]>
Commit | Line | Data |
---|---|---|
e2641e09 | 1 | #include "redis.h" |
2 | ||
3 | #include <sys/time.h> | |
4 | #include <unistd.h> | |
5 | #include <fcntl.h> | |
d310fbed | 6 | #include <sys/socket.h> |
e2641e09 | 7 | #include <sys/stat.h> |
8 | ||
f4aa600b | 9 | /* ---------------------------------- MASTER -------------------------------- */ |
10 | ||
e2641e09 | 11 | void replicationFeedSlaves(list *slaves, int dictid, robj **argv, int argc) { |
12 | listNode *ln; | |
13 | listIter li; | |
632e4c09 | 14 | int j; |
e2641e09 | 15 | |
e2641e09 | 16 | listRewind(slaves,&li); |
17 | while((ln = listNext(&li))) { | |
18 | redisClient *slave = ln->value; | |
19 | ||
20 | /* Don't feed slaves that are still waiting for BGSAVE to start */ | |
21 | if (slave->replstate == REDIS_REPL_WAIT_BGSAVE_START) continue; | |
22 | ||
632e4c09 PN |
23 | /* Feed slaves that are waiting for the initial SYNC (so these commands |
24 | * are queued in the output buffer until the intial SYNC completes), | |
25 | * or are already in sync with the master. */ | |
e2641e09 | 26 | if (slave->slaveseldb != dictid) { |
27 | robj *selectcmd; | |
28 | ||
f892797e | 29 | if (dictid >= 0 && dictid < REDIS_SHARED_SELECT_CMDS) { |
f892797e | 30 | selectcmd = shared.select[dictid]; |
c2672a06 | 31 | incrRefCount(selectcmd); |
f892797e | 32 | } else { |
e2641e09 | 33 | selectcmd = createObject(REDIS_STRING, |
34 | sdscatprintf(sdsempty(),"select %d\r\n",dictid)); | |
e2641e09 | 35 | } |
36 | addReply(slave,selectcmd); | |
f892797e | 37 | decrRefCount(selectcmd); |
e2641e09 | 38 | slave->slaveseldb = dictid; |
39 | } | |
632e4c09 PN |
40 | addReplyMultiBulkLen(slave,argc); |
41 | for (j = 0; j < argc; j++) addReplyBulk(slave,argv[j]); | |
e2641e09 | 42 | } |
e2641e09 | 43 | } |
44 | ||
e31b615e | 45 | void replicationFeedMonitors(redisClient *c, list *monitors, int dictid, robj **argv, int argc) { |
e2641e09 | 46 | listNode *ln; |
47 | listIter li; | |
e31b615e | 48 | int j, port; |
e2641e09 | 49 | sds cmdrepr = sdsnew("+"); |
50 | robj *cmdobj; | |
e31b615e | 51 | char ip[32]; |
e2641e09 | 52 | struct timeval tv; |
53 | ||
54 | gettimeofday(&tv,NULL); | |
2b2eca1f | 55 | cmdrepr = sdscatprintf(cmdrepr,"%ld.%06ld ",(long)tv.tv_sec,(long)tv.tv_usec); |
e31b615e | 56 | if (c->flags & REDIS_LUA_CLIENT) { |
2ea41242 | 57 | cmdrepr = sdscatprintf(cmdrepr,"[%d lua] ",dictid); |
58 | } else if (c->flags & REDIS_UNIX_SOCKET) { | |
59 | cmdrepr = sdscatprintf(cmdrepr,"[%d unix:%s] ",dictid,server.unixsocket); | |
e31b615e | 60 | } else { |
61 | anetPeerToString(c->fd,ip,&port); | |
2ea41242 | 62 | cmdrepr = sdscatprintf(cmdrepr,"[%d %s:%d] ",dictid,ip,port); |
e31b615e | 63 | } |
e2641e09 | 64 | |
65 | for (j = 0; j < argc; j++) { | |
66 | if (argv[j]->encoding == REDIS_ENCODING_INT) { | |
d3b958c3 | 67 | cmdrepr = sdscatprintf(cmdrepr, "\"%ld\"", (long)argv[j]->ptr); |
e2641e09 | 68 | } else { |
69 | cmdrepr = sdscatrepr(cmdrepr,(char*)argv[j]->ptr, | |
70 | sdslen(argv[j]->ptr)); | |
71 | } | |
72 | if (j != argc-1) | |
73 | cmdrepr = sdscatlen(cmdrepr," ",1); | |
74 | } | |
75 | cmdrepr = sdscatlen(cmdrepr,"\r\n",2); | |
76 | cmdobj = createObject(REDIS_STRING,cmdrepr); | |
77 | ||
78 | listRewind(monitors,&li); | |
79 | while((ln = listNext(&li))) { | |
80 | redisClient *monitor = ln->value; | |
81 | addReply(monitor,cmdobj); | |
82 | } | |
83 | decrRefCount(cmdobj); | |
84 | } | |
85 | ||
e2641e09 | 86 | void syncCommand(redisClient *c) { |
87 | /* ignore SYNC if aleady slave or in monitor mode */ | |
88 | if (c->flags & REDIS_SLAVE) return; | |
89 | ||
778b2210 | 90 | /* Refuse SYNC requests if we are a slave but the link with our master |
91 | * is not ok... */ | |
1844f990 | 92 | if (server.masterhost && server.repl_state != REDIS_REPL_CONNECTED) { |
3ab20376 | 93 | addReplyError(c,"Can't SYNC while not connected with my master"); |
778b2210 | 94 | return; |
95 | } | |
96 | ||
e2641e09 | 97 | /* SYNC can't be issued when the server has pending data to send to |
98 | * the client about already issued commands. We need a fresh reply | |
99 | * buffer registering the differences between the BGSAVE and the current | |
100 | * dataset, so that we can copy to other slaves if needed. */ | |
101 | if (listLength(c->reply) != 0) { | |
3ab20376 | 102 | addReplyError(c,"SYNC is invalid with pending input"); |
e2641e09 | 103 | return; |
104 | } | |
105 | ||
106 | redisLog(REDIS_NOTICE,"Slave ask for synchronization"); | |
107 | /* Here we need to check if there is a background saving operation | |
108 | * in progress, or if it is required to start one */ | |
f48cd4b9 | 109 | if (server.rdb_child_pid != -1) { |
e2641e09 | 110 | /* Ok a background save is in progress. Let's check if it is a good |
111 | * one for replication, i.e. if there is another slave that is | |
112 | * registering differences since the server forked to save */ | |
113 | redisClient *slave; | |
114 | listNode *ln; | |
115 | listIter li; | |
116 | ||
117 | listRewind(server.slaves,&li); | |
118 | while((ln = listNext(&li))) { | |
119 | slave = ln->value; | |
120 | if (slave->replstate == REDIS_REPL_WAIT_BGSAVE_END) break; | |
121 | } | |
122 | if (ln) { | |
123 | /* Perfect, the server is already registering differences for | |
124 | * another slave. Set the right state, and copy the buffer. */ | |
1824e3a3 | 125 | copyClientOutputBuffer(c,slave); |
e2641e09 | 126 | c->replstate = REDIS_REPL_WAIT_BGSAVE_END; |
127 | redisLog(REDIS_NOTICE,"Waiting for end of BGSAVE for SYNC"); | |
128 | } else { | |
129 | /* No way, we need to wait for the next BGSAVE in order to | |
130 | * register differences */ | |
131 | c->replstate = REDIS_REPL_WAIT_BGSAVE_START; | |
132 | redisLog(REDIS_NOTICE,"Waiting for next BGSAVE for SYNC"); | |
133 | } | |
134 | } else { | |
135 | /* Ok we don't have a BGSAVE in progress, let's start one */ | |
136 | redisLog(REDIS_NOTICE,"Starting BGSAVE for SYNC"); | |
f48cd4b9 | 137 | if (rdbSaveBackground(server.rdb_filename) != REDIS_OK) { |
e2641e09 | 138 | redisLog(REDIS_NOTICE,"Replication failed, can't BGSAVE"); |
3ab20376 | 139 | addReplyError(c,"Unable to perform background save"); |
e2641e09 | 140 | return; |
141 | } | |
142 | c->replstate = REDIS_REPL_WAIT_BGSAVE_END; | |
143 | } | |
144 | c->repldbfd = -1; | |
145 | c->flags |= REDIS_SLAVE; | |
146 | c->slaveseldb = 0; | |
147 | listAddNodeTail(server.slaves,c); | |
148 | return; | |
149 | } | |
150 | ||
3a328978 | 151 | /* REPLCONF <option> <value> <option> <value> ... |
152 | * This command is used by a slave in order to configure the replication | |
153 | * process before starting it with the SYNC command. | |
154 | * | |
155 | * Currently the only use of this command is to communicate to the master | |
156 | * what is the listening port of the Slave redis instance, so that the | |
157 | * master can accurately list slaves and their listening ports in | |
158 | * the INFO output. | |
159 | * | |
160 | * In the future the same command can be used in order to configure | |
161 | * the replication to initiate an incremental replication instead of a | |
162 | * full resync. */ | |
163 | void replconfCommand(redisClient *c) { | |
164 | int j; | |
165 | ||
166 | if ((c->argc % 2) == 0) { | |
167 | /* Number of arguments must be odd to make sure that every | |
168 | * option has a corresponding value. */ | |
169 | addReply(c,shared.syntaxerr); | |
170 | return; | |
171 | } | |
172 | ||
173 | /* Process every option-value pair. */ | |
174 | for (j = 1; j < c->argc; j+=2) { | |
175 | if (!strcasecmp(c->argv[j]->ptr,"listening-port")) { | |
176 | long port; | |
177 | ||
178 | if ((getLongFromObjectOrReply(c,c->argv[j+1], | |
179 | &port,NULL) != REDIS_OK)) | |
180 | return; | |
181 | c->slave_listening_port = port; | |
182 | } else { | |
183 | addReplyErrorFormat(c,"Unrecognized REPLCONF option: %s", | |
184 | (char*)c->argv[j]->ptr); | |
185 | return; | |
186 | } | |
187 | } | |
188 | addReply(c,shared.ok); | |
189 | } | |
190 | ||
e2641e09 | 191 | void sendBulkToSlave(aeEventLoop *el, int fd, void *privdata, int mask) { |
192 | redisClient *slave = privdata; | |
193 | REDIS_NOTUSED(el); | |
194 | REDIS_NOTUSED(mask); | |
195 | char buf[REDIS_IOBUF_LEN]; | |
196 | ssize_t nwritten, buflen; | |
197 | ||
198 | if (slave->repldboff == 0) { | |
199 | /* Write the bulk write count before to transfer the DB. In theory here | |
200 | * we don't know how much room there is in the output buffer of the | |
201 | * socket, but in pratice SO_SNDLOWAT (the minimum count for output | |
202 | * operations) will never be smaller than the few bytes we need. */ | |
203 | sds bulkcount; | |
204 | ||
205 | bulkcount = sdscatprintf(sdsempty(),"$%lld\r\n",(unsigned long long) | |
206 | slave->repldbsize); | |
207 | if (write(fd,bulkcount,sdslen(bulkcount)) != (signed)sdslen(bulkcount)) | |
208 | { | |
209 | sdsfree(bulkcount); | |
210 | freeClient(slave); | |
211 | return; | |
212 | } | |
213 | sdsfree(bulkcount); | |
214 | } | |
215 | lseek(slave->repldbfd,slave->repldboff,SEEK_SET); | |
216 | buflen = read(slave->repldbfd,buf,REDIS_IOBUF_LEN); | |
217 | if (buflen <= 0) { | |
218 | redisLog(REDIS_WARNING,"Read error sending DB to slave: %s", | |
219 | (buflen == 0) ? "premature EOF" : strerror(errno)); | |
220 | freeClient(slave); | |
221 | return; | |
222 | } | |
223 | if ((nwritten = write(fd,buf,buflen)) == -1) { | |
224 | redisLog(REDIS_VERBOSE,"Write error sending DB to slave: %s", | |
225 | strerror(errno)); | |
226 | freeClient(slave); | |
227 | return; | |
228 | } | |
229 | slave->repldboff += nwritten; | |
230 | if (slave->repldboff == slave->repldbsize) { | |
231 | close(slave->repldbfd); | |
232 | slave->repldbfd = -1; | |
233 | aeDeleteFileEvent(server.el,slave->fd,AE_WRITABLE); | |
234 | slave->replstate = REDIS_REPL_ONLINE; | |
235 | if (aeCreateFileEvent(server.el, slave->fd, AE_WRITABLE, | |
236 | sendReplyToClient, slave) == AE_ERR) { | |
237 | freeClient(slave); | |
238 | return; | |
239 | } | |
e2641e09 | 240 | redisLog(REDIS_NOTICE,"Synchronization with slave succeeded"); |
241 | } | |
242 | } | |
243 | ||
244 | /* This function is called at the end of every backgrond saving. | |
245 | * The argument bgsaveerr is REDIS_OK if the background saving succeeded | |
246 | * otherwise REDIS_ERR is passed to the function. | |
247 | * | |
248 | * The goal of this function is to handle slaves waiting for a successful | |
249 | * background saving in order to perform non-blocking synchronization. */ | |
250 | void updateSlavesWaitingBgsave(int bgsaveerr) { | |
251 | listNode *ln; | |
252 | int startbgsave = 0; | |
253 | listIter li; | |
254 | ||
255 | listRewind(server.slaves,&li); | |
256 | while((ln = listNext(&li))) { | |
257 | redisClient *slave = ln->value; | |
258 | ||
259 | if (slave->replstate == REDIS_REPL_WAIT_BGSAVE_START) { | |
260 | startbgsave = 1; | |
261 | slave->replstate = REDIS_REPL_WAIT_BGSAVE_END; | |
262 | } else if (slave->replstate == REDIS_REPL_WAIT_BGSAVE_END) { | |
263 | struct redis_stat buf; | |
264 | ||
265 | if (bgsaveerr != REDIS_OK) { | |
266 | freeClient(slave); | |
267 | redisLog(REDIS_WARNING,"SYNC failed. BGSAVE child returned an error"); | |
268 | continue; | |
269 | } | |
f48cd4b9 | 270 | if ((slave->repldbfd = open(server.rdb_filename,O_RDONLY)) == -1 || |
e2641e09 | 271 | redis_fstat(slave->repldbfd,&buf) == -1) { |
272 | freeClient(slave); | |
273 | redisLog(REDIS_WARNING,"SYNC failed. Can't open/stat DB after BGSAVE: %s", strerror(errno)); | |
274 | continue; | |
275 | } | |
276 | slave->repldboff = 0; | |
277 | slave->repldbsize = buf.st_size; | |
278 | slave->replstate = REDIS_REPL_SEND_BULK; | |
279 | aeDeleteFileEvent(server.el,slave->fd,AE_WRITABLE); | |
280 | if (aeCreateFileEvent(server.el, slave->fd, AE_WRITABLE, sendBulkToSlave, slave) == AE_ERR) { | |
281 | freeClient(slave); | |
282 | continue; | |
283 | } | |
284 | } | |
285 | } | |
286 | if (startbgsave) { | |
f48cd4b9 | 287 | if (rdbSaveBackground(server.rdb_filename) != REDIS_OK) { |
e2641e09 | 288 | listIter li; |
289 | ||
290 | listRewind(server.slaves,&li); | |
291 | redisLog(REDIS_WARNING,"SYNC failed. BGSAVE failed"); | |
292 | while((ln = listNext(&li))) { | |
293 | redisClient *slave = ln->value; | |
294 | ||
295 | if (slave->replstate == REDIS_REPL_WAIT_BGSAVE_START) | |
296 | freeClient(slave); | |
297 | } | |
298 | } | |
299 | } | |
300 | } | |
301 | ||
f4aa600b | 302 | /* ----------------------------------- SLAVE -------------------------------- */ |
303 | ||
304 | /* Abort the async download of the bulk dataset while SYNC-ing with master */ | |
305 | void replicationAbortSyncTransfer(void) { | |
1844f990 | 306 | redisAssert(server.repl_state == REDIS_REPL_TRANSFER); |
f4aa600b | 307 | |
308 | aeDeleteFileEvent(server.el,server.repl_transfer_s,AE_READABLE); | |
309 | close(server.repl_transfer_s); | |
310 | close(server.repl_transfer_fd); | |
311 | unlink(server.repl_transfer_tmpfile); | |
312 | zfree(server.repl_transfer_tmpfile); | |
1844f990 | 313 | server.repl_state = REDIS_REPL_CONNECT; |
f4aa600b | 314 | } |
315 | ||
316 | /* Asynchronously read the SYNC payload we receive from a master */ | |
784b9308 | 317 | #define REPL_MAX_WRITTEN_BEFORE_FSYNC (1024*1024*8) /* 8 MB */ |
f4aa600b | 318 | void readSyncBulkPayload(aeEventLoop *el, int fd, void *privdata, int mask) { |
26b33669 | 319 | char buf[4096]; |
62ec599c | 320 | ssize_t nread, readlen; |
784b9308 | 321 | off_t left; |
62ec599c | 322 | REDIS_NOTUSED(el); |
323 | REDIS_NOTUSED(privdata); | |
324 | REDIS_NOTUSED(mask); | |
f4aa600b | 325 | |
784b9308 | 326 | /* If repl_transfer_size == -1 we still have to read the bulk length |
26b33669 | 327 | * from the master reply. */ |
784b9308 | 328 | if (server.repl_transfer_size == -1) { |
9157549f | 329 | if (syncReadLine(fd,buf,1024,server.repl_syncio_timeout*1000) == -1) { |
26b33669 | 330 | redisLog(REDIS_WARNING, |
331 | "I/O error reading bulk count from MASTER: %s", | |
332 | strerror(errno)); | |
b075621f | 333 | goto error; |
26b33669 | 334 | } |
b075621f | 335 | |
26b33669 | 336 | if (buf[0] == '-') { |
337 | redisLog(REDIS_WARNING, | |
338 | "MASTER aborted replication with an error: %s", | |
339 | buf+1); | |
b075621f | 340 | goto error; |
89a1433e | 341 | } else if (buf[0] == '\0') { |
342 | /* At this stage just a newline works as a PING in order to take | |
343 | * the connection live. So we refresh our last interaction | |
344 | * timestamp. */ | |
d1949054 | 345 | server.repl_transfer_lastio = server.unixtime; |
89a1433e | 346 | return; |
26b33669 | 347 | } else if (buf[0] != '$') { |
348 | redisLog(REDIS_WARNING,"Bad protocol from MASTER, the first byte is not '$', are you sure the host and port are right?"); | |
b075621f | 349 | goto error; |
26b33669 | 350 | } |
784b9308 | 351 | server.repl_transfer_size = strtol(buf+1,NULL,10); |
26b33669 | 352 | redisLog(REDIS_NOTICE, |
353 | "MASTER <-> SLAVE sync: receiving %ld bytes from master", | |
784b9308 | 354 | server.repl_transfer_size); |
26b33669 | 355 | return; |
356 | } | |
357 | ||
358 | /* Read bulk data */ | |
784b9308 | 359 | left = server.repl_transfer_size - server.repl_transfer_read; |
360 | readlen = (left < (signed)sizeof(buf)) ? left : (signed)sizeof(buf); | |
f4aa600b | 361 | nread = read(fd,buf,readlen); |
362 | if (nread <= 0) { | |
363 | redisLog(REDIS_WARNING,"I/O error trying to sync with MASTER: %s", | |
364 | (nread == -1) ? strerror(errno) : "connection lost"); | |
365 | replicationAbortSyncTransfer(); | |
366 | return; | |
367 | } | |
d1949054 | 368 | server.repl_transfer_lastio = server.unixtime; |
f4aa600b | 369 | if (write(server.repl_transfer_fd,buf,nread) != nread) { |
31788f50 | 370 | redisLog(REDIS_WARNING,"Write error or short write writing to the DB dump file needed for MASTER <-> SLAVE synchronization: %s", strerror(errno)); |
b075621f | 371 | goto error; |
f4aa600b | 372 | } |
784b9308 | 373 | server.repl_transfer_read += nread; |
374 | ||
375 | /* Sync data on disk from time to time, otherwise at the end of the transfer | |
376 | * we may suffer a big delay as the memory buffers are copied into the | |
377 | * actual disk. */ | |
378 | if (server.repl_transfer_read >= | |
379 | server.repl_transfer_last_fsync_off + REPL_MAX_WRITTEN_BEFORE_FSYNC) | |
380 | { | |
381 | off_t sync_size = server.repl_transfer_read - | |
382 | server.repl_transfer_last_fsync_off; | |
383 | rdb_fsync_range(server.repl_transfer_fd, | |
384 | server.repl_transfer_last_fsync_off, sync_size); | |
385 | server.repl_transfer_last_fsync_off += sync_size; | |
386 | } | |
387 | ||
f4aa600b | 388 | /* Check if the transfer is now complete */ |
784b9308 | 389 | if (server.repl_transfer_read == server.repl_transfer_size) { |
f48cd4b9 | 390 | if (rename(server.repl_transfer_tmpfile,server.rdb_filename) == -1) { |
f4aa600b | 391 | redisLog(REDIS_WARNING,"Failed trying to rename the temp DB into dump.rdb in MASTER <-> SLAVE synchronization: %s", strerror(errno)); |
392 | replicationAbortSyncTransfer(); | |
393 | return; | |
394 | } | |
f6433915 | 395 | redisLog(REDIS_NOTICE, "MASTER <-> SLAVE sync: Loading DB in memory"); |
f4aa600b | 396 | emptyDb(); |
9fd01051 | 397 | /* Before loading the DB into memory we need to delete the readable |
398 | * handler, otherwise it will get called recursively since | |
399 | * rdbLoad() will call the event loop to process events from time to | |
400 | * time for non blocking loading. */ | |
401 | aeDeleteFileEvent(server.el,server.repl_transfer_s,AE_READABLE); | |
f48cd4b9 | 402 | if (rdbLoad(server.rdb_filename) != REDIS_OK) { |
f4aa600b | 403 | redisLog(REDIS_WARNING,"Failed trying to load the MASTER synchronization DB from disk"); |
404 | replicationAbortSyncTransfer(); | |
405 | return; | |
406 | } | |
407 | /* Final setup of the connected slave <- master link */ | |
f4aa600b | 408 | zfree(server.repl_transfer_tmpfile); |
409 | close(server.repl_transfer_fd); | |
410 | server.master = createClient(server.repl_transfer_s); | |
411 | server.master->flags |= REDIS_MASTER; | |
412 | server.master->authenticated = 1; | |
1844f990 | 413 | server.repl_state = REDIS_REPL_CONNECTED; |
26b33669 | 414 | redisLog(REDIS_NOTICE, "MASTER <-> SLAVE sync: Finished with success"); |
e7a2e7c1 | 415 | /* Restart the AOF subsystem now that we finished the sync. This |
416 | * will trigger an AOF rewrite, and when done will start appending | |
417 | * to the new file. */ | |
e394114d | 418 | if (server.aof_state != REDIS_AOF_OFF) { |
e7a2e7c1 | 419 | int retry = 10; |
420 | ||
421 | stopAppendOnly(); | |
422 | while (retry-- && startAppendOnly() == REDIS_ERR) { | |
423 | redisLog(REDIS_WARNING,"Failed enabling the AOF after successful master synchrnization! Trying it again in one second."); | |
424 | sleep(1); | |
425 | } | |
426 | if (!retry) { | |
427 | redisLog(REDIS_WARNING,"FATAL: this slave instance finished the synchronization with its master, but the AOF can't be turned on. Exiting now."); | |
428 | exit(1); | |
429 | } | |
430 | } | |
f4aa600b | 431 | } |
b075621f PN |
432 | |
433 | return; | |
434 | ||
435 | error: | |
436 | replicationAbortSyncTransfer(); | |
437 | return; | |
f4aa600b | 438 | } |
439 | ||
3a328978 | 440 | /* Send a synchronous command to the master. Used to send AUTH and |
36def8fd | 441 | * REPLCONF commands before starting the replication with SYNC. |
3a328978 | 442 | * |
443 | * On success NULL is returned. | |
444 | * On error an sds string describing the error is returned. | |
445 | */ | |
446 | char *sendSynchronousCommand(int fd, ...) { | |
447 | va_list ap; | |
448 | sds cmd = sdsempty(); | |
449 | char *arg, buf[256]; | |
450 | ||
451 | /* Create the command to send to the master, we use simple inline | |
452 | * protocol for simplicity as currently we only send simple strings. */ | |
453 | va_start(ap,fd); | |
454 | while(1) { | |
455 | arg = va_arg(ap, char*); | |
456 | if (arg == NULL) break; | |
457 | ||
458 | if (sdslen(cmd) != 0) cmd = sdscatlen(cmd," ",1); | |
459 | cmd = sdscat(cmd,arg); | |
460 | } | |
461 | cmd = sdscatlen(cmd,"\r\n",2); | |
462 | ||
463 | /* Transfer command to the server. */ | |
464 | if (syncWrite(fd,cmd,sdslen(cmd),server.repl_syncio_timeout*1000) == -1) { | |
465 | sdsfree(cmd); | |
466 | return sdscatprintf(sdsempty(),"Writing to master: %s", | |
467 | strerror(errno)); | |
468 | } | |
469 | sdsfree(cmd); | |
470 | ||
471 | /* Read the reply from the server. */ | |
472 | if (syncReadLine(fd,buf,sizeof(buf),server.repl_syncio_timeout*1000) == -1) | |
473 | { | |
474 | return sdscatprintf(sdsempty(),"Reading from master: %s", | |
475 | strerror(errno)); | |
476 | } | |
477 | ||
478 | /* Check for errors from the server. */ | |
479 | if (buf[0] != '+') { | |
480 | return sdscatprintf(sdsempty(),"Error from master: %s", buf); | |
481 | } | |
482 | ||
483 | return NULL; /* No errors. */ | |
484 | } | |
485 | ||
a3309139 | 486 | void syncWithMaster(aeEventLoop *el, int fd, void *privdata, int mask) { |
3a328978 | 487 | char tmpfile[256], *err; |
e2641e09 | 488 | int dfd, maxtries = 5; |
bb66fc31 | 489 | int sockerr = 0; |
490 | socklen_t errlen = sizeof(sockerr); | |
a3309139 PN |
491 | REDIS_NOTUSED(el); |
492 | REDIS_NOTUSED(privdata); | |
493 | REDIS_NOTUSED(mask); | |
e2641e09 | 494 | |
76e772f3 | 495 | /* If this event fired after the user turned the instance into a master |
496 | * with SLAVEOF NO ONE we must just return ASAP. */ | |
1844f990 | 497 | if (server.repl_state == REDIS_REPL_NONE) { |
76e772f3 | 498 | close(fd); |
499 | return; | |
500 | } | |
501 | ||
bb66fc31 | 502 | /* Check for errors in the socket. */ |
503 | if (getsockopt(fd, SOL_SOCKET, SO_ERROR, &sockerr, &errlen) == -1) | |
504 | sockerr = errno; | |
505 | if (sockerr) { | |
506 | aeDeleteFileEvent(server.el,fd,AE_READABLE|AE_WRITABLE); | |
507 | redisLog(REDIS_WARNING,"Error condition on socket for SYNC: %s", | |
508 | strerror(sockerr)); | |
509 | goto error; | |
510 | } | |
511 | ||
512 | /* If we were connecting, it's time to send a non blocking PING, we want to | |
513 | * make sure the master is able to reply before going into the actual | |
514 | * replication process where we have long timeouts in the order of | |
515 | * seconds (in the meantime the slave would block). */ | |
516 | if (server.repl_state == REDIS_REPL_CONNECTING) { | |
517 | redisLog(REDIS_NOTICE,"Non blocking connect for SYNC fired the event."); | |
518 | /* Delete the writable event so that the readable event remains | |
519 | * registered and we can wait for the PONG reply. */ | |
520 | aeDeleteFileEvent(server.el,fd,AE_WRITABLE); | |
521 | server.repl_state = REDIS_REPL_RECEIVE_PONG; | |
522 | /* Send the PING, don't check for errors at all, we have the timeout | |
523 | * that will take care about this. */ | |
524 | syncWrite(fd,"PING\r\n",6,100); | |
525 | return; | |
526 | } | |
527 | ||
528 | /* Receive the PONG command. */ | |
529 | if (server.repl_state == REDIS_REPL_RECEIVE_PONG) { | |
530 | char buf[1024]; | |
531 | ||
532 | /* Delete the readable event, we no longer need it now that there is | |
533 | * the PING reply to read. */ | |
534 | aeDeleteFileEvent(server.el,fd,AE_READABLE); | |
535 | ||
536 | /* Read the reply with explicit timeout. */ | |
537 | buf[0] = '\0'; | |
538 | if (syncReadLine(fd,buf,sizeof(buf), | |
539 | server.repl_syncio_timeout*1000) == -1) | |
540 | { | |
541 | redisLog(REDIS_WARNING, | |
542 | "I/O error reading PING reply from master: %s", | |
543 | strerror(errno)); | |
544 | goto error; | |
545 | } | |
546 | ||
547 | /* We don't care about the reply, it can be +PONG or an error since | |
548 | * the server requires AUTH. As long as it replies correctly, it's | |
549 | * fine from our point of view. */ | |
550 | if (buf[0] != '-' && buf[0] != '+') { | |
551 | redisLog(REDIS_WARNING,"Unexpected reply to PING from master."); | |
552 | goto error; | |
553 | } else { | |
554 | redisLog(REDIS_NOTICE, | |
555 | "Master replied to PING, replication can continue..."); | |
556 | } | |
557 | } | |
e2641e09 | 558 | |
559 | /* AUTH with the master if required. */ | |
560 | if(server.masterauth) { | |
3a328978 | 561 | err = sendSynchronousCommand(fd,"AUTH",server.masterauth,NULL); |
562 | if (err) { | |
563 | redisLog(REDIS_WARNING,"Unable to AUTH to MASTER: %s",err); | |
564 | sdsfree(err); | |
a3309139 PN |
565 | goto error; |
566 | } | |
3a328978 | 567 | } |
568 | ||
569 | /* Set the slave port, so that Master's INFO command can list the | |
570 | * slave listening port correctly. */ | |
571 | { | |
572 | sds port = sdsfromlonglong(server.port); | |
573 | err = sendSynchronousCommand(fd,"REPLCONF","listening-port",port, | |
574 | NULL); | |
575 | sdsfree(port); | |
576 | /* Ignore the error if any, not all the Redis versions support | |
577 | * REPLCONF listening-port. */ | |
578 | if (err) { | |
579 | redisLog(REDIS_NOTICE,"(non critical): Master does not understand REPLCONF listening-port: %s", err); | |
580 | sdsfree(err); | |
e2641e09 | 581 | } |
582 | } | |
583 | ||
584 | /* Issue the SYNC command */ | |
299290d3 | 585 | if (syncWrite(fd,"SYNC\r\n",6,server.repl_syncio_timeout*1000) == -1) { |
e2641e09 | 586 | redisLog(REDIS_WARNING,"I/O error writing to MASTER: %s", |
587 | strerror(errno)); | |
a3309139 | 588 | goto error; |
e2641e09 | 589 | } |
26b33669 | 590 | |
591 | /* Prepare a suitable temp file for bulk transfer */ | |
e2641e09 | 592 | while(maxtries--) { |
593 | snprintf(tmpfile,256, | |
d1949054 | 594 | "temp-%d.%ld.rdb",(int)server.unixtime,(long int)getpid()); |
e2641e09 | 595 | dfd = open(tmpfile,O_CREAT|O_WRONLY|O_EXCL,0644); |
596 | if (dfd != -1) break; | |
597 | sleep(1); | |
598 | } | |
599 | if (dfd == -1) { | |
e2641e09 | 600 | redisLog(REDIS_WARNING,"Opening the temp file needed for MASTER <-> SLAVE synchronization: %s",strerror(errno)); |
a3309139 | 601 | goto error; |
e2641e09 | 602 | } |
e2641e09 | 603 | |
f4aa600b | 604 | /* Setup the non blocking download of the bulk file. */ |
a3309139 | 605 | if (aeCreateFileEvent(server.el,fd, AE_READABLE,readSyncBulkPayload,NULL) |
62ec599c | 606 | == AE_ERR) |
f4aa600b | 607 | { |
f4aa600b | 608 | redisLog(REDIS_WARNING,"Can't create readable event for SYNC"); |
a3309139 | 609 | goto error; |
e2641e09 | 610 | } |
a3309139 | 611 | |
1844f990 | 612 | server.repl_state = REDIS_REPL_TRANSFER; |
784b9308 | 613 | server.repl_transfer_size = -1; |
614 | server.repl_transfer_read = 0; | |
615 | server.repl_transfer_last_fsync_off = 0; | |
f4aa600b | 616 | server.repl_transfer_fd = dfd; |
d1949054 | 617 | server.repl_transfer_lastio = server.unixtime; |
f4aa600b | 618 | server.repl_transfer_tmpfile = zstrdup(tmpfile); |
a3309139 PN |
619 | return; |
620 | ||
621 | error: | |
a3309139 | 622 | close(fd); |
bb66fc31 | 623 | server.repl_transfer_s = -1; |
624 | server.repl_state = REDIS_REPL_CONNECT; | |
a3309139 PN |
625 | return; |
626 | } | |
627 | ||
628 | int connectWithMaster(void) { | |
629 | int fd; | |
630 | ||
631 | fd = anetTcpNonBlockConnect(NULL,server.masterhost,server.masterport); | |
632 | if (fd == -1) { | |
633 | redisLog(REDIS_WARNING,"Unable to connect to MASTER: %s", | |
634 | strerror(errno)); | |
635 | return REDIS_ERR; | |
636 | } | |
637 | ||
45029d37 | 638 | if (aeCreateFileEvent(server.el,fd,AE_READABLE|AE_WRITABLE,syncWithMaster,NULL) == |
b075621f | 639 | AE_ERR) |
a3309139 PN |
640 | { |
641 | close(fd); | |
642 | redisLog(REDIS_WARNING,"Can't create readable event for SYNC"); | |
643 | return REDIS_ERR; | |
644 | } | |
645 | ||
d1949054 | 646 | server.repl_transfer_lastio = server.unixtime; |
a3309139 | 647 | server.repl_transfer_s = fd; |
1844f990 | 648 | server.repl_state = REDIS_REPL_CONNECTING; |
e2641e09 | 649 | return REDIS_OK; |
650 | } | |
651 | ||
27acd7aa | 652 | /* This function can be called when a non blocking connection is currently |
653 | * in progress to undo it. */ | |
654 | void undoConnectWithMaster(void) { | |
655 | int fd = server.repl_transfer_s; | |
656 | ||
bb66fc31 | 657 | redisAssert(server.repl_state == REDIS_REPL_CONNECTING || |
658 | server.repl_state == REDIS_REPL_RECEIVE_PONG); | |
27acd7aa | 659 | aeDeleteFileEvent(server.el,fd,AE_READABLE|AE_WRITABLE); |
660 | close(fd); | |
661 | server.repl_transfer_s = -1; | |
1844f990 | 662 | server.repl_state = REDIS_REPL_CONNECT; |
27acd7aa | 663 | } |
664 | ||
e2641e09 | 665 | void slaveofCommand(redisClient *c) { |
666 | if (!strcasecmp(c->argv[1]->ptr,"no") && | |
667 | !strcasecmp(c->argv[2]->ptr,"one")) { | |
668 | if (server.masterhost) { | |
669 | sdsfree(server.masterhost); | |
670 | server.masterhost = NULL; | |
671 | if (server.master) freeClient(server.master); | |
1844f990 | 672 | if (server.repl_state == REDIS_REPL_TRANSFER) |
f4aa600b | 673 | replicationAbortSyncTransfer(); |
bb66fc31 | 674 | else if (server.repl_state == REDIS_REPL_CONNECTING || |
675 | server.repl_state == REDIS_REPL_RECEIVE_PONG) | |
27acd7aa | 676 | undoConnectWithMaster(); |
1844f990 | 677 | server.repl_state = REDIS_REPL_NONE; |
e2641e09 | 678 | redisLog(REDIS_NOTICE,"MASTER MODE enabled (user request)"); |
679 | } | |
680 | } else { | |
ebdfad69 | 681 | long port; |
682 | ||
683 | if ((getLongFromObjectOrReply(c, c->argv[2], &port, NULL) != REDIS_OK)) | |
684 | return; | |
685 | ||
686 | /* Check if we are already attached to the specified slave */ | |
687 | if (server.masterhost && !strcasecmp(server.masterhost,c->argv[1]->ptr) | |
688 | && server.masterport == port) { | |
689 | redisLog(REDIS_NOTICE,"SLAVE OF would result into synchronization with the master we are already connected with. No operation performed."); | |
690 | addReplySds(c,sdsnew("+OK Already connected to specified master\r\n")); | |
691 | return; | |
692 | } | |
693 | /* There was no previous master or the user specified a different one, | |
694 | * we can continue. */ | |
e2641e09 | 695 | sdsfree(server.masterhost); |
696 | server.masterhost = sdsdup(c->argv[1]->ptr); | |
ebdfad69 | 697 | server.masterport = port; |
e2641e09 | 698 | if (server.master) freeClient(server.master); |
179e54d2 | 699 | disconnectSlaves(); /* Force our slaves to resync with us as well. */ |
1844f990 | 700 | if (server.repl_state == REDIS_REPL_TRANSFER) |
f4aa600b | 701 | replicationAbortSyncTransfer(); |
1844f990 | 702 | server.repl_state = REDIS_REPL_CONNECT; |
e2641e09 | 703 | redisLog(REDIS_NOTICE,"SLAVE OF %s:%d enabled (user request)", |
704 | server.masterhost, server.masterport); | |
705 | } | |
706 | addReply(c,shared.ok); | |
707 | } | |
f4aa600b | 708 | |
709 | /* --------------------------- REPLICATION CRON ---------------------------- */ | |
710 | ||
f4aa600b | 711 | void replicationCron(void) { |
27acd7aa | 712 | /* Non blocking connection timeout? */ |
bb66fc31 | 713 | if (server.masterhost && |
714 | (server.repl_state == REDIS_REPL_CONNECTING || | |
715 | server.repl_state == REDIS_REPL_RECEIVE_PONG) && | |
27acd7aa | 716 | (time(NULL)-server.repl_transfer_lastio) > server.repl_timeout) |
717 | { | |
718 | redisLog(REDIS_WARNING,"Timeout connecting to the MASTER..."); | |
719 | undoConnectWithMaster(); | |
720 | } | |
721 | ||
f4aa600b | 722 | /* Bulk transfer I/O timeout? */ |
1844f990 | 723 | if (server.masterhost && server.repl_state == REDIS_REPL_TRANSFER && |
8996bf77 | 724 | (time(NULL)-server.repl_transfer_lastio) > server.repl_timeout) |
f4aa600b | 725 | { |
f0b9f803 | 726 | redisLog(REDIS_WARNING,"Timeout receiving bulk data from MASTER... If the problem persists try to set the 'repl-timeout' parameter in redis.conf to a larger value."); |
f4aa600b | 727 | replicationAbortSyncTransfer(); |
728 | } | |
729 | ||
89a1433e | 730 | /* Timed out master when we are an already connected slave? */ |
1844f990 | 731 | if (server.masterhost && server.repl_state == REDIS_REPL_CONNECTED && |
8996bf77 | 732 | (time(NULL)-server.master->lastinteraction) > server.repl_timeout) |
89a1433e | 733 | { |
734 | redisLog(REDIS_WARNING,"MASTER time out: no data nor PING received..."); | |
735 | freeClient(server.master); | |
736 | } | |
737 | ||
f4aa600b | 738 | /* Check if we should connect to a MASTER */ |
1844f990 | 739 | if (server.repl_state == REDIS_REPL_CONNECT) { |
f4aa600b | 740 | redisLog(REDIS_NOTICE,"Connecting to MASTER..."); |
a3309139 PN |
741 | if (connectWithMaster() == REDIS_OK) { |
742 | redisLog(REDIS_NOTICE,"MASTER <-> SLAVE sync started"); | |
f4aa600b | 743 | } |
744 | } | |
89a1433e | 745 | |
746 | /* If we have attached slaves, PING them from time to time. | |
747 | * So slaves can implement an explicit timeout to masters, and will | |
748 | * be able to detect a link disconnection even if the TCP connection | |
749 | * will not actually go down. */ | |
9edfe635 | 750 | if (!(server.cronloops % (server.repl_ping_slave_period * REDIS_HZ))) { |
89a1433e | 751 | listIter li; |
752 | listNode *ln; | |
753 | ||
754 | listRewind(server.slaves,&li); | |
755 | while((ln = listNext(&li))) { | |
756 | redisClient *slave = ln->value; | |
757 | ||
758 | /* Don't ping slaves that are in the middle of a bulk transfer | |
759 | * with the master for first synchronization. */ | |
760 | if (slave->replstate == REDIS_REPL_SEND_BULK) continue; | |
761 | if (slave->replstate == REDIS_REPL_ONLINE) { | |
762 | /* If the slave is online send a normal ping */ | |
a950a843 | 763 | addReplySds(slave,sdsnew("*1\r\n$4\r\nPING\r\n")); |
89a1433e | 764 | } else { |
765 | /* Otherwise we are in the pre-synchronization stage. | |
766 | * Just a newline will do the work of refreshing the | |
767 | * connection last interaction time, and at the same time | |
768 | * we'll be sure that being a single char there are no | |
769 | * short-write problems. */ | |
f96a9f82 | 770 | if (write(slave->fd, "\n", 1) == -1) { |
771 | /* Don't worry, it's just a ping. */ | |
772 | } | |
89a1433e | 773 | } |
774 | } | |
775 | } | |
f4aa600b | 776 | } |