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