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 void aofUpdateCurrentSize(void);
  12
  13 /* Called when the user switches from "appendonly yes" to "appendonly no"
  14  * at runtime using the CONFIG command. */
  15 void 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 */
  24     if (server.bgrewritechildpid != -1) {
  25         int statloc;
  26
  27         if (kill(server.bgrewritechildpid,SIGKILL) != -1)
  28             wait3(&statloc,0,NULL);
  29         /* reset the buffer accumulating changes while the child saves */
  30         sdsfree(server.bgrewritebuf);
  31         server.bgrewritebuf = sdsempty();
  32         server.bgrewritechildpid = -1;
  33     }
  34 }
  35
  36 /* Called when the user switches from "appendonly no" to "appendonly yes"
  37  * at runtime using the CONFIG command. */
  38 int 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. */
  62 void 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();
  87     server.appendonly_current_size += nwritten;
  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
 107 sds 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
 120 sds 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
 140 void 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. */
 187 struct 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;
 195     c->bufpos = 0;
 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();
 201     c->watched_keys = listCreate();
 202     listSetFreeMethod(c->reply,decrRefCount);
 203     listSetDupMethod(c->reply,dupClientReplyValue);
 204     initClientMultiState(c);
 205     return c;
 206 }
 207
 208 void freeFakeClient(struct redisClient *c) {
 209     sdsfree(c->querybuf);
 210     listRelease(c->reply);
 211     listRelease(c->watched_keys);
 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. */
 219 int loadAppendOnlyFile(char *filename) {
 220     struct redisClient *fakeClient;
 221     FILE *fp = fopen(filename,"r");
 222     struct redis_stat sb;
 223     int appendonly = server.appendonly;
 224     long loops = 0;
 225
 226     if (fp && redis_fstat(fileno(fp),&sb) != -1 && sb.st_size == 0) {
 227         server.appendonly_current_size = 0;
 228         fclose(fp);
 229         return REDIS_ERR;
 230     }
 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();
 242     startLoading(fp);
 243
 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;
 251
 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
 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         }
 283         /* Run the command in the context of a fake client */
 284         fakeClient->argc = argc;
 285         fakeClient->argv = argv;
 286         cmd->proc(fakeClient);
 287
 288         /* The fake client should not have a reply */
 289         redisAssert(fakeClient->bufpos == 0 && listLength(fakeClient->reply) == 0);
 290
 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);
 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;
 305     stopLoading();
 306     aofUpdateCurrentSize();
 307     server.auto_aofrewrite_base_size = server.appendonly_current_size;
 308     return REDIS_OK;
 309
 310 readerr:
 311     if (feof(fp)) {
 312         redisLog(REDIS_WARNING,"Unexpected end of file reading the append only file");
 313     } else {
 314         redisLog(REDIS_WARNING,"Unrecoverable error reading the append only file: %s", strerror(errno));
 315     }
 316     exit(1);
 317 fmterr:
 318     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>");
 319     exit(1);
 320 }
 321
 322 /* Write a sequence of commands able to fully rebuild the dataset into
 323  * "filename". Used both by REWRITEAOF and BGREWRITEAOF. */
 324 int rewriteAppendOnlyFile(char *filename) {
 325     dictIterator *di = NULL;
 326     dictEntry *de;
 327     FILE *fp;
 328     char tmpfile[256];
 329     int j;
 330     time_t now = time(NULL);
 331
 332     /* Note that we have to use a different temp name here compared to the
 333      * one used by rewriteAppendOnlyFileBackground() function. */
 334     snprintf(tmpfile,256,"temp-rewriteaof-%d.aof", (int) getpid());
 335     fp = fopen(tmpfile,"w");
 336     if (!fp) {
 337         redisLog(REDIS_WARNING, "Failed rewriting the append only file: %s", strerror(errno));
 338         return REDIS_ERR;
 339     }
 340     for (j = 0; j < server.dbnum; j++) {
 341         char selectcmd[] = "*2\r\n$6\r\nSELECT\r\n";
 342         redisDb *db = server.db+j;
 343         dict *d = db->dict;
 344         if (dictSize(d) == 0) continue;
 345         di = dictGetSafeIterator(d);
 346         if (!di) {
 347             fclose(fp);
 348             return REDIS_ERR;
 349         }
 350
 351         /* SELECT the new DB */
 352         if (fwrite(selectcmd,sizeof(selectcmd)-1,1,fp) == 0) goto werr;
 353         if (fwriteBulkLongLong(fp,j) == 0) goto werr;
 354
 355         /* Iterate this DB writing every entry */
 356         while((de = dictNext(di)) != NULL) {
 357             sds keystr;
 358             robj key, *o;
 359             time_t expiretime;
 360
 361             keystr = dictGetEntryKey(de);
 362             o = dictGetEntryVal(de);
 363             initStaticStringObject(key,keystr);
 364
 365             expiretime = getExpire(db,&key);
 366
 367             /* Save the key and associated value */
 368             if (o->type == REDIS_STRING) {
 369                 /* Emit a SET command */
 370                 char cmd[]="*3\r\n$3\r\nSET\r\n";
 371                 if (fwrite(cmd,sizeof(cmd)-1,1,fp) == 0) goto werr;
 372                 /* Key and value */
 373                 if (fwriteBulkObject(fp,&key) == 0) goto werr;
 374                 if (fwriteBulkObject(fp,o) == 0) goto werr;
 375             } else if (o->type == REDIS_LIST) {
 376                 /* Emit the RPUSHes needed to rebuild the list */
 377                 char cmd[]="*3\r\n$5\r\nRPUSH\r\n";
 378                 if (o->encoding == REDIS_ENCODING_ZIPLIST) {
 379                     unsigned char *zl = o->ptr;
 380                     unsigned char *p = ziplistIndex(zl,0);
 381                     unsigned char *vstr;
 382                     unsigned int vlen;
 383                     long long vlong;
 384
 385                     while(ziplistGet(p,&vstr,&vlen,&vlong)) {
 386                         if (fwrite(cmd,sizeof(cmd)-1,1,fp) == 0) goto werr;
 387                         if (fwriteBulkObject(fp,&key) == 0) goto werr;
 388                         if (vstr) {
 389                             if (fwriteBulkString(fp,(char*)vstr,vlen) == 0)
 390                                 goto werr;
 391                         } else {
 392                             if (fwriteBulkLongLong(fp,vlong) == 0)
 393                                 goto werr;
 394                         }
 395                         p = ziplistNext(zl,p);
 396                     }
 397                 } else if (o->encoding == REDIS_ENCODING_LINKEDLIST) {
 398                     list *list = o->ptr;
 399                     listNode *ln;
 400                     listIter li;
 401
 402                     listRewind(list,&li);
 403                     while((ln = listNext(&li))) {
 404                         robj *eleobj = listNodeValue(ln);
 405
 406                         if (fwrite(cmd,sizeof(cmd)-1,1,fp) == 0) goto werr;
 407                         if (fwriteBulkObject(fp,&key) == 0) goto werr;
 408                         if (fwriteBulkObject(fp,eleobj) == 0) goto werr;
 409                     }
 410                 } else {
 411                     redisPanic("Unknown list encoding");
 412                 }
 413             } else if (o->type == REDIS_SET) {
 414                 char cmd[]="*3\r\n$4\r\nSADD\r\n";
 415
 416                 /* Emit the SADDs needed to rebuild the set */
 417                 if (o->encoding == REDIS_ENCODING_INTSET) {
 418                     int ii = 0;
 419                     int64_t llval;
 420                     while(intsetGet(o->ptr,ii++,&llval)) {
 421                         if (fwrite(cmd,sizeof(cmd)-1,1,fp) == 0) goto werr;
 422                         if (fwriteBulkObject(fp,&key) == 0) goto werr;
 423                         if (fwriteBulkLongLong(fp,llval) == 0) goto werr;
 424                     }
 425                 } else if (o->encoding == REDIS_ENCODING_HT) {
 426                     dictIterator *di = dictGetIterator(o->ptr);
 427                     dictEntry *de;
 428                     while((de = dictNext(di)) != NULL) {
 429                         robj *eleobj = dictGetEntryKey(de);
 430                         if (fwrite(cmd,sizeof(cmd)-1,1,fp) == 0) goto werr;
 431                         if (fwriteBulkObject(fp,&key) == 0) goto werr;
 432                         if (fwriteBulkObject(fp,eleobj) == 0) goto werr;
 433                     }
 434                     dictReleaseIterator(di);
 435                 } else {
 436                     redisPanic("Unknown set encoding");
 437                 }
 438             } else if (o->type == REDIS_ZSET) {
 439                 /* Emit the ZADDs needed to rebuild the sorted set */
 440                 char cmd[]="*4\r\n$4\r\nZADD\r\n";
 441
 442                 if (o->encoding == REDIS_ENCODING_ZIPLIST) {
 443                     unsigned char *zl = o->ptr;
 444                     unsigned char *eptr, *sptr;
 445                     unsigned char *vstr;
 446                     unsigned int vlen;
 447                     long long vll;
 448                     double score;
 449
 450                     eptr = ziplistIndex(zl,0);
 451                     redisAssert(eptr != NULL);
 452                     sptr = ziplistNext(zl,eptr);
 453                     redisAssert(sptr != NULL);
 454
 455                     while (eptr != NULL) {
 456                         redisAssert(ziplistGet(eptr,&vstr,&vlen,&vll));
 457                         score = zzlGetScore(sptr);
 458
 459                         if (fwrite(cmd,sizeof(cmd)-1,1,fp) == 0) goto werr;
 460                         if (fwriteBulkObject(fp,&key) == 0) goto werr;
 461                         if (fwriteBulkDouble(fp,score) == 0) goto werr;
 462                         if (vstr != NULL) {
 463                             if (fwriteBulkString(fp,(char*)vstr,vlen) == 0)
 464                                 goto werr;
 465                         } else {
 466                             if (fwriteBulkLongLong(fp,vll) == 0)
 467                                 goto werr;
 468                         }
 469                         zzlNext(zl,&eptr,&sptr);
 470                     }
 471                 } else if (o->encoding == REDIS_ENCODING_SKIPLIST) {
 472                     zset *zs = o->ptr;
 473                     dictIterator *di = dictGetIterator(zs->dict);
 474                     dictEntry *de;
 475
 476                     while((de = dictNext(di)) != NULL) {
 477                         robj *eleobj = dictGetEntryKey(de);
 478                         double *score = dictGetEntryVal(de);
 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 (fwriteBulkObject(fp,eleobj) == 0) goto werr;
 484                     }
 485                     dictReleaseIterator(di);
 486                 } else {
 487                     redisPanic("Unknown sorted set encoding");
 488                 }
 489             } else if (o->type == REDIS_HASH) {
 490                 char cmd[]="*4\r\n$4\r\nHSET\r\n";
 491
 492                 /* Emit the HSETs needed to rebuild the hash */
 493                 if (o->encoding == REDIS_ENCODING_ZIPMAP) {
 494                     unsigned char *p = zipmapRewind(o->ptr);
 495                     unsigned char *field, *val;
 496                     unsigned int flen, vlen;
 497
 498                     while((p = zipmapNext(p,&field,&flen,&val,&vlen)) != NULL) {
 499                         if (fwrite(cmd,sizeof(cmd)-1,1,fp) == 0) goto werr;
 500                         if (fwriteBulkObject(fp,&key) == 0) goto werr;
 501                         if (fwriteBulkString(fp,(char*)field,flen) == 0)
 502                             goto werr;
 503                         if (fwriteBulkString(fp,(char*)val,vlen) == 0)
 504                             goto werr;
 505                     }
 506                 } else {
 507                     dictIterator *di = dictGetIterator(o->ptr);
 508                     dictEntry *de;
 509
 510                     while((de = dictNext(di)) != NULL) {
 511                         robj *field = dictGetEntryKey(de);
 512                         robj *val = dictGetEntryVal(de);
 513
 514                         if (fwrite(cmd,sizeof(cmd)-1,1,fp) == 0) goto werr;
 515                         if (fwriteBulkObject(fp,&key) == 0) goto werr;
 516                         if (fwriteBulkObject(fp,field) == 0) goto werr;
 517                         if (fwriteBulkObject(fp,val) == 0) goto werr;
 518                     }
 519                     dictReleaseIterator(di);
 520                 }
 521             } else {
 522                 redisPanic("Unknown object type");
 523             }
 524             /* Save the expire time */
 525             if (expiretime != -1) {
 526                 char cmd[]="*3\r\n$8\r\nEXPIREAT\r\n";
 527                 /* If this key is already expired skip it */
 528                 if (expiretime < now) continue;
 529                 if (fwrite(cmd,sizeof(cmd)-1,1,fp) == 0) goto werr;
 530                 if (fwriteBulkObject(fp,&key) == 0) goto werr;
 531                 if (fwriteBulkLongLong(fp,expiretime) == 0) goto werr;
 532             }
 533         }
 534         dictReleaseIterator(di);
 535     }
 536
 537     /* Make sure data will not remain on the OS's output buffers */
 538     fflush(fp);
 539     aof_fsync(fileno(fp));
 540     fclose(fp);
 541
 542     /* Use RENAME to make sure the DB file is changed atomically only
 543      * if the generate DB file is ok. */
 544     if (rename(tmpfile,filename) == -1) {
 545         redisLog(REDIS_WARNING,"Error moving temp append only file on the final destination: %s", strerror(errno));
 546         unlink(tmpfile);
 547         return REDIS_ERR;
 548     }
 549     redisLog(REDIS_NOTICE,"SYNC append only file rewrite performed");
 550     return REDIS_OK;
 551
 552 werr:
 553     fclose(fp);
 554     unlink(tmpfile);
 555     redisLog(REDIS_WARNING,"Write error writing append only file on disk: %s", strerror(errno));
 556     if (di) dictReleaseIterator(di);
 557     return REDIS_ERR;
 558 }
 559
 560 /* This is how rewriting of the append only file in background works:
 561  *
 562  * 1) The user calls BGREWRITEAOF
 563  * 2) Redis calls this function, that forks():
 564  *    2a) the child rewrite the append only file in a temp file.
 565  *    2b) the parent accumulates differences in server.bgrewritebuf.
 566  * 3) When the child finished '2a' exists.
 567  * 4) The parent will trap the exit code, if it's OK, will append the
 568  *    data accumulated into server.bgrewritebuf into the temp file, and
 569  *    finally will rename(2) the temp file in the actual file name.
 570  *    The the new file is reopened as the new append only file. Profit!
 571  */
 572 int rewriteAppendOnlyFileBackground(void) {
 573     pid_t childpid;
 574     long long start;
 575
 576     if (server.bgrewritechildpid != -1) return REDIS_ERR;
 577     start = ustime();
 578     if ((childpid = fork()) == 0) {
 579         char tmpfile[256];
 580
 581         /* Child */
 582         if (server.ipfd > 0) close(server.ipfd);
 583         if (server.sofd > 0) close(server.sofd);
 584         snprintf(tmpfile,256,"temp-rewriteaof-bg-%d.aof", (int) getpid());
 585         if (rewriteAppendOnlyFile(tmpfile) == REDIS_OK) {
 586             _exit(0);
 587         } else {
 588             _exit(1);
 589         }
 590     } else {
 591         /* Parent */
 592         server.stat_fork_time = ustime()-start;
 593         if (childpid == -1) {
 594             redisLog(REDIS_WARNING,
 595                 "Can't rewrite append only file in background: fork: %s",
 596                 strerror(errno));
 597             return REDIS_ERR;
 598         }
 599         redisLog(REDIS_NOTICE,
 600             "Background append only file rewriting started by pid %d",childpid);
 601         server.bgrewritechildpid = childpid;
 602         updateDictResizePolicy();
 603         /* We set appendseldb to -1 in order to force the next call to the
 604          * feedAppendOnlyFile() to issue a SELECT command, so the differences
 605          * accumulated by the parent into server.bgrewritebuf will start
 606          * with a SELECT statement and it will be safe to merge. */
 607         server.appendseldb = -1;
 608         return REDIS_OK;
 609     }
 610     return REDIS_OK; /* unreached */
 611 }
 612
 613 void bgrewriteaofCommand(redisClient *c) {
 614     if (server.bgrewritechildpid != -1) {
 615         addReplyError(c,"Background append only file rewriting already in progress");
 616     } else if (server.bgsavechildpid != -1) {
 617         server.aofrewrite_scheduled = 1;
 618         addReplyStatus(c,"Background append only file rewriting scheduled");
 619     } else if (rewriteAppendOnlyFileBackground() == REDIS_OK) {
 620         addReplyStatus(c,"Background append only file rewriting started");
 621     } else {
 622         addReply(c,shared.err);
 623     }
 624 }
 625
 626 void aofRemoveTempFile(pid_t childpid) {
 627     char tmpfile[256];
 628
 629     snprintf(tmpfile,256,"temp-rewriteaof-bg-%d.aof", (int) childpid);
 630     unlink(tmpfile);
 631 }
 632
 633 /* Update the server.appendonly_current_size filed explicitly using stat(2)
 634  * to check the size of the file. This is useful after a rewrite or after
 635  * a restart, normally the size is updated just adding the write length
 636  * to the current lenght, that is much faster. */
 637 void aofUpdateCurrentSize(void) {
 638     struct redis_stat sb;
 639
 640     if (redis_fstat(server.appendfd,&sb) == -1) {
 641         redisLog(REDIS_WARNING,"Unable to check the AOF length: %s",
 642             strerror(errno));
 643     } else {
 644         server.appendonly_current_size = sb.st_size;
 645     }
 646 }
 647
 648 /* A background append only file rewriting (BGREWRITEAOF) terminated its work.
 649  * Handle this. */
 650 void backgroundRewriteDoneHandler(int exitcode, int bysignal) {
 651     if (!bysignal && exitcode == 0) {
 652         int fd;
 653         char tmpfile[256];
 654
 655         redisLog(REDIS_NOTICE,
 656             "Background append only file rewriting terminated with success");
 657         /* Now it's time to flush the differences accumulated by the parent */
 658         snprintf(tmpfile,256,"temp-rewriteaof-bg-%d.aof", (int) server.bgrewritechildpid);
 659         fd = open(tmpfile,O_WRONLY|O_APPEND);
 660         if (fd == -1) {
 661             redisLog(REDIS_WARNING, "Not able to open the temp append only file produced by the child: %s", strerror(errno));
 662             goto cleanup;
 663         }
 664         /* Flush our data... */
 665         if (write(fd,server.bgrewritebuf,sdslen(server.bgrewritebuf)) !=
 666                 (signed) sdslen(server.bgrewritebuf)) {
 667             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));
 668             close(fd);
 669             goto cleanup;
 670         }
 671         redisLog(REDIS_NOTICE,"Parent diff flushed into the new append log file with success (%lu bytes)",sdslen(server.bgrewritebuf));
 672         /* Now our work is to rename the temp file into the stable file. And
 673          * switch the file descriptor used by the server for append only. */
 674         if (rename(tmpfile,server.appendfilename) == -1) {
 675             redisLog(REDIS_WARNING,"Can't rename the temp append only file into the stable one: %s", strerror(errno));
 676             close(fd);
 677             goto cleanup;
 678         }
 679         /* Mission completed... almost */
 680         redisLog(REDIS_NOTICE,"Append only file successfully rewritten.");
 681         if (server.appendfd != -1) {
 682             /* If append only is actually enabled... */
 683             close(server.appendfd);
 684             server.appendfd = fd;
 685             if (server.appendfsync != APPENDFSYNC_NO) aof_fsync(fd);
 686             server.appendseldb = -1; /* Make sure it will issue SELECT */
 687             redisLog(REDIS_NOTICE,"The new append only file was selected for future appends.");
 688             aofUpdateCurrentSize();
 689             server.auto_aofrewrite_base_size = server.appendonly_current_size;
 690         } else {
 691             /* If append only is disabled we just generate a dump in this
 692              * format. Why not? */
 693             close(fd);
 694         }
 695     } else if (!bysignal && exitcode != 0) {
 696         redisLog(REDIS_WARNING, "Background append only file rewriting error");
 697     } else {
 698         redisLog(REDIS_WARNING,
 699             "Background append only file rewriting terminated by signal %d",
 700             bysignal);
 701     }
 702 cleanup:
 703     sdsfree(server.bgrewritebuf);
 704     server.bgrewritebuf = sdsempty();
 705     aofRemoveTempFile(server.bgrewritechildpid);
 706     server.bgrewritechildpid = -1;
 707 }