]> git.saurik.com Git - redis.git/blob - src/aof.c
Remove backgroud jobs from the queue only when the processing was completed, and...
[redis.git] / src / aof.c
1 #include "redis.h"
2 #include "bio.h"
3
4 #include <signal.h>
5 #include <fcntl.h>
6 #include <sys/stat.h>
7 #include <sys/types.h>
8 #include <sys/time.h>
9 #include <sys/resource.h>
10 #include <sys/wait.h>
11
12 void aofUpdateCurrentSize(void);
13
14 /* Called when the user switches from "appendonly yes" to "appendonly no"
15 * at runtime using the CONFIG command. */
16 void stopAppendOnly(void) {
17 flushAppendOnlyFile();
18 aof_fsync(server.appendfd);
19 close(server.appendfd);
20
21 server.appendfd = -1;
22 server.appendseldb = -1;
23 server.appendonly = 0;
24 /* rewrite operation in progress? kill it, wait child exit */
25 if (server.bgrewritechildpid != -1) {
26 int statloc;
27
28 if (kill(server.bgrewritechildpid,SIGKILL) != -1)
29 wait3(&statloc,0,NULL);
30 /* reset the buffer accumulating changes while the child saves */
31 sdsfree(server.bgrewritebuf);
32 server.bgrewritebuf = sdsempty();
33 server.bgrewritechildpid = -1;
34 }
35 }
36
37 /* Called when the user switches from "appendonly no" to "appendonly yes"
38 * at runtime using the CONFIG command. */
39 int startAppendOnly(void) {
40 server.appendonly = 1;
41 server.lastfsync = time(NULL);
42 server.appendfd = open(server.appendfilename,O_WRONLY|O_APPEND|O_CREAT,0644);
43 if (server.appendfd == -1) {
44 redisLog(REDIS_WARNING,"Used tried to switch on AOF via CONFIG, but I can't open the AOF file: %s",strerror(errno));
45 return REDIS_ERR;
46 }
47 if (rewriteAppendOnlyFileBackground() == REDIS_ERR) {
48 server.appendonly = 0;
49 close(server.appendfd);
50 redisLog(REDIS_WARNING,"Used tried to switch on AOF via CONFIG, I can't trigger a background AOF rewrite operation. Check the above logs for more info about the error.",strerror(errno));
51 return REDIS_ERR;
52 }
53 return REDIS_OK;
54 }
55
56 /* Write the append only file buffer on disk.
57 *
58 * Since we are required to write the AOF before replying to the client,
59 * and the only way the client socket can get a write is entering when the
60 * the event loop, we accumulate all the AOF writes in a memory
61 * buffer and write it on disk using this function just before entering
62 * the event loop again. */
63 void flushAppendOnlyFile(void) {
64 ssize_t nwritten;
65
66 if (sdslen(server.aofbuf) == 0) return;
67
68 /* We want to perform a single write. This should be guaranteed atomic
69 * at least if the filesystem we are writing is a real physical one.
70 * While this will save us against the server being killed I don't think
71 * there is much to do about the whole server stopping for power problems
72 * or alike */
73 nwritten = write(server.appendfd,server.aofbuf,sdslen(server.aofbuf));
74 if (nwritten != (signed)sdslen(server.aofbuf)) {
75 /* Ooops, we are in troubles. The best thing to do for now is
76 * aborting instead of giving the illusion that everything is
77 * working as expected. */
78 if (nwritten == -1) {
79 redisLog(REDIS_WARNING,"Exiting on error writing to the append-only file: %s",strerror(errno));
80 } else {
81 redisLog(REDIS_WARNING,"Exiting on short write while writing to the append-only file: %s",strerror(errno));
82 }
83 exit(1);
84 }
85 server.appendonly_current_size += nwritten;
86
87 /* Re-use AOF buffer when it is small enough. The maximum comes from the
88 * arena size of 4k minus some overhead (but is otherwise arbitrary). */
89 if ((sdslen(server.aofbuf)+sdsavail(server.aofbuf)) < 4000) {
90 sdsclear(server.aofbuf);
91 } else {
92 sdsfree(server.aofbuf);
93 server.aofbuf = sdsempty();
94 }
95
96 /* Don't fsync if no-appendfsync-on-rewrite is set to yes and there are
97 * children doing I/O in the background. */
98 if (server.no_appendfsync_on_rewrite &&
99 (server.bgrewritechildpid != -1 || server.bgsavechildpid != -1))
100 return;
101
102 /* Perform the fsync if needed. */
103 if (server.appendfsync == APPENDFSYNC_ALWAYS ||
104 (server.appendfsync == APPENDFSYNC_EVERYSEC &&
105 server.unixtime > server.lastfsync))
106 {
107 /* aof_fsync is defined as fdatasync() for Linux in order to avoid
108 * flushing metadata. */
109 aof_fsync(server.appendfd); /* Let's try to get this data on the disk */
110 server.lastfsync = server.unixtime;
111 }
112 }
113
114 sds catAppendOnlyGenericCommand(sds dst, int argc, robj **argv) {
115 char buf[32];
116 int len, j;
117 robj *o;
118
119 buf[0] = '*';
120 len = 1+ll2string(buf+1,sizeof(buf)-1,argc);
121 buf[len++] = '\r';
122 buf[len++] = '\n';
123 dst = sdscatlen(dst,buf,len);
124
125 for (j = 0; j < argc; j++) {
126 o = getDecodedObject(argv[j]);
127 buf[0] = '$';
128 len = 1+ll2string(buf+1,sizeof(buf)-1,sdslen(o->ptr));
129 buf[len++] = '\r';
130 buf[len++] = '\n';
131 dst = sdscatlen(dst,buf,len);
132 dst = sdscatlen(dst,o->ptr,sdslen(o->ptr));
133 dst = sdscatlen(dst,"\r\n",2);
134 decrRefCount(o);
135 }
136 return dst;
137 }
138
139 sds catAppendOnlyExpireAtCommand(sds buf, robj *key, robj *seconds) {
140 int argc = 3;
141 long when;
142 robj *argv[3];
143
144 /* Make sure we can use strtol */
145 seconds = getDecodedObject(seconds);
146 when = time(NULL)+strtol(seconds->ptr,NULL,10);
147 decrRefCount(seconds);
148
149 argv[0] = createStringObject("EXPIREAT",8);
150 argv[1] = key;
151 argv[2] = createObject(REDIS_STRING,
152 sdscatprintf(sdsempty(),"%ld",when));
153 buf = catAppendOnlyGenericCommand(buf, argc, argv);
154 decrRefCount(argv[0]);
155 decrRefCount(argv[2]);
156 return buf;
157 }
158
159 void feedAppendOnlyFile(struct redisCommand *cmd, int dictid, robj **argv, int argc) {
160 sds buf = sdsempty();
161 robj *tmpargv[3];
162
163 /* The DB this command was targetting is not the same as the last command
164 * we appendend. To issue a SELECT command is needed. */
165 if (dictid != server.appendseldb) {
166 char seldb[64];
167
168 snprintf(seldb,sizeof(seldb),"%d",dictid);
169 buf = sdscatprintf(buf,"*2\r\n$6\r\nSELECT\r\n$%lu\r\n%s\r\n",
170 (unsigned long)strlen(seldb),seldb);
171 server.appendseldb = dictid;
172 }
173
174 if (cmd->proc == expireCommand) {
175 /* Translate EXPIRE into EXPIREAT */
176 buf = catAppendOnlyExpireAtCommand(buf,argv[1],argv[2]);
177 } else if (cmd->proc == setexCommand) {
178 /* Translate SETEX to SET and EXPIREAT */
179 tmpargv[0] = createStringObject("SET",3);
180 tmpargv[1] = argv[1];
181 tmpargv[2] = argv[3];
182 buf = catAppendOnlyGenericCommand(buf,3,tmpargv);
183 decrRefCount(tmpargv[0]);
184 buf = catAppendOnlyExpireAtCommand(buf,argv[1],argv[2]);
185 } else {
186 buf = catAppendOnlyGenericCommand(buf,argc,argv);
187 }
188
189 /* Append to the AOF buffer. This will be flushed on disk just before
190 * of re-entering the event loop, so before the client will get a
191 * positive reply about the operation performed. */
192 server.aofbuf = sdscatlen(server.aofbuf,buf,sdslen(buf));
193
194 /* If a background append only file rewriting is in progress we want to
195 * accumulate the differences between the child DB and the current one
196 * in a buffer, so that when the child process will do its work we
197 * can append the differences to the new append only file. */
198 if (server.bgrewritechildpid != -1)
199 server.bgrewritebuf = sdscatlen(server.bgrewritebuf,buf,sdslen(buf));
200
201 sdsfree(buf);
202 }
203
204 /* In Redis commands are always executed in the context of a client, so in
205 * order to load the append only file we need to create a fake client. */
206 struct redisClient *createFakeClient(void) {
207 struct redisClient *c = zmalloc(sizeof(*c));
208
209 selectDb(c,0);
210 c->fd = -1;
211 c->querybuf = sdsempty();
212 c->argc = 0;
213 c->argv = NULL;
214 c->bufpos = 0;
215 c->flags = 0;
216 /* We set the fake client as a slave waiting for the synchronization
217 * so that Redis will not try to send replies to this client. */
218 c->replstate = REDIS_REPL_WAIT_BGSAVE_START;
219 c->reply = listCreate();
220 c->watched_keys = listCreate();
221 listSetFreeMethod(c->reply,decrRefCount);
222 listSetDupMethod(c->reply,dupClientReplyValue);
223 initClientMultiState(c);
224 return c;
225 }
226
227 void freeFakeClient(struct redisClient *c) {
228 sdsfree(c->querybuf);
229 listRelease(c->reply);
230 listRelease(c->watched_keys);
231 freeClientMultiState(c);
232 zfree(c);
233 }
234
235 /* Replay the append log file. On error REDIS_OK is returned. On non fatal
236 * error (the append only file is zero-length) REDIS_ERR is returned. On
237 * fatal error an error message is logged and the program exists. */
238 int loadAppendOnlyFile(char *filename) {
239 struct redisClient *fakeClient;
240 FILE *fp = fopen(filename,"r");
241 struct redis_stat sb;
242 int appendonly = server.appendonly;
243 long loops = 0;
244
245 if (fp && redis_fstat(fileno(fp),&sb) != -1 && sb.st_size == 0) {
246 server.appendonly_current_size = 0;
247 fclose(fp);
248 return REDIS_ERR;
249 }
250
251 if (fp == NULL) {
252 redisLog(REDIS_WARNING,"Fatal error: can't open the append log file for reading: %s",strerror(errno));
253 exit(1);
254 }
255
256 /* Temporarily disable AOF, to prevent EXEC from feeding a MULTI
257 * to the same file we're about to read. */
258 server.appendonly = 0;
259
260 fakeClient = createFakeClient();
261 startLoading(fp);
262
263 while(1) {
264 int argc, j;
265 unsigned long len;
266 robj **argv;
267 char buf[128];
268 sds argsds;
269 struct redisCommand *cmd;
270
271 /* Serve the clients from time to time */
272 if (!(loops++ % 1000)) {
273 loadingProgress(ftello(fp));
274 aeProcessEvents(server.el, AE_FILE_EVENTS|AE_DONT_WAIT);
275 }
276
277 if (fgets(buf,sizeof(buf),fp) == NULL) {
278 if (feof(fp))
279 break;
280 else
281 goto readerr;
282 }
283 if (buf[0] != '*') goto fmterr;
284 argc = atoi(buf+1);
285 argv = zmalloc(sizeof(robj*)*argc);
286 for (j = 0; j < argc; j++) {
287 if (fgets(buf,sizeof(buf),fp) == NULL) goto readerr;
288 if (buf[0] != '$') goto fmterr;
289 len = strtol(buf+1,NULL,10);
290 argsds = sdsnewlen(NULL,len);
291 if (len && fread(argsds,len,1,fp) == 0) goto fmterr;
292 argv[j] = createObject(REDIS_STRING,argsds);
293 if (fread(buf,2,1,fp) == 0) goto fmterr; /* discard CRLF */
294 }
295
296 /* Command lookup */
297 cmd = lookupCommand(argv[0]->ptr);
298 if (!cmd) {
299 redisLog(REDIS_WARNING,"Unknown command '%s' reading the append only file", argv[0]->ptr);
300 exit(1);
301 }
302 /* Run the command in the context of a fake client */
303 fakeClient->argc = argc;
304 fakeClient->argv = argv;
305 cmd->proc(fakeClient);
306
307 /* The fake client should not have a reply */
308 redisAssert(fakeClient->bufpos == 0 && listLength(fakeClient->reply) == 0);
309 /* The fake client should never get blocked */
310 redisAssert((fakeClient->flags & REDIS_BLOCKED) == 0);
311
312 /* Clean up. Command code may have changed argv/argc so we use the
313 * argv/argc of the client instead of the local variables. */
314 for (j = 0; j < fakeClient->argc; j++)
315 decrRefCount(fakeClient->argv[j]);
316 zfree(fakeClient->argv);
317 }
318
319 /* This point can only be reached when EOF is reached without errors.
320 * If the client is in the middle of a MULTI/EXEC, log error and quit. */
321 if (fakeClient->flags & REDIS_MULTI) goto readerr;
322
323 fclose(fp);
324 freeFakeClient(fakeClient);
325 server.appendonly = appendonly;
326 stopLoading();
327 aofUpdateCurrentSize();
328 server.auto_aofrewrite_base_size = server.appendonly_current_size;
329 return REDIS_OK;
330
331 readerr:
332 if (feof(fp)) {
333 redisLog(REDIS_WARNING,"Unexpected end of file reading the append only file");
334 } else {
335 redisLog(REDIS_WARNING,"Unrecoverable error reading the append only file: %s", strerror(errno));
336 }
337 exit(1);
338 fmterr:
339 redisLog(REDIS_WARNING,"Bad file format reading the append only file: make a backup of your AOF file, then use ./redis-check-aof --fix <filename>");
340 exit(1);
341 }
342
343 /* Write a sequence of commands able to fully rebuild the dataset into
344 * "filename". Used both by REWRITEAOF and BGREWRITEAOF. */
345 int rewriteAppendOnlyFile(char *filename) {
346 dictIterator *di = NULL;
347 dictEntry *de;
348 FILE *fp;
349 char tmpfile[256];
350 int j;
351 time_t now = time(NULL);
352
353 /* Note that we have to use a different temp name here compared to the
354 * one used by rewriteAppendOnlyFileBackground() function. */
355 snprintf(tmpfile,256,"temp-rewriteaof-%d.aof", (int) getpid());
356 fp = fopen(tmpfile,"w");
357 if (!fp) {
358 redisLog(REDIS_WARNING, "Failed rewriting the append only file: %s", strerror(errno));
359 return REDIS_ERR;
360 }
361 for (j = 0; j < server.dbnum; j++) {
362 char selectcmd[] = "*2\r\n$6\r\nSELECT\r\n";
363 redisDb *db = server.db+j;
364 dict *d = db->dict;
365 if (dictSize(d) == 0) continue;
366 di = dictGetSafeIterator(d);
367 if (!di) {
368 fclose(fp);
369 return REDIS_ERR;
370 }
371
372 /* SELECT the new DB */
373 if (fwrite(selectcmd,sizeof(selectcmd)-1,1,fp) == 0) goto werr;
374 if (fwriteBulkLongLong(fp,j) == 0) goto werr;
375
376 /* Iterate this DB writing every entry */
377 while((de = dictNext(di)) != NULL) {
378 sds keystr;
379 robj key, *o;
380 time_t expiretime;
381
382 keystr = dictGetEntryKey(de);
383 o = dictGetEntryVal(de);
384 initStaticStringObject(key,keystr);
385
386 expiretime = getExpire(db,&key);
387
388 /* Save the key and associated value */
389 if (o->type == REDIS_STRING) {
390 /* Emit a SET command */
391 char cmd[]="*3\r\n$3\r\nSET\r\n";
392 if (fwrite(cmd,sizeof(cmd)-1,1,fp) == 0) goto werr;
393 /* Key and value */
394 if (fwriteBulkObject(fp,&key) == 0) goto werr;
395 if (fwriteBulkObject(fp,o) == 0) goto werr;
396 } else if (o->type == REDIS_LIST) {
397 /* Emit the RPUSHes needed to rebuild the list */
398 char cmd[]="*3\r\n$5\r\nRPUSH\r\n";
399 if (o->encoding == REDIS_ENCODING_ZIPLIST) {
400 unsigned char *zl = o->ptr;
401 unsigned char *p = ziplistIndex(zl,0);
402 unsigned char *vstr;
403 unsigned int vlen;
404 long long vlong;
405
406 while(ziplistGet(p,&vstr,&vlen,&vlong)) {
407 if (fwrite(cmd,sizeof(cmd)-1,1,fp) == 0) goto werr;
408 if (fwriteBulkObject(fp,&key) == 0) goto werr;
409 if (vstr) {
410 if (fwriteBulkString(fp,(char*)vstr,vlen) == 0)
411 goto werr;
412 } else {
413 if (fwriteBulkLongLong(fp,vlong) == 0)
414 goto werr;
415 }
416 p = ziplistNext(zl,p);
417 }
418 } else if (o->encoding == REDIS_ENCODING_LINKEDLIST) {
419 list *list = o->ptr;
420 listNode *ln;
421 listIter li;
422
423 listRewind(list,&li);
424 while((ln = listNext(&li))) {
425 robj *eleobj = listNodeValue(ln);
426
427 if (fwrite(cmd,sizeof(cmd)-1,1,fp) == 0) goto werr;
428 if (fwriteBulkObject(fp,&key) == 0) goto werr;
429 if (fwriteBulkObject(fp,eleobj) == 0) goto werr;
430 }
431 } else {
432 redisPanic("Unknown list encoding");
433 }
434 } else if (o->type == REDIS_SET) {
435 char cmd[]="*3\r\n$4\r\nSADD\r\n";
436
437 /* Emit the SADDs needed to rebuild the set */
438 if (o->encoding == REDIS_ENCODING_INTSET) {
439 int ii = 0;
440 int64_t llval;
441 while(intsetGet(o->ptr,ii++,&llval)) {
442 if (fwrite(cmd,sizeof(cmd)-1,1,fp) == 0) goto werr;
443 if (fwriteBulkObject(fp,&key) == 0) goto werr;
444 if (fwriteBulkLongLong(fp,llval) == 0) goto werr;
445 }
446 } else if (o->encoding == REDIS_ENCODING_HT) {
447 dictIterator *di = dictGetIterator(o->ptr);
448 dictEntry *de;
449 while((de = dictNext(di)) != NULL) {
450 robj *eleobj = dictGetEntryKey(de);
451 if (fwrite(cmd,sizeof(cmd)-1,1,fp) == 0) goto werr;
452 if (fwriteBulkObject(fp,&key) == 0) goto werr;
453 if (fwriteBulkObject(fp,eleobj) == 0) goto werr;
454 }
455 dictReleaseIterator(di);
456 } else {
457 redisPanic("Unknown set encoding");
458 }
459 } else if (o->type == REDIS_ZSET) {
460 /* Emit the ZADDs needed to rebuild the sorted set */
461 char cmd[]="*4\r\n$4\r\nZADD\r\n";
462
463 if (o->encoding == REDIS_ENCODING_ZIPLIST) {
464 unsigned char *zl = o->ptr;
465 unsigned char *eptr, *sptr;
466 unsigned char *vstr;
467 unsigned int vlen;
468 long long vll;
469 double score;
470
471 eptr = ziplistIndex(zl,0);
472 redisAssert(eptr != NULL);
473 sptr = ziplistNext(zl,eptr);
474 redisAssert(sptr != NULL);
475
476 while (eptr != NULL) {
477 redisAssert(ziplistGet(eptr,&vstr,&vlen,&vll));
478 score = zzlGetScore(sptr);
479
480 if (fwrite(cmd,sizeof(cmd)-1,1,fp) == 0) goto werr;
481 if (fwriteBulkObject(fp,&key) == 0) goto werr;
482 if (fwriteBulkDouble(fp,score) == 0) goto werr;
483 if (vstr != NULL) {
484 if (fwriteBulkString(fp,(char*)vstr,vlen) == 0)
485 goto werr;
486 } else {
487 if (fwriteBulkLongLong(fp,vll) == 0)
488 goto werr;
489 }
490 zzlNext(zl,&eptr,&sptr);
491 }
492 } else if (o->encoding == REDIS_ENCODING_SKIPLIST) {
493 zset *zs = o->ptr;
494 dictIterator *di = dictGetIterator(zs->dict);
495 dictEntry *de;
496
497 while((de = dictNext(di)) != NULL) {
498 robj *eleobj = dictGetEntryKey(de);
499 double *score = dictGetEntryVal(de);
500
501 if (fwrite(cmd,sizeof(cmd)-1,1,fp) == 0) goto werr;
502 if (fwriteBulkObject(fp,&key) == 0) goto werr;
503 if (fwriteBulkDouble(fp,*score) == 0) goto werr;
504 if (fwriteBulkObject(fp,eleobj) == 0) goto werr;
505 }
506 dictReleaseIterator(di);
507 } else {
508 redisPanic("Unknown sorted set encoding");
509 }
510 } else if (o->type == REDIS_HASH) {
511 char cmd[]="*4\r\n$4\r\nHSET\r\n";
512
513 /* Emit the HSETs needed to rebuild the hash */
514 if (o->encoding == REDIS_ENCODING_ZIPMAP) {
515 unsigned char *p = zipmapRewind(o->ptr);
516 unsigned char *field, *val;
517 unsigned int flen, vlen;
518
519 while((p = zipmapNext(p,&field,&flen,&val,&vlen)) != NULL) {
520 if (fwrite(cmd,sizeof(cmd)-1,1,fp) == 0) goto werr;
521 if (fwriteBulkObject(fp,&key) == 0) goto werr;
522 if (fwriteBulkString(fp,(char*)field,flen) == 0)
523 goto werr;
524 if (fwriteBulkString(fp,(char*)val,vlen) == 0)
525 goto werr;
526 }
527 } else {
528 dictIterator *di = dictGetIterator(o->ptr);
529 dictEntry *de;
530
531 while((de = dictNext(di)) != NULL) {
532 robj *field = dictGetEntryKey(de);
533 robj *val = dictGetEntryVal(de);
534
535 if (fwrite(cmd,sizeof(cmd)-1,1,fp) == 0) goto werr;
536 if (fwriteBulkObject(fp,&key) == 0) goto werr;
537 if (fwriteBulkObject(fp,field) == 0) goto werr;
538 if (fwriteBulkObject(fp,val) == 0) goto werr;
539 }
540 dictReleaseIterator(di);
541 }
542 } else {
543 redisPanic("Unknown object type");
544 }
545 /* Save the expire time */
546 if (expiretime != -1) {
547 char cmd[]="*3\r\n$8\r\nEXPIREAT\r\n";
548 /* If this key is already expired skip it */
549 if (expiretime < now) continue;
550 if (fwrite(cmd,sizeof(cmd)-1,1,fp) == 0) goto werr;
551 if (fwriteBulkObject(fp,&key) == 0) goto werr;
552 if (fwriteBulkLongLong(fp,expiretime) == 0) goto werr;
553 }
554 }
555 dictReleaseIterator(di);
556 }
557
558 /* Make sure data will not remain on the OS's output buffers */
559 fflush(fp);
560 aof_fsync(fileno(fp));
561 fclose(fp);
562
563 /* Use RENAME to make sure the DB file is changed atomically only
564 * if the generate DB file is ok. */
565 if (rename(tmpfile,filename) == -1) {
566 redisLog(REDIS_WARNING,"Error moving temp append only file on the final destination: %s", strerror(errno));
567 unlink(tmpfile);
568 return REDIS_ERR;
569 }
570 redisLog(REDIS_NOTICE,"SYNC append only file rewrite performed");
571 return REDIS_OK;
572
573 werr:
574 fclose(fp);
575 unlink(tmpfile);
576 redisLog(REDIS_WARNING,"Write error writing append only file on disk: %s", strerror(errno));
577 if (di) dictReleaseIterator(di);
578 return REDIS_ERR;
579 }
580
581 /* This is how rewriting of the append only file in background works:
582 *
583 * 1) The user calls BGREWRITEAOF
584 * 2) Redis calls this function, that forks():
585 * 2a) the child rewrite the append only file in a temp file.
586 * 2b) the parent accumulates differences in server.bgrewritebuf.
587 * 3) When the child finished '2a' exists.
588 * 4) The parent will trap the exit code, if it's OK, will append the
589 * data accumulated into server.bgrewritebuf into the temp file, and
590 * finally will rename(2) the temp file in the actual file name.
591 * The the new file is reopened as the new append only file. Profit!
592 */
593 int rewriteAppendOnlyFileBackground(void) {
594 pid_t childpid;
595 long long start;
596
597 if (server.bgrewritechildpid != -1) return REDIS_ERR;
598 start = ustime();
599 if ((childpid = fork()) == 0) {
600 char tmpfile[256];
601
602 /* Child */
603 if (server.ipfd > 0) close(server.ipfd);
604 if (server.sofd > 0) close(server.sofd);
605 snprintf(tmpfile,256,"temp-rewriteaof-bg-%d.aof", (int) getpid());
606 if (rewriteAppendOnlyFile(tmpfile) == REDIS_OK) {
607 _exit(0);
608 } else {
609 _exit(1);
610 }
611 } else {
612 /* Parent */
613 server.stat_fork_time = ustime()-start;
614 if (childpid == -1) {
615 redisLog(REDIS_WARNING,
616 "Can't rewrite append only file in background: fork: %s",
617 strerror(errno));
618 return REDIS_ERR;
619 }
620 redisLog(REDIS_NOTICE,
621 "Background append only file rewriting started by pid %d",childpid);
622 server.bgrewritechildpid = childpid;
623 updateDictResizePolicy();
624 /* We set appendseldb to -1 in order to force the next call to the
625 * feedAppendOnlyFile() to issue a SELECT command, so the differences
626 * accumulated by the parent into server.bgrewritebuf will start
627 * with a SELECT statement and it will be safe to merge. */
628 server.appendseldb = -1;
629 return REDIS_OK;
630 }
631 return REDIS_OK; /* unreached */
632 }
633
634 void bgrewriteaofCommand(redisClient *c) {
635 if (server.bgrewritechildpid != -1) {
636 addReplyError(c,"Background append only file rewriting already in progress");
637 } else if (server.bgsavechildpid != -1) {
638 server.aofrewrite_scheduled = 1;
639 addReplyStatus(c,"Background append only file rewriting scheduled");
640 } else if (rewriteAppendOnlyFileBackground() == REDIS_OK) {
641 addReplyStatus(c,"Background append only file rewriting started");
642 } else {
643 addReply(c,shared.err);
644 }
645 }
646
647 void aofRemoveTempFile(pid_t childpid) {
648 char tmpfile[256];
649
650 snprintf(tmpfile,256,"temp-rewriteaof-bg-%d.aof", (int) childpid);
651 unlink(tmpfile);
652 }
653
654 /* Update the server.appendonly_current_size filed explicitly using stat(2)
655 * to check the size of the file. This is useful after a rewrite or after
656 * a restart, normally the size is updated just adding the write length
657 * to the current lenght, that is much faster. */
658 void aofUpdateCurrentSize(void) {
659 struct redis_stat sb;
660
661 if (redis_fstat(server.appendfd,&sb) == -1) {
662 redisLog(REDIS_WARNING,"Unable to check the AOF length: %s",
663 strerror(errno));
664 } else {
665 server.appendonly_current_size = sb.st_size;
666 }
667 }
668
669 /* A background append only file rewriting (BGREWRITEAOF) terminated its work.
670 * Handle this. */
671 void backgroundRewriteDoneHandler(int exitcode, int bysignal) {
672 if (!bysignal && exitcode == 0) {
673 int newfd, oldfd;
674 int nwritten;
675 char tmpfile[256];
676 long long now = ustime();
677
678 redisLog(REDIS_NOTICE,
679 "Background AOF rewrite terminated with success");
680
681 /* Flush the differences accumulated by the parent to the
682 * rewritten AOF. */
683 snprintf(tmpfile,256,"temp-rewriteaof-bg-%d.aof",
684 (int)server.bgrewritechildpid);
685 newfd = open(tmpfile,O_WRONLY|O_APPEND);
686 if (newfd == -1) {
687 redisLog(REDIS_WARNING,
688 "Unable to open the temporary AOF produced by the child: %s", strerror(errno));
689 goto cleanup;
690 }
691
692 nwritten = write(newfd,server.bgrewritebuf,sdslen(server.bgrewritebuf));
693 if (nwritten != (signed)sdslen(server.bgrewritebuf)) {
694 if (nwritten == -1) {
695 redisLog(REDIS_WARNING,
696 "Error trying to flush the parent diff to the rewritten AOF: %s", strerror(errno));
697 } else {
698 redisLog(REDIS_WARNING,
699 "Short write trying to flush the parent diff to the rewritten AOF: %s", strerror(errno));
700 }
701 close(newfd);
702 goto cleanup;
703 }
704
705 redisLog(REDIS_NOTICE,
706 "Parent diff successfully flushed to the rewritten AOF (%lu bytes)", nwritten);
707
708 /* The only remaining thing to do is to rename the temporary file to
709 * the configured file and switch the file descriptor used to do AOF
710 * writes. We don't want close(2) or rename(2) calls to block the
711 * server on old file deletion.
712 *
713 * There are two possible scenarios:
714 *
715 * 1) AOF is DISABLED and this was a one time rewrite. The temporary
716 * file will be renamed to the configured file. When this file already
717 * exists, it will be unlinked, which may block the server.
718 *
719 * 2) AOF is ENABLED and the rewritten AOF will immediately start
720 * receiving writes. After the temporary file is renamed to the
721 * configured file, the original AOF file descriptor will be closed.
722 * Since this will be the last reference to that file, closing it
723 * causes the underlying file to be unlinked, which may block the
724 * server.
725 *
726 * To mitigate the blocking effect of the unlink operation (either
727 * caused by rename(2) in scenario 1, or by close(2) in scenario 2), we
728 * use a background thread to take care of this. First, we
729 * make scenario 1 identical to scenario 2 by opening the target file
730 * when it exists. The unlink operation after the rename(2) will then
731 * be executed upon calling close(2) for its descriptor. Everything to
732 * guarantee atomicity for this switch has already happened by then, so
733 * we don't care what the outcome or duration of that close operation
734 * is, as long as the file descriptor is released again. */
735 if (server.appendfd == -1) {
736 /* AOF disabled */
737
738 /* Don't care if this fails: oldfd will be -1 and we handle that.
739 * One notable case of -1 return is if the old file does
740 * not exist. */
741 oldfd = open(server.appendfilename,O_RDONLY|O_NONBLOCK);
742 } else {
743 /* AOF enabled */
744 oldfd = -1; /* We'll set this to the current AOF filedes later. */
745 }
746
747 /* Rename the temporary file. This will not unlink the target file if
748 * it exists, because we reference it with "oldfd". */
749 if (rename(tmpfile,server.appendfilename) == -1) {
750 redisLog(REDIS_WARNING,
751 "Error trying to rename the temporary AOF: %s", strerror(errno));
752 close(newfd);
753 if (oldfd != -1) close(oldfd);
754 goto cleanup;
755 }
756
757 if (server.appendfd == -1) {
758 /* AOF disabled, we don't need to set the AOF file descriptor
759 * to this new file, so we can close it. */
760 close(newfd);
761 } else {
762 /* AOF enabled, replace the old fd with the new one. */
763 oldfd = server.appendfd;
764 server.appendfd = newfd;
765 if (server.appendfsync != APPENDFSYNC_NO) aof_fsync(newfd);
766 server.appendseldb = -1; /* Make sure SELECT is re-issued */
767 aofUpdateCurrentSize();
768 server.auto_aofrewrite_base_size = server.appendonly_current_size;
769 }
770
771 redisLog(REDIS_NOTICE, "Background AOF rewrite successful");
772
773 /* Asynchronously close the overwritten AOF. */
774 if (oldfd != -1) bioCreateBackgroundJob(REDIS_BIO_CLOSE_FILE,(void*)(long)oldfd,NULL,NULL);
775
776 redisLog(REDIS_VERBOSE,
777 "Background AOF rewrite signal handler took %lldus", ustime()-now);
778 } else if (!bysignal && exitcode != 0) {
779 redisLog(REDIS_WARNING,
780 "Background AOF rewrite terminated with error");
781 } else {
782 redisLog(REDIS_WARNING,
783 "Background AOF rewrite terminated by signal %d", bysignal);
784 }
785
786 cleanup:
787 sdsfree(server.bgrewritebuf);
788 server.bgrewritebuf = sdsempty();
789 aofRemoveTempFile(server.bgrewritechildpid);
790 server.bgrewritechildpid = -1;
791 }