src/aof.c

   1 #include "redis.h"
   2 #include "bio.h"
   3 #include "rio.h"
   4
   5 #include <signal.h>
   6 #include <fcntl.h>
   7 #include <sys/stat.h>
   8 #include <sys/types.h>
   9 #include <sys/time.h>
  10 #include <sys/resource.h>
  11 #include <sys/wait.h>
  12
  13 void aofUpdateCurrentSize(void);
  14
  15 void aof_background_fsync(int fd) {
  16     bioCreateBackgroundJob(REDIS_BIO_AOF_FSYNC,(void*)(long)fd,NULL,NULL);
  17 }
  18
  19 /* Called when the user switches from "appendonly yes" to "appendonly no"
  20  * at runtime using the CONFIG command. */
  21 void stopAppendOnly(void) {
  22     flushAppendOnlyFile(1);
  23     aof_fsync(server.appendfd);
  24     close(server.appendfd);
  25
  26     server.appendfd = -1;
  27     server.appendseldb = -1;
  28     server.appendonly = 0;
  29     /* rewrite operation in progress? kill it, wait child exit */
  30     if (server.bgrewritechildpid != -1) {
  31         int statloc;
  32
  33         if (kill(server.bgrewritechildpid,SIGKILL) != -1)
  34             wait3(&statloc,0,NULL);
  35         /* reset the buffer accumulating changes while the child saves */
  36         sdsfree(server.bgrewritebuf);
  37         server.bgrewritebuf = sdsempty();
  38         server.bgrewritechildpid = -1;
  39     }
  40 }
  41
  42 /* Called when the user switches from "appendonly no" to "appendonly yes"
  43  * at runtime using the CONFIG command. */
  44 int startAppendOnly(void) {
  45     server.appendonly = 1;
  46     server.lastfsync = time(NULL);
  47     server.appendfd = open(server.appendfilename,O_WRONLY|O_APPEND|O_CREAT,0644);
  48     if (server.appendfd == -1) {
  49         redisLog(REDIS_WARNING,"Used tried to switch on AOF via CONFIG, but I can't open the AOF file: %s",strerror(errno));
  50         return REDIS_ERR;
  51     }
  52     if (rewriteAppendOnlyFileBackground() == REDIS_ERR) {
  53         server.appendonly = 0;
  54         close(server.appendfd);
  55         redisLog(REDIS_WARNING,"User tried turning on AOF with CONFIG SET but I can't trigger a background AOF rewrite operation. Check the above logs for more info about the error.");
  56         return REDIS_ERR;
  57     }
  58     return REDIS_OK;
  59 }
  60
  61 /* Write the append only file buffer on disk.
  62  *
  63  * Since we are required to write the AOF before replying to the client,
  64  * and the only way the client socket can get a write is entering when the
  65  * the event loop, we accumulate all the AOF writes in a memory
  66  * buffer and write it on disk using this function just before entering
  67  * the event loop again.
  68  *
  69  * About the 'force' argument:
  70  *
  71  * When the fsync policy is set to 'everysec' we may delay the flush if there
  72  * is still an fsync() going on in the background thread, since for instance
  73  * on Linux write(2) will be blocked by the background fsync anyway.
  74  * When this happens we remember that there is some aof buffer to be
  75  * flushed ASAP, and will try to do that in the serverCron() function.
  76  *
  77  * However if force is set to 1 we'll write regardless of the background
  78  * fsync. */
  79 void flushAppendOnlyFile(int force) {
  80     ssize_t nwritten;
  81     int sync_in_progress = 0;
  82
  83     if (sdslen(server.aofbuf) == 0) return;
  84
  85     if (server.appendfsync == APPENDFSYNC_EVERYSEC)
  86         sync_in_progress = bioPendingJobsOfType(REDIS_BIO_AOF_FSYNC) != 0;
  87
  88     if (server.appendfsync == APPENDFSYNC_EVERYSEC && !force) {
  89         /* With this append fsync policy we do background fsyncing.
  90          * If the fsync is still in progress we can try to delay
  91          * the write for a couple of seconds. */
  92         if (sync_in_progress) {
  93             if (server.aof_flush_postponed_start == 0) {
  94                 /* No previous write postponinig, remember that we are
  95                  * postponing the flush and return. */
  96                 server.aof_flush_postponed_start = server.unixtime;
  97                 return;
  98             } else if (server.unixtime - server.aof_flush_postponed_start < 2) {
  99                 /* We were already waiting for fsync to finish, but for less
 100                  * than two seconds this is still ok. Postpone again. */
 101                 return;
 102             }
 103             /* Otherwise fall trough, and go write since we can't wait
 104              * over two seconds. */
 105             redisLog(REDIS_NOTICE,"Asynchronous AOF fsync is taking too long (disk is busy?). Writing the AOF buffer without waiting for fsync to complete, this may slow down Redis.");
 106         }
 107     }
 108     /* If you are following this code path, then we are going to write so
 109      * set reset the postponed flush sentinel to zero. */
 110     server.aof_flush_postponed_start = 0;
 111
 112     /* We want to perform a single write. This should be guaranteed atomic
 113      * at least if the filesystem we are writing is a real physical one.
 114      * While this will save us against the server being killed I don't think
 115      * there is much to do about the whole server stopping for power problems
 116      * or alike */
 117     nwritten = write(server.appendfd,server.aofbuf,sdslen(server.aofbuf));
 118     if (nwritten != (signed)sdslen(server.aofbuf)) {
 119         /* Ooops, we are in troubles. The best thing to do for now is
 120          * aborting instead of giving the illusion that everything is
 121          * working as expected. */
 122         if (nwritten == -1) {
 123             redisLog(REDIS_WARNING,"Exiting on error writing to the append-only file: %s",strerror(errno));
 124         } else {
 125             redisLog(REDIS_WARNING,"Exiting on short write while writing to the append-only file: %s",strerror(errno));
 126         }
 127         exit(1);
 128     }
 129     server.appendonly_current_size += nwritten;
 130
 131     /* Re-use AOF buffer when it is small enough. The maximum comes from the
 132      * arena size of 4k minus some overhead (but is otherwise arbitrary). */
 133     if ((sdslen(server.aofbuf)+sdsavail(server.aofbuf)) < 4000) {
 134         sdsclear(server.aofbuf);
 135     } else {
 136         sdsfree(server.aofbuf);
 137         server.aofbuf = sdsempty();
 138     }
 139
 140     /* Don't fsync if no-appendfsync-on-rewrite is set to yes and there are
 141      * children doing I/O in the background. */
 142     if (server.no_appendfsync_on_rewrite &&
 143         (server.bgrewritechildpid != -1 || server.bgsavechildpid != -1))
 144             return;
 145
 146     /* Perform the fsync if needed. */
 147     if (server.appendfsync == APPENDFSYNC_ALWAYS) {
 148         /* aof_fsync is defined as fdatasync() for Linux in order to avoid
 149          * flushing metadata. */
 150         aof_fsync(server.appendfd); /* Let's try to get this data on the disk */
 151         server.lastfsync = server.unixtime;
 152     } else if ((server.appendfsync == APPENDFSYNC_EVERYSEC &&
 153                 server.unixtime > server.lastfsync)) {
 154         if (!sync_in_progress) aof_background_fsync(server.appendfd);
 155         server.lastfsync = server.unixtime;
 156     }
 157 }
 158
 159 sds catAppendOnlyGenericCommand(sds dst, int argc, robj **argv) {
 160     char buf[32];
 161     int len, j;
 162     robj *o;
 163
 164     buf[0] = '*';
 165     len = 1+ll2string(buf+1,sizeof(buf)-1,argc);
 166     buf[len++] = '\r';
 167     buf[len++] = '\n';
 168     dst = sdscatlen(dst,buf,len);
 169
 170     for (j = 0; j < argc; j++) {
 171         o = getDecodedObject(argv[j]);
 172         buf[0] = '$';
 173         len = 1+ll2string(buf+1,sizeof(buf)-1,sdslen(o->ptr));
 174         buf[len++] = '\r';
 175         buf[len++] = '\n';
 176         dst = sdscatlen(dst,buf,len);
 177         dst = sdscatlen(dst,o->ptr,sdslen(o->ptr));
 178         dst = sdscatlen(dst,"\r\n",2);
 179         decrRefCount(o);
 180     }
 181     return dst;
 182 }
 183
 184 /* Create the sds representation of an PEXPIREAT command, using
 185  * 'seconds' as time to live and 'cmd' to understand what command
 186  * we are translating into a PEXPIREAT.
 187  *
 188  * This command is used in order to translate EXPIRE and PEXPIRE commands
 189  * into PEXPIREAT command so that we retain precision in the append only
 190  * file, and the time is always absolute and not relative. */
 191 sds catAppendOnlyExpireAtCommand(sds buf, struct redisCommand *cmd, robj *key, robj *seconds) {
 192     long long when;
 193     robj *argv[3];
 194
 195     /* Make sure we can use strtol */
 196     seconds = getDecodedObject(seconds);
 197     when = strtoll(seconds->ptr,NULL,10);
 198     /* Convert argument into milliseconds for EXPIRE, SETEX, EXPIREAT */
 199     if (cmd->proc == expireCommand || cmd->proc == setexCommand ||
 200         cmd->proc == expireatCommand)
 201     {
 202         when *= 1000;
 203     }
 204     /* Convert into absolute time for EXPIRE, PEXPIRE, SETEX, PSETEX */
 205     if (cmd->proc == expireCommand || cmd->proc == pexpireCommand ||
 206         cmd->proc == setexCommand || cmd->proc == psetexCommand)
 207     {
 208         when += mstime();
 209     }
 210     decrRefCount(seconds);
 211
 212     argv[0] = createStringObject("PEXPIREAT",9);
 213     argv[1] = key;
 214     argv[2] = createStringObjectFromLongLong(when);
 215     buf = catAppendOnlyGenericCommand(buf, 3, argv);
 216     decrRefCount(argv[0]);
 217     decrRefCount(argv[2]);
 218     return buf;
 219 }
 220
 221 void feedAppendOnlyFile(struct redisCommand *cmd, int dictid, robj **argv, int argc) {
 222     sds buf = sdsempty();
 223     robj *tmpargv[3];
 224
 225     /* The DB this command was targetting is not the same as the last command
 226      * we appendend. To issue a SELECT command is needed. */
 227     if (dictid != server.appendseldb) {
 228         char seldb[64];
 229
 230         snprintf(seldb,sizeof(seldb),"%d",dictid);
 231         buf = sdscatprintf(buf,"*2\r\n$6\r\nSELECT\r\n$%lu\r\n%s\r\n",
 232             (unsigned long)strlen(seldb),seldb);
 233         server.appendseldb = dictid;
 234     }
 235
 236     if (cmd->proc == expireCommand || cmd->proc == pexpireCommand ||
 237         cmd->proc == expireatCommand) {
 238         /* Translate EXPIRE/PEXPIRE/EXPIREAT into PEXPIREAT */
 239         buf = catAppendOnlyExpireAtCommand(buf,cmd,argv[1],argv[2]);
 240     } else if (cmd->proc == setexCommand || cmd->proc == psetexCommand) {
 241         /* Translate SETEX/PSETEX to SET and PEXPIREAT */
 242         tmpargv[0] = createStringObject("SET",3);
 243         tmpargv[1] = argv[1];
 244         tmpargv[2] = argv[3];
 245         buf = catAppendOnlyGenericCommand(buf,3,tmpargv);
 246         decrRefCount(tmpargv[0]);
 247         buf = catAppendOnlyExpireAtCommand(buf,cmd,argv[1],argv[2]);
 248     } else {
 249         /* All the other commands don't need translation or need the
 250          * same translation already operated in the command vector
 251          * for the replication itself. */
 252         buf = catAppendOnlyGenericCommand(buf,argc,argv);
 253     }
 254
 255     /* Append to the AOF buffer. This will be flushed on disk just before
 256      * of re-entering the event loop, so before the client will get a
 257      * positive reply about the operation performed. */
 258     server.aofbuf = sdscatlen(server.aofbuf,buf,sdslen(buf));
 259
 260     /* If a background append only file rewriting is in progress we want to
 261      * accumulate the differences between the child DB and the current one
 262      * in a buffer, so that when the child process will do its work we
 263      * can append the differences to the new append only file. */
 264     if (server.bgrewritechildpid != -1)
 265         server.bgrewritebuf = sdscatlen(server.bgrewritebuf,buf,sdslen(buf));
 266
 267     sdsfree(buf);
 268 }
 269
 270 /* In Redis commands are always executed in the context of a client, so in
 271  * order to load the append only file we need to create a fake client. */
 272 struct redisClient *createFakeClient(void) {
 273     struct redisClient *c = zmalloc(sizeof(*c));
 274
 275     selectDb(c,0);
 276     c->fd = -1;
 277     c->querybuf = sdsempty();
 278     c->argc = 0;
 279     c->argv = NULL;
 280     c->bufpos = 0;
 281     c->flags = 0;
 282     /* We set the fake client as a slave waiting for the synchronization
 283      * so that Redis will not try to send replies to this client. */
 284     c->replstate = REDIS_REPL_WAIT_BGSAVE_START;
 285     c->reply = listCreate();
 286     c->watched_keys = listCreate();
 287     listSetFreeMethod(c->reply,decrRefCount);
 288     listSetDupMethod(c->reply,dupClientReplyValue);
 289     initClientMultiState(c);
 290     return c;
 291 }
 292
 293 void freeFakeClient(struct redisClient *c) {
 294     sdsfree(c->querybuf);
 295     listRelease(c->reply);
 296     listRelease(c->watched_keys);
 297     freeClientMultiState(c);
 298     zfree(c);
 299 }
 300
 301 /* Replay the append log file. On error REDIS_OK is returned. On non fatal
 302  * error (the append only file is zero-length) REDIS_ERR is returned. On
 303  * fatal error an error message is logged and the program exists. */
 304 int loadAppendOnlyFile(char *filename) {
 305     struct redisClient *fakeClient;
 306     FILE *fp = fopen(filename,"r");
 307     struct redis_stat sb;
 308     int appendonly = server.appendonly;
 309     long loops = 0;
 310
 311     if (fp && redis_fstat(fileno(fp),&sb) != -1 && sb.st_size == 0) {
 312         server.appendonly_current_size = 0;
 313         fclose(fp);
 314         return REDIS_ERR;
 315     }
 316
 317     if (fp == NULL) {
 318         redisLog(REDIS_WARNING,"Fatal error: can't open the append log file for reading: %s",strerror(errno));
 319         exit(1);
 320     }
 321
 322     /* Temporarily disable AOF, to prevent EXEC from feeding a MULTI
 323      * to the same file we're about to read. */
 324     server.appendonly = 0;
 325
 326     fakeClient = createFakeClient();
 327     startLoading(fp);
 328
 329     while(1) {
 330         int argc, j;
 331         unsigned long len;
 332         robj **argv;
 333         char buf[128];
 334         sds argsds;
 335         struct redisCommand *cmd;
 336
 337         /* Serve the clients from time to time */
 338         if (!(loops++ % 1000)) {
 339             loadingProgress(ftello(fp));
 340             aeProcessEvents(server.el, AE_FILE_EVENTS|AE_DONT_WAIT);
 341         }
 342
 343         if (fgets(buf,sizeof(buf),fp) == NULL) {
 344             if (feof(fp))
 345                 break;
 346             else
 347                 goto readerr;
 348         }
 349         if (buf[0] != '*') goto fmterr;
 350         argc = atoi(buf+1);
 351         if (argc < 1) goto fmterr;
 352
 353         argv = zmalloc(sizeof(robj*)*argc);
 354         for (j = 0; j < argc; j++) {
 355             if (fgets(buf,sizeof(buf),fp) == NULL) goto readerr;
 356             if (buf[0] != '$') goto fmterr;
 357             len = strtol(buf+1,NULL,10);
 358             argsds = sdsnewlen(NULL,len);
 359             if (len && fread(argsds,len,1,fp) == 0) goto fmterr;
 360             argv[j] = createObject(REDIS_STRING,argsds);
 361             if (fread(buf,2,1,fp) == 0) goto fmterr; /* discard CRLF */
 362         }
 363
 364         /* Command lookup */
 365         cmd = lookupCommand(argv[0]->ptr);
 366         if (!cmd) {
 367             redisLog(REDIS_WARNING,"Unknown command '%s' reading the append only file", argv[0]->ptr);
 368             exit(1);
 369         }
 370         /* Run the command in the context of a fake client */
 371         fakeClient->argc = argc;
 372         fakeClient->argv = argv;
 373         cmd->proc(fakeClient);
 374
 375         /* The fake client should not have a reply */
 376         redisAssert(fakeClient->bufpos == 0 && listLength(fakeClient->reply) == 0);
 377         /* The fake client should never get blocked */
 378         redisAssert((fakeClient->flags & REDIS_BLOCKED) == 0);
 379
 380         /* Clean up. Command code may have changed argv/argc so we use the
 381          * argv/argc of the client instead of the local variables. */
 382         for (j = 0; j < fakeClient->argc; j++)
 383             decrRefCount(fakeClient->argv[j]);
 384         zfree(fakeClient->argv);
 385     }
 386
 387     /* This point can only be reached when EOF is reached without errors.
 388      * If the client is in the middle of a MULTI/EXEC, log error and quit. */
 389     if (fakeClient->flags & REDIS_MULTI) goto readerr;
 390
 391     fclose(fp);
 392     freeFakeClient(fakeClient);
 393     server.appendonly = appendonly;
 394     stopLoading();
 395     aofUpdateCurrentSize();
 396     server.auto_aofrewrite_base_size = server.appendonly_current_size;
 397     return REDIS_OK;
 398
 399 readerr:
 400     if (feof(fp)) {
 401         redisLog(REDIS_WARNING,"Unexpected end of file reading the append only file");
 402     } else {
 403         redisLog(REDIS_WARNING,"Unrecoverable error reading the append only file: %s", strerror(errno));
 404     }
 405     exit(1);
 406 fmterr:
 407     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>");
 408     exit(1);
 409 }
 410
 411 /* Delegate writing an object to writing a bulk string or bulk long long.
 412  * This is not placed in rio.c since that adds the redis.h dependency. */
 413 int rioWriteBulkObject(rio *r, robj *obj) {
 414     /* Avoid using getDecodedObject to help copy-on-write (we are often
 415      * in a child process when this function is called). */
 416     if (obj->encoding == REDIS_ENCODING_INT) {
 417         return rioWriteBulkLongLong(r,(long)obj->ptr);
 418     } else if (obj->encoding == REDIS_ENCODING_RAW) {
 419         return rioWriteBulkString(r,obj->ptr,sdslen(obj->ptr));
 420     } else {
 421         redisPanic("Unknown string encoding");
 422     }
 423 }
 424
 425 /* Emit the commands needed to rebuild a list object.
 426  * The function returns 0 on error, 1 on success. */
 427 int rewriteListObject(rio *r, robj *key, robj *o) {
 428     long long count = 0, items = listTypeLength(o);
 429
 430     if (o->encoding == REDIS_ENCODING_ZIPLIST) {
 431         unsigned char *zl = o->ptr;
 432         unsigned char *p = ziplistIndex(zl,0);
 433         unsigned char *vstr;
 434         unsigned int vlen;
 435         long long vlong;
 436
 437         while(ziplistGet(p,&vstr,&vlen,&vlong)) {
 438             if (count == 0) {
 439                 int cmd_items = (items > REDIS_AOFREWRITE_ITEMS_PER_CMD) ?
 440                     REDIS_AOFREWRITE_ITEMS_PER_CMD : items;
 441
 442                 if (rioWriteBulkCount(r,'*',2+cmd_items) == 0) return 0;
 443                 if (rioWriteBulkString(r,"RPUSH",5) == 0) return 0;
 444                 if (rioWriteBulkObject(r,key) == 0) return 0;
 445             }
 446             if (vstr) {
 447                 if (rioWriteBulkString(r,(char*)vstr,vlen) == 0) return 0;
 448             } else {
 449                 if (rioWriteBulkLongLong(r,vlong) == 0) return 0;
 450             }
 451             p = ziplistNext(zl,p);
 452             if (++count == REDIS_AOFREWRITE_ITEMS_PER_CMD) count = 0;
 453             items--;
 454         }
 455     } else if (o->encoding == REDIS_ENCODING_LINKEDLIST) {
 456         list *list = o->ptr;
 457         listNode *ln;
 458         listIter li;
 459
 460         listRewind(list,&li);
 461         while((ln = listNext(&li))) {
 462             robj *eleobj = listNodeValue(ln);
 463
 464             if (count == 0) {
 465                 int cmd_items = (items > REDIS_AOFREWRITE_ITEMS_PER_CMD) ?
 466                     REDIS_AOFREWRITE_ITEMS_PER_CMD : items;
 467
 468                 if (rioWriteBulkCount(r,'*',2+cmd_items) == 0) return 0;
 469                 if (rioWriteBulkString(r,"RPUSH",5) == 0) return 0;
 470                 if (rioWriteBulkObject(r,key) == 0) return 0;
 471             }
 472             if (rioWriteBulkObject(r,eleobj) == 0) return 0;
 473             if (++count == REDIS_AOFREWRITE_ITEMS_PER_CMD) count = 0;
 474             items--;
 475         }
 476     } else {
 477         redisPanic("Unknown list encoding");
 478     }
 479     return 1;
 480 }
 481
 482 /* Emit the commands needed to rebuild a set object.
 483  * The function returns 0 on error, 1 on success. */
 484 int rewriteSetObject(rio *r, robj *key, robj *o) {
 485     long long count = 0, items = setTypeSize(o);
 486
 487     if (o->encoding == REDIS_ENCODING_INTSET) {
 488         int ii = 0;
 489         int64_t llval;
 490
 491         while(intsetGet(o->ptr,ii++,&llval)) {
 492             if (count == 0) {
 493                 int cmd_items = (items > REDIS_AOFREWRITE_ITEMS_PER_CMD) ?
 494                     REDIS_AOFREWRITE_ITEMS_PER_CMD : items;
 495
 496                 if (rioWriteBulkCount(r,'*',2+cmd_items) == 0) return 0;
 497                 if (rioWriteBulkString(r,"SADD",4) == 0) return 0;
 498                 if (rioWriteBulkObject(r,key) == 0) return 0;
 499             }
 500             if (rioWriteBulkLongLong(r,llval) == 0) return 0;
 501             if (++count == REDIS_AOFREWRITE_ITEMS_PER_CMD) count = 0;
 502             items--;
 503         }
 504     } else if (o->encoding == REDIS_ENCODING_HT) {
 505         dictIterator *di = dictGetIterator(o->ptr);
 506         dictEntry *de;
 507
 508         while((de = dictNext(di)) != NULL) {
 509             robj *eleobj = dictGetKey(de);
 510             if (count == 0) {
 511                 int cmd_items = (items > REDIS_AOFREWRITE_ITEMS_PER_CMD) ?
 512                     REDIS_AOFREWRITE_ITEMS_PER_CMD : items;
 513
 514                 if (rioWriteBulkCount(r,'*',2+cmd_items) == 0) return 0;
 515                 if (rioWriteBulkString(r,"SADD",4) == 0) return 0;
 516                 if (rioWriteBulkObject(r,key) == 0) return 0;
 517             }
 518             if (rioWriteBulkObject(r,eleobj) == 0) return 0;
 519             if (++count == REDIS_AOFREWRITE_ITEMS_PER_CMD) count = 0;
 520             items--;
 521         }
 522         dictReleaseIterator(di);
 523     } else {
 524         redisPanic("Unknown set encoding");
 525     }
 526     return 1;
 527 }
 528
 529 /* Emit the commands needed to rebuild a sorted set object.
 530  * The function returns 0 on error, 1 on success. */
 531 int rewriteSortedSetObject(rio *r, robj *key, robj *o) {
 532     long long count = 0, items = zsetLength(o);
 533
 534     if (o->encoding == REDIS_ENCODING_ZIPLIST) {
 535         unsigned char *zl = o->ptr;
 536         unsigned char *eptr, *sptr;
 537         unsigned char *vstr;
 538         unsigned int vlen;
 539         long long vll;
 540         double score;
 541
 542         eptr = ziplistIndex(zl,0);
 543         redisAssert(eptr != NULL);
 544         sptr = ziplistNext(zl,eptr);
 545         redisAssert(sptr != NULL);
 546
 547         while (eptr != NULL) {
 548             redisAssert(ziplistGet(eptr,&vstr,&vlen,&vll));
 549             score = zzlGetScore(sptr);
 550
 551             if (count == 0) {
 552                 int cmd_items = (items > REDIS_AOFREWRITE_ITEMS_PER_CMD) ?
 553                     REDIS_AOFREWRITE_ITEMS_PER_CMD : items;
 554
 555                 if (rioWriteBulkCount(r,'*',2+cmd_items*2) == 0) return 0;
 556                 if (rioWriteBulkString(r,"ZADD",4) == 0) return 0;
 557                 if (rioWriteBulkObject(r,key) == 0) return 0;
 558             }
 559             if (rioWriteBulkDouble(r,score) == 0) return 0;
 560             if (vstr != NULL) {
 561                 if (rioWriteBulkString(r,(char*)vstr,vlen) == 0) return 0;
 562             } else {
 563                 if (rioWriteBulkLongLong(r,vll) == 0) return 0;
 564             }
 565             zzlNext(zl,&eptr,&sptr);
 566             if (++count == REDIS_AOFREWRITE_ITEMS_PER_CMD) count = 0;
 567             items--;
 568         }
 569     } else if (o->encoding == REDIS_ENCODING_SKIPLIST) {
 570         zset *zs = o->ptr;
 571         dictIterator *di = dictGetIterator(zs->dict);
 572         dictEntry *de;
 573
 574         while((de = dictNext(di)) != NULL) {
 575             robj *eleobj = dictGetKey(de);
 576             double *score = dictGetVal(de);
 577
 578             if (count == 0) {
 579                 int cmd_items = (items > REDIS_AOFREWRITE_ITEMS_PER_CMD) ?
 580                     REDIS_AOFREWRITE_ITEMS_PER_CMD : items;
 581
 582                 if (rioWriteBulkCount(r,'*',2+cmd_items*2) == 0) return 0;
 583                 if (rioWriteBulkString(r,"ZADD",4) == 0) return 0;
 584                 if (rioWriteBulkObject(r,key) == 0) return 0;
 585             }
 586             if (rioWriteBulkDouble(r,*score) == 0) return 0;
 587             if (rioWriteBulkObject(r,eleobj) == 0) return 0;
 588             if (++count == REDIS_AOFREWRITE_ITEMS_PER_CMD) count = 0;
 589             items--;
 590         }
 591         dictReleaseIterator(di);
 592     } else {
 593         redisPanic("Unknown sorted zset encoding");
 594     }
 595     return 1;
 596 }
 597
 598 /* Emit the commands needed to rebuild a hash object.
 599  * The function returns 0 on error, 1 on success. */
 600 int rewriteHashObject(rio *r, robj *key, robj *o) {
 601     long long count = 0, items = hashTypeLength(o);
 602
 603     if (o->encoding == REDIS_ENCODING_ZIPMAP) {
 604         unsigned char *p = zipmapRewind(o->ptr);
 605         unsigned char *field, *val;
 606         unsigned int flen, vlen;
 607
 608         while((p = zipmapNext(p,&field,&flen,&val,&vlen)) != NULL) {
 609             if (count == 0) {
 610                 int cmd_items = (items > REDIS_AOFREWRITE_ITEMS_PER_CMD) ?
 611                     REDIS_AOFREWRITE_ITEMS_PER_CMD : items;
 612
 613                 if (rioWriteBulkCount(r,'*',2+cmd_items*2) == 0) return 0;
 614                 if (rioWriteBulkString(r,"HMSET",5) == 0) return 0;
 615                 if (rioWriteBulkObject(r,key) == 0) return 0;
 616             }
 617             if (rioWriteBulkString(r,(char*)field,flen) == 0) return 0;
 618             if (rioWriteBulkString(r,(char*)val,vlen) == 0) return 0;
 619             if (++count == REDIS_AOFREWRITE_ITEMS_PER_CMD) count = 0;
 620             items--;
 621         }
 622     } else {
 623         dictIterator *di = dictGetIterator(o->ptr);
 624         dictEntry *de;
 625
 626         while((de = dictNext(di)) != NULL) {
 627             robj *field = dictGetKey(de);
 628             robj *val = dictGetVal(de);
 629
 630             if (count == 0) {
 631                 int cmd_items = (items > REDIS_AOFREWRITE_ITEMS_PER_CMD) ?
 632                     REDIS_AOFREWRITE_ITEMS_PER_CMD : items;
 633
 634                 if (rioWriteBulkCount(r,'*',2+cmd_items*2) == 0) return 0;
 635                 if (rioWriteBulkString(r,"HMSET",5) == 0) return 0;
 636                 if (rioWriteBulkObject(r,key) == 0) return 0;
 637             }
 638             if (rioWriteBulkObject(r,field) == 0) return 0;
 639             if (rioWriteBulkObject(r,val) == 0) return 0;
 640             if (++count == REDIS_AOFREWRITE_ITEMS_PER_CMD) count = 0;
 641             items--;
 642         }
 643         dictReleaseIterator(di);
 644     }
 645     return 1;
 646 }
 647
 648 /* Write a sequence of commands able to fully rebuild the dataset into
 649  * "filename". Used both by REWRITEAOF and BGREWRITEAOF.
 650  *
 651  * In order to minimize the number of commands needed in the rewritten
 652  * log Redis uses variadic commands when possible, such as RPUSH, SADD
 653  * and ZADD. However at max REDIS_AOFREWRITE_ITEMS_PER_CMD items per time
 654  * are inserted using a single command. */
 655 int rewriteAppendOnlyFile(char *filename) {
 656     dictIterator *di = NULL;
 657     dictEntry *de;
 658     rio aof;
 659     FILE *fp;
 660     char tmpfile[256];
 661     int j;
 662     long long now = mstime();
 663
 664     /* Note that we have to use a different temp name here compared to the
 665      * one used by rewriteAppendOnlyFileBackground() function. */
 666     snprintf(tmpfile,256,"temp-rewriteaof-%d.aof", (int) getpid());
 667     fp = fopen(tmpfile,"w");
 668     if (!fp) {
 669         redisLog(REDIS_WARNING, "Failed rewriting the append only file: %s", strerror(errno));
 670         return REDIS_ERR;
 671     }
 672
 673     rioInitWithFile(&aof,fp);
 674     for (j = 0; j < server.dbnum; j++) {
 675         char selectcmd[] = "*2\r\n$6\r\nSELECT\r\n";
 676         redisDb *db = server.db+j;
 677         dict *d = db->dict;
 678         if (dictSize(d) == 0) continue;
 679         di = dictGetSafeIterator(d);
 680         if (!di) {
 681             fclose(fp);
 682             return REDIS_ERR;
 683         }
 684
 685         /* SELECT the new DB */
 686         if (rioWrite(&aof,selectcmd,sizeof(selectcmd)-1) == 0) goto werr;
 687         if (rioWriteBulkLongLong(&aof,j) == 0) goto werr;
 688
 689         /* Iterate this DB writing every entry */
 690         while((de = dictNext(di)) != NULL) {
 691             sds keystr;
 692             robj key, *o;
 693             long long expiretime;
 694
 695             keystr = dictGetKey(de);
 696             o = dictGetVal(de);
 697             initStaticStringObject(key,keystr);
 698
 699             expiretime = getExpire(db,&key);
 700
 701             /* Save the key and associated value */
 702             if (o->type == REDIS_STRING) {
 703                 /* Emit a SET command */
 704                 char cmd[]="*3\r\n$3\r\nSET\r\n";
 705                 if (rioWrite(&aof,cmd,sizeof(cmd)-1) == 0) goto werr;
 706                 /* Key and value */
 707                 if (rioWriteBulkObject(&aof,&key) == 0) goto werr;
 708                 if (rioWriteBulkObject(&aof,o) == 0) goto werr;
 709             } else if (o->type == REDIS_LIST) {
 710                 if (rewriteListObject(&aof,&key,o) == 0) goto werr;
 711             } else if (o->type == REDIS_SET) {
 712                 if (rewriteSetObject(&aof,&key,o) == 0) goto werr;
 713             } else if (o->type == REDIS_ZSET) {
 714                 if (rewriteSortedSetObject(&aof,&key,o) == 0) goto werr;
 715             } else if (o->type == REDIS_HASH) {
 716                 if (rewriteHashObject(&aof,&key,o) == 0) goto werr;
 717             } else {
 718                 redisPanic("Unknown object type");
 719             }
 720             /* Save the expire time */
 721             if (expiretime != -1) {
 722                 char cmd[]="*3\r\n$9\r\nPEXPIREAT\r\n";
 723                 /* If this key is already expired skip it */
 724                 if (expiretime < now) continue;
 725                 if (rioWrite(&aof,cmd,sizeof(cmd)-1) == 0) goto werr;
 726                 if (rioWriteBulkObject(&aof,&key) == 0) goto werr;
 727                 if (rioWriteBulkLongLong(&aof,expiretime) == 0) goto werr;
 728             }
 729         }
 730         dictReleaseIterator(di);
 731     }
 732
 733     /* Make sure data will not remain on the OS's output buffers */
 734     fflush(fp);
 735     aof_fsync(fileno(fp));
 736     fclose(fp);
 737
 738     /* Use RENAME to make sure the DB file is changed atomically only
 739      * if the generate DB file is ok. */
 740     if (rename(tmpfile,filename) == -1) {
 741         redisLog(REDIS_WARNING,"Error moving temp append only file on the final destination: %s", strerror(errno));
 742         unlink(tmpfile);
 743         return REDIS_ERR;
 744     }
 745     redisLog(REDIS_NOTICE,"SYNC append only file rewrite performed");
 746     return REDIS_OK;
 747
 748 werr:
 749     fclose(fp);
 750     unlink(tmpfile);
 751     redisLog(REDIS_WARNING,"Write error writing append only file on disk: %s", strerror(errno));
 752     if (di) dictReleaseIterator(di);
 753     return REDIS_ERR;
 754 }
 755
 756 /* This is how rewriting of the append only file in background works:
 757  *
 758  * 1) The user calls BGREWRITEAOF
 759  * 2) Redis calls this function, that forks():
 760  *    2a) the child rewrite the append only file in a temp file.
 761  *    2b) the parent accumulates differences in server.bgrewritebuf.
 762  * 3) When the child finished '2a' exists.
 763  * 4) The parent will trap the exit code, if it's OK, will append the
 764  *    data accumulated into server.bgrewritebuf into the temp file, and
 765  *    finally will rename(2) the temp file in the actual file name.
 766  *    The the new file is reopened as the new append only file. Profit!
 767  */
 768 int rewriteAppendOnlyFileBackground(void) {
 769     pid_t childpid;
 770     long long start;
 771
 772     if (server.bgrewritechildpid != -1) return REDIS_ERR;
 773     start = ustime();
 774     if ((childpid = fork()) == 0) {
 775         char tmpfile[256];
 776
 777         /* Child */
 778         if (server.ipfd > 0) close(server.ipfd);
 779         if (server.sofd > 0) close(server.sofd);
 780         snprintf(tmpfile,256,"temp-rewriteaof-bg-%d.aof", (int) getpid());
 781         if (rewriteAppendOnlyFile(tmpfile) == REDIS_OK) {
 782             _exit(0);
 783         } else {
 784             _exit(1);
 785         }
 786     } else {
 787         /* Parent */
 788         server.stat_fork_time = ustime()-start;
 789         if (childpid == -1) {
 790             redisLog(REDIS_WARNING,
 791                 "Can't rewrite append only file in background: fork: %s",
 792                 strerror(errno));
 793             return REDIS_ERR;
 794         }
 795         redisLog(REDIS_NOTICE,
 796             "Background append only file rewriting started by pid %d",childpid);
 797         server.aofrewrite_scheduled = 0;
 798         server.bgrewritechildpid = childpid;
 799         updateDictResizePolicy();
 800         /* We set appendseldb to -1 in order to force the next call to the
 801          * feedAppendOnlyFile() to issue a SELECT command, so the differences
 802          * accumulated by the parent into server.bgrewritebuf will start
 803          * with a SELECT statement and it will be safe to merge. */
 804         server.appendseldb = -1;
 805         return REDIS_OK;
 806     }
 807     return REDIS_OK; /* unreached */
 808 }
 809
 810 void bgrewriteaofCommand(redisClient *c) {
 811     if (server.bgrewritechildpid != -1) {
 812         addReplyError(c,"Background append only file rewriting already in progress");
 813     } else if (server.bgsavechildpid != -1) {
 814         server.aofrewrite_scheduled = 1;
 815         addReplyStatus(c,"Background append only file rewriting scheduled");
 816     } else if (rewriteAppendOnlyFileBackground() == REDIS_OK) {
 817         addReplyStatus(c,"Background append only file rewriting started");
 818     } else {
 819         addReply(c,shared.err);
 820     }
 821 }
 822
 823 void aofRemoveTempFile(pid_t childpid) {
 824     char tmpfile[256];
 825
 826     snprintf(tmpfile,256,"temp-rewriteaof-bg-%d.aof", (int) childpid);
 827     unlink(tmpfile);
 828 }
 829
 830 /* Update the server.appendonly_current_size filed explicitly using stat(2)
 831  * to check the size of the file. This is useful after a rewrite or after
 832  * a restart, normally the size is updated just adding the write length
 833  * to the current lenght, that is much faster. */
 834 void aofUpdateCurrentSize(void) {
 835     struct redis_stat sb;
 836
 837     if (redis_fstat(server.appendfd,&sb) == -1) {
 838         redisLog(REDIS_WARNING,"Unable to check the AOF length: %s",
 839             strerror(errno));
 840     } else {
 841         server.appendonly_current_size = sb.st_size;
 842     }
 843 }
 844
 845 /* A background append only file rewriting (BGREWRITEAOF) terminated its work.
 846  * Handle this. */
 847 void backgroundRewriteDoneHandler(int exitcode, int bysignal) {
 848     if (!bysignal && exitcode == 0) {
 849         int newfd, oldfd;
 850         int nwritten;
 851         char tmpfile[256];
 852         long long now = ustime();
 853
 854         redisLog(REDIS_NOTICE,
 855             "Background AOF rewrite terminated with success");
 856
 857         /* Flush the differences accumulated by the parent to the
 858          * rewritten AOF. */
 859         snprintf(tmpfile,256,"temp-rewriteaof-bg-%d.aof",
 860             (int)server.bgrewritechildpid);
 861         newfd = open(tmpfile,O_WRONLY|O_APPEND);
 862         if (newfd == -1) {
 863             redisLog(REDIS_WARNING,
 864                 "Unable to open the temporary AOF produced by the child: %s", strerror(errno));
 865             goto cleanup;
 866         }
 867
 868         nwritten = write(newfd,server.bgrewritebuf,sdslen(server.bgrewritebuf));
 869         if (nwritten != (signed)sdslen(server.bgrewritebuf)) {
 870             if (nwritten == -1) {
 871                 redisLog(REDIS_WARNING,
 872                     "Error trying to flush the parent diff to the rewritten AOF: %s", strerror(errno));
 873             } else {
 874                 redisLog(REDIS_WARNING,
 875                     "Short write trying to flush the parent diff to the rewritten AOF: %s", strerror(errno));
 876             }
 877             close(newfd);
 878             goto cleanup;
 879         }
 880
 881         redisLog(REDIS_NOTICE,
 882             "Parent diff successfully flushed to the rewritten AOF (%lu bytes)", nwritten);
 883
 884         /* The only remaining thing to do is to rename the temporary file to
 885          * the configured file and switch the file descriptor used to do AOF
 886          * writes. We don't want close(2) or rename(2) calls to block the
 887          * server on old file deletion.
 888          *
 889          * There are two possible scenarios:
 890          *
 891          * 1) AOF is DISABLED and this was a one time rewrite. The temporary
 892          * file will be renamed to the configured file. When this file already
 893          * exists, it will be unlinked, which may block the server.
 894          *
 895          * 2) AOF is ENABLED and the rewritten AOF will immediately start
 896          * receiving writes. After the temporary file is renamed to the
 897          * configured file, the original AOF file descriptor will be closed.
 898          * Since this will be the last reference to that file, closing it
 899          * causes the underlying file to be unlinked, which may block the
 900          * server.
 901          *
 902          * To mitigate the blocking effect of the unlink operation (either
 903          * caused by rename(2) in scenario 1, or by close(2) in scenario 2), we
 904          * use a background thread to take care of this. First, we
 905          * make scenario 1 identical to scenario 2 by opening the target file
 906          * when it exists. The unlink operation after the rename(2) will then
 907          * be executed upon calling close(2) for its descriptor. Everything to
 908          * guarantee atomicity for this switch has already happened by then, so
 909          * we don't care what the outcome or duration of that close operation
 910          * is, as long as the file descriptor is released again. */
 911         if (server.appendfd == -1) {
 912             /* AOF disabled */
 913
 914              /* Don't care if this fails: oldfd will be -1 and we handle that.
 915               * One notable case of -1 return is if the old file does
 916               * not exist. */
 917              oldfd = open(server.appendfilename,O_RDONLY|O_NONBLOCK);
 918         } else {
 919             /* AOF enabled */
 920             oldfd = -1; /* We'll set this to the current AOF filedes later. */
 921         }
 922
 923         /* Rename the temporary file. This will not unlink the target file if
 924          * it exists, because we reference it with "oldfd". */
 925         if (rename(tmpfile,server.appendfilename) == -1) {
 926             redisLog(REDIS_WARNING,
 927                 "Error trying to rename the temporary AOF: %s", strerror(errno));
 928             close(newfd);
 929             if (oldfd != -1) close(oldfd);
 930             goto cleanup;
 931         }
 932
 933         if (server.appendfd == -1) {
 934             /* AOF disabled, we don't need to set the AOF file descriptor
 935              * to this new file, so we can close it. */
 936             close(newfd);
 937         } else {
 938             /* AOF enabled, replace the old fd with the new one. */
 939             oldfd = server.appendfd;
 940             server.appendfd = newfd;
 941             if (server.appendfsync == APPENDFSYNC_ALWAYS)
 942                 aof_fsync(newfd);
 943             else if (server.appendfsync == APPENDFSYNC_EVERYSEC)
 944                 aof_background_fsync(newfd);
 945             server.appendseldb = -1; /* Make sure SELECT is re-issued */
 946             aofUpdateCurrentSize();
 947             server.auto_aofrewrite_base_size = server.appendonly_current_size;
 948
 949             /* Clear regular AOF buffer since its contents was just written to
 950              * the new AOF from the background rewrite buffer. */
 951             sdsfree(server.aofbuf);
 952             server.aofbuf = sdsempty();
 953         }
 954
 955         redisLog(REDIS_NOTICE, "Background AOF rewrite successful");
 956
 957         /* Asynchronously close the overwritten AOF. */
 958         if (oldfd != -1) bioCreateBackgroundJob(REDIS_BIO_CLOSE_FILE,(void*)(long)oldfd,NULL,NULL);
 959
 960         redisLog(REDIS_VERBOSE,
 961             "Background AOF rewrite signal handler took %lldus", ustime()-now);
 962     } else if (!bysignal && exitcode != 0) {
 963         redisLog(REDIS_WARNING,
 964             "Background AOF rewrite terminated with error");
 965     } else {
 966         redisLog(REDIS_WARNING,
 967             "Background AOF rewrite terminated by signal %d", bysignal);
 968     }
 969
 970 cleanup:
 971     sdsfree(server.bgrewritebuf);
 972     server.bgrewritebuf = sdsempty();
 973     aofRemoveTempFile(server.bgrewritechildpid);
 974     server.bgrewritechildpid = -1;
 975 }