1 /* 2 * Copyright (c) 2009-2012, Salvatore Sanfilippo <antirez at gmail dot com> 3 * All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions are met: 7 * 8 * * Redistributions of source code must retain the above copyright notice, 9 * this list of conditions and the following disclaimer. 10 * * Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * * Neither the name of Redis nor the names of its contributors may be used 14 * to endorse or promote products derived from this software without 15 * specific prior written permission. 16 * 17 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" 18 * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 20 * ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE 21 * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 22 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 23 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 24 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 25 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 26 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 27 * POSSIBILITY OF SUCH DAMAGE. 28 */ 29 30 #include "redis.h" 31 #include "bio.h" 32 #include "rio.h" 33 34 #include <signal.h> 35 #include <fcntl.h> 36 #include <sys/stat.h> 37 #include <sys/types.h> 38 #include <sys/time.h> 39 #include <sys/resource.h> 40 #include <sys/wait.h> 41 42 void aofUpdateCurrentSize(void); 43 void aofClosePipes(void); 44 45 /* ---------------------------------------------------------------------------- 46 * AOF rewrite buffer implementation. 47 * 48 * The following code implement a simple buffer used in order to accumulate 49 * changes while the background process is rewriting the AOF file. 50 * 51 * We only need to append, but can't just use realloc with a large block 52 * because 'huge' reallocs are not always handled as one could expect 53 * (via remapping of pages at OS level) but may involve copying data. 54 * 55 * For this reason we use a list of blocks, every block is 56 * AOF_RW_BUF_BLOCK_SIZE bytes. 57 * ------------------------------------------------------------------------- */ 58 59 #define AOF_RW_BUF_BLOCK_SIZE (1024*1024*10) /* 10 MB per block */ 60 61 typedef struct aofrwblock { 62 unsigned long used, free; 63 char buf[AOF_RW_BUF_BLOCK_SIZE]; 64 } aofrwblock; 65 66 /* This function free the old AOF rewrite buffer if needed, and initialize 67 * a fresh new one. It tests for server.aof_rewrite_buf_blocks equal to NULL 68 * so can be used for the first initialization as well. */ 69 void aofRewriteBufferReset(void) { 70 if (server.aof_rewrite_buf_blocks) 71 listRelease(server.aof_rewrite_buf_blocks); 72 73 server.aof_rewrite_buf_blocks = listCreate(); 74 listSetFreeMethod(server.aof_rewrite_buf_blocks,zfree); 75 } 76 77 /* Return the current size of the AOF rerwite buffer. */ 78 unsigned long aofRewriteBufferSize(void) { 79 listNode *ln; 80 listIter li; 81 unsigned long size = 0; 82 83 listRewind(server.aof_rewrite_buf_blocks,&li); 84 while((ln = listNext(&li))) { 85 aofrwblock *block = listNodeValue(ln); 86 size += block->used; 87 } 88 return size; 89 } 90 91 /* Event handler used to send data to the child process doing the AOF 92 * rewrite. We send pieces of our AOF differences buffer so that the final 93 * write when the child finishes the rewrite will be small. */ 94 void aofChildWriteDiffData(aeEventLoop *el, int fd, void *privdata, int mask) { 95 listNode *ln; 96 aofrwblock *block; 97 ssize_t nwritten; 98 REDIS_NOTUSED(el); 99 REDIS_NOTUSED(fd); 100 REDIS_NOTUSED(privdata); 101 REDIS_NOTUSED(mask); 102 103 while(1) { 104 ln = listFirst(server.aof_rewrite_buf_blocks); 105 block = ln ? ln->value : NULL; 106 if (server.aof_stop_sending_diff || !block) { 107 aeDeleteFileEvent(server.el,server.aof_pipe_write_data_to_child, 108 AE_WRITABLE); 109 return; 110 } 111 if (block->used > 0) { 112 nwritten = write(server.aof_pipe_write_data_to_child, 113 block->buf,block->used); 114 if (nwritten <= 0) return; 115 memmove(block->buf,block->buf+nwritten,block->used-nwritten); 116 block->used -= nwritten; 117 } 118 if (block->used == 0) listDelNode(server.aof_rewrite_buf_blocks,ln); 119 } 120 } 121 122 /* Append data to the AOF rewrite buffer, allocating new blocks if needed. */ 123 void aofRewriteBufferAppend(unsigned char *s, unsigned long len) { 124 listNode *ln = listLast(server.aof_rewrite_buf_blocks); 125 aofrwblock *block = ln ? ln->value : NULL; 126 127 while(len) { 128 /* If we already got at least an allocated block, try appending 129 * at least some piece into it. */ 130 if (block) { 131 unsigned long thislen = (block->free < len) ? block->free : len; 132 if (thislen) { /* The current block is not already full. */ 133 memcpy(block->buf+block->used, s, thislen); 134 block->used += thislen; 135 block->free -= thislen; 136 s += thislen; 137 len -= thislen; 138 } 139 } 140 141 if (len) { /* First block to allocate, or need another block. */ 142 int numblocks; 143 144 block = zmalloc(sizeof(*block)); 145 block->free = AOF_RW_BUF_BLOCK_SIZE; 146 block->used = 0; 147 listAddNodeTail(server.aof_rewrite_buf_blocks,block); 148 149 /* Log every time we cross more 10 or 100 blocks, respectively 150 * as a notice or warning. */ 151 numblocks = listLength(server.aof_rewrite_buf_blocks); 152 if (((numblocks+1) % 10) == 0) { 153 int level = ((numblocks+1) % 100) == 0 ? REDIS_WARNING : 154 REDIS_NOTICE; 155 redisLog(level,"Background AOF buffer size: %lu MB", 156 aofRewriteBufferSize()/(1024*1024)); 157 } 158 } 159 } 160 161 /* Install a file event to send data to the rewrite child if there is 162 * not one already. */ 163 if (aeGetFileEvents(server.el,server.aof_pipe_write_data_to_child) == 0) { 164 aeCreateFileEvent(server.el, server.aof_pipe_write_data_to_child, 165 AE_WRITABLE, aofChildWriteDiffData, NULL); 166 } 167 } 168 169 /* Write the buffer (possibly composed of multiple blocks) into the specified 170 * fd. If a short write or any other error happens -1 is returned, 171 * otherwise the number of bytes written is returned. */ 172 ssize_t aofRewriteBufferWrite(int fd) { 173 listNode *ln; 174 listIter li; 175 ssize_t count = 0; 176 177 listRewind(server.aof_rewrite_buf_blocks,&li); 178 while((ln = listNext(&li))) { 179 aofrwblock *block = listNodeValue(ln); 180 ssize_t nwritten; 181 182 if (block->used) { 183 nwritten = write(fd,block->buf,block->used); 184 if (nwritten != (ssize_t)block->used) { 185 if (nwritten == 0) errno = EIO; 186 return -1; 187 } 188 count += nwritten; 189 } 190 } 191 return count; 192 } 193 194 /* ---------------------------------------------------------------------------- 195 * AOF file implementation 196 * ------------------------------------------------------------------------- */ 197 198 /* Starts a background task that performs fsync() against the specified 199 * file descriptor (the one of the AOF file) in another thread. */ 200 void aof_background_fsync(int fd) { 201 bioCreateBackgroundJob(REDIS_BIO_AOF_FSYNC,(void*)(long)fd,NULL,NULL); 202 } 203 204 /* Called when the user switches from "appendonly yes" to "appendonly no" 205 * at runtime using the CONFIG command. */ 206 void stopAppendOnly(void) { 207 redisAssert(server.aof_state != REDIS_AOF_OFF); 208 flushAppendOnlyFile(1); 209 aof_fsync(server.aof_fd); 210 close(server.aof_fd); 211 212 server.aof_fd = -1; 213 server.aof_selected_db = -1; 214 server.aof_state = REDIS_AOF_OFF; 215 /* rewrite operation in progress? kill it, wait child exit */ 216 if (server.aof_child_pid != -1) { 217 int statloc; 218 219 redisLog(REDIS_NOTICE,"Killing running AOF rewrite child: %ld", 220 (long) server.aof_child_pid); 221 if (kill(server.aof_child_pid,SIGUSR1) != -1) 222 wait3(&statloc,0,NULL); 223 /* reset the buffer accumulating changes while the child saves */ 224 aofRewriteBufferReset(); 225 aofRemoveTempFile(server.aof_child_pid); 226 server.aof_child_pid = -1; 227 server.aof_rewrite_time_start = -1; 228 /* close pipes used for IPC between the two processes. */ 229 aofClosePipes(); 230 } 231 } 232 233 /* Called when the user switches from "appendonly no" to "appendonly yes" 234 * at runtime using the CONFIG command. */ 235 int startAppendOnly(void) { 236 server.aof_last_fsync = server.unixtime; 237 server.aof_fd = open(server.aof_filename,O_WRONLY|O_APPEND|O_CREAT,0644); 238 redisAssert(server.aof_state == REDIS_AOF_OFF); 239 if (server.aof_fd == -1) { 240 redisLog(REDIS_WARNING,"Redis needs to enable the AOF but can't open the append only file: %s",strerror(errno)); 241 return REDIS_ERR; 242 } 243 if (rewriteAppendOnlyFileBackground() == REDIS_ERR) { 244 close(server.aof_fd); 245 redisLog(REDIS_WARNING,"Redis needs to enable the AOF but can't trigger a background AOF rewrite operation. Check the above logs for more info about the error."); 246 return REDIS_ERR; 247 } 248 /* We correctly switched on AOF, now wait for the rerwite to be complete 249 * in order to append data on disk. */ 250 server.aof_state = REDIS_AOF_WAIT_REWRITE; 251 return REDIS_OK; 252 } 253 254 /* Write the append only file buffer on disk. 255 * 256 * Since we are required to write the AOF before replying to the client, 257 * and the only way the client socket can get a write is entering when the 258 * the event loop, we accumulate all the AOF writes in a memory 259 * buffer and write it on disk using this function just before entering 260 * the event loop again. 261 * 262 * About the 'force' argument: 263 * 264 * When the fsync policy is set to 'everysec' we may delay the flush if there 265 * is still an fsync() going on in the background thread, since for instance 266 * on Linux write(2) will be blocked by the background fsync anyway. 267 * When this happens we remember that there is some aof buffer to be 268 * flushed ASAP, and will try to do that in the serverCron() function. 269 * 270 * However if force is set to 1 we'll write regardless of the background 271 * fsync. */ 272 #define AOF_WRITE_LOG_ERROR_RATE 30 /* Seconds between errors logging. */ 273 void flushAppendOnlyFile(int force) { 274 ssize_t nwritten; 275 int sync_in_progress = 0; 276 mstime_t latency; 277 278 if (sdslen(server.aof_buf) == 0) return; 279 280 if (server.aof_fsync == AOF_FSYNC_EVERYSEC) 281 sync_in_progress = bioPendingJobsOfType(REDIS_BIO_AOF_FSYNC) != 0; 282 283 if (server.aof_fsync == AOF_FSYNC_EVERYSEC && !force) { 284 /* With this append fsync policy we do background fsyncing. 285 * If the fsync is still in progress we can try to delay 286 * the write for a couple of seconds. */ 287 if (sync_in_progress) { 288 if (server.aof_flush_postponed_start == 0) { 289 /* No previous write postponinig, remember that we are 290 * postponing the flush and return. */ 291 server.aof_flush_postponed_start = server.unixtime; 292 return; 293 } else if (server.unixtime - server.aof_flush_postponed_start < 2) { 294 /* We were already waiting for fsync to finish, but for less 295 * than two seconds this is still ok. Postpone again. */ 296 return; 297 } 298 /* Otherwise fall trough, and go write since we can't wait 299 * over two seconds. */ 300 server.aof_delayed_fsync++; 301 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."); 302 } 303 } 304 /* We want to perform a single write. This should be guaranteed atomic 305 * at least if the filesystem we are writing is a real physical one. 306 * While this will save us against the server being killed I don't think 307 * there is much to do about the whole server stopping for power problems 308 * or alike */ 309 310 latencyStartMonitor(latency); 311 nwritten = write(server.aof_fd,server.aof_buf,sdslen(server.aof_buf)); 312 latencyEndMonitor(latency); 313 /* We want to capture different events for delayed writes: 314 * when the delay happens with a pending fsync, or with a saving child 315 * active, and when the above two conditions are missing. 316 * We also use an additional event name to save all samples which is 317 * useful for graphing / monitoring purposes. */ 318 if (sync_in_progress) { 319 latencyAddSampleIfNeeded("aof-write-pending-fsync",latency); 320 } else if (server.aof_child_pid != -1 || server.rdb_child_pid != -1) { 321 latencyAddSampleIfNeeded("aof-write-active-child",latency); 322 } else { 323 latencyAddSampleIfNeeded("aof-write-alone",latency); 324 } 325 latencyAddSampleIfNeeded("aof-write",latency); 326 327 /* We performed the write so reset the postponed flush sentinel to zero. */ 328 server.aof_flush_postponed_start = 0; 329 330 if (nwritten != (signed)sdslen(server.aof_buf)) { 331 static time_t last_write_error_log = 0; 332 int can_log = 0; 333 334 /* Limit logging rate to 1 line per AOF_WRITE_LOG_ERROR_RATE seconds. */ 335 if ((server.unixtime - last_write_error_log) > AOF_WRITE_LOG_ERROR_RATE) { 336 can_log = 1; 337 last_write_error_log = server.unixtime; 338 } 339 340 /* Lof the AOF write error and record the error code. */ 341 if (nwritten == -1) { 342 if (can_log) { 343 redisLog(REDIS_WARNING,"Error writing to the AOF file: %s", 344 strerror(errno)); 345 server.aof_last_write_errno = errno; 346 } 347 } else { 348 if (can_log) { 349 redisLog(REDIS_WARNING,"Short write while writing to " 350 "the AOF file: (nwritten=%lld, " 351 "expected=%lld)", 352 (long long)nwritten, 353 (long long)sdslen(server.aof_buf)); 354 } 355 356 if (ftruncate(server.aof_fd, server.aof_current_size) == -1) { 357 if (can_log) { 358 redisLog(REDIS_WARNING, "Could not remove short write " 359 "from the append-only file. Redis may refuse " 360 "to load the AOF the next time it starts. " 361 "ftruncate: %s", strerror(errno)); 362 } 363 } else { 364 /* If the ftrunacate() succeeded we can set nwritten to 365 * -1 since there is no longer partial data into the AOF. */ 366 nwritten = -1; 367 } 368 server.aof_last_write_errno = ENOSPC; 369 } 370 371 /* Handle the AOF write error. */ 372 if (server.aof_fsync == AOF_FSYNC_ALWAYS) { 373 /* We can't recover when the fsync policy is ALWAYS since the 374 * reply for the client is already in the output buffers, and we 375 * have the contract with the user that on acknowledged write data 376 * is synched on disk. */ 377 redisLog(REDIS_WARNING,"Can't recover from AOF write error when the AOF fsync policy is 'always'. Exiting..."); 378 exit(1); 379 } else { 380 /* Recover from failed write leaving data into the buffer. However 381 * set an error to stop accepting writes as long as the error 382 * condition is not cleared. */ 383 server.aof_last_write_status = REDIS_ERR; 384 385 /* Trim the sds buffer if there was a partial write, and there 386 * was no way to undo it with ftruncate(2). */ 387 if (nwritten > 0) { 388 server.aof_current_size += nwritten; 389 sdsrange(server.aof_buf,nwritten,-1); 390 } 391 return; /* We'll try again on the next call... */ 392 } 393 } else { 394 /* Successful write(2). If AOF was in error state, restore the 395 * OK state and log the event. */ 396 if (server.aof_last_write_status == REDIS_ERR) { 397 redisLog(REDIS_WARNING, 398 "AOF write error looks solved, Redis can write again."); 399 server.aof_last_write_status = REDIS_OK; 400 } 401 } 402 server.aof_current_size += nwritten; 403 404 /* Re-use AOF buffer when it is small enough. The maximum comes from the 405 * arena size of 4k minus some overhead (but is otherwise arbitrary). */ 406 if ((sdslen(server.aof_buf)+sdsavail(server.aof_buf)) < 4000) { 407 sdsclear(server.aof_buf); 408 } else { 409 sdsfree(server.aof_buf); 410 server.aof_buf = sdsempty(); 411 } 412 413 /* Don't fsync if no-appendfsync-on-rewrite is set to yes and there are 414 * children doing I/O in the background. */ 415 if (server.aof_no_fsync_on_rewrite && 416 (server.aof_child_pid != -1 || server.rdb_child_pid != -1)) 417 return; 418 419 /* Perform the fsync if needed. */ 420 if (server.aof_fsync == AOF_FSYNC_ALWAYS) { 421 /* aof_fsync is defined as fdatasync() for Linux in order to avoid 422 * flushing metadata. */ 423 latencyStartMonitor(latency); 424 aof_fsync(server.aof_fd); /* Let's try to get this data on the disk */ 425 latencyEndMonitor(latency); 426 latencyAddSampleIfNeeded("aof-fsync-always",latency); 427 server.aof_last_fsync = server.unixtime; 428 } else if ((server.aof_fsync == AOF_FSYNC_EVERYSEC && 429 server.unixtime > server.aof_last_fsync)) { 430 if (!sync_in_progress) aof_background_fsync(server.aof_fd); 431 server.aof_last_fsync = server.unixtime; 432 } 433 } 434 435 sds catAppendOnlyGenericCommand(sds dst, int argc, robj **argv) { 436 char buf[32]; 437 int len, j; 438 robj *o; 439 440 buf[0] = '*'; 441 len = 1+ll2string(buf+1,sizeof(buf)-1,argc); 442 buf[len++] = '\r'; 443 buf[len++] = '\n'; 444 dst = sdscatlen(dst,buf,len); 445 446 for (j = 0; j < argc; j++) { 447 o = getDecodedObject(argv[j]); 448 buf[0] = '$'; 449 len = 1+ll2string(buf+1,sizeof(buf)-1,sdslen(o->ptr)); 450 buf[len++] = '\r'; 451 buf[len++] = '\n'; 452 dst = sdscatlen(dst,buf,len); 453 dst = sdscatlen(dst,o->ptr,sdslen(o->ptr)); 454 dst = sdscatlen(dst,"\r\n",2); 455 decrRefCount(o); 456 } 457 return dst; 458 } 459 460 /* Create the sds representation of an PEXPIREAT command, using 461 * 'seconds' as time to live and 'cmd' to understand what command 462 * we are translating into a PEXPIREAT. 463 * 464 * This command is used in order to translate EXPIRE and PEXPIRE commands 465 * into PEXPIREAT command so that we retain precision in the append only 466 * file, and the time is always absolute and not relative. */ 467 sds catAppendOnlyExpireAtCommand(sds buf, struct redisCommand *cmd, robj *key, robj *seconds) { 468 long long when; 469 robj *argv[3]; 470 471 /* Make sure we can use strtol */ 472 seconds = getDecodedObject(seconds); 473 when = strtoll(seconds->ptr,NULL,10); 474 /* Convert argument into milliseconds for EXPIRE, SETEX, EXPIREAT */ 475 if (cmd->proc == expireCommand || cmd->proc == setexCommand || 476 cmd->proc == expireatCommand) 477 { 478 when *= 1000; 479 } 480 /* Convert into absolute time for EXPIRE, PEXPIRE, SETEX, PSETEX */ 481 if (cmd->proc == expireCommand || cmd->proc == pexpireCommand || 482 cmd->proc == setexCommand || cmd->proc == psetexCommand) 483 { 484 when += mstime(); 485 } 486 decrRefCount(seconds); 487 488 argv[0] = createStringObject("PEXPIREAT",9); 489 argv[1] = key; 490 argv[2] = createStringObjectFromLongLong(when); 491 buf = catAppendOnlyGenericCommand(buf, 3, argv); 492 decrRefCount(argv[0]); 493 decrRefCount(argv[2]); 494 return buf; 495 } 496 497 void feedAppendOnlyFile(struct redisCommand *cmd, int dictid, robj **argv, int argc) { 498 sds buf = sdsempty(); 499 robj *tmpargv[3]; 500 501 /* The DB this command was targeting is not the same as the last command 502 * we appendend. To issue a SELECT command is needed. */ 503 if (dictid != server.aof_selected_db) { 504 char seldb[64]; 505 506 snprintf(seldb,sizeof(seldb),"%d",dictid); 507 buf = sdscatprintf(buf,"*2\r\n$6\r\nSELECT\r\n$%lu\r\n%s\r\n", 508 (unsigned long)strlen(seldb),seldb); 509 server.aof_selected_db = dictid; 510 } 511 512 if (cmd->proc == expireCommand || cmd->proc == pexpireCommand || 513 cmd->proc == expireatCommand) { 514 /* Translate EXPIRE/PEXPIRE/EXPIREAT into PEXPIREAT */ 515 buf = catAppendOnlyExpireAtCommand(buf,cmd,argv[1],argv[2]); 516 } else if (cmd->proc == setexCommand || cmd->proc == psetexCommand) { 517 /* Translate SETEX/PSETEX to SET and PEXPIREAT */ 518 tmpargv[0] = createStringObject("SET",3); 519 tmpargv[1] = argv[1]; 520 tmpargv[2] = argv[3]; 521 buf = catAppendOnlyGenericCommand(buf,3,tmpargv); 522 decrRefCount(tmpargv[0]); 523 buf = catAppendOnlyExpireAtCommand(buf,cmd,argv[1],argv[2]); 524 } else { 525 /* All the other commands don't need translation or need the 526 * same translation already operated in the command vector 527 * for the replication itself. */ 528 buf = catAppendOnlyGenericCommand(buf,argc,argv); 529 } 530 531 /* Append to the AOF buffer. This will be flushed on disk just before 532 * of re-entering the event loop, so before the client will get a 533 * positive reply about the operation performed. */ 534 if (server.aof_state == REDIS_AOF_ON) 535 server.aof_buf = sdscatlen(server.aof_buf,buf,sdslen(buf)); 536 537 /* If a background append only file rewriting is in progress we want to 538 * accumulate the differences between the child DB and the current one 539 * in a buffer, so that when the child process will do its work we 540 * can append the differences to the new append only file. */ 541 if (server.aof_child_pid != -1) 542 aofRewriteBufferAppend((unsigned char*)buf,sdslen(buf)); 543 544 sdsfree(buf); 545 } 546 547 /* ---------------------------------------------------------------------------- 548 * AOF loading 549 * ------------------------------------------------------------------------- */ 550 551 /* In Redis commands are always executed in the context of a client, so in 552 * order to load the append only file we need to create a fake client. */ 553 struct redisClient *createFakeClient(void) { 554 struct redisClient *c = zmalloc(sizeof(*c)); 555 556 selectDb(c,0); 557 c->fd = -1; 558 c->name = NULL; 559 c->querybuf = sdsempty(); 560 c->querybuf_peak = 0; 561 c->argc = 0; 562 c->argv = NULL; 563 c->bufpos = 0; 564 c->flags = 0; 565 c->btype = REDIS_BLOCKED_NONE; 566 /* We set the fake client as a slave waiting for the synchronization 567 * so that Redis will not try to send replies to this client. */ 568 c->replstate = REDIS_REPL_WAIT_BGSAVE_START; 569 c->reply = listCreate(); 570 c->reply_bytes = 0; 571 c->obuf_soft_limit_reached_time = 0; 572 c->watched_keys = listCreate(); 573 c->peerid = NULL; 574 listSetFreeMethod(c->reply,decrRefCountVoid); 575 listSetDupMethod(c->reply,dupClientReplyValue); 576 initClientMultiState(c); 577 return c; 578 } 579 580 void freeFakeClientArgv(struct redisClient *c) { 581 int j; 582 583 for (j = 0; j < c->argc; j++) 584 decrRefCount(c->argv[j]); 585 zfree(c->argv); 586 } 587 588 void freeFakeClient(struct redisClient *c) { 589 sdsfree(c->querybuf); 590 listRelease(c->reply); 591 listRelease(c->watched_keys); 592 freeClientMultiState(c); 593 zfree(c); 594 } 595 596 /* Replay the append log file. On error REDIS_OK is returned. On non fatal 597 * error (the append only file is zero-length) REDIS_ERR is returned. On 598 * fatal error an error message is logged and the program exists. */ 599 int loadAppendOnlyFile(char *filename) { 600 struct redisClient *fakeClient; 601 FILE *fp = fopen(filename,"r"); 602 struct redis_stat sb; 603 int old_aof_state = server.aof_state; 604 long loops = 0; 605 606 if (fp && redis_fstat(fileno(fp),&sb) != -1 && sb.st_size == 0) { 607 server.aof_current_size = 0; 608 fclose(fp); 609 return REDIS_ERR; 610 } 611 612 if (fp == NULL) { 613 redisLog(REDIS_WARNING,"Fatal error: can't open the append log file for reading: %s",strerror(errno)); 614 exit(1); 615 } 616 617 /* Temporarily disable AOF, to prevent EXEC from feeding a MULTI 618 * to the same file we're about to read. */ 619 server.aof_state = REDIS_AOF_OFF; 620 621 fakeClient = createFakeClient(); 622 startLoading(fp); 623 624 while(1) { 625 int argc, j; 626 unsigned long len; 627 robj **argv; 628 char buf[128]; 629 sds argsds; 630 struct redisCommand *cmd; 631 632 /* Serve the clients from time to time */ 633 if (!(loops++ % 1000)) { 634 loadingProgress(ftello(fp)); 635 processEventsWhileBlocked(); 636 } 637 638 if (fgets(buf,sizeof(buf),fp) == NULL) { 639 if (feof(fp)) 640 break; 641 else 642 goto readerr; 643 } 644 if (buf[0] != '*') goto fmterr; 645 if (buf[1] == '\0') goto readerr; 646 argc = atoi(buf+1); 647 if (argc < 1) goto fmterr; 648 649 argv = zmalloc(sizeof(robj*)*argc); 650 fakeClient->argc = argc; 651 fakeClient->argv = argv; 652 653 for (j = 0; j < argc; j++) { 654 if (fgets(buf,sizeof(buf),fp) == NULL) { 655 fakeClient->argc = j; /* Free up to j-1. */ 656 freeFakeClientArgv(fakeClient); 657 goto readerr; 658 } 659 if (buf[0] != '$') goto fmterr; 660 len = strtol(buf+1,NULL,10); 661 argsds = sdsnewlen(NULL,len); 662 if (len && fread(argsds,len,1,fp) == 0) { 663 sdsfree(argsds); 664 fakeClient->argc = j; /* Free up to j-1. */ 665 freeFakeClientArgv(fakeClient); 666 goto readerr; 667 } 668 argv[j] = createObject(REDIS_STRING,argsds); 669 if (fread(buf,2,1,fp) == 0) { 670 fakeClient->argc = j+1; /* Free up to j. */ 671 freeFakeClientArgv(fakeClient); 672 goto readerr; /* discard CRLF */ 673 } 674 } 675 676 /* Command lookup */ 677 cmd = lookupCommand(argv[0]->ptr); 678 if (!cmd) { 679 redisLog(REDIS_WARNING,"Unknown command '%s' reading the append only file", (char*)argv[0]->ptr); 680 exit(1); 681 } 682 683 /* Run the command in the context of a fake client */ 684 cmd->proc(fakeClient); 685 686 /* The fake client should not have a reply */ 687 redisAssert(fakeClient->bufpos == 0 && listLength(fakeClient->reply) == 0); 688 /* The fake client should never get blocked */ 689 redisAssert((fakeClient->flags & REDIS_BLOCKED) == 0); 690 691 /* Clean up. Command code may have changed argv/argc so we use the 692 * argv/argc of the client instead of the local variables. */ 693 freeFakeClientArgv(fakeClient); 694 } 695 696 /* This point can only be reached when EOF is reached without errors. 697 * If the client is in the middle of a MULTI/EXEC, log error and quit. */ 698 if (fakeClient->flags & REDIS_MULTI) goto uxeof; 699 700 loaded_ok: /* DB loaded, cleanup and return REDIS_OK to the caller. */ 701 fclose(fp); 702 freeFakeClient(fakeClient); 703 server.aof_state = old_aof_state; 704 stopLoading(); 705 aofUpdateCurrentSize(); 706 server.aof_rewrite_base_size = server.aof_current_size; 707 return REDIS_OK; 708 709 readerr: /* Read error. If feof(fp) is true, fall through to unexpected EOF. */ 710 if (!feof(fp)) { 711 redisLog(REDIS_WARNING,"Unrecoverable error reading the append only file: %s", strerror(errno)); 712 exit(1); 713 } 714 715 uxeof: /* Unexpected AOF end of file. */ 716 if (server.aof_load_truncated) { 717 redisLog(REDIS_WARNING,"!!! Warning: short read while loading the AOF file !!!"); 718 redisLog(REDIS_WARNING, 719 "AOF loaded anyway because aof-load-truncated is enabled"); 720 goto loaded_ok; 721 } 722 redisLog(REDIS_WARNING,"Unexpected end of file reading the append only file. You can: 1) Make a backup of your AOF file, then use ./redis-check-aof --fix <filename>. 2) Alternatively you can set the 'aof-load-truncated' configuration option to yes and restart the server."); 723 exit(1); 724 725 fmterr: /* Format error. */ 726 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>"); 727 exit(1); 728 } 729 730 /* ---------------------------------------------------------------------------- 731 * AOF rewrite 732 * ------------------------------------------------------------------------- */ 733 734 /* Delegate writing an object to writing a bulk string or bulk long long. 735 * This is not placed in rio.c since that adds the redis.h dependency. */ 736 int rioWriteBulkObject(rio *r, robj *obj) { 737 /* Avoid using getDecodedObject to help copy-on-write (we are often 738 * in a child process when this function is called). */ 739 if (obj->encoding == REDIS_ENCODING_INT) { 740 return rioWriteBulkLongLong(r,(long)obj->ptr); 741 } else if (sdsEncodedObject(obj)) { 742 return rioWriteBulkString(r,obj->ptr,sdslen(obj->ptr)); 743 } else { 744 redisPanic("Unknown string encoding"); 745 } 746 } 747 748 /* Emit the commands needed to rebuild a list object. 749 * The function returns 0 on error, 1 on success. */ 750 int rewriteListObject(rio *r, robj *key, robj *o) { 751 long long count = 0, items = listTypeLength(o); 752 753 if (o->encoding == REDIS_ENCODING_ZIPLIST) { 754 unsigned char *zl = o->ptr; 755 unsigned char *p = ziplistIndex(zl,0); 756 unsigned char *vstr; 757 unsigned int vlen; 758 long long vlong; 759 760 while(ziplistGet(p,&vstr,&vlen,&vlong)) { 761 if (count == 0) { 762 int cmd_items = (items > REDIS_AOF_REWRITE_ITEMS_PER_CMD) ? 763 REDIS_AOF_REWRITE_ITEMS_PER_CMD : items; 764 765 if (rioWriteBulkCount(r,'*',2+cmd_items) == 0) return 0; 766 if (rioWriteBulkString(r,"RPUSH",5) == 0) return 0; 767 if (rioWriteBulkObject(r,key) == 0) return 0; 768 } 769 if (vstr) { 770 if (rioWriteBulkString(r,(char*)vstr,vlen) == 0) return 0; 771 } else { 772 if (rioWriteBulkLongLong(r,vlong) == 0) return 0; 773 } 774 p = ziplistNext(zl,p); 775 if (++count == REDIS_AOF_REWRITE_ITEMS_PER_CMD) count = 0; 776 items--; 777 } 778 } else if (o->encoding == REDIS_ENCODING_LINKEDLIST) { 779 list *list = o->ptr; 780 listNode *ln; 781 listIter li; 782 783 listRewind(list,&li); 784 while((ln = listNext(&li))) { 785 robj *eleobj = listNodeValue(ln); 786 787 if (count == 0) { 788 int cmd_items = (items > REDIS_AOF_REWRITE_ITEMS_PER_CMD) ? 789 REDIS_AOF_REWRITE_ITEMS_PER_CMD : items; 790 791 if (rioWriteBulkCount(r,'*',2+cmd_items) == 0) return 0; 792 if (rioWriteBulkString(r,"RPUSH",5) == 0) return 0; 793 if (rioWriteBulkObject(r,key) == 0) return 0; 794 } 795 if (rioWriteBulkObject(r,eleobj) == 0) return 0; 796 if (++count == REDIS_AOF_REWRITE_ITEMS_PER_CMD) count = 0; 797 items--; 798 } 799 } else { 800 redisPanic("Unknown list encoding"); 801 } 802 return 1; 803 } 804 805 /* Emit the commands needed to rebuild a set object. 806 * The function returns 0 on error, 1 on success. */ 807 int rewriteSetObject(rio *r, robj *key, robj *o) { 808 long long count = 0, items = setTypeSize(o); 809 810 if (o->encoding == REDIS_ENCODING_INTSET) { 811 int ii = 0; 812 int64_t llval; 813 814 while(intsetGet(o->ptr,ii++,&llval)) { 815 if (count == 0) { 816 int cmd_items = (items > REDIS_AOF_REWRITE_ITEMS_PER_CMD) ? 817 REDIS_AOF_REWRITE_ITEMS_PER_CMD : items; 818 819 if (rioWriteBulkCount(r,'*',2+cmd_items) == 0) return 0; 820 if (rioWriteBulkString(r,"SADD",4) == 0) return 0; 821 if (rioWriteBulkObject(r,key) == 0) return 0; 822 } 823 if (rioWriteBulkLongLong(r,llval) == 0) return 0; 824 if (++count == REDIS_AOF_REWRITE_ITEMS_PER_CMD) count = 0; 825 items--; 826 } 827 } else if (o->encoding == REDIS_ENCODING_HT) { 828 dictIterator *di = dictGetIterator(o->ptr); 829 dictEntry *de; 830 831 while((de = dictNext(di)) != NULL) { 832 robj *eleobj = dictGetKey(de); 833 if (count == 0) { 834 int cmd_items = (items > REDIS_AOF_REWRITE_ITEMS_PER_CMD) ? 835 REDIS_AOF_REWRITE_ITEMS_PER_CMD : items; 836 837 if (rioWriteBulkCount(r,'*',2+cmd_items) == 0) return 0; 838 if (rioWriteBulkString(r,"SADD",4) == 0) return 0; 839 if (rioWriteBulkObject(r,key) == 0) return 0; 840 } 841 if (rioWriteBulkObject(r,eleobj) == 0) return 0; 842 if (++count == REDIS_AOF_REWRITE_ITEMS_PER_CMD) count = 0; 843 items--; 844 } 845 dictReleaseIterator(di); 846 } else { 847 redisPanic("Unknown set encoding"); 848 } 849 return 1; 850 } 851 852 /* Emit the commands needed to rebuild a sorted set object. 853 * The function returns 0 on error, 1 on success. */ 854 int rewriteSortedSetObject(rio *r, robj *key, robj *o) { 855 long long count = 0, items = zsetLength(o); 856 857 if (o->encoding == REDIS_ENCODING_ZIPLIST) { 858 unsigned char *zl = o->ptr; 859 unsigned char *eptr, *sptr; 860 unsigned char *vstr; 861 unsigned int vlen; 862 long long vll; 863 double score; 864 865 eptr = ziplistIndex(zl,0); 866 redisAssert(eptr != NULL); 867 sptr = ziplistNext(zl,eptr); 868 redisAssert(sptr != NULL); 869 870 while (eptr != NULL) { 871 redisAssert(ziplistGet(eptr,&vstr,&vlen,&vll)); 872 score = zzlGetScore(sptr); 873 874 if (count == 0) { 875 int cmd_items = (items > REDIS_AOF_REWRITE_ITEMS_PER_CMD) ? 876 REDIS_AOF_REWRITE_ITEMS_PER_CMD : items; 877 878 if (rioWriteBulkCount(r,'*',2+cmd_items*2) == 0) return 0; 879 if (rioWriteBulkString(r,"ZADD",4) == 0) return 0; 880 if (rioWriteBulkObject(r,key) == 0) return 0; 881 } 882 if (rioWriteBulkDouble(r,score) == 0) return 0; 883 if (vstr != NULL) { 884 if (rioWriteBulkString(r,(char*)vstr,vlen) == 0) return 0; 885 } else { 886 if (rioWriteBulkLongLong(r,vll) == 0) return 0; 887 } 888 zzlNext(zl,&eptr,&sptr); 889 if (++count == REDIS_AOF_REWRITE_ITEMS_PER_CMD) count = 0; 890 items--; 891 } 892 } else if (o->encoding == REDIS_ENCODING_SKIPLIST) { 893 zset *zs = o->ptr; 894 dictIterator *di = dictGetIterator(zs->dict); 895 dictEntry *de; 896 897 while((de = dictNext(di)) != NULL) { 898 robj *eleobj = dictGetKey(de); 899 double *score = dictGetVal(de); 900 901 if (count == 0) { 902 int cmd_items = (items > REDIS_AOF_REWRITE_ITEMS_PER_CMD) ? 903 REDIS_AOF_REWRITE_ITEMS_PER_CMD : items; 904 905 if (rioWriteBulkCount(r,'*',2+cmd_items*2) == 0) return 0; 906 if (rioWriteBulkString(r,"ZADD",4) == 0) return 0; 907 if (rioWriteBulkObject(r,key) == 0) return 0; 908 } 909 if (rioWriteBulkDouble(r,*score) == 0) return 0; 910 if (rioWriteBulkObject(r,eleobj) == 0) return 0; 911 if (++count == REDIS_AOF_REWRITE_ITEMS_PER_CMD) count = 0; 912 items--; 913 } 914 dictReleaseIterator(di); 915 } else { 916 redisPanic("Unknown sorted zset encoding"); 917 } 918 return 1; 919 } 920 921 /* Write either the key or the value of the currently selected item of a hash. 922 * The 'hi' argument passes a valid Redis hash iterator. 923 * The 'what' filed specifies if to write a key or a value and can be 924 * either REDIS_HASH_KEY or REDIS_HASH_VALUE. 925 * 926 * The function returns 0 on error, non-zero on success. */ 927 static int rioWriteHashIteratorCursor(rio *r, hashTypeIterator *hi, int what) { 928 if (hi->encoding == REDIS_ENCODING_ZIPLIST) { 929 unsigned char *vstr = NULL; 930 unsigned int vlen = UINT_MAX; 931 long long vll = LLONG_MAX; 932 933 hashTypeCurrentFromZiplist(hi, what, &vstr, &vlen, &vll); 934 if (vstr) { 935 return rioWriteBulkString(r, (char*)vstr, vlen); 936 } else { 937 return rioWriteBulkLongLong(r, vll); 938 } 939 940 } else if (hi->encoding == REDIS_ENCODING_HT) { 941 robj *value; 942 943 hashTypeCurrentFromHashTable(hi, what, &value); 944 return rioWriteBulkObject(r, value); 945 } 946 947 redisPanic("Unknown hash encoding"); 948 return 0; 949 } 950 951 /* Emit the commands needed to rebuild a hash object. 952 * The function returns 0 on error, 1 on success. */ 953 int rewriteHashObject(rio *r, robj *key, robj *o) { 954 hashTypeIterator *hi; 955 long long count = 0, items = hashTypeLength(o); 956 957 hi = hashTypeInitIterator(o); 958 while (hashTypeNext(hi) != REDIS_ERR) { 959 if (count == 0) { 960 int cmd_items = (items > REDIS_AOF_REWRITE_ITEMS_PER_CMD) ? 961 REDIS_AOF_REWRITE_ITEMS_PER_CMD : items; 962 963 if (rioWriteBulkCount(r,'*',2+cmd_items*2) == 0) return 0; 964 if (rioWriteBulkString(r,"HMSET",5) == 0) return 0; 965 if (rioWriteBulkObject(r,key) == 0) return 0; 966 } 967 968 if (rioWriteHashIteratorCursor(r, hi, REDIS_HASH_KEY) == 0) return 0; 969 if (rioWriteHashIteratorCursor(r, hi, REDIS_HASH_VALUE) == 0) return 0; 970 if (++count == REDIS_AOF_REWRITE_ITEMS_PER_CMD) count = 0; 971 items--; 972 } 973 974 hashTypeReleaseIterator(hi); 975 976 return 1; 977 } 978 979 /* This function is called by the child rewriting the AOF file to read 980 * the difference accumulated from the parent into a buffer, that is 981 * concatenated at the end of the rewrite. */ 982 ssize_t aofReadDiffFromParent(void) { 983 char buf[65536]; /* Default pipe buffer size on most Linux sytems. */ 984 ssize_t nread, total = 0; 985 986 while ((nread = 987 read(server.aof_pipe_read_data_from_parent,buf,sizeof(buf))) > 0) { 988 server.aof_child_diff = sdscatlen(server.aof_child_diff,buf,nread); 989 total += nread; 990 } 991 return total; 992 } 993 994 /* Write a sequence of commands able to fully rebuild the dataset into 995 * "filename". Used both by REWRITEAOF and BGREWRITEAOF. 996 * 997 * In order to minimize the number of commands needed in the rewritten 998 * log Redis uses variadic commands when possible, such as RPUSH, SADD 999 * and ZADD. However at max REDIS_AOF_REWRITE_ITEMS_PER_CMD items per time 1000 * are inserted using a single command. */ 1001 int rewriteAppendOnlyFile(char *filename) { 1002 dictIterator *di = NULL; 1003 dictEntry *de; 1004 rio aof; 1005 FILE *fp; 1006 char tmpfile[256]; 1007 int j; 1008 long long now = mstime(); 1009 char byte; 1010 size_t processed = 0; 1011 1012 /* Note that we have to use a different temp name here compared to the 1013 * one used by rewriteAppendOnlyFileBackground() function. */ 1014 snprintf(tmpfile,256,"temp-rewriteaof-%d.aof", (int) getpid()); 1015 fp = fopen(tmpfile,"w"); 1016 if (!fp) { 1017 redisLog(REDIS_WARNING, "Opening the temp file for AOF rewrite in rewriteAppendOnlyFile(): %s", strerror(errno)); 1018 return REDIS_ERR; 1019 } 1020 1021 server.aof_child_diff = sdsempty(); 1022 rioInitWithFile(&aof,fp); 1023 if (server.aof_rewrite_incremental_fsync) 1024 rioSetAutoSync(&aof,REDIS_AOF_AUTOSYNC_BYTES); 1025 for (j = 0; j < server.dbnum; j++) { 1026 char selectcmd[] = "*2\r\n$6\r\nSELECT\r\n"; 1027 redisDb *db = server.db+j; 1028 dict *d = db->dict; 1029 if (dictSize(d) == 0) continue; 1030 di = dictGetSafeIterator(d); 1031 if (!di) { 1032 fclose(fp); 1033 return REDIS_ERR; 1034 } 1035 1036 /* SELECT the new DB */ 1037 if (rioWrite(&aof,selectcmd,sizeof(selectcmd)-1) == 0) goto werr; 1038 if (rioWriteBulkLongLong(&aof,j) == 0) goto werr; 1039 1040 /* Iterate this DB writing every entry */ 1041 while((de = dictNext(di)) != NULL) { 1042 sds keystr; 1043 robj key, *o; 1044 long long expiretime; 1045 1046 keystr = dictGetKey(de); 1047 o = dictGetVal(de); 1048 initStaticStringObject(key,keystr); 1049 1050 expiretime = getExpire(db,&key); 1051 1052 /* If this key is already expired skip it */ 1053 if (expiretime != -1 && expiretime < now) continue; 1054 1055 /* Save the key and associated value */ 1056 if (o->type == REDIS_STRING) { 1057 /* Emit a SET command */ 1058 char cmd[]="*3\r\n$3\r\nSET\r\n"; 1059 if (rioWrite(&aof,cmd,sizeof(cmd)-1) == 0) goto werr; 1060 /* Key and value */ 1061 if (rioWriteBulkObject(&aof,&key) == 0) goto werr; 1062 if (rioWriteBulkObject(&aof,o) == 0) goto werr; 1063 } else if (o->type == REDIS_LIST) { 1064 if (rewriteListObject(&aof,&key,o) == 0) goto werr; 1065 } else if (o->type == REDIS_SET) { 1066 if (rewriteSetObject(&aof,&key,o) == 0) goto werr; 1067 } else if (o->type == REDIS_ZSET) { 1068 if (rewriteSortedSetObject(&aof,&key,o) == 0) goto werr; 1069 } else if (o->type == REDIS_HASH) { 1070 if (rewriteHashObject(&aof,&key,o) == 0) goto werr; 1071 } else { 1072 redisPanic("Unknown object type"); 1073 } 1074 /* Save the expire time */ 1075 if (expiretime != -1) { 1076 char cmd[]="*3\r\n$9\r\nPEXPIREAT\r\n"; 1077 if (rioWrite(&aof,cmd,sizeof(cmd)-1) == 0) goto werr; 1078 if (rioWriteBulkObject(&aof,&key) == 0) goto werr; 1079 if (rioWriteBulkLongLong(&aof,expiretime) == 0) goto werr; 1080 } 1081 /* Read some diff from the parent process from time to time. */ 1082 if (aof.processed_bytes > processed+1024*10) { 1083 processed = aof.processed_bytes; 1084 aofReadDiffFromParent(); 1085 } 1086 } 1087 dictReleaseIterator(di); 1088 } 1089 1090 /* Do an initial slow fsync here while the parent is still sending 1091 * data, in order to make the next final fsync faster. */ 1092 if (fflush(fp) == EOF) goto werr; 1093 if (fsync(fileno(fp)) == -1) goto werr; 1094 1095 /* Read again a few times to get more data from the parent. 1096 * We can't read forever (the server may receive data from clients 1097 * fater than it is able to send data to the child), so we try to read 1098 * some more data in a loop as soon as there is a good chance more data 1099 * will come. If it looks like we are wasting time, we abort (this 1100 * happens after 20 ms without new data). */ 1101 int nodata = 0; 1102 mstime_t start = mstime(); 1103 while(mstime()-start < 1000 && nodata < 20) { 1104 if (aeWait(server.aof_pipe_read_data_from_parent, AE_READABLE, 1) <= 0) 1105 { 1106 nodata++; 1107 continue; 1108 } 1109 nodata = 0; /* Start counting from zero, we stop on N *contiguous* 1110 timeouts. */ 1111 aofReadDiffFromParent(); 1112 } 1113 1114 /* Ask the master to stop sending diffs. */ 1115 if (write(server.aof_pipe_write_ack_to_parent,"!",1) != 1) goto werr; 1116 if (anetNonBlock(NULL,server.aof_pipe_read_ack_from_parent) != ANET_OK) 1117 goto werr; 1118 /* We read the ACK from the server using a 10 seconds timeout. Normally 1119 * it should reply ASAP, but just in case we lose its reply, we are sure 1120 * the child will eventually get terminated. */ 1121 if (syncRead(server.aof_pipe_read_ack_from_parent,&byte,1,5000) != 1 || 1122 byte != '!') goto werr; 1123 redisLog(REDIS_NOTICE,"Parent agreed to stop sending diffs. Finalizing AOF..."); 1124 1125 /* Read the final diff if any. */ 1126 aofReadDiffFromParent(); 1127 1128 /* Write the received diff to the file. */ 1129 redisLog(REDIS_NOTICE, 1130 "Concatenating %.2f MB of AOF diff received from parent.", 1131 (double) sdslen(server.aof_child_diff) / (1024*1024)); 1132 if (rioWrite(&aof,server.aof_child_diff,sdslen(server.aof_child_diff)) == 0) 1133 goto werr; 1134 1135 /* Make sure data will not remain on the OS's output buffers */ 1136 if (fflush(fp) == EOF) goto werr; 1137 if (fsync(fileno(fp)) == -1) goto werr; 1138 if (fclose(fp) == EOF) goto werr; 1139 1140 /* Use RENAME to make sure the DB file is changed atomically only 1141 * if the generate DB file is ok. */ 1142 if (rename(tmpfile,filename) == -1) { 1143 redisLog(REDIS_WARNING,"Error moving temp append only file on the final destination: %s", strerror(errno)); 1144 unlink(tmpfile); 1145 return REDIS_ERR; 1146 } 1147 redisLog(REDIS_NOTICE,"SYNC append only file rewrite performed"); 1148 return REDIS_OK; 1149 1150 werr: 1151 fclose(fp); 1152 unlink(tmpfile); 1153 redisLog(REDIS_WARNING,"Write error writing append only file on disk: %s", strerror(errno)); 1154 if (di) dictReleaseIterator(di); 1155 return REDIS_ERR; 1156 } 1157 1158 /* ---------------------------------------------------------------------------- 1159 * AOF rewrite pipes for IPC 1160 * -------------------------------------------------------------------------- */ 1161 1162 /* This event handler is called when the AOF rewriting child sends us a 1163 * single '!' char to signal we should stop sending buffer diffs. The 1164 * parent sends a '!' as well to acknowledge. */ 1165 void aofChildPipeReadable(aeEventLoop *el, int fd, void *privdata, int mask) { 1166 char byte; 1167 REDIS_NOTUSED(el); 1168 REDIS_NOTUSED(privdata); 1169 REDIS_NOTUSED(mask); 1170 1171 if (read(fd,&byte,1) == 1 && byte == '!') { 1172 redisLog(REDIS_NOTICE,"AOF rewrite child asks to stop sending diffs."); 1173 server.aof_stop_sending_diff = 1; 1174 if (write(server.aof_pipe_write_ack_to_child,"!",1) != 1) { 1175 /* If we can't send the ack, inform the user, but don't try again 1176 * since in the other side the children will use a timeout if the 1177 * kernel can't buffer our write, or, the children was 1178 * terminated. */ 1179 redisLog(REDIS_WARNING,"Can't send ACK to AOF child: %s", 1180 strerror(errno)); 1181 } 1182 } 1183 /* Remove the handler since this can be called only one time during a 1184 * rewrite. */ 1185 aeDeleteFileEvent(server.el,server.aof_pipe_read_ack_from_child,AE_READABLE); 1186 } 1187 1188 /* Create the pipes used for parent - child process IPC during rewrite. 1189 * We have a data pipe used to send AOF incremental diffs to the child, 1190 * and two other pipes used by the children to signal it finished with 1191 * the rewrite so no more data should be written, and another for the 1192 * parent to acknowledge it understood this new condition. */ 1193 int aofCreatePipes(void) { 1194 int fds[6] = {-1, -1, -1, -1, -1, -1}; 1195 int j; 1196 1197 if (pipe(fds) == -1) goto error; /* parent -> children data. */ 1198 if (pipe(fds+2) == -1) goto error; /* children -> parent ack. */ 1199 if (pipe(fds+4) == -1) goto error; /* children -> parent ack. */ 1200 /* Parent -> children data is non blocking. */ 1201 if (anetNonBlock(NULL,fds[0]) != ANET_OK) goto error; 1202 if (anetNonBlock(NULL,fds[1]) != ANET_OK) goto error; 1203 if (aeCreateFileEvent(server.el, fds[2], AE_READABLE, aofChildPipeReadable, NULL) == AE_ERR) goto error; 1204 1205 server.aof_pipe_write_data_to_child = fds[1]; 1206 server.aof_pipe_read_data_from_parent = fds[0]; 1207 server.aof_pipe_write_ack_to_parent = fds[3]; 1208 server.aof_pipe_read_ack_from_child = fds[2]; 1209 server.aof_pipe_write_ack_to_child = fds[5]; 1210 server.aof_pipe_read_ack_from_parent = fds[4]; 1211 server.aof_stop_sending_diff = 0; 1212 return REDIS_OK; 1213 1214 error: 1215 redisLog(REDIS_WARNING,"Error opening /setting AOF rewrite IPC pipes: %s", 1216 strerror(errno)); 1217 for (j = 0; j < 6; j++) if(fds[j] != -1) close(fds[j]); 1218 return REDIS_ERR; 1219 } 1220 1221 void aofClosePipes(void) { 1222 aeDeleteFileEvent(server.el,server.aof_pipe_read_ack_from_child,AE_READABLE); 1223 aeDeleteFileEvent(server.el,server.aof_pipe_write_data_to_child,AE_WRITABLE); 1224 close(server.aof_pipe_write_data_to_child); 1225 close(server.aof_pipe_read_data_from_parent); 1226 close(server.aof_pipe_write_ack_to_parent); 1227 close(server.aof_pipe_read_ack_from_child); 1228 close(server.aof_pipe_write_ack_to_child); 1229 close(server.aof_pipe_read_ack_from_parent); 1230 } 1231 1232 /* ---------------------------------------------------------------------------- 1233 * AOF backgorund rewrite 1234 * ------------------------------------------------------------------------- */ 1235 1236 /* This is how rewriting of the append only file in background works: 1237 * 1238 * 1) The user calls BGREWRITEAOF 1239 * 2) Redis calls this function, that forks(): 1240 * 2a) the child rewrite the append only file in a temp file. 1241 * 2b) the parent accumulates differences in server.aof_rewrite_buf. 1242 * 3) When the child finished '2a' exists. 1243 * 4) The parent will trap the exit code, if it's OK, will append the 1244 * data accumulated into server.aof_rewrite_buf into the temp file, and 1245 * finally will rename(2) the temp file in the actual file name. 1246 * The the new file is reopened as the new append only file. Profit! 1247 */ 1248 int rewriteAppendOnlyFileBackground(void) { 1249 pid_t childpid; 1250 long long start; 1251 1252 if (server.aof_child_pid != -1) return REDIS_ERR; 1253 if (aofCreatePipes() != REDIS_OK) return REDIS_ERR; 1254 start = ustime(); 1255 if ((childpid = fork()) == 0) { 1256 char tmpfile[256]; 1257 1258 /* Child */ 1259 closeListeningSockets(0); 1260 redisSetProcTitle("redis-aof-rewrite"); 1261 snprintf(tmpfile,256,"temp-rewriteaof-bg-%d.aof", (int) getpid()); 1262 if (rewriteAppendOnlyFile(tmpfile) == REDIS_OK) { 1263 size_t private_dirty = zmalloc_get_private_dirty(); 1264 1265 if (private_dirty) { 1266 redisLog(REDIS_NOTICE, 1267 "AOF rewrite: %zu MB of memory used by copy-on-write", 1268 private_dirty/(1024*1024)); 1269 } 1270 exitFromChild(0); 1271 } else { 1272 exitFromChild(1); 1273 } 1274 } else { 1275 /* Parent */ 1276 server.stat_fork_time = ustime()-start; 1277 server.stat_fork_rate = (double) zmalloc_used_memory() * 1000000 / server.stat_fork_time / (1024*1024*1024); /* GB per second. */ 1278 latencyAddSampleIfNeeded("fork",server.stat_fork_time/1000); 1279 if (childpid == -1) { 1280 redisLog(REDIS_WARNING, 1281 "Can't rewrite append only file in background: fork: %s", 1282 strerror(errno)); 1283 return REDIS_ERR; 1284 } 1285 redisLog(REDIS_NOTICE, 1286 "Background append only file rewriting started by pid %d",childpid); 1287 server.aof_rewrite_scheduled = 0; 1288 server.aof_rewrite_time_start = time(NULL); 1289 server.aof_child_pid = childpid; 1290 updateDictResizePolicy(); 1291 /* We set appendseldb to -1 in order to force the next call to the 1292 * feedAppendOnlyFile() to issue a SELECT command, so the differences 1293 * accumulated by the parent into server.aof_rewrite_buf will start 1294 * with a SELECT statement and it will be safe to merge. */ 1295 server.aof_selected_db = -1; 1296 replicationScriptCacheFlush(); 1297 return REDIS_OK; 1298 } 1299 return REDIS_OK; /* unreached */ 1300 } 1301 1302 void bgrewriteaofCommand(redisClient *c) { 1303 if (server.aof_child_pid != -1) { 1304 addReplyError(c,"Background append only file rewriting already in progress"); 1305 } else if (server.rdb_child_pid != -1) { 1306 server.aof_rewrite_scheduled = 1; 1307 addReplyStatus(c,"Background append only file rewriting scheduled"); 1308 } else if (rewriteAppendOnlyFileBackground() == REDIS_OK) { 1309 addReplyStatus(c,"Background append only file rewriting started"); 1310 } else { 1311 addReply(c,shared.err); 1312 } 1313 } 1314 1315 void aofRemoveTempFile(pid_t childpid) { 1316 char tmpfile[256]; 1317 1318 snprintf(tmpfile,256,"temp-rewriteaof-bg-%d.aof", (int) childpid); 1319 unlink(tmpfile); 1320 } 1321 1322 /* Update the server.aof_current_size field explicitly using stat(2) 1323 * to check the size of the file. This is useful after a rewrite or after 1324 * a restart, normally the size is updated just adding the write length 1325 * to the current length, that is much faster. */ 1326 void aofUpdateCurrentSize(void) { 1327 struct redis_stat sb; 1328 mstime_t latency; 1329 1330 latencyStartMonitor(latency); 1331 if (redis_fstat(server.aof_fd,&sb) == -1) { 1332 redisLog(REDIS_WARNING,"Unable to obtain the AOF file length. stat: %s", 1333 strerror(errno)); 1334 } else { 1335 server.aof_current_size = sb.st_size; 1336 } 1337 latencyEndMonitor(latency); 1338 latencyAddSampleIfNeeded("aof-fstat",latency); 1339 } 1340 1341 /* A background append only file rewriting (BGREWRITEAOF) terminated its work. 1342 * Handle this. */ 1343 void backgroundRewriteDoneHandler(int exitcode, int bysignal) { 1344 if (!bysignal && exitcode == 0) { 1345 int newfd, oldfd; 1346 char tmpfile[256]; 1347 long long now = ustime(); 1348 mstime_t latency; 1349 1350 redisLog(REDIS_NOTICE, 1351 "Background AOF rewrite terminated with success"); 1352 1353 /* Flush the differences accumulated by the parent to the 1354 * rewritten AOF. */ 1355 latencyStartMonitor(latency); 1356 snprintf(tmpfile,256,"temp-rewriteaof-bg-%d.aof", 1357 (int)server.aof_child_pid); 1358 newfd = open(tmpfile,O_WRONLY|O_APPEND); 1359 if (newfd == -1) { 1360 redisLog(REDIS_WARNING, 1361 "Unable to open the temporary AOF produced by the child: %s", strerror(errno)); 1362 goto cleanup; 1363 } 1364 1365 if (aofRewriteBufferWrite(newfd) == -1) { 1366 redisLog(REDIS_WARNING, 1367 "Error trying to flush the parent diff to the rewritten AOF: %s", strerror(errno)); 1368 close(newfd); 1369 goto cleanup; 1370 } 1371 latencyEndMonitor(latency); 1372 latencyAddSampleIfNeeded("aof-rewrite-diff-write",latency); 1373 1374 redisLog(REDIS_NOTICE, 1375 "Redidual parent diff successfully flushed to the rewritten AOF (%.2f MB)", (double) aofRewriteBufferSize() / (1024*1024)); 1376 1377 /* The only remaining thing to do is to rename the temporary file to 1378 * the configured file and switch the file descriptor used to do AOF 1379 * writes. We don't want close(2) or rename(2) calls to block the 1380 * server on old file deletion. 1381 * 1382 * There are two possible scenarios: 1383 * 1384 * 1) AOF is DISABLED and this was a one time rewrite. The temporary 1385 * file will be renamed to the configured file. When this file already 1386 * exists, it will be unlinked, which may block the server. 1387 * 1388 * 2) AOF is ENABLED and the rewritten AOF will immediately start 1389 * receiving writes. After the temporary file is renamed to the 1390 * configured file, the original AOF file descriptor will be closed. 1391 * Since this will be the last reference to that file, closing it 1392 * causes the underlying file to be unlinked, which may block the 1393 * server. 1394 * 1395 * To mitigate the blocking effect of the unlink operation (either 1396 * caused by rename(2) in scenario 1, or by close(2) in scenario 2), we 1397 * use a background thread to take care of this. First, we 1398 * make scenario 1 identical to scenario 2 by opening the target file 1399 * when it exists. The unlink operation after the rename(2) will then 1400 * be executed upon calling close(2) for its descriptor. Everything to 1401 * guarantee atomicity for this switch has already happened by then, so 1402 * we don't care what the outcome or duration of that close operation 1403 * is, as long as the file descriptor is released again. */ 1404 if (server.aof_fd == -1) { 1405 /* AOF disabled */ 1406 1407 /* Don't care if this fails: oldfd will be -1 and we handle that. 1408 * One notable case of -1 return is if the old file does 1409 * not exist. */ 1410 oldfd = open(server.aof_filename,O_RDONLY|O_NONBLOCK); 1411 } else { 1412 /* AOF enabled */ 1413 oldfd = -1; /* We'll set this to the current AOF filedes later. */ 1414 } 1415 1416 /* Rename the temporary file. This will not unlink the target file if 1417 * it exists, because we reference it with "oldfd". */ 1418 latencyStartMonitor(latency); 1419 if (rename(tmpfile,server.aof_filename) == -1) { 1420 redisLog(REDIS_WARNING, 1421 "Error trying to rename the temporary AOF file: %s", strerror(errno)); 1422 close(newfd); 1423 if (oldfd != -1) close(oldfd); 1424 goto cleanup; 1425 } 1426 latencyEndMonitor(latency); 1427 latencyAddSampleIfNeeded("aof-rename",latency); 1428 1429 if (server.aof_fd == -1) { 1430 /* AOF disabled, we don't need to set the AOF file descriptor 1431 * to this new file, so we can close it. */ 1432 close(newfd); 1433 } else { 1434 /* AOF enabled, replace the old fd with the new one. */ 1435 oldfd = server.aof_fd; 1436 server.aof_fd = newfd; 1437 if (server.aof_fsync == AOF_FSYNC_ALWAYS) 1438 aof_fsync(newfd); 1439 else if (server.aof_fsync == AOF_FSYNC_EVERYSEC) 1440 aof_background_fsync(newfd); 1441 server.aof_selected_db = -1; /* Make sure SELECT is re-issued */ 1442 aofUpdateCurrentSize(); 1443 server.aof_rewrite_base_size = server.aof_current_size; 1444 1445 /* Clear regular AOF buffer since its contents was just written to 1446 * the new AOF from the background rewrite buffer. */ 1447 sdsfree(server.aof_buf); 1448 server.aof_buf = sdsempty(); 1449 } 1450 1451 server.aof_lastbgrewrite_status = REDIS_OK; 1452 1453 redisLog(REDIS_NOTICE, "Background AOF rewrite finished successfully"); 1454 /* Change state from WAIT_REWRITE to ON if needed */ 1455 if (server.aof_state == REDIS_AOF_WAIT_REWRITE) 1456 server.aof_state = REDIS_AOF_ON; 1457 1458 /* Asynchronously close the overwritten AOF. */ 1459 if (oldfd != -1) bioCreateBackgroundJob(REDIS_BIO_CLOSE_FILE,(void*)(long)oldfd,NULL,NULL); 1460 1461 redisLog(REDIS_VERBOSE, 1462 "Background AOF rewrite signal handler took %lldus", ustime()-now); 1463 } else if (!bysignal && exitcode != 0) { 1464 server.aof_lastbgrewrite_status = REDIS_ERR; 1465 1466 redisLog(REDIS_WARNING, 1467 "Background AOF rewrite terminated with error"); 1468 } else { 1469 server.aof_lastbgrewrite_status = REDIS_ERR; 1470 1471 redisLog(REDIS_WARNING, 1472 "Background AOF rewrite terminated by signal %d", bysignal); 1473 } 1474 1475 cleanup: 1476 aofClosePipes(); 1477 aofRewriteBufferReset(); 1478 aofRemoveTempFile(server.aof_child_pid); 1479 server.aof_child_pid = -1; 1480 server.aof_rewrite_time_last = time(NULL)-server.aof_rewrite_time_start; 1481 server.aof_rewrite_time_start = -1; 1482 /* Schedule a new rewrite if we are waiting for it to switch the AOF ON. */ 1483 if (server.aof_state == REDIS_AOF_WAIT_REWRITE) 1484 server.aof_rewrite_scheduled = 1; 1485 } 1486