]> git.saurik.com Git - redis.git/blame - src/aof.c
automatic AOF rewrite first implementation. Still to be tested.
[redis.git] / src / aof.c
CommitLineData
e2641e09 1#include "redis.h"
2
3#include <signal.h>
4#include <fcntl.h>
5#include <sys/stat.h>
3688d7f3 6#include <sys/types.h>
7#include <sys/time.h>
8#include <sys/resource.h>
9#include <sys/wait.h>
e2641e09 10
b333e239 11void aofUpdateCurrentSize(void);
12
e2641e09 13/* Called when the user switches from "appendonly yes" to "appendonly no"
14 * at runtime using the CONFIG command. */
15void stopAppendOnly(void) {
16 flushAppendOnlyFile();
17 aof_fsync(server.appendfd);
18 close(server.appendfd);
19
20 server.appendfd = -1;
21 server.appendseldb = -1;
22 server.appendonly = 0;
23 /* rewrite operation in progress? kill it, wait child exit */
b333e239 24 if (server.bgrewritechildpid != -1) {
e2641e09 25 int statloc;
26
b333e239 27 if (kill(server.bgrewritechildpid,SIGKILL) != -1)
e2641e09 28 wait3(&statloc,0,NULL);
29 /* reset the buffer accumulating changes while the child saves */
30 sdsfree(server.bgrewritebuf);
31 server.bgrewritebuf = sdsempty();
b333e239 32 server.bgrewritechildpid = -1;
e2641e09 33 }
34}
35
36/* Called when the user switches from "appendonly no" to "appendonly yes"
37 * at runtime using the CONFIG command. */
38int startAppendOnly(void) {
39 server.appendonly = 1;
40 server.lastfsync = time(NULL);
41 server.appendfd = open(server.appendfilename,O_WRONLY|O_APPEND|O_CREAT,0644);
42 if (server.appendfd == -1) {
43 redisLog(REDIS_WARNING,"Used tried to switch on AOF via CONFIG, but I can't open the AOF file: %s",strerror(errno));
44 return REDIS_ERR;
45 }
46 if (rewriteAppendOnlyFileBackground() == REDIS_ERR) {
47 server.appendonly = 0;
48 close(server.appendfd);
49 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));
50 return REDIS_ERR;
51 }
52 return REDIS_OK;
53}
54
55/* Write the append only file buffer on disk.
56 *
57 * Since we are required to write the AOF before replying to the client,
58 * and the only way the client socket can get a write is entering when the
59 * the event loop, we accumulate all the AOF writes in a memory
60 * buffer and write it on disk using this function just before entering
61 * the event loop again. */
62void flushAppendOnlyFile(void) {
63 time_t now;
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 sdsfree(server.aofbuf);
86 server.aofbuf = sdsempty();
b333e239 87 server.appendonly_current_size += nwritten;
e2641e09 88
89 /* Don't Fsync if no-appendfsync-on-rewrite is set to yes and we have
90 * childs performing heavy I/O on disk. */
91 if (server.no_appendfsync_on_rewrite &&
92 (server.bgrewritechildpid != -1 || server.bgsavechildpid != -1))
93 return;
94 /* Fsync if needed */
95 now = time(NULL);
96 if (server.appendfsync == APPENDFSYNC_ALWAYS ||
97 (server.appendfsync == APPENDFSYNC_EVERYSEC &&
98 now-server.lastfsync > 1))
99 {
100 /* aof_fsync is defined as fdatasync() for Linux in order to avoid
101 * flushing metadata. */
102 aof_fsync(server.appendfd); /* Let's try to get this data on the disk */
103 server.lastfsync = now;
104 }
105}
106
107sds catAppendOnlyGenericCommand(sds buf, int argc, robj **argv) {
108 int j;
109 buf = sdscatprintf(buf,"*%d\r\n",argc);
110 for (j = 0; j < argc; j++) {
111 robj *o = getDecodedObject(argv[j]);
112 buf = sdscatprintf(buf,"$%lu\r\n",(unsigned long)sdslen(o->ptr));
113 buf = sdscatlen(buf,o->ptr,sdslen(o->ptr));
114 buf = sdscatlen(buf,"\r\n",2);
115 decrRefCount(o);
116 }
117 return buf;
118}
119
120sds catAppendOnlyExpireAtCommand(sds buf, robj *key, robj *seconds) {
121 int argc = 3;
122 long when;
123 robj *argv[3];
124
125 /* Make sure we can use strtol */
126 seconds = getDecodedObject(seconds);
127 when = time(NULL)+strtol(seconds->ptr,NULL,10);
128 decrRefCount(seconds);
129
130 argv[0] = createStringObject("EXPIREAT",8);
131 argv[1] = key;
132 argv[2] = createObject(REDIS_STRING,
133 sdscatprintf(sdsempty(),"%ld",when));
134 buf = catAppendOnlyGenericCommand(buf, argc, argv);
135 decrRefCount(argv[0]);
136 decrRefCount(argv[2]);
137 return buf;
138}
139
140void feedAppendOnlyFile(struct redisCommand *cmd, int dictid, robj **argv, int argc) {
141 sds buf = sdsempty();
142 robj *tmpargv[3];
143
144 /* The DB this command was targetting is not the same as the last command
145 * we appendend. To issue a SELECT command is needed. */
146 if (dictid != server.appendseldb) {
147 char seldb[64];
148
149 snprintf(seldb,sizeof(seldb),"%d",dictid);
150 buf = sdscatprintf(buf,"*2\r\n$6\r\nSELECT\r\n$%lu\r\n%s\r\n",
151 (unsigned long)strlen(seldb),seldb);
152 server.appendseldb = dictid;
153 }
154
155 if (cmd->proc == expireCommand) {
156 /* Translate EXPIRE into EXPIREAT */
157 buf = catAppendOnlyExpireAtCommand(buf,argv[1],argv[2]);
158 } else if (cmd->proc == setexCommand) {
159 /* Translate SETEX to SET and EXPIREAT */
160 tmpargv[0] = createStringObject("SET",3);
161 tmpargv[1] = argv[1];
162 tmpargv[2] = argv[3];
163 buf = catAppendOnlyGenericCommand(buf,3,tmpargv);
164 decrRefCount(tmpargv[0]);
165 buf = catAppendOnlyExpireAtCommand(buf,argv[1],argv[2]);
166 } else {
167 buf = catAppendOnlyGenericCommand(buf,argc,argv);
168 }
169
170 /* Append to the AOF buffer. This will be flushed on disk just before
171 * of re-entering the event loop, so before the client will get a
172 * positive reply about the operation performed. */
173 server.aofbuf = sdscatlen(server.aofbuf,buf,sdslen(buf));
174
175 /* If a background append only file rewriting is in progress we want to
176 * accumulate the differences between the child DB and the current one
177 * in a buffer, so that when the child process will do its work we
178 * can append the differences to the new append only file. */
179 if (server.bgrewritechildpid != -1)
180 server.bgrewritebuf = sdscatlen(server.bgrewritebuf,buf,sdslen(buf));
181
182 sdsfree(buf);
183}
184
185/* In Redis commands are always executed in the context of a client, so in
186 * order to load the append only file we need to create a fake client. */
187struct redisClient *createFakeClient(void) {
188 struct redisClient *c = zmalloc(sizeof(*c));
189
190 selectDb(c,0);
191 c->fd = -1;
192 c->querybuf = sdsempty();
193 c->argc = 0;
194 c->argv = NULL;
2403fc9f 195 c->bufpos = 0;
e2641e09 196 c->flags = 0;
197 /* We set the fake client as a slave waiting for the synchronization
198 * so that Redis will not try to send replies to this client. */
199 c->replstate = REDIS_REPL_WAIT_BGSAVE_START;
200 c->reply = listCreate();
b67d2345 201 c->watched_keys = listCreate();
e2641e09 202 listSetFreeMethod(c->reply,decrRefCount);
203 listSetDupMethod(c->reply,dupClientReplyValue);
204 initClientMultiState(c);
205 return c;
206}
207
208void freeFakeClient(struct redisClient *c) {
209 sdsfree(c->querybuf);
210 listRelease(c->reply);
b67d2345 211 listRelease(c->watched_keys);
e2641e09 212 freeClientMultiState(c);
213 zfree(c);
214}
215
216/* Replay the append log file. On error REDIS_OK is returned. On non fatal
217 * error (the append only file is zero-length) REDIS_ERR is returned. On
218 * fatal error an error message is logged and the program exists. */
219int loadAppendOnlyFile(char *filename) {
220 struct redisClient *fakeClient;
221 FILE *fp = fopen(filename,"r");
222 struct redis_stat sb;
223 int appendonly = server.appendonly;
97e7f8ae 224 long loops = 0;
e2641e09 225
4aec2ec8 226 if (fp && redis_fstat(fileno(fp),&sb) != -1 && sb.st_size == 0) {
b333e239 227 server.appendonly_current_size = 0;
4aec2ec8 228 fclose(fp);
e2641e09 229 return REDIS_ERR;
4aec2ec8 230 }
e2641e09 231
232 if (fp == NULL) {
233 redisLog(REDIS_WARNING,"Fatal error: can't open the append log file for reading: %s",strerror(errno));
234 exit(1);
235 }
236
237 /* Temporarily disable AOF, to prevent EXEC from feeding a MULTI
238 * to the same file we're about to read. */
239 server.appendonly = 0;
240
241 fakeClient = createFakeClient();
97e7f8ae 242 startLoading(fp);
243
e2641e09 244 while(1) {
245 int argc, j;
246 unsigned long len;
247 robj **argv;
248 char buf[128];
249 sds argsds;
250 struct redisCommand *cmd;
e2641e09 251
97e7f8ae 252 /* Serve the clients from time to time */
253 if (!(loops++ % 1000)) {
254 loadingProgress(ftello(fp));
255 aeProcessEvents(server.el, AE_FILE_EVENTS|AE_DONT_WAIT);
256 }
257
e2641e09 258 if (fgets(buf,sizeof(buf),fp) == NULL) {
259 if (feof(fp))
260 break;
261 else
262 goto readerr;
263 }
264 if (buf[0] != '*') goto fmterr;
265 argc = atoi(buf+1);
266 argv = zmalloc(sizeof(robj*)*argc);
267 for (j = 0; j < argc; j++) {
268 if (fgets(buf,sizeof(buf),fp) == NULL) goto readerr;
269 if (buf[0] != '$') goto fmterr;
270 len = strtol(buf+1,NULL,10);
271 argsds = sdsnewlen(NULL,len);
272 if (len && fread(argsds,len,1,fp) == 0) goto fmterr;
273 argv[j] = createObject(REDIS_STRING,argsds);
274 if (fread(buf,2,1,fp) == 0) goto fmterr; /* discard CRLF */
275 }
276
277 /* Command lookup */
278 cmd = lookupCommand(argv[0]->ptr);
279 if (!cmd) {
280 redisLog(REDIS_WARNING,"Unknown command '%s' reading the append only file", argv[0]->ptr);
281 exit(1);
282 }
e2641e09 283 /* Run the command in the context of a fake client */
284 fakeClient->argc = argc;
285 fakeClient->argv = argv;
286 cmd->proc(fakeClient);
57b07380
PN
287
288 /* The fake client should not have a reply */
289 redisAssert(fakeClient->bufpos == 0 && listLength(fakeClient->reply) == 0);
290
45b0f6fb
PN
291 /* Clean up. Command code may have changed argv/argc so we use the
292 * argv/argc of the client instead of the local variables. */
293 for (j = 0; j < fakeClient->argc; j++)
294 decrRefCount(fakeClient->argv[j]);
295 zfree(fakeClient->argv);
e2641e09 296 }
297
298 /* This point can only be reached when EOF is reached without errors.
299 * If the client is in the middle of a MULTI/EXEC, log error and quit. */
300 if (fakeClient->flags & REDIS_MULTI) goto readerr;
301
302 fclose(fp);
303 freeFakeClient(fakeClient);
304 server.appendonly = appendonly;
97e7f8ae 305 stopLoading();
b333e239 306 aofUpdateCurrentSize();
e2641e09 307 return REDIS_OK;
308
309readerr:
310 if (feof(fp)) {
311 redisLog(REDIS_WARNING,"Unexpected end of file reading the append only file");
312 } else {
313 redisLog(REDIS_WARNING,"Unrecoverable error reading the append only file: %s", strerror(errno));
314 }
315 exit(1);
316fmterr:
412e457c 317 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>");
e2641e09 318 exit(1);
319}
320
e2641e09 321/* Write a sequence of commands able to fully rebuild the dataset into
322 * "filename". Used both by REWRITEAOF and BGREWRITEAOF. */
323int rewriteAppendOnlyFile(char *filename) {
324 dictIterator *di = NULL;
325 dictEntry *de;
326 FILE *fp;
327 char tmpfile[256];
328 int j;
329 time_t now = time(NULL);
330
331 /* Note that we have to use a different temp name here compared to the
332 * one used by rewriteAppendOnlyFileBackground() function. */
333 snprintf(tmpfile,256,"temp-rewriteaof-%d.aof", (int) getpid());
334 fp = fopen(tmpfile,"w");
335 if (!fp) {
336 redisLog(REDIS_WARNING, "Failed rewriting the append only file: %s", strerror(errno));
337 return REDIS_ERR;
338 }
339 for (j = 0; j < server.dbnum; j++) {
340 char selectcmd[] = "*2\r\n$6\r\nSELECT\r\n";
341 redisDb *db = server.db+j;
342 dict *d = db->dict;
343 if (dictSize(d) == 0) continue;
344 di = dictGetIterator(d);
345 if (!di) {
346 fclose(fp);
347 return REDIS_ERR;
348 }
349
350 /* SELECT the new DB */
351 if (fwrite(selectcmd,sizeof(selectcmd)-1,1,fp) == 0) goto werr;
352 if (fwriteBulkLongLong(fp,j) == 0) goto werr;
353
354 /* Iterate this DB writing every entry */
355 while((de = dictNext(di)) != NULL) {
6901fe77 356 sds keystr;
e2641e09 357 robj key, *o;
358 time_t expiretime;
e2641e09 359
360 keystr = dictGetEntryKey(de);
361 o = dictGetEntryVal(de);
362 initStaticStringObject(key,keystr);
16d77878 363
e2641e09 364 expiretime = getExpire(db,&key);
365
366 /* Save the key and associated value */
367 if (o->type == REDIS_STRING) {
368 /* Emit a SET command */
369 char cmd[]="*3\r\n$3\r\nSET\r\n";
370 if (fwrite(cmd,sizeof(cmd)-1,1,fp) == 0) goto werr;
371 /* Key and value */
372 if (fwriteBulkObject(fp,&key) == 0) goto werr;
373 if (fwriteBulkObject(fp,o) == 0) goto werr;
374 } else if (o->type == REDIS_LIST) {
375 /* Emit the RPUSHes needed to rebuild the list */
376 char cmd[]="*3\r\n$5\r\nRPUSH\r\n";
377 if (o->encoding == REDIS_ENCODING_ZIPLIST) {
378 unsigned char *zl = o->ptr;
379 unsigned char *p = ziplistIndex(zl,0);
380 unsigned char *vstr;
381 unsigned int vlen;
382 long long vlong;
383
384 while(ziplistGet(p,&vstr,&vlen,&vlong)) {
385 if (fwrite(cmd,sizeof(cmd)-1,1,fp) == 0) goto werr;
386 if (fwriteBulkObject(fp,&key) == 0) goto werr;
387 if (vstr) {
388 if (fwriteBulkString(fp,(char*)vstr,vlen) == 0)
389 goto werr;
390 } else {
391 if (fwriteBulkLongLong(fp,vlong) == 0)
392 goto werr;
393 }
394 p = ziplistNext(zl,p);
395 }
396 } else if (o->encoding == REDIS_ENCODING_LINKEDLIST) {
397 list *list = o->ptr;
398 listNode *ln;
399 listIter li;
400
401 listRewind(list,&li);
402 while((ln = listNext(&li))) {
403 robj *eleobj = listNodeValue(ln);
404
405 if (fwrite(cmd,sizeof(cmd)-1,1,fp) == 0) goto werr;
406 if (fwriteBulkObject(fp,&key) == 0) goto werr;
407 if (fwriteBulkObject(fp,eleobj) == 0) goto werr;
408 }
409 } else {
410 redisPanic("Unknown list encoding");
411 }
412 } else if (o->type == REDIS_SET) {
2767f1c0 413 char cmd[]="*3\r\n$4\r\nSADD\r\n";
e2641e09 414
2767f1c0
PN
415 /* Emit the SADDs needed to rebuild the set */
416 if (o->encoding == REDIS_ENCODING_INTSET) {
417 int ii = 0;
23c64fe5 418 int64_t llval;
2767f1c0
PN
419 while(intsetGet(o->ptr,ii++,&llval)) {
420 if (fwrite(cmd,sizeof(cmd)-1,1,fp) == 0) goto werr;
421 if (fwriteBulkObject(fp,&key) == 0) goto werr;
422 if (fwriteBulkLongLong(fp,llval) == 0) goto werr;
423 }
424 } else if (o->encoding == REDIS_ENCODING_HT) {
425 dictIterator *di = dictGetIterator(o->ptr);
426 dictEntry *de;
427 while((de = dictNext(di)) != NULL) {
428 robj *eleobj = dictGetEntryKey(de);
429 if (fwrite(cmd,sizeof(cmd)-1,1,fp) == 0) goto werr;
430 if (fwriteBulkObject(fp,&key) == 0) goto werr;
431 if (fwriteBulkObject(fp,eleobj) == 0) goto werr;
432 }
433 dictReleaseIterator(di);
434 } else {
435 redisPanic("Unknown set encoding");
e2641e09 436 }
e2641e09 437 } else if (o->type == REDIS_ZSET) {
438 /* Emit the ZADDs needed to rebuild the sorted set */
dddf5335
PN
439 char cmd[]="*4\r\n$4\r\nZADD\r\n";
440
441 if (o->encoding == REDIS_ENCODING_ZIPLIST) {
442 unsigned char *zl = o->ptr;
443 unsigned char *eptr, *sptr;
444 unsigned char *vstr;
445 unsigned int vlen;
446 long long vll;
447 double score;
448
449 eptr = ziplistIndex(zl,0);
450 redisAssert(eptr != NULL);
451 sptr = ziplistNext(zl,eptr);
452 redisAssert(sptr != NULL);
453
454 while (eptr != NULL) {
455 redisAssert(ziplistGet(eptr,&vstr,&vlen,&vll));
456 score = zzlGetScore(sptr);
457
458 if (fwrite(cmd,sizeof(cmd)-1,1,fp) == 0) goto werr;
459 if (fwriteBulkObject(fp,&key) == 0) goto werr;
460 if (fwriteBulkDouble(fp,score) == 0) goto werr;
461 if (vstr != NULL) {
462 if (fwriteBulkString(fp,(char*)vstr,vlen) == 0)
463 goto werr;
464 } else {
465 if (fwriteBulkLongLong(fp,vll) == 0)
466 goto werr;
467 }
468 zzlNext(zl,&eptr,&sptr);
469 }
100ed062 470 } else if (o->encoding == REDIS_ENCODING_SKIPLIST) {
dddf5335
PN
471 zset *zs = o->ptr;
472 dictIterator *di = dictGetIterator(zs->dict);
473 dictEntry *de;
474
475 while((de = dictNext(di)) != NULL) {
476 robj *eleobj = dictGetEntryKey(de);
477 double *score = dictGetEntryVal(de);
478
479 if (fwrite(cmd,sizeof(cmd)-1,1,fp) == 0) goto werr;
480 if (fwriteBulkObject(fp,&key) == 0) goto werr;
481 if (fwriteBulkDouble(fp,*score) == 0) goto werr;
482 if (fwriteBulkObject(fp,eleobj) == 0) goto werr;
483 }
484 dictReleaseIterator(di);
485 } else {
486 redisPanic("Unknown sorted set encoding");
e2641e09 487 }
e2641e09 488 } else if (o->type == REDIS_HASH) {
489 char cmd[]="*4\r\n$4\r\nHSET\r\n";
490
491 /* Emit the HSETs needed to rebuild the hash */
492 if (o->encoding == REDIS_ENCODING_ZIPMAP) {
493 unsigned char *p = zipmapRewind(o->ptr);
494 unsigned char *field, *val;
495 unsigned int flen, vlen;
496
497 while((p = zipmapNext(p,&field,&flen,&val,&vlen)) != NULL) {
498 if (fwrite(cmd,sizeof(cmd)-1,1,fp) == 0) goto werr;
499 if (fwriteBulkObject(fp,&key) == 0) goto werr;
daf2049d 500 if (fwriteBulkString(fp,(char*)field,flen) == 0)
5bd09cd4 501 goto werr;
daf2049d 502 if (fwriteBulkString(fp,(char*)val,vlen) == 0)
5bd09cd4 503 goto werr;
e2641e09 504 }
505 } else {
506 dictIterator *di = dictGetIterator(o->ptr);
507 dictEntry *de;
508
509 while((de = dictNext(di)) != NULL) {
510 robj *field = dictGetEntryKey(de);
511 robj *val = dictGetEntryVal(de);
512
513 if (fwrite(cmd,sizeof(cmd)-1,1,fp) == 0) goto werr;
514 if (fwriteBulkObject(fp,&key) == 0) goto werr;
5bd09cd4 515 if (fwriteBulkObject(fp,field) == 0) goto werr;
516 if (fwriteBulkObject(fp,val) == 0) goto werr;
e2641e09 517 }
518 dictReleaseIterator(di);
519 }
520 } else {
521 redisPanic("Unknown object type");
522 }
523 /* Save the expire time */
524 if (expiretime != -1) {
525 char cmd[]="*3\r\n$8\r\nEXPIREAT\r\n";
526 /* If this key is already expired skip it */
527 if (expiretime < now) continue;
528 if (fwrite(cmd,sizeof(cmd)-1,1,fp) == 0) goto werr;
529 if (fwriteBulkObject(fp,&key) == 0) goto werr;
530 if (fwriteBulkLongLong(fp,expiretime) == 0) goto werr;
531 }
e2641e09 532 }
533 dictReleaseIterator(di);
534 }
535
536 /* Make sure data will not remain on the OS's output buffers */
537 fflush(fp);
538 aof_fsync(fileno(fp));
539 fclose(fp);
540
541 /* Use RENAME to make sure the DB file is changed atomically only
542 * if the generate DB file is ok. */
543 if (rename(tmpfile,filename) == -1) {
544 redisLog(REDIS_WARNING,"Error moving temp append only file on the final destination: %s", strerror(errno));
545 unlink(tmpfile);
546 return REDIS_ERR;
547 }
548 redisLog(REDIS_NOTICE,"SYNC append only file rewrite performed");
549 return REDIS_OK;
550
551werr:
552 fclose(fp);
553 unlink(tmpfile);
554 redisLog(REDIS_WARNING,"Write error writing append only file on disk: %s", strerror(errno));
555 if (di) dictReleaseIterator(di);
556 return REDIS_ERR;
557}
558
559/* This is how rewriting of the append only file in background works:
560 *
561 * 1) The user calls BGREWRITEAOF
562 * 2) Redis calls this function, that forks():
563 * 2a) the child rewrite the append only file in a temp file.
564 * 2b) the parent accumulates differences in server.bgrewritebuf.
565 * 3) When the child finished '2a' exists.
566 * 4) The parent will trap the exit code, if it's OK, will append the
567 * data accumulated into server.bgrewritebuf into the temp file, and
568 * finally will rename(2) the temp file in the actual file name.
569 * The the new file is reopened as the new append only file. Profit!
570 */
571int rewriteAppendOnlyFileBackground(void) {
572 pid_t childpid;
615e414c 573 long long start;
e2641e09 574
575 if (server.bgrewritechildpid != -1) return REDIS_ERR;
69bfffb4 576 if (server.ds_enabled != 0) {
577 redisLog(REDIS_WARNING,"BGREWRITEAOF called with diskstore enabled: AOF is not supported when diskstore is enabled. Operation not performed.");
578 return REDIS_ERR;
579 }
615e414c 580 start = ustime();
e2641e09 581 if ((childpid = fork()) == 0) {
e2641e09 582 char tmpfile[256];
583
615e414c 584 /* Child */
a5639e7d
PN
585 if (server.ipfd > 0) close(server.ipfd);
586 if (server.sofd > 0) close(server.sofd);
e2641e09 587 snprintf(tmpfile,256,"temp-rewriteaof-bg-%d.aof", (int) getpid());
588 if (rewriteAppendOnlyFile(tmpfile) == REDIS_OK) {
589 _exit(0);
590 } else {
591 _exit(1);
592 }
593 } else {
594 /* Parent */
615e414c 595 server.stat_fork_time = ustime()-start;
e2641e09 596 if (childpid == -1) {
597 redisLog(REDIS_WARNING,
598 "Can't rewrite append only file in background: fork: %s",
599 strerror(errno));
600 return REDIS_ERR;
601 }
602 redisLog(REDIS_NOTICE,
603 "Background append only file rewriting started by pid %d",childpid);
604 server.bgrewritechildpid = childpid;
605 updateDictResizePolicy();
606 /* We set appendseldb to -1 in order to force the next call to the
607 * feedAppendOnlyFile() to issue a SELECT command, so the differences
608 * accumulated by the parent into server.bgrewritebuf will start
609 * with a SELECT statement and it will be safe to merge. */
610 server.appendseldb = -1;
611 return REDIS_OK;
612 }
613 return REDIS_OK; /* unreached */
614}
615
616void bgrewriteaofCommand(redisClient *c) {
617 if (server.bgrewritechildpid != -1) {
3ab20376 618 addReplyError(c,"Background append only file rewriting already in progress");
b333e239 619 } else if (server.bgsavechildpid != -1) {
620 server.aofrewrite_scheduled = 1;
621 addReplyStatus(c,"Background append only file rewriting started");
622 } else if (rewriteAppendOnlyFileBackground() == REDIS_OK) {
3ab20376 623 addReplyStatus(c,"Background append only file rewriting started");
e2641e09 624 } else {
625 addReply(c,shared.err);
626 }
627}
628
629void aofRemoveTempFile(pid_t childpid) {
630 char tmpfile[256];
631
632 snprintf(tmpfile,256,"temp-rewriteaof-bg-%d.aof", (int) childpid);
633 unlink(tmpfile);
634}
635
b333e239 636/* Update the server.appendonly_current_size filed explicitly using stat(2)
637 * to check the size of the file. This is useful after a rewrite or after
638 * a restart, normally the size is updated just adding the write length
639 * to the current lenght, that is much faster. */
640void aofUpdateCurrentSize(void) {
641 struct redis_stat sb;
642
643 if (redis_fstat(server.appendfd,&sb) == -1) {
644 redisLog(REDIS_WARNING,"Unable to check the AOF length: %s",
645 strerror(errno));
646 } else {
647 server.appendonly_current_size = sb.st_size;
648 }
649}
650
e2641e09 651/* A background append only file rewriting (BGREWRITEAOF) terminated its work.
652 * Handle this. */
36c17a53 653void backgroundRewriteDoneHandler(int exitcode, int bysignal) {
e2641e09 654 if (!bysignal && exitcode == 0) {
655 int fd;
656 char tmpfile[256];
657
658 redisLog(REDIS_NOTICE,
659 "Background append only file rewriting terminated with success");
660 /* Now it's time to flush the differences accumulated by the parent */
661 snprintf(tmpfile,256,"temp-rewriteaof-bg-%d.aof", (int) server.bgrewritechildpid);
662 fd = open(tmpfile,O_WRONLY|O_APPEND);
663 if (fd == -1) {
664 redisLog(REDIS_WARNING, "Not able to open the temp append only file produced by the child: %s", strerror(errno));
665 goto cleanup;
666 }
667 /* Flush our data... */
668 if (write(fd,server.bgrewritebuf,sdslen(server.bgrewritebuf)) !=
669 (signed) sdslen(server.bgrewritebuf)) {
670 redisLog(REDIS_WARNING, "Error or short write trying to flush the parent diff of the append log file in the child temp file: %s", strerror(errno));
671 close(fd);
672 goto cleanup;
673 }
674 redisLog(REDIS_NOTICE,"Parent diff flushed into the new append log file with success (%lu bytes)",sdslen(server.bgrewritebuf));
675 /* Now our work is to rename the temp file into the stable file. And
676 * switch the file descriptor used by the server for append only. */
677 if (rename(tmpfile,server.appendfilename) == -1) {
678 redisLog(REDIS_WARNING,"Can't rename the temp append only file into the stable one: %s", strerror(errno));
679 close(fd);
680 goto cleanup;
681 }
682 /* Mission completed... almost */
683 redisLog(REDIS_NOTICE,"Append only file successfully rewritten.");
684 if (server.appendfd != -1) {
685 /* If append only is actually enabled... */
686 close(server.appendfd);
687 server.appendfd = fd;
688 if (server.appendfsync != APPENDFSYNC_NO) aof_fsync(fd);
689 server.appendseldb = -1; /* Make sure it will issue SELECT */
690 redisLog(REDIS_NOTICE,"The new append only file was selected for future appends.");
b333e239 691 aofUpdateCurrentSize();
e2641e09 692 } else {
693 /* If append only is disabled we just generate a dump in this
694 * format. Why not? */
695 close(fd);
696 }
697 } else if (!bysignal && exitcode != 0) {
698 redisLog(REDIS_WARNING, "Background append only file rewriting error");
699 } else {
700 redisLog(REDIS_WARNING,
701 "Background append only file rewriting terminated by signal %d",
5b8ce853 702 bysignal);
e2641e09 703 }
704cleanup:
705 sdsfree(server.bgrewritebuf);
706 server.bgrewritebuf = sdsempty();
707 aofRemoveTempFile(server.bgrewritechildpid);
708 server.bgrewritechildpid = -1;
709}