src/aof.c

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