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 (redis_fstat(fileno(fp),&sb) != -1 && sb.st_size == 0)
 224         return REDIS_ERR;
 225
 226     if (fp == NULL) {
 227         redisLog(REDIS_WARNING,"Fatal error: can't open the append log file for reading: %s",strerror(errno));
 228         exit(1);
 229     }
 230
 231     /* Temporarily disable AOF, to prevent EXEC from feeding a MULTI
 232      * to the same file we're about to read. */
 233     server.appendonly = 0;
 234
 235     fakeClient = createFakeClient();
 236     startLoading(fp);
 237
 238     while(1) {
 239         int argc, j;
 240         unsigned long len;
 241         robj **argv;
 242         char buf[128];
 243         sds argsds;
 244         struct redisCommand *cmd;
 245
 246         /* Serve the clients from time to time */
 247         if (!(loops++ % 1000)) {
 248             loadingProgress(ftello(fp));
 249             aeProcessEvents(server.el, AE_FILE_EVENTS|AE_DONT_WAIT);
 250         }
 251
 252         if (fgets(buf,sizeof(buf),fp) == NULL) {
 253             if (feof(fp))
 254                 break;
 255             else
 256                 goto readerr;
 257         }
 258         if (buf[0] != '*') goto fmterr;
 259         argc = atoi(buf+1);
 260         argv = zmalloc(sizeof(robj*)*argc);
 261         for (j = 0; j < argc; j++) {
 262             if (fgets(buf,sizeof(buf),fp) == NULL) goto readerr;
 263             if (buf[0] != '$') goto fmterr;
 264             len = strtol(buf+1,NULL,10);
 265             argsds = sdsnewlen(NULL,len);
 266             if (len && fread(argsds,len,1,fp) == 0) goto fmterr;
 267             argv[j] = createObject(REDIS_STRING,argsds);
 268             if (fread(buf,2,1,fp) == 0) goto fmterr; /* discard CRLF */
 269         }
 270
 271         /* Command lookup */
 272         cmd = lookupCommand(argv[0]->ptr);
 273         if (!cmd) {
 274             redisLog(REDIS_WARNING,"Unknown command '%s' reading the append only file", argv[0]->ptr);
 275             exit(1);
 276         }
 277         /* Run the command in the context of a fake client */
 278         fakeClient->argc = argc;
 279         fakeClient->argv = argv;
 280         cmd->proc(fakeClient);
 281
 282         /* The fake client should not have a reply */
 283         redisAssert(fakeClient->bufpos == 0 && listLength(fakeClient->reply) == 0);
 284
 285         /* Clean up, ready for the next command */
 286         for (j = 0; j < argc; j++) decrRefCount(argv[j]);
 287         zfree(argv);
 288     }
 289
 290     /* This point can only be reached when EOF is reached without errors.
 291      * If the client is in the middle of a MULTI/EXEC, log error and quit. */
 292     if (fakeClient->flags & REDIS_MULTI) goto readerr;
 293
 294     fclose(fp);
 295     freeFakeClient(fakeClient);
 296     server.appendonly = appendonly;
 297     stopLoading();
 298     return REDIS_OK;
 299
 300 readerr:
 301     if (feof(fp)) {
 302         redisLog(REDIS_WARNING,"Unexpected end of file reading the append only file");
 303     } else {
 304         redisLog(REDIS_WARNING,"Unrecoverable error reading the append only file: %s", strerror(errno));
 305     }
 306     exit(1);
 307 fmterr:
 308     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>");
 309     exit(1);
 310 }
 311
 312 /* Write a sequence of commands able to fully rebuild the dataset into
 313  * "filename". Used both by REWRITEAOF and BGREWRITEAOF. */
 314 int rewriteAppendOnlyFile(char *filename) {
 315     dictIterator *di = NULL;
 316     dictEntry *de;
 317     FILE *fp;
 318     char tmpfile[256];
 319     int j;
 320     time_t now = time(NULL);
 321
 322     /* Note that we have to use a different temp name here compared to the
 323      * one used by rewriteAppendOnlyFileBackground() function. */
 324     snprintf(tmpfile,256,"temp-rewriteaof-%d.aof", (int) getpid());
 325     fp = fopen(tmpfile,"w");
 326     if (!fp) {
 327         redisLog(REDIS_WARNING, "Failed rewriting the append only file: %s", strerror(errno));
 328         return REDIS_ERR;
 329     }
 330     for (j = 0; j < server.dbnum; j++) {
 331         char selectcmd[] = "*2\r\n$6\r\nSELECT\r\n";
 332         redisDb *db = server.db+j;
 333         dict *d = db->dict;
 334         if (dictSize(d) == 0) continue;
 335         di = dictGetIterator(d);
 336         if (!di) {
 337             fclose(fp);
 338             return REDIS_ERR;
 339         }
 340
 341         /* SELECT the new DB */
 342         if (fwrite(selectcmd,sizeof(selectcmd)-1,1,fp) == 0) goto werr;
 343         if (fwriteBulkLongLong(fp,j) == 0) goto werr;
 344
 345         /* Iterate this DB writing every entry */
 346         while((de = dictNext(di)) != NULL) {
 347             sds keystr = dictGetEntryKey(de);
 348             robj key, *o;
 349             time_t expiretime;
 350
 351             keystr = dictGetEntryKey(de);
 352             o = dictGetEntryVal(de);
 353             initStaticStringObject(key,keystr);
 354
 355             expiretime = getExpire(db,&key);
 356
 357             /* Save the key and associated value */
 358             if (o->type == REDIS_STRING) {
 359                 /* Emit a SET command */
 360                 char cmd[]="*3\r\n$3\r\nSET\r\n";
 361                 if (fwrite(cmd,sizeof(cmd)-1,1,fp) == 0) goto werr;
 362                 /* Key and value */
 363                 if (fwriteBulkObject(fp,&key) == 0) goto werr;
 364                 if (fwriteBulkObject(fp,o) == 0) goto werr;
 365             } else if (o->type == REDIS_LIST) {
 366                 /* Emit the RPUSHes needed to rebuild the list */
 367                 char cmd[]="*3\r\n$5\r\nRPUSH\r\n";
 368                 if (o->encoding == REDIS_ENCODING_ZIPLIST) {
 369                     unsigned char *zl = o->ptr;
 370                     unsigned char *p = ziplistIndex(zl,0);
 371                     unsigned char *vstr;
 372                     unsigned int vlen;
 373                     long long vlong;
 374
 375                     while(ziplistGet(p,&vstr,&vlen,&vlong)) {
 376                         if (fwrite(cmd,sizeof(cmd)-1,1,fp) == 0) goto werr;
 377                         if (fwriteBulkObject(fp,&key) == 0) goto werr;
 378                         if (vstr) {
 379                             if (fwriteBulkString(fp,(char*)vstr,vlen) == 0)
 380                                 goto werr;
 381                         } else {
 382                             if (fwriteBulkLongLong(fp,vlong) == 0)
 383                                 goto werr;
 384                         }
 385                         p = ziplistNext(zl,p);
 386                     }
 387                 } else if (o->encoding == REDIS_ENCODING_LINKEDLIST) {
 388                     list *list = o->ptr;
 389                     listNode *ln;
 390                     listIter li;
 391
 392                     listRewind(list,&li);
 393                     while((ln = listNext(&li))) {
 394                         robj *eleobj = listNodeValue(ln);
 395
 396                         if (fwrite(cmd,sizeof(cmd)-1,1,fp) == 0) goto werr;
 397                         if (fwriteBulkObject(fp,&key) == 0) goto werr;
 398                         if (fwriteBulkObject(fp,eleobj) == 0) goto werr;
 399                     }
 400                 } else {
 401                     redisPanic("Unknown list encoding");
 402                 }
 403             } else if (o->type == REDIS_SET) {
 404                 char cmd[]="*3\r\n$4\r\nSADD\r\n";
 405
 406                 /* Emit the SADDs needed to rebuild the set */
 407                 if (o->encoding == REDIS_ENCODING_INTSET) {
 408                     int ii = 0;
 409                     int64_t llval;
 410                     while(intsetGet(o->ptr,ii++,&llval)) {
 411                         if (fwrite(cmd,sizeof(cmd)-1,1,fp) == 0) goto werr;
 412                         if (fwriteBulkObject(fp,&key) == 0) goto werr;
 413                         if (fwriteBulkLongLong(fp,llval) == 0) goto werr;
 414                     }
 415                 } else if (o->encoding == REDIS_ENCODING_HT) {
 416                     dictIterator *di = dictGetIterator(o->ptr);
 417                     dictEntry *de;
 418                     while((de = dictNext(di)) != NULL) {
 419                         robj *eleobj = dictGetEntryKey(de);
 420                         if (fwrite(cmd,sizeof(cmd)-1,1,fp) == 0) goto werr;
 421                         if (fwriteBulkObject(fp,&key) == 0) goto werr;
 422                         if (fwriteBulkObject(fp,eleobj) == 0) goto werr;
 423                     }
 424                     dictReleaseIterator(di);
 425                 } else {
 426                     redisPanic("Unknown set encoding");
 427                 }
 428             } else if (o->type == REDIS_ZSET) {
 429                 /* Emit the ZADDs needed to rebuild the sorted set */
 430                 zset *zs = o->ptr;
 431                 dictIterator *di = dictGetIterator(zs->dict);
 432                 dictEntry *de;
 433
 434                 while((de = dictNext(di)) != NULL) {
 435                     char cmd[]="*4\r\n$4\r\nZADD\r\n";
 436                     robj *eleobj = dictGetEntryKey(de);
 437                     double *score = dictGetEntryVal(de);
 438
 439                     if (fwrite(cmd,sizeof(cmd)-1,1,fp) == 0) goto werr;
 440                     if (fwriteBulkObject(fp,&key) == 0) goto werr;
 441                     if (fwriteBulkDouble(fp,*score) == 0) goto werr;
 442                     if (fwriteBulkObject(fp,eleobj) == 0) goto werr;
 443                 }
 444                 dictReleaseIterator(di);
 445             } else if (o->type == REDIS_HASH) {
 446                 char cmd[]="*4\r\n$4\r\nHSET\r\n";
 447
 448                 /* Emit the HSETs needed to rebuild the hash */
 449                 if (o->encoding == REDIS_ENCODING_ZIPMAP) {
 450                     unsigned char *p = zipmapRewind(o->ptr);
 451                     unsigned char *field, *val;
 452                     unsigned int flen, vlen;
 453
 454                     while((p = zipmapNext(p,&field,&flen,&val,&vlen)) != NULL) {
 455                         if (fwrite(cmd,sizeof(cmd)-1,1,fp) == 0) goto werr;
 456                         if (fwriteBulkObject(fp,&key) == 0) goto werr;
 457                         if (fwriteBulkString(fp,(char*)field,flen) == 0)
 458                             goto werr;
 459                         if (fwriteBulkString(fp,(char*)val,vlen) == 0)
 460                             goto werr;
 461                     }
 462                 } else {
 463                     dictIterator *di = dictGetIterator(o->ptr);
 464                     dictEntry *de;
 465
 466                     while((de = dictNext(di)) != NULL) {
 467                         robj *field = dictGetEntryKey(de);
 468                         robj *val = dictGetEntryVal(de);
 469
 470                         if (fwrite(cmd,sizeof(cmd)-1,1,fp) == 0) goto werr;
 471                         if (fwriteBulkObject(fp,&key) == 0) goto werr;
 472                         if (fwriteBulkObject(fp,field) == 0) goto werr;
 473                         if (fwriteBulkObject(fp,val) == 0) goto werr;
 474                     }
 475                     dictReleaseIterator(di);
 476                 }
 477             } else {
 478                 redisPanic("Unknown object type");
 479             }
 480             /* Save the expire time */
 481             if (expiretime != -1) {
 482                 char cmd[]="*3\r\n$8\r\nEXPIREAT\r\n";
 483                 /* If this key is already expired skip it */
 484                 if (expiretime < now) continue;
 485                 if (fwrite(cmd,sizeof(cmd)-1,1,fp) == 0) goto werr;
 486                 if (fwriteBulkObject(fp,&key) == 0) goto werr;
 487                 if (fwriteBulkLongLong(fp,expiretime) == 0) goto werr;
 488             }
 489         }
 490         dictReleaseIterator(di);
 491     }
 492
 493     /* Make sure data will not remain on the OS's output buffers */
 494     fflush(fp);
 495     aof_fsync(fileno(fp));
 496     fclose(fp);
 497
 498     /* Use RENAME to make sure the DB file is changed atomically only
 499      * if the generate DB file is ok. */
 500     if (rename(tmpfile,filename) == -1) {
 501         redisLog(REDIS_WARNING,"Error moving temp append only file on the final destination: %s", strerror(errno));
 502         unlink(tmpfile);
 503         return REDIS_ERR;
 504     }
 505     redisLog(REDIS_NOTICE,"SYNC append only file rewrite performed");
 506     return REDIS_OK;
 507
 508 werr:
 509     fclose(fp);
 510     unlink(tmpfile);
 511     redisLog(REDIS_WARNING,"Write error writing append only file on disk: %s", strerror(errno));
 512     if (di) dictReleaseIterator(di);
 513     return REDIS_ERR;
 514 }
 515
 516 /* This is how rewriting of the append only file in background works:
 517  *
 518  * 1) The user calls BGREWRITEAOF
 519  * 2) Redis calls this function, that forks():
 520  *    2a) the child rewrite the append only file in a temp file.
 521  *    2b) the parent accumulates differences in server.bgrewritebuf.
 522  * 3) When the child finished '2a' exists.
 523  * 4) The parent will trap the exit code, if it's OK, will append the
 524  *    data accumulated into server.bgrewritebuf into the temp file, and
 525  *    finally will rename(2) the temp file in the actual file name.
 526  *    The the new file is reopened as the new append only file. Profit!
 527  */
 528 int rewriteAppendOnlyFileBackground(void) {
 529     pid_t childpid;
 530
 531     if (server.bgrewritechildpid != -1) return REDIS_ERR;
 532     redisAssert(server.ds_enabled == 0);
 533     if ((childpid = fork()) == 0) {
 534         /* Child */
 535         char tmpfile[256];
 536
 537         if (server.ipfd > 0) close(server.ipfd);
 538         if (server.sofd > 0) close(server.sofd);
 539         snprintf(tmpfile,256,"temp-rewriteaof-bg-%d.aof", (int) getpid());
 540         if (rewriteAppendOnlyFile(tmpfile) == REDIS_OK) {
 541             _exit(0);
 542         } else {
 543             _exit(1);
 544         }
 545     } else {
 546         /* Parent */
 547         if (childpid == -1) {
 548             redisLog(REDIS_WARNING,
 549                 "Can't rewrite append only file in background: fork: %s",
 550                 strerror(errno));
 551             return REDIS_ERR;
 552         }
 553         redisLog(REDIS_NOTICE,
 554             "Background append only file rewriting started by pid %d",childpid);
 555         server.bgrewritechildpid = childpid;
 556         updateDictResizePolicy();
 557         /* We set appendseldb to -1 in order to force the next call to the
 558          * feedAppendOnlyFile() to issue a SELECT command, so the differences
 559          * accumulated by the parent into server.bgrewritebuf will start
 560          * with a SELECT statement and it will be safe to merge. */
 561         server.appendseldb = -1;
 562         return REDIS_OK;
 563     }
 564     return REDIS_OK; /* unreached */
 565 }
 566
 567 void bgrewriteaofCommand(redisClient *c) {
 568     if (server.bgrewritechildpid != -1) {
 569         addReplyError(c,"Background append only file rewriting already in progress");
 570         return;
 571     }
 572     if (rewriteAppendOnlyFileBackground() == REDIS_OK) {
 573         addReplyStatus(c,"Background append only file rewriting started");
 574     } else {
 575         addReply(c,shared.err);
 576     }
 577 }
 578
 579 void aofRemoveTempFile(pid_t childpid) {
 580     char tmpfile[256];
 581
 582     snprintf(tmpfile,256,"temp-rewriteaof-bg-%d.aof", (int) childpid);
 583     unlink(tmpfile);
 584 }
 585
 586 /* A background append only file rewriting (BGREWRITEAOF) terminated its work.
 587  * Handle this. */
 588 void backgroundRewriteDoneHandler(int statloc) {
 589     int exitcode = WEXITSTATUS(statloc);
 590     int bysignal = WIFSIGNALED(statloc);
 591
 592     if (!bysignal && exitcode == 0) {
 593         int fd;
 594         char tmpfile[256];
 595
 596         redisLog(REDIS_NOTICE,
 597             "Background append only file rewriting terminated with success");
 598         /* Now it's time to flush the differences accumulated by the parent */
 599         snprintf(tmpfile,256,"temp-rewriteaof-bg-%d.aof", (int) server.bgrewritechildpid);
 600         fd = open(tmpfile,O_WRONLY|O_APPEND);
 601         if (fd == -1) {
 602             redisLog(REDIS_WARNING, "Not able to open the temp append only file produced by the child: %s", strerror(errno));
 603             goto cleanup;
 604         }
 605         /* Flush our data... */
 606         if (write(fd,server.bgrewritebuf,sdslen(server.bgrewritebuf)) !=
 607                 (signed) sdslen(server.bgrewritebuf)) {
 608             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));
 609             close(fd);
 610             goto cleanup;
 611         }
 612         redisLog(REDIS_NOTICE,"Parent diff flushed into the new append log file with success (%lu bytes)",sdslen(server.bgrewritebuf));
 613         /* Now our work is to rename the temp file into the stable file. And
 614          * switch the file descriptor used by the server for append only. */
 615         if (rename(tmpfile,server.appendfilename) == -1) {
 616             redisLog(REDIS_WARNING,"Can't rename the temp append only file into the stable one: %s", strerror(errno));
 617             close(fd);
 618             goto cleanup;
 619         }
 620         /* Mission completed... almost */
 621         redisLog(REDIS_NOTICE,"Append only file successfully rewritten.");
 622         if (server.appendfd != -1) {
 623             /* If append only is actually enabled... */
 624             close(server.appendfd);
 625             server.appendfd = fd;
 626             if (server.appendfsync != APPENDFSYNC_NO) aof_fsync(fd);
 627             server.appendseldb = -1; /* Make sure it will issue SELECT */
 628             redisLog(REDIS_NOTICE,"The new append only file was selected for future appends.");
 629         } else {
 630             /* If append only is disabled we just generate a dump in this
 631              * format. Why not? */
 632             close(fd);
 633         }
 634     } else if (!bysignal && exitcode != 0) {
 635         redisLog(REDIS_WARNING, "Background append only file rewriting error");
 636     } else {
 637         redisLog(REDIS_WARNING,
 638             "Background append only file rewriting terminated by signal %d",
 639             WTERMSIG(statloc));
 640     }
 641 cleanup:
 642     sdsfree(server.bgrewritebuf);
 643     server.bgrewritebuf = sdsempty();
 644     aofRemoveTempFile(server.bgrewritechildpid);
 645     server.bgrewritechildpid = -1;
 646 }