub_time) > SENTINEL_PUBLISH_PERIOD) | |
1300 | { | |
1301 | /* PUBLISH hello messages only to masters. */ | |
1302 | struct sockaddr_in sa; | |
1303 | socklen_t salen = sizeof(sa); | |
1304 | ||
1305 | if (getsockname(ri->cc->c.fd,(struct sockaddr*)&sa,&salen) != -1) { | |
1306 | char myaddr[128]; | |
1307 | ||
1308 | snprintf(myaddr,sizeof(myaddr),"%s:%d:%s:%d", | |
1309 | inet_ntoa(sa.sin_addr), server.port, server.runid, | |
1310 | (ri->flags & SRI_CAN_FAILOVER) != 0); | |
1311 | retval = redisAsyncCommand(ri->cc, | |
1312 | sentinelPublishReplyCallback, NULL, "PUBLISH %s %s", | |
1313 | SENTINEL_HELLO_CHANNEL,myaddr); | |
1314 | if (retval != REDIS_OK) return; | |
1315 | ri->pending_commands++; | |
1316 | } | |
1317 | } | |
1318 | } | |
1319 | ||
1320 | /* =========================== SENTINEL command ============================= */ | |
1321 | ||
1322 | const char *sentinelFailoverStateStr(int state) { | |
1323 | switch(state) { | |
1324 | case SENTINEL_FAILOVER_STATE_NONE: return "none"; | |
1325 | case SENTINEL_FAILOVER_STATE_WAIT_START: return "wait_start"; | |
1326 | case SENTINEL_FAILOVER_STATE_SELECT_SLAVE: return "select_slave"; | |
1327 | case SENTINEL_FAILOVER_STATE_SEND_SLAVEOF_NOONE: return "send_slaveof_noone"; | |
1328 | case SENTINEL_FAILOVER_STATE_WAIT_PROMOTION: return "wait_promotion"; | |
1329 | case SENTINEL_FAILOVER_STATE_RECONF_SLAVES: return "reconf_slaves"; | |
1330 | case SENTINEL_FAILOVER_STATE_ALERT_CLIENTS: return "alert_clients"; | |
1331 | case SENTINEL_FAILOVER_STATE_DETECT_END: return "detect_end"; | |
1332 | case SENTINEL_FAILOVER_STATE_UPDATE_CONFIG: return "update_config"; | |
1333 | default: return "unknown"; | |
1334 | } | |
1335 | } | |
1336 | ||
1337 | /* Redis instance to Redis protocol representation. */ | |
1338 | void addReplySentinelRedisInstance(redisClient *c, sentinelRedisInstance *ri) { | |
1339 | char *flags = sdsempty(); | |
1340 | void *mbl; | |
1341 | int fields = 0; | |
1342 | ||
1343 | mbl = addDeferredMultiBulkLength(c); | |
1344 | ||
1345 | addReplyBulkCString(c,"name"); | |
1346 | addReplyBulkCString(c,ri->name); | |
1347 | fields++; | |
1348 | ||
1349 | addReplyBulkCString(c,"ip"); | |
1350 | addReplyBulkCString(c,ri->addr->ip); | |
1351 | fields++; | |
1352 | ||
1353 | addReplyBulkCString(c,"port"); | |
1354 | addReplyBulkLongLong(c,ri->addr->port); | |
1355 | fields++; | |
1356 | ||
1357 | addReplyBulkCString(c,"runid"); | |
1358 | addReplyBulkCString(c,ri->runid ? ri->runid : ""); | |
1359 | fields++; | |
1360 | ||
1361 | addReplyBulkCString(c,"flags"); | |
1362 | if (ri->flags & SRI_S_DOWN) flags = sdscat(flags,"s_down,"); | |
1363 | if (ri->flags & SRI_O_DOWN) flags = sdscat(flags,"o_down,"); | |
1364 | if (ri->flags & SRI_MASTER) flags = sdscat(flags,"master,"); | |
1365 | if (ri->flags & SRI_SLAVE) flags = sdscat(flags,"slave,"); | |
1366 | if (ri->flags & SRI_SENTINEL) flags = sdscat(flags,"sentinel,"); | |
1367 | if (ri->flags & SRI_DISCONNECTED) flags = sdscat(flags,"disconnected,"); | |
1368 | if (ri->flags & SRI_MASTER_DOWN) flags = sdscat(flags,"master_down,"); | |
1369 | if (ri->flags & SRI_FAILOVER_IN_PROGRESS) | |
1370 | flags = sdscat(flags,"failover_in_progress,"); | |
1371 | if (ri->flags & SRI_I_AM_THE_LEADER) | |
1372 | flags = sdscat(flags,"i_am_the_leader,"); | |
1373 | if (ri->flags & SRI_PROMOTED) flags = sdscat(flags,"promoted,"); | |
1374 | if (ri->flags & SRI_RECONF_SENT) flags = sdscat(flags,"reconf_sent,"); | |
1375 | if (ri->flags & SRI_RECONF_INPROG) flags = sdscat(flags,"reconf_inprog,"); | |
1376 | if (ri->flags & SRI_RECONF_DONE) flags = sdscat(flags,"reconf_done,"); | |
1377 | ||
1378 | if (sdslen(flags) != 0) flags = sdsrange(flags,0,-2); /* remove last "," */ | |
1379 | addReplyBulkCString(c,flags); | |
1380 | sdsfree(flags); | |
1381 | fields++; | |
1382 | ||
1383 | addReplyBulkCString(c,"pending-commands"); | |
1384 | addReplyBulkLongLong(c,ri->pending_commands); | |
1385 | fields++; | |
1386 | ||
1387 | if (ri->flags & SRI_FAILOVER_IN_PROGRESS) { | |
1388 | addReplyBulkCString(c,"failover-state"); | |
1389 | addReplyBulkCString(c,(char*)sentinelFailoverStateStr(ri->failover_state)); | |
1390 | fields++; | |
1391 | } | |
1392 | ||
1393 | addReplyBulkCString(c,"last-ok-ping-reply"); | |
1394 | addReplyBulkLongLong(c,mstime() - ri->last_avail_time); | |
1395 | fields++; | |
1396 | ||
1397 | addReplyBulkCString(c,"last-ping-reply"); | |
1398 | addReplyBulkLongLong(c,mstime() - ri->last_pong_time); | |
1399 | fields++; | |
1400 | ||
1401 | if (ri->flags & SRI_S_DOWN) { | |
1402 | addReplyBulkCString(c,"s-down-time"); | |
1403 | addReplyBulkLongLong(c,mstime()-ri->s_down_since_time); | |
1404 | fields++; | |
1405 | } | |
1406 | ||
1407 | if (ri->flags & SRI_O_DOWN) { | |
1408 | addReplyBulkCString(c,"o-down-time"); | |
1409 | addReplyBulkLongLong(c,mstime()-ri->o_down_since_time); | |
1410 | fields++; | |
1411 | } | |
1412 | ||
1413 | /* Masters and Slaves */ | |
1414 | if (ri->flags & (SRI_MASTER|SRI_SLAVE)) { | |
1415 | addReplyBulkCString(c,"info-refresh"); | |
1416 | addReplyBulkLongLong(c,mstime() - ri->info_refresh); | |
1417 | fields++; | |
1418 | } | |
1419 | ||
1420 | /* Only masters */ | |
1421 | if (ri->flags & SRI_MASTER) { | |
1422 | addReplyBulkCString(c,"num-slaves"); | |
1423 | addReplyBulkLongLong(c,dictSize(ri->slaves)); | |
1424 | fields++; | |
1425 | ||
1426 | addReplyBulkCString(c,"num-other-sentinels"); | |
1427 | addReplyBulkLongLong(c,dictSize(ri->sentinels)); | |
1428 | fields++; | |
1429 | ||
1430 | addReplyBulkCString(c,"quorum"); | |
1431 | addReplyBulkLongLong(c,ri->quorum); | |
1432 | fields++; | |
1433 | } | |
1434 | ||
1435 | /* Only slaves */ | |
1436 | if (ri->flags & SRI_SLAVE) { | |
1437 | addReplyBulkCString(c,"master-link-down-time"); | |
1438 | addReplyBulkLongLong(c,ri->master_link_down_time); | |
1439 | fields++; | |
1440 | ||
1441 | addReplyBulkCString(c,"master-link-status"); | |
1442 | addReplyBulkCString(c, | |
1443 | (ri->slave_master_link_status == SENTINEL_MASTER_LINK_STATUS_UP) ? | |
1444 | "ok" : "err"); | |
1445 | fields++; | |
1446 | ||
1447 | addReplyBulkCString(c,"master-host"); | |
1448 | addReplyBulkCString(c, | |
1449 | ri->slave_master_host ? ri->slave_master_host : "?"); | |
1450 | fields++; | |
1451 | ||
1452 | addReplyBulkCString(c,"master-port"); | |
1453 | addReplyBulkLongLong(c,ri->slave_master_port); | |
1454 | fields++; | |
1455 | } | |
1456 | ||
1457 | /* Only sentinels */ | |
1458 | if (ri->flags & SRI_SENTINEL) { | |
1459 | addReplyBulkCString(c,"last-hello-message"); | |
1460 | addReplyBulkLongLong(c,mstime() - ri->last_hello_time); | |
1461 | fields++; | |
1462 | ||
1463 | addReplyBulkCString(c,"can-failover-its-master"); | |
1464 | addReplyBulkLongLong(c,(ri->flags & SRI_CAN_FAILOVER) != 0); | |
1465 | fields++; | |
1466 | ||
1467 | if (ri->flags & SRI_MASTER_DOWN) { | |
1468 | addReplyBulkCString(c,"subjective-leader"); | |
1469 | addReplyBulkCString(c,ri->leader ? ri->leader : "?"); | |
1470 | fields++; | |
1471 | } | |
1472 | } | |
1473 | ||
1474 | setDeferredMultiBulkLength(c,mbl,fields*2); | |
1475 | } | |
1476 | ||
1477 | /* Output a number of instances contanined inside a dictionary as | |
1478 | * Redis protocol. */ | |
1479 | void addReplyDictOfRedisInstances(redisClient *c, dict *instances) { | |
1480 | dictIterator *di; | |
1481 | dictEntry *de; | |
1482 | ||
1483 | di = dictGetIterator(instances); | |
1484 | addReplyMultiBulkLen(c,dictSize(instances)); | |
1485 | while((de = dictNext(di)) != NULL) { | |
1486 | sentinelRedisInstance *ri = dictGetVal(de); | |
1487 | ||
1488 | addReplySentinelRedisInstance(c,ri); | |
1489 | } | |
1490 | dictReleaseIterator(di); | |
1491 | } | |
1492 | ||
1493 | /* Lookup the named master into sentinel.masters. | |
1494 | * If the master is not found reply to the client with an error and returns | |
1495 | * NULL. */ | |
1496 | sentinelRedisInstance *sentinelGetMasterByNameOrReplyError(redisClient *c, | |
1497 | robj *name) | |
1498 | { | |
1499 | sentinelRedisInstance *ri; | |
1500 | ||
1501 | ri = dictFetchValue(sentinel.masters,c->argv[2]->ptr); | |
1502 | if (!ri) { | |
1503 | addReplyError(c,"No such master with that name"); | |
1504 | return NULL; | |
1505 | } | |
1506 | return ri; | |
1507 | } | |
1508 | ||
1509 | void sentinelCommand(redisClient *c) { | |
1510 | if (!strcasecmp(c->argv[1]->ptr,"masters")) { | |
1511 | /* SENTINEL MASTERS */ | |
1512 | if (c->argc != 2) goto numargserr; | |
1513 | ||
1514 | addReplyDictOfRedisInstances(c,sentinel.masters); | |
1515 | } else if (!strcasecmp(c->argv[1]->ptr,"slaves")) { | |
1516 | /* SENTINEL SLAVES <master-name> */ | |
1517 | sentinelRedisInstance *ri; | |
1518 | ||
1519 | if (c->argc != 3) goto numargserr; | |
1520 | if ((ri = sentinelGetMasterByNameOrReplyError(c,c->argv[2])) == NULL) | |
1521 | return; | |
1522 | addReplyDictOfRedisInstances(c,ri->slaves); | |
1523 | } else if (!strcasecmp(c->argv[1]->ptr,"sentinels")) { | |
1524 | /* SENTINEL SENTINELS <master-name> */ | |
1525 | sentinelRedisInstance *ri; | |
1526 | ||
1527 | if (c->argc != 3) goto numargserr; | |
1528 | if ((ri = sentinelGetMasterByNameOrReplyError(c,c->argv[2])) == NULL) | |
1529 | return; | |
1530 | addReplyDictOfRedisInstances(c,ri->sentinels); | |
1531 | } else if (!strcasecmp(c->argv[1]->ptr,"is-master-down-by-addr")) { | |
1532 | /* SENTINEL IS-MASTER-DOWN-BY-ADDR <ip> <port> */ | |
1533 | sentinelRedisInstance *ri; | |
1534 | char *leader = NULL; | |
1535 | long port; | |
1536 | int isdown = 0; | |
1537 | ||
1538 | if (c->argc != 4) goto numargserr; | |
1539 | if (getLongFromObjectOrReply(c,c->argv[3],&port,NULL) != REDIS_OK) | |
1540 | return; | |
1541 | ri = getSentinelRedisInstanceByAddrAndRunID(sentinel.masters, | |
1542 | c->argv[2]->ptr,port,NULL); | |
1543 | ||
1544 | /* It exists? Is actually a master? Is subjectively down? It's down. | |
1545 | * Note: if we are in tilt mode we always reply with "0". */ | |
1546 | if (!sentinel.tilt && ri && (ri->flags & SRI_S_DOWN) && | |
1547 | (ri->flags & SRI_MASTER)) | |
1548 | isdown = 1; | |
1549 | if (ri) leader = sentinelGetSubjectiveLeader(ri); | |
1550 | ||
1551 | /* Reply with a two-elements multi-bulk reply: down state, leader. */ | |
1552 | addReplyMultiBulkLen(c,2); | |
1553 | addReply(c, isdown ? shared.cone : shared.czero); | |
1554 | addReplyBulkCString(c, leader ? leader : "?"); | |
1555 | if (leader) sdsfree(leader); | |
1556 | } else if (!strcasecmp(c->argv[1]->ptr,"reset")) { | |
1557 | /* SENTINEL RESET <pattern> */ | |
1558 | if (c->argc != 3) goto numargserr; | |
1559 | addReplyLongLong(c,sentinelResetMastersByPattern(c->argv[2]->ptr)); | |
1560 | } else if (!strcasecmp(c->argv[1]->ptr,"get-master-addr-by-name")) { | |
1561 | /* SENTINEL GET-MASTER-ADDR-BY-NAME <master-name> */ | |
1562 | sentinelRedisInstance *ri; | |
1563 | ||
1564 | if (c->argc != 3) goto numargserr; | |
1565 | ri = sentinelGetMasterByName(c->argv[2]->ptr); | |
1566 | if (ri == NULL) { | |
1567 | addReply(c,shared.nullmultibulk); | |
1568 | } else { | |
1569 | sentinelAddr *addr = ri->addr; | |
1570 | ||
1571 | if ((ri->flags & SRI_FAILOVER_IN_PROGRESS) && ri->promoted_slave) | |
1572 | addr = ri->promoted_slave->addr; | |
1573 | addReplyMultiBulkLen(c,2); | |
1574 | addReplyBulkCString(c,addr->ip); | |
1575 | addReplyBulkLongLong(c,addr->port); | |
1576 | } | |
1577 | } else { | |
1578 | addReplyErrorFormat(c,"Unknown sentinel subcommand '%s'", | |
1579 | (char*)c->argv[1]->ptr); | |
1580 | } | |
1581 | return; | |
1582 | ||
1583 | numargserr: | |
1584 | addReplyErrorFormat(c,"Wrong number of commands for 'sentinel %s'", | |
1585 | (char*)c->argv[1]->ptr); | |
1586 | } | |
1587 | ||
1588 | /* ===================== SENTINEL availability checks ======================= */ | |
1589 | ||
1590 | /* Is this instance down from our point of view? */ | |
1591 | void sentinelCheckSubjectivelyDown(sentinelRedisInstance *ri) { | |
1592 | mstime_t elapsed = mstime() - ri->last_avail_time; | |
1593 | ||
1594 | /* Check if we are in need for a reconnection of one of the | |
1595 | * links, because we are detecting low activity. | |
1596 | * | |
1597 | * 1) Check if the command link seems connected, was connected not less | |
1598 | * than SENTINEL_MIN_LINK_RECONNECT_PERIOD, but still we have an | |
1599 | * idle time that is greater than down_after_period / 2 seconds. */ | |
1600 | if (ri->cc && | |
1601 | (mstime() - ri->cc_conn_time) > SENTINEL_MIN_LINK_RECONNECT_PERIOD && | |
1602 | (mstime() - ri->last_pong_time) > (ri->down_after_period/2)) | |
1603 | { | |
1604 | redisAsyncFree(ri->cc); /* will call the disconnection callback */ | |
1605 | } | |
1606 | ||
1607 | /* 2) Check if the pubsub link seems connected, was connected not less | |
1608 | * than SENTINEL_MIN_LINK_RECONNECT_PERIOD, but still we have no | |
1609 | * activity in the Pub/Sub channel for more than | |
1610 | * SENTINEL_PUBLISH_PERIOD * 3. | |
1611 | */ | |
1612 | if (ri->pc && | |
1613 | (mstime() - ri->pc_conn_time) > SENTINEL_MIN_LINK_RECONNECT_PERIOD && | |
1614 | (mstime() - ri->pc_last_activity) > (SENTINEL_PUBLISH_PERIOD*3)) | |
1615 | { | |
1616 | redisAsyncFree(ri->pc); /* will call the disconnection callback */ | |
1617 | } | |
1618 | ||
1619 | /* Update the subjectively down flag. */ | |
1620 | if (elapsed > ri->down_after_period) { | |
1621 | /* Is subjectively down */ | |
1622 | if ((ri->flags & SRI_S_DOWN) == 0) { | |
1623 | sentinelEvent(REDIS_WARNING,"+sdown",ri,"%@"); | |
1624 | ri->s_down_since_time = mstime(); | |
1625 | ri->flags |= SRI_S_DOWN; | |
1626 | } | |
1627 | } else { | |
1628 | /* Is subjectively up */ | |
1629 | if (ri->flags & SRI_S_DOWN) { | |
1630 | sentinelEvent(REDIS_WARNING,"-sdown",ri,"%@"); | |
1631 | ri->flags &= ~SRI_S_DOWN; | |
1632 | } | |
1633 | } | |
1634 | } | |
1635 | ||
1636 | /* Is this instance down accordingly to the configured quorum? */ | |
1637 | void sentinelCheckObjectivelyDown(sentinelRedisInstance *master) { | |
1638 | dictIterator *di; | |
1639 | dictEntry *de; | |
1640 | int quorum = 0, odown = 0; | |
1641 | ||
1642 | if (master->flags & SRI_S_DOWN) { | |
1643 | /* Is down for enough sentinels? */ | |
1644 | quorum = 1; /* the current sentinel. */ | |
1645 | /* Count all the other sentinels. */ | |
1646 | di = dictGetIterator(master->sentinels); | |
1647 | while((de = dictNext(di)) != NULL) { | |
1648 | sentinelRedisInstance *ri = dictGetVal(de); | |
1649 | ||
1650 | if (ri->flags & SRI_MASTER_DOWN) quorum++; | |
1651 | } | |
1652 | dictReleaseIterator(di); | |
1653 | if (quorum >= master->quorum) odown = 1; | |
1654 | } | |
1655 | ||
1656 | /* Set the flag accordingly to the outcome. */ | |
1657 | if (odown) { | |
1658 | if ((master->flags & SRI_O_DOWN) == 0) { | |
1659 | sentinelEvent(REDIS_WARNING,"+odown",master,"%@ #quorum %d/%d", | |
1660 | quorum, master->quorum); | |
1661 | master->flags |= SRI_O_DOWN; | |
1662 | master->o_down_since_time = mstime(); | |
1663 | } | |
1664 | } else { | |
1665 | if (master->flags & SRI_O_DOWN) { | |
1666 | sentinelEvent(REDIS_WARNING,"-odown",master,"%@"); | |
1667 | master->flags &= ~SRI_O_DOWN; | |
1668 | } | |
1669 | } | |
1670 | } | |
1671 | ||
1672 | /* Receive the SENTINEL is-master-down-by-addr reply, see the | |
1673 | * sentinelAskMasterStateToOtherSentinels() function for more information. */ | |
1674 | void sentinelReceiveIsMasterDownReply(redisAsyncContext *c, void *reply, void *privdata) { | |
1675 | sentinelRedisInstance *ri = c->data; | |
1676 | redisReply *r; | |
1677 | ||
1678 | ri->pending_commands--; | |
1679 | if (!reply) return; | |
1680 | r = reply; | |
1681 | ||
1682 | /* Ignore every error or unexpected reply. | |
1683 | * Note that if the command returns an error for any reason we'll | |
1684 | * end clearing the SRI_MASTER_DOWN flag for timeout anyway. */ | |
1685 | if (r->type == REDIS_REPLY_ARRAY && r->elements == 2 && | |
1686 | r->element[0]->type == REDIS_REPLY_INTEGER && | |
1687 | r->element[1]->type == REDIS_REPLY_STRING) | |
1688 | { | |
1689 | ri->last_master_down_reply_time = mstime(); | |
1690 | if (r->element[0]->integer == 1) { | |
1691 | ri->flags |= SRI_MASTER_DOWN; | |
1692 | } else { | |
1693 | ri->flags &= ~SRI_MASTER_DOWN; | |
1694 | } | |
1695 | sdsfree(ri->leader); | |
1696 | ri->leader = sdsnew(r->element[1]->str); | |
1697 | } | |
1698 | } | |
1699 | ||
1700 | /* If we think (subjectively) the master is down, we start sending | |
1701 | * SENTINEL IS-MASTER-DOWN-BY-ADDR requests to other sentinels | |
1702 | * in order to get the replies that allow to reach the quorum and | |
1703 | * possibly also mark the master as objectively down. */ | |
1704 | void sentinelAskMasterStateToOtherSentinels(sentinelRedisInstance *master) { | |
1705 | dictIterator *di; | |
1706 | dictEntry *de; | |
1707 | ||
1708 | di = dictGetIterator(master->sentinels); | |
1709 | while((de = dictNext(di)) != NULL) { | |
1710 | sentinelRedisInstance *ri = dictGetVal(de); | |
1711 | mstime_t elapsed = mstime() - ri->last_master_down_reply_time; | |
1712 | char port[32]; | |
1713 | int retval; | |
1714 | ||
1715 | /* If the master state from other sentinel is too old, we clear it. */ | |
1716 | if (elapsed > SENTINEL_INFO_VALIDITY_TIME) { | |
1717 | ri->flags &= ~SRI_MASTER_DOWN; | |
1718 | sdsfree(ri->leader); | |
1719 | ri->leader = NULL; | |
1720 | } | |
1721 | ||
1722 | /* Only ask if master is down to other sentinels if: | |
1723 | * | |
1724 | * 1) We believe it is down, or there is a failover in progress. | |
1725 | * 2) Sentinel is connected. | |
1726 | * 3) We did not received the info within SENTINEL_ASK_PERIOD ms. */ | |
1727 | if ((master->flags & (SRI_S_DOWN|SRI_FAILOVER_IN_PROGRESS)) == 0) | |
1728 | continue; | |
1729 | if (ri->flags & SRI_DISCONNECTED) continue; | |
1730 | if (mstime() - ri->last_master_down_reply_time < SENTINEL_ASK_PERIOD) | |
1731 | continue; | |
1732 | ||
1733 | /* Ask */ | |
1734 | ll2string(port,sizeof(port),master->addr->port); | |
1735 | retval = redisAsyncCommand(ri->cc, | |
1736 | sentinelReceiveIsMasterDownReply, NULL, | |
1737 | "SENTINEL is-master-down-by-addr %s %s", | |
1738 | master->addr->ip, port); | |
1739 | if (retval == REDIS_OK) ri->pending_commands++; | |
1740 | } | |
1741 | dictReleaseIterator(di); | |
1742 | } | |
1743 | ||
1744 | /* =============================== FAILOVER ================================= */ | |
1745 | ||
1746 | /* Given a master get the "subjective leader", that is, among all the sentinels | |
1747 | * with given characteristics, the one with the lexicographically smaller | |
1748 | * runid. The characteristics required are: | |
1749 | * | |
1750 | * 1) Has SRI_CAN_FAILOVER flag. | |
1751 | * 2) Is not disconnected. | |
1752 | * 3) Recently answered to our ping (no longer than | |
1753 | * SENTINEL_INFO_VALIDITY_TIME milliseconds ago). | |
1754 | * | |
1755 | * The function returns a pointer to an sds string representing the runid of the | |
1756 | * leader sentinel instance (from our point of view). Otherwise NULL is | |
1757 | * returned if there are no suitable sentinels. | |
1758 | */ | |
1759 | ||
1760 | int compareRunID(const void *a, const void *b) { | |
1761 | char **aptrptr = (char**)a, **bptrptr = (char**)b; | |
1762 | return strcasecmp(*aptrptr, *bptrptr); | |
1763 | } | |
1764 | ||
1765 | char *sentinelGetSubjectiveLeader(sentinelRedisInstance *master) { | |
1766 | dictIterator *di; | |
1767 | dictEntry *de; | |
1768 | char **instance = | |
1769 | zmalloc(sizeof(char*)*(dictSize(master->sentinels)+1)); | |
1770 | int instances = 0; | |
1771 | char *leader = NULL; | |
1772 | ||
1773 | if (master->flags & SRI_CAN_FAILOVER) { | |
1774 | /* Add myself if I'm a Sentinel that can failover this master. */ | |
1775 | instance[instances++] = server.runid; | |
1776 | } | |
1777 | ||
1778 | di = dictGetIterator(master->sentinels); | |
1779 | while((de = dictNext(di)) != NULL) { | |
1780 | sentinelRedisInstance *ri = dictGetVal(de); | |
1781 | mstime_t lag = mstime() - ri->last_avail_time; | |
1782 | ||
1783 | if (lag > SENTINEL_INFO_VALIDITY_TIME || | |
1784 | !(ri->flags & SRI_CAN_FAILOVER) || | |
1785 | (ri->flags & SRI_DISCONNECTED) || | |
1786 | ri->runid == NULL) | |
1787 | continue; | |
1788 | instance[instances++] = ri->runid; | |
1789 | } | |
1790 | dictReleaseIterator(di); | |
1791 | ||
1792 | /* If we have at least one instance passing our checks, order the array | |
1793 | * by runid. */ | |
1794 | if (instances) { | |
1795 | qsort(instance,instances,sizeof(char*),compareRunID); | |
1796 | leader = sdsnew(instance[0]); | |
1797 | } | |
1798 | zfree(instance); | |
1799 | return leader; | |
1800 | } | |
1801 | ||
1802 | struct sentinelLeader { | |
1803 | char *runid; | |
1804 | unsigned long votes; | |
1805 | }; | |
1806 | ||
1807 | /* Helper function for sentinelGetObjectiveLeader, increment the counter | |
1808 | * relative to the specified runid. */ | |
1809 | void sentinelObjectiveLeaderIncr(dict *counters, char *runid) { | |
1810 | dictEntry *de = dictFind(counters,runid); | |
1811 | uint64_t oldval; | |
1812 | ||
1813 | if (de) { | |
1814 | oldval = dictGetUnsignedIntegerVal(de); | |
1815 | dictSetUnsignedIntegerVal(de,oldval+1); | |
1816 | } else { | |
1817 | de = dictAddRaw(counters,runid); | |
1818 | redisAssert(de != NULL); | |
1819 | dictSetUnsignedIntegerVal(de,1); | |
1820 | } | |
1821 | } | |
1822 | ||
1823 | /* Scan all the Sentinels attached to this master to check what is the | |
1824 | * most voted leader among Sentinels. */ | |
1825 | char *sentinelGetObjectiveLeader(sentinelRedisInstance *master) { | |
1826 | dict *counters; | |
1827 | dictIterator *di; | |
1828 | dictEntry *de; | |
1829 | unsigned int voters = 0, voters_quorum; | |
1830 | char *myvote; | |
1831 | char *winner = NULL; | |
1832 | ||
1833 | redisAssert(master->flags & (SRI_O_DOWN|SRI_FAILOVER_IN_PROGRESS)); | |
1834 | counters = dictCreate(&leaderVotesDictType,NULL); | |
1835 | ||
1836 | /* Count my vote. */ | |
1837 | myvote = sentinelGetSubjectiveLeader(master); | |
1838 | if (myvote) { | |
1839 | sentinelObjectiveLeaderIncr(counters,myvote); | |
1840 | voters++; | |
1841 | } | |
1842 | ||
1843 | /* Count other sentinels votes */ | |
1844 | di = dictGetIterator(master->sentinels); | |
1845 | while((de = dictNext(di)) != NULL) { | |
1846 | sentinelRedisInstance *ri = dictGetVal(de); | |
1847 | if (ri->leader == NULL) continue; | |
1848 | /* If the failover is not already in progress we are only interested | |
1849 | * in Sentinels that believe the master is down. Otherwise the leader | |
1850 | * selection is useful for the "failover-takedown" when the original | |
1851 | * leader fails. In that case we consider all the voters. */ | |
1852 | if (!(master->flags & SRI_FAILOVER_IN_PROGRESS) && | |
1853 | !(ri->flags & SRI_MASTER_DOWN)) continue; | |
1854 | sentinelObjectiveLeaderIncr(counters,ri->leader); | |
1855 | voters++; | |
1856 | } | |
1857 | dictReleaseIterator(di); | |
1858 | voters_quorum = voters/2+1; | |
1859 | ||
1860 | /* Check what's the winner. For the winner to win, it needs two conditions: | |
1861 | * 1) Absolute majority between voters (50% + 1). | |
1862 | * 2) And anyway at least master->quorum votes. */ | |
1863 | { | |
1864 | uint64_t max_votes = 0; /* Max votes so far. */ | |
1865 | ||
1866 | di = dictGetIterator(counters); | |
1867 | while((de = dictNext(di)) != NULL) { | |
1868 | uint64_t votes = dictGetUnsignedIntegerVal(de); | |
1869 | ||
1870 | if (max_votes < votes) { | |
1871 | max_votes = votes; | |
1872 | winner = dictGetKey(de); | |
1873 | } | |
1874 | } | |
1875 | dictReleaseIterator(di); | |
1876 | if (winner && (max_votes < voters_quorum || max_votes < master->quorum)) | |
1877 | winner = NULL; | |
1878 | } | |
1879 | winner = winner ? sdsnew(winner) : NULL; | |
1880 | sdsfree(myvote); | |
1881 | dictRelease(counters); | |
1882 | return winner; | |
1883 | } | |
1884 | ||
1885 | /* This function checks if there are the conditions to start the failover, | |
1886 | * that is: | |
1887 | * | |
1888 | * 1) Enough time has passed since O_DOWN. | |
1889 | * 2) The master is marked as SRI_CAN_FAILOVER, so we can failover it. | |
1890 | * 3) We are the objectively leader for this master. | |
1891 | * | |
1892 | * If the conditions are met we flag the master as SRI_FAILOVER_IN_PROGRESS | |
1893 | * and SRI_I_AM_THE_LEADER. | |
1894 | */ | |
1895 | void sentinelStartFailover(sentinelRedisInstance *master) { | |
1896 | char *leader; | |
1897 | int isleader; | |
1898 | ||
1899 | /* We can't failover if the master is not in O_DOWN state or if | |
1900 | * there is not already a failover in progress (to perform the | |
1901 | * takedown if the leader died) or if this Sentinel is not allowed | |
1902 | * to start a failover. */ | |
1903 | if (!(master->flags & SRI_CAN_FAILOVER) || | |
1904 | !(master->flags & (SRI_O_DOWN|SRI_FAILOVER_IN_PROGRESS))) return; | |
1905 | ||
1906 | leader = sentinelGetObjectiveLeader(master); | |
1907 | isleader = leader && strcasecmp(leader,server.runid) == 0; | |
1908 | sdsfree(leader); | |
1909 | ||
1910 | /* If I'm not the leader, I can't failover for sure. */ | |
1911 | if (!isleader) return; | |
1912 | ||
1913 | /* If the failover is already in progress there are two options... */ | |
1914 | if (master->flags & SRI_FAILOVER_IN_PROGRESS) { | |
1915 | if (master->flags & SRI_I_AM_THE_LEADER) { | |
1916 | /* 1) I'm flagged as leader so I already started the failover. | |
1917 | * Just return. */ | |
1918 | return; | |
1919 | } else { | |
1920 | mstime_t elapsed = mstime() - master->failover_state_change_time; | |
1921 | ||
1922 | /* 2) I'm the new leader, but I'm not flagged as leader in the | |
1923 | * master: I did not started the failover, but the original | |
1924 | * leader has no longer the leadership. | |
1925 | * | |
1926 | * In this case if the failover appears to be lagging | |
1927 | * for at least 25% of the configured failover timeout, | |
1928 | * I can assume I can take control. Otherwise | |
1929 | * it's better to return and wait more. */ | |
1930 | if (elapsed < (master->failover_timeout/4)) return; | |
1931 | sentinelEvent(REDIS_WARNING,"+failover-takedown",master,"%@"); | |
1932 | /* We have already an elected slave if we are in | |
1933 | * FAILOVER_IN_PROGRESS state, that is, the slave that we | |
1934 | * observed turning into a master. */ | |
1935 | master->failover_state = SENTINEL_FAILOVER_STATE_RECONF_SLAVES; | |
1936 | /* As an observer we flagged all the slaves as RECONF_SENT but | |
1937 | * now we are in charge of actually sending the reconfiguration | |
1938 | * command so let's clear this flag for all the instances. */ | |
1939 | sentinelDelFlagsToDictOfRedisInstances(master->slaves, | |
1940 | SRI_RECONF_SENT); | |
1941 | } | |
1942 | } else { | |
1943 | /* Brand new failover as SRI_FAILOVER_IN_PROGRESS was not set. */ | |
1944 | master->failover_state = SENTINEL_FAILOVER_STATE_WAIT_START; | |
1945 | } | |
1946 | ||
1947 | master->flags |= SRI_FAILOVER_IN_PROGRESS|SRI_I_AM_THE_LEADER; | |
1948 | sentinelEvent(REDIS_WARNING,"+failover-triggered",master,"%@"); | |
1949 | ||
1950 | /* Pick a random delay if it's a fresh failover (WAIT_START), and not | |
1951 | * a recovery of a failover started by another sentinel. */ | |
1952 | if (master->failover_state == SENTINEL_FAILOVER_STATE_WAIT_START) { | |
1953 | master->failover_start_time = mstime() + | |
1954 | SENTINEL_FAILOVER_FIXED_DELAY + | |
1955 | (rand() % SENTINEL_FAILOVER_MAX_RANDOM_DELAY); | |
1956 | sentinelEvent(REDIS_WARNING,"+failover-state-wait-start",master, | |
1957 | "%@ #starting in %lld milliseconds", | |
1958 | master->failover_start_time-mstime()); | |
1959 | } | |
1960 | master->failover_state_change_time = mstime(); | |
1961 | } | |
1962 | ||
1963 | /* Select a suitable slave to promote. The current algorithm only uses | |
1964 | * the following parameters: | |
1965 | * | |
1966 | * 1) None of the following conditions: S_DOWN, O_DOWN, DISCONNECTED. | |
1967 | * 2) last_avail_time more recent than SENTINEL_INFO_VALIDITY_TIME. | |
1968 | * 3) info_refresh more recent than SENTINEL_INFO_VALIDITY_TIME. | |
1969 | * 4) master_link_down_time no more than: | |
1970 | * (now - master->s_down_since_time) + (master->down_after_period * 10). | |
1971 | * | |
1972 | * Among all the slaves matching the above conditions we select the slave | |
1973 | * with lower slave_priority. If priority is the same we select the slave | |
1974 | * with lexicographically smaller runid. | |
1975 | * | |
1976 | * The function returns the pointer to the selected slave, otherwise | |
1977 | * NULL if no suitable slave was found. | |
1978 | */ | |
1979 | ||
1980 | int compareSlavesForPromotion(const void *a, const void *b) { | |
1981 | sentinelRedisInstance **sa = (sentinelRedisInstance **)a, | |
1982 | **sb = (sentinelRedisInstance **)b; | |
1983 | if ((*sa)->slave_priority != (*sb)->slave_priority) | |
1984 | return (*sa)->slave_priority - (*sb)->slave_priority; | |
1985 | return strcasecmp((*sa)->runid,(*sb)->runid); | |
1986 | } | |
1987 | ||
1988 | sentinelRedisInstance *sentinelSelectSlave(sentinelRedisInstance *master) { | |
1989 | sentinelRedisInstance **instance = | |
1990 | zmalloc(sizeof(instance[0])*dictSize(master->slaves)); | |
1991 | sentinelRedisInstance *selected = NULL; | |
1992 | int instances = 0; | |
1993 | dictIterator *di; | |
1994 | dictEntry *de; | |
1995 | mstime_t max_master_down_time; | |
1996 | ||
1997 | max_master_down_time = (mstime() - master->s_down_since_time) + | |
1998 | (master->down_after_period * 10); | |
1999 | ||
2000 | di = dictGetIterator(master->slaves); | |
2001 | while((de = dictNext(di)) != NULL) { | |
2002 | sentinelRedisInstance *slave = dictGetVal(de); | |
2003 | mstime_t info_validity_time = mstime()-SENTINEL_INFO_VALIDITY_TIME; | |
2004 | ||
2005 | if (slave->flags & (SRI_S_DOWN|SRI_O_DOWN|SRI_DISCONNECTED)) continue; | |
2006 | if (slave->last_avail_time < info_validity_time) continue; | |
2007 | if (slave->info_refresh < info_validity_time) continue; | |
2008 | if (slave->master_link_down_time > max_master_down_time) continue; | |
2009 | instance[instances++] = slave; | |
2010 | } | |
2011 | dictReleaseIterator(di); | |
2012 | if (instances) { | |
2013 | qsort(instance,instances,sizeof(sentinelRedisInstance*), | |
2014 | compareSlavesForPromotion); | |
2015 | selected = instance[0]; | |
2016 | } | |
2017 | zfree(instance); | |
2018 | return selected; | |
2019 | } | |
2020 | ||
2021 | /* ---------------- Failover state machine implementation ------------------- */ | |
2022 | void sentinelFailoverWaitStart(sentinelRedisInstance *ri) { | |
2023 | if (mstime() >= ri->failover_start_time) { | |
2024 | ri->failover_state = SENTINEL_FAILOVER_STATE_SELECT_SLAVE; | |
2025 | ri->failover_state_change_time = mstime(); | |
2026 | sentinelEvent(REDIS_WARNING,"+failover-state-select-slave",ri,"%@"); | |
2027 | } | |
2028 | } | |
2029 | ||
2030 | void sentinelFailoverSelectSlave(sentinelRedisInstance *ri) { | |
2031 | sentinelRedisInstance *slave = sentinelSelectSlave(ri); | |
2032 | ||
2033 | if (slave == NULL) { | |
2034 | sentinelEvent(REDIS_WARNING,"-no-good-slave",ri, | |
2035 | "%@ #retrying in %d seconds", | |
2036 | (SENTINEL_FAILOVER_FIXED_DELAY+ | |
2037 | SENTINEL_FAILOVER_MAX_RANDOM_DELAY)/1000); | |
2038 | ri->failover_state = SENTINEL_FAILOVER_STATE_WAIT_START; | |
2039 | ri->failover_start_time = mstime() + SENTINEL_FAILOVER_FIXED_DELAY + | |
2040 | SENTINEL_FAILOVER_MAX_RANDOM_DELAY; | |
2041 | } else { | |
2042 | sentinelEvent(REDIS_WARNING,"+selected-slave",slave,"%@"); | |
2043 | slave->flags |= SRI_PROMOTED; | |
2044 | ri->promoted_slave = slave; | |
2045 | ri->failover_state = SENTINEL_FAILOVER_STATE_SEND_SLAVEOF_NOONE; | |
2046 | ri->failover_state_change_time = mstime(); | |
2047 | sentinelEvent(REDIS_NOTICE,"+failover-state-send-slaveof-noone", | |
2048 | slave, "%@"); | |
2049 | } | |
2050 | } | |
2051 | ||
2052 | void sentinelFailoverSendSlaveOfNoOne(sentinelRedisInstance *ri) { | |
2053 | int retval; | |
2054 | ||
2055 | if (ri->promoted_slave->flags & SRI_DISCONNECTED) return; | |
2056 | ||
2057 | /* Send SLAVEOF NO ONE command to turn the slave into a master. | |
2058 | * We actually register a generic callback for this command as we don't | |
2059 | * really care about the reply. We check if it worked indirectly observing | |
2060 | * if INFO returns a different role (master instead of slave). */ | |
2061 | retval = redisAsyncCommand(ri->promoted_slave->cc, | |
2062 | sentinelDiscardReplyCallback, NULL, "SLAVEOF NO ONE"); | |
2063 | if (retval != REDIS_OK) return; | |
2064 | ri->promoted_slave->pending_commands++; | |
2065 | sentinelEvent(REDIS_NOTICE, "+failover-state-wait-promotion", | |
2066 | ri->promoted_slave,"%@"); | |
2067 | ri->failover_state = SENTINEL_FAILOVER_STATE_WAIT_PROMOTION; | |
2068 | ri->failover_state_change_time = mstime(); | |
2069 | } | |
2070 | ||
2071 | /* We actually wait for promotion indirectly checking with INFO when the | |
2072 | * slave turns into a master. */ | |
2073 | void sentinelFailoverWaitPromotion(sentinelRedisInstance *ri) { | |
2074 | mstime_t elapsed = mstime() - ri->failover_state_change_time; | |
2075 | ||
2076 | if (elapsed >= SENTINEL_PROMOTION_RETRY_PERIOD) { | |
2077 | sentinelEvent(REDIS_WARNING,"-promotion-timeout",ri->promoted_slave, | |
2078 | "%@"); | |
2079 | sentinelEvent(REDIS_WARNING,"+failover-state-select-slave",ri,"%@"); | |
2080 | ri->failover_state = SENTINEL_FAILOVER_STATE_SELECT_SLAVE; | |
2081 | ri->failover_state_change_time = mstime(); | |
2082 | ri->promoted_slave->flags &= ~SRI_PROMOTED; | |
2083 | ri->promoted_slave = NULL; | |
2084 | } | |
2085 | } | |
2086 | ||
2087 | void sentinelFailoverDetectEnd(sentinelRedisInstance *master) { | |
2088 | int not_reconfigured = 0, timeout = 0; | |
2089 | dictIterator *di; | |
2090 | dictEntry *de; | |
2091 | mstime_t elapsed = mstime() - master->failover_state_change_time; | |
2092 | ||
2093 | /* We can't consider failover finished if the promoted slave is | |
2094 | * not reachable. */ | |
2095 | if (master->promoted_slave == NULL || | |
2096 | master->promoted_slave->flags & SRI_S_DOWN) return; | |
2097 | ||
2098 | /* The failover terminates once all the reachable slaves are properly | |
2099 | * configured. */ | |
2100 | di = dictGetIterator(master->slaves); | |
2101 | while((de = dictNext(di)) != NULL) { | |
2102 | sentinelRedisInstance *slave = dictGetVal(de); | |
2103 | ||
2104 | if (slave->flags & (SRI_PROMOTED|SRI_RECONF_DONE)) continue; | |
2105 | if (slave->flags & SRI_S_DOWN) continue; | |
2106 | not_reconfigured++; | |
2107 | } | |
2108 | dictReleaseIterator(di); | |
2109 | ||
2110 | /* Force end of failover on timeout. */ | |
2111 | if (elapsed > master->failover_timeout) { | |
2112 | not_reconfigured = 0; | |
2113 | timeout = 1; | |
2114 | sentinelEvent(REDIS_WARNING,"+failover-end-for-timeout",master,"%@"); | |
2115 | } | |
2116 | ||
2117 | if (not_reconfigured == 0) { | |
2118 | sentinelEvent(REDIS_WARNING,"+failover-end",master,"%@"); | |
2119 | master->failover_state = SENTINEL_FAILOVER_STATE_UPDATE_CONFIG; | |
2120 | master->failover_state_change_time = mstime(); | |
2121 | } | |
2122 | ||
2123 | /* If I'm the leader it is a good idea to send a best effort SLAVEOF | |
2124 | * command to all the slaves still not reconfigured to replicate with | |
2125 | * the new master. */ | |
2126 | if (timeout && (master->flags & SRI_I_AM_THE_LEADER)) { | |
2127 | dictIterator *di; | |
2128 | dictEntry *de; | |
2129 | char master_port[32]; | |
2130 | ||
2131 | ll2string(master_port,sizeof(master_port), | |
2132 | master->promoted_slave->addr->port); | |
2133 | ||
2134 | di = dictGetIterator(master->slaves); | |
2135 | while((de = dictNext(di)) != NULL) { | |
2136 | sentinelRedisInstance *slave = dictGetVal(de); | |
2137 | int retval; | |
2138 | ||
2139 | if (slave->flags & | |
2140 | (SRI_RECONF_DONE|SRI_RECONF_SENT|SRI_DISCONNECTED)) continue; | |
2141 | ||
2142 | retval = redisAsyncCommand(slave->cc, | |
2143 | sentinelDiscardReplyCallback, NULL, "SLAVEOF %s %s", | |
2144 | master->promoted_slave->addr->ip, | |
2145 | master_port); | |
2146 | if (retval == REDIS_OK) { | |
2147 | sentinelEvent(REDIS_NOTICE,"+slave-reconf-sent-be",slave,"%@"); | |
2148 | slave->flags |= SRI_RECONF_SENT; | |
2149 | } | |
2150 | } | |
2151 | dictReleaseIterator(di); | |
2152 | } | |
2153 | } | |
2154 | ||
2155 | /* Send SLAVE OF <new master address> to all the remaining slaves that | |
2156 | * still don't appear to have the configuration updated. */ | |
2157 | void sentinelFailoverReconfNextSlave(sentinelRedisInstance *master) { | |
2158 | dictIterator *di; | |
2159 | dictEntry *de; | |
2160 | int in_progress = 0; | |
2161 | ||
2162 | di = dictGetIterator(master->slaves); | |
2163 | while((de = dictNext(di)) != NULL) { | |
2164 | sentinelRedisInstance *slave = dictGetVal(de); | |
2165 | ||
2166 | if (slave->flags & (SRI_RECONF_SENT|SRI_RECONF_INPROG)) | |
2167 | in_progress++; | |
2168 | } | |
2169 | dictReleaseIterator(di); | |
2170 | ||
2171 | di = dictGetIterator(master->slaves); | |
2172 | while(in_progress < master->parallel_syncs && | |
2173 | (de = dictNext(di)) != NULL) | |
2174 | { | |
2175 | sentinelRedisInstance *slave = dictGetVal(de); | |
2176 | int retval; | |
2177 | char master_port[32]; | |
2178 | ||
2179 | /* Skip the promoted slave, and already configured slaves. */ | |
2180 | if (slave->flags & (SRI_PROMOTED|SRI_RECONF_DONE)) continue; | |
2181 | ||
2182 | /* Clear the SRI_RECONF_SENT flag if too much time elapsed without | |
2183 | * the slave moving forward to the next state. */ | |
2184 | if ((slave->flags & SRI_RECONF_SENT) && | |
2185 | (mstime() - slave->slave_reconf_sent_time) > | |
2186 | SENTINEL_SLAVE_RECONF_RETRY_PERIOD) | |
2187 | { | |
2188 | sentinelEvent(REDIS_NOTICE,"-slave-reconf-sent-timeout",slave,"%@"); | |
2189 | slave->flags &= ~SRI_RECONF_SENT; | |
2190 | } | |
2191 | ||
2192 | /* Nothing to do for instances that are disconnected or already | |
2193 | * in RECONF_SENT state. */ | |
2194 | if (slave->flags & (SRI_DISCONNECTED|SRI_RECONF_SENT|SRI_RECONF_INPROG)) | |
2195 | continue; | |
2196 | ||
2197 | /* Send SLAVEOF <new master>. */ | |
2198 | ll2string(master_port,sizeof(master_port), | |
2199 | master->promoted_slave->addr->port); | |
2200 | retval = redisAsyncCommand(slave->cc, | |
2201 | sentinelDiscardReplyCallback, NULL, "SLAVEOF %s %s", | |
2202 | master->promoted_slave->addr->ip, | |
2203 | master_port); | |
2204 | if (retval == REDIS_OK) { | |
2205 | slave->flags |= SRI_RECONF_SENT; | |
2206 | slave->pending_commands++; | |
2207 | slave->slave_reconf_sent_time = mstime(); | |
2208 | sentinelEvent(REDIS_NOTICE,"+slave-reconf-sent",slave,"%@"); | |
2209 | in_progress++; | |
2210 | } | |
2211 | } | |
2212 | dictReleaseIterator(di); | |
2213 | sentinelFailoverDetectEnd(master); | |
2214 | } | |
2215 | ||
2216 | /* This function is called when the slave is in | |
2217 | * SENTINEL_FAILOVER_STATE_UPDATE_CONFIG state. In this state we need | |
2218 | * to remove it from the master table and add the promoted slave instead. | |
2219 | * | |
2220 | * If there are no promoted slaves as this instance is unique, we remove | |
2221 | * and re-add it with the same address to trigger a complete state | |
2222 | * refresh. */ | |
2223 | void sentinelFailoverSwitchToPromotedSlave(sentinelRedisInstance *master) { | |
2224 | sentinelRedisInstance *new, *ref = master->promoted_slave ? | |
2225 | master->promoted_slave : master; | |
2226 | int quorum = ref->quorum, parallel_syncs = ref->parallel_syncs; | |
2227 | char *name = sdsnew(master->name); | |
2228 | char *ip = sdsnew(ref->addr->ip), *oldip = sdsnew(master->addr->ip); | |
2229 | int port = ref->addr->port, oldport = master->addr->port; | |
2230 | int retval, oldflags = master->flags; | |
2231 | mstime_t old_down_after_period = master->down_after_period; | |
2232 | mstime_t old_failover_timeout = master->failover_timeout; | |
2233 | ||
2234 | retval = dictDelete(sentinel.masters,master->name); | |
2235 | redisAssert(retval == DICT_OK); | |
2236 | new = createSentinelRedisInstance(name,SRI_MASTER,ip,port,quorum,NULL); | |
2237 | redisAssert(new != NULL); | |
2238 | new->parallel_syncs = parallel_syncs; | |
2239 | new->flags |= (oldflags & SRI_CAN_FAILOVER); | |
2240 | new->down_after_period = old_down_after_period; | |
2241 | new->failover_timeout = old_failover_timeout; | |
2242 | /* TODO: ... set the scripts as well. */ | |
2243 | sentinelEvent(REDIS_WARNING,"+switch-master",new,"%s %s %d %s %d", | |
2244 | name, oldip, oldport, ip, port); | |
2245 | sdsfree(name); | |
2246 | sdsfree(ip); | |
2247 | sdsfree(oldip); | |
2248 | } | |
2249 | ||
2250 | void sentinelFailoverStateMachine(sentinelRedisInstance *ri) { | |
2251 | redisAssert(ri->flags & SRI_MASTER); | |
2252 | ||
2253 | if (!(ri->flags & SRI_FAILOVER_IN_PROGRESS)) return; | |
2254 | ||
2255 | switch(ri->failover_state) { | |
2256 | case SENTINEL_FAILOVER_STATE_WAIT_START: | |
2257 | sentinelFailoverWaitStart(ri); | |
2258 | break; | |
2259 | case SENTINEL_FAILOVER_STATE_SELECT_SLAVE: | |
2260 | sentinelFailoverSelectSlave(ri); | |
2261 | break; | |
2262 | case SENTINEL_FAILOVER_STATE_SEND_SLAVEOF_NOONE: | |
2263 | sentinelFailoverSendSlaveOfNoOne(ri); | |
2264 | break; | |
2265 | case SENTINEL_FAILOVER_STATE_WAIT_PROMOTION: | |
2266 | sentinelFailoverWaitPromotion(ri); | |
2267 | break; | |
2268 | case SENTINEL_FAILOVER_STATE_RECONF_SLAVES: | |
2269 | sentinelFailoverReconfNextSlave(ri); | |
2270 | break; | |
2271 | case SENTINEL_FAILOVER_STATE_DETECT_END: | |
2272 | sentinelFailoverDetectEnd(ri); | |
2273 | break; | |
2274 | } | |
2275 | } | |
2276 | ||
2277 | /* The following is called only for master instances and will abort the | |
2278 | * failover process if: | |
2279 | * | |
2280 | * 1) The failover is in progress. | |
2281 | * 2) We already promoted a slave. | |
2282 | * 3) The promoted slave is in extended SDOWN condition. | |
2283 | */ | |
2284 | void sentinelAbortFailoverIfNeeded(sentinelRedisInstance *ri) { | |
2285 | dictIterator *di; | |
2286 | dictEntry *de; | |
2287 | ||
2288 | /* Failover is in progress? Do we have a promoted slave? */ | |
2289 | if (!(ri->flags & SRI_FAILOVER_IN_PROGRESS) || !ri->promoted_slave) return; | |
2290 | ||
2291 | /* Is the promoted slave into an extended SDOWN state? */ | |
2292 | if (!(ri->promoted_slave->flags & SRI_S_DOWN) || | |
2293 | (mstime() - ri->promoted_slave->s_down_since_time) < | |
2294 | (ri->down_after_period * SENTINEL_EXTENDED_SDOWN_MULTIPLIER)) return; | |
2295 | ||
2296 | sentinelEvent(REDIS_WARNING,"-failover-abort-x-sdown",ri->promoted_slave,"%@"); | |
2297 | ||
2298 | /* Clear failover related flags from slaves. | |
2299 | * Also if we are the leader make sure to send SLAVEOF commands to all the | |
2300 | * already reconfigured slaves in order to turn them back into slaves of | |
2301 | * the original master. */ | |
2302 | ||
2303 | di = dictGetIterator(ri->slaves); | |
2304 | while((de = dictNext(di)) != NULL) { | |
2305 | sentinelRedisInstance *slave = dictGetVal(de); | |
2306 | if (ri->flags & SRI_I_AM_THE_LEADER) { | |
2307 | char master_port[32]; | |
2308 | int retval; | |
2309 | ||
2310 | ll2string(master_port,sizeof(master_port),ri->addr->port); | |
2311 | retval = redisAsyncCommand(slave->cc, | |
2312 | sentinelDiscardReplyCallback, NULL, "SLAVEOF %s %s", | |
2313 | ri->addr->ip, | |
2314 | master_port); | |
2315 | if (retval == REDIS_OK) | |
2316 | sentinelEvent(REDIS_NOTICE,"-slave-reconf-undo",slave,"%@"); | |
2317 | } | |
2318 | slave->flags &= ~(SRI_RECONF_SENT|SRI_RECONF_INPROG|SRI_RECONF_DONE); | |
2319 | } | |
2320 | dictReleaseIterator(di); | |
2321 | ||
2322 | ri->flags &= ~(SRI_FAILOVER_IN_PROGRESS|SRI_I_AM_THE_LEADER); | |
2323 | ri->failover_state = SENTINEL_FAILOVER_STATE_NONE; | |
2324 | ri->failover_state_change_time = mstime(); | |
2325 | ri->promoted_slave->flags &= ~SRI_PROMOTED; | |
2326 | ri->promoted_slave = NULL; | |
2327 | } | |
2328 | ||
2329 | /* ======================== SENTINEL timer handler ========================== | |
2330 | * This is the "main" our Sentinel, being sentinel completely non blocking | |
2331 | * in design. The function is called every second. | |
2332 | * -------------------------------------------------------------------------- */ | |
2333 | ||
2334 | /* Perform scheduled operations for the specified Redis instance. */ | |
2335 | void sentinelHandleRedisInstance(sentinelRedisInstance *ri) { | |
2336 | /* ========== MONITORING HALF ============ */ | |
2337 | /* Every kind of instance */ | |
2338 | sentinelReconnectInstance(ri); | |
2339 | sentinelPingInstance(ri); | |
2340 | ||
2341 | /* Masters and slaves */ | |
2342 | if (ri->flags & (SRI_MASTER|SRI_SLAVE)) { | |
2343 | /* Nothing so far. */ | |
2344 | } | |
2345 | ||
2346 | /* Only masters */ | |
2347 | if (ri->flags & SRI_MASTER) { | |
2348 | sentinelAskMasterStateToOtherSentinels(ri); | |
2349 | } | |
2350 | ||
2351 | /* ============== ACTING HALF ============= */ | |
2352 | /* We don't proceed with the acting half if we are in TILT mode. | |
2353 | * TILT happens when we find something odd with the time, like a | |
2354 | * sudden change in the clock. */ | |
2355 | if (sentinel.tilt) { | |
2356 | if (mstime()-sentinel.tilt_start_time < SENTINEL_TILT_PERIOD) return; | |
2357 | sentinel.tilt = 0; | |
2358 | sentinelEvent(REDIS_WARNING,"-tilt",NULL,"#tilt mode exited"); | |
2359 | } | |
2360 | ||
2361 | /* Every kind of instance */ | |
2362 | sentinelCheckSubjectivelyDown(ri); | |
2363 | ||
2364 | /* Masters and slaves */ | |
2365 | if (ri->flags & (SRI_MASTER|SRI_SLAVE)) { | |
2366 | /* Nothing so far. */ | |
2367 | } | |
2368 | ||
2369 | /* Only masters */ | |
2370 | if (ri->flags & SRI_MASTER) { | |
2371 | sentinelCheckObjectivelyDown(ri); | |
2372 | sentinelStartFailover(ri); | |
2373 | sentinelFailoverStateMachine(ri); | |
2374 | sentinelAbortFailoverIfNeeded(ri); | |
2375 | } | |
2376 | } | |
2377 | ||
2378 | /* Perform scheduled operations for all the instances in the dictionary. | |
2379 | * Recursively call the function against dictionaries of slaves. */ | |
2380 | void sentinelHandleDictOfRedisInstances(dict *instances) { | |
2381 | dictIterator *di; | |
2382 | dictEntry *de; | |
2383 | sentinelRedisInstance *switch_to_promoted = NULL; | |
2384 | ||
2385 | /* There are a number of things we need to perform against every master. */ | |
2386 | di = dictGetIterator(instances); | |
2387 | while((de = dictNext(di)) != NULL) { | |
2388 | sentinelRedisInstance *ri = dictGetVal(de); | |
2389 | ||
2390 | sentinelHandleRedisInstance(ri); | |
2391 | if (ri->flags & SRI_MASTER) { | |
2392 | sentinelHandleDictOfRedisInstances(ri->slaves); | |
2393 | sentinelHandleDictOfRedisInstances(ri->sentinels); | |
2394 | if (ri->failover_state == SENTINEL_FAILOVER_STATE_UPDATE_CONFIG) { | |
2395 | switch_to_promoted = ri; | |
2396 | } | |
2397 | } | |
2398 | } | |
2399 | if (switch_to_promoted) | |
2400 | sentinelFailoverSwitchToPromotedSlave(switch_to_promoted); | |
2401 | dictReleaseIterator(di); | |
2402 | } | |
2403 | ||
2404 | /* This function checks if we need to enter the TITL mode. | |
2405 | * | |
2406 | * The TILT mode is entered if we detect that between two invocations of the | |
2407 | * timer interrupt, a negative amount of time, or too much time has passed. | |
2408 | * Note that we expect that more or less just 100 milliseconds will pass | |
2409 | * if everything is fine. However we'll see a negative number or a | |
2410 | * difference bigger than SENTINEL_TILT_TRIGGER milliseconds if one of the | |
2411 | * following conditions happen: | |
2412 | * | |
2413 | * 1) The Sentiel process for some time is blocked, for every kind of | |
2414 | * random reason: the load is huge, the computer was freezed for some time | |
2415 | * in I/O or alike, the process was stopped by a signal. Everything. | |
2416 | * 2) The system clock was altered significantly. | |
2417 | * | |
2418 | * Under both this conditions we'll see everything as timed out and failing | |
2419 | * without good reasons. Instead we enter the TILT mode and wait | |
2420 | * for SENTIENL_TILT_PERIOD to elapse before starting to act again. | |
2421 | * | |
2422 | * During TILT time we still collect information, we just do not act. */ | |
2423 | void sentinelCheckTiltCondition(void) { | |
2424 | mstime_t now = mstime(); | |
2425 | mstime_t delta = now - sentinel.previous_time; | |
2426 | ||
2427 | if (delta < 0 || delta > SENTINEL_TILT_TRIGGER) { | |
2428 | sentinel.tilt = 1; | |
2429 | sentinel.tilt_start_time = mstime(); | |
2430 | sentinelEvent(REDIS_WARNING,"+tilt",NULL,"#tilt mode entered"); | |
2431 | } | |
2432 | sentinel.previous_time = mstime(); | |
2433 | } | |
2434 | ||
2435 | void sentinelTimer(void) { | |
2436 | sentinelCheckTiltCondition(); | |
2437 | sentinelHandleDictOfRedisInstances(sentinel.masters); | |
2438 | } | |
2439 |