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 "server.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 #include <sys/param.h>
42
43 void aofUpdateCurrentSize(void);
44 void aofClosePipes(void);
45
46 /* ----------------------------------------------------------------------------
47 * AOF rewrite buffer implementation.
48 *
49 * The following code implement a simple buffer used in order to accumulate
50 * changes while the background process is rewriting the AOF file.
51 *
52 * We only need to append, but can't just use realloc with a large block
53 * because 'huge' reallocs are not always handled as one could expect
54 * (via remapping of pages at OS level) but may involve copying data.
55 *
56 * For this reason we use a list of blocks, every block is
57 * AOF_RW_BUF_BLOCK_SIZE bytes.
58 * ------------------------------------------------------------------------- */
59
60 #define AOF_RW_BUF_BLOCK_SIZE (1024*1024*10) /* 10 MB per block */
61
62 typedef struct aofrwblock {
63 unsigned long used, free;
64 char buf[AOF_RW_BUF_BLOCK_SIZE];
65 } aofrwblock;
66
67 /* This function free the old AOF rewrite buffer if needed, and initialize
68 * a fresh new one. It tests for server.aof_rewrite_buf_blocks equal to NULL
69 * so can be used for the first initialization as well. */
aofRewriteBufferReset(void)70 void aofRewriteBufferReset(void) {
71 if (server.aof_rewrite_buf_blocks)
72 listRelease(server.aof_rewrite_buf_blocks);
73
74 server.aof_rewrite_buf_blocks = listCreate();
75 listSetFreeMethod(server.aof_rewrite_buf_blocks,zfree);
76 }
77
78 /* Return the current size of the AOF rewrite buffer. */
aofRewriteBufferSize(void)79 unsigned long aofRewriteBufferSize(void) {
80 listNode *ln;
81 listIter li;
82 unsigned long size = 0;
83
84 listRewind(server.aof_rewrite_buf_blocks,&li);
85 while((ln = listNext(&li))) {
86 aofrwblock *block = listNodeValue(ln);
87 size += block->used;
88 }
89 return size;
90 }
91
92 /* Event handler used to send data to the child process doing the AOF
93 * rewrite. We send pieces of our AOF differences buffer so that the final
94 * write when the child finishes the rewrite will be small. */
aofChildWriteDiffData(aeEventLoop * el,int fd,void * privdata,int mask)95 void aofChildWriteDiffData(aeEventLoop *el, int fd, void *privdata, int mask) {
96 listNode *ln;
97 aofrwblock *block;
98 ssize_t nwritten;
99 UNUSED(el);
100 UNUSED(fd);
101 UNUSED(privdata);
102 UNUSED(mask);
103
104 while(1) {
105 ln = listFirst(server.aof_rewrite_buf_blocks);
106 block = ln ? ln->value : NULL;
107 if (server.aof_stop_sending_diff || !block) {
108 aeDeleteFileEvent(server.el,server.aof_pipe_write_data_to_child,
109 AE_WRITABLE);
110 return;
111 }
112 if (block->used > 0) {
113 nwritten = write(server.aof_pipe_write_data_to_child,
114 block->buf,block->used);
115 if (nwritten <= 0) return;
116 memmove(block->buf,block->buf+nwritten,block->used-nwritten);
117 block->used -= nwritten;
118 block->free += nwritten;
119 }
120 if (block->used == 0) listDelNode(server.aof_rewrite_buf_blocks,ln);
121 }
122 }
123
124 /* Append data to the AOF rewrite buffer, allocating new blocks if needed. */
aofRewriteBufferAppend(unsigned char * s,unsigned long len)125 void aofRewriteBufferAppend(unsigned char *s, unsigned long len) {
126 listNode *ln = listLast(server.aof_rewrite_buf_blocks);
127 aofrwblock *block = ln ? ln->value : NULL;
128
129 while(len) {
130 /* If we already got at least an allocated block, try appending
131 * at least some piece into it. */
132 if (block) {
133 unsigned long thislen = (block->free < len) ? block->free : len;
134 if (thislen) { /* The current block is not already full. */
135 memcpy(block->buf+block->used, s, thislen);
136 block->used += thislen;
137 block->free -= thislen;
138 s += thislen;
139 len -= thislen;
140 }
141 }
142
143 if (len) { /* First block to allocate, or need another block. */
144 int numblocks;
145
146 block = zmalloc(sizeof(*block));
147 block->free = AOF_RW_BUF_BLOCK_SIZE;
148 block->used = 0;
149 listAddNodeTail(server.aof_rewrite_buf_blocks,block);
150
151 /* Log every time we cross more 10 or 100 blocks, respectively
152 * as a notice or warning. */
153 numblocks = listLength(server.aof_rewrite_buf_blocks);
154 if (((numblocks+1) % 10) == 0) {
155 int level = ((numblocks+1) % 100) == 0 ? LL_WARNING :
156 LL_NOTICE;
157 serverLog(level,"Background AOF buffer size: %lu MB",
158 aofRewriteBufferSize()/(1024*1024));
159 }
160 }
161 }
162
163 /* Install a file event to send data to the rewrite child if there is
164 * not one already. */
165 if (aeGetFileEvents(server.el,server.aof_pipe_write_data_to_child) == 0) {
166 aeCreateFileEvent(server.el, server.aof_pipe_write_data_to_child,
167 AE_WRITABLE, aofChildWriteDiffData, NULL);
168 }
169 }
170
171 /* Write the buffer (possibly composed of multiple blocks) into the specified
172 * fd. If a short write or any other error happens -1 is returned,
173 * otherwise the number of bytes written is returned. */
aofRewriteBufferWrite(int fd)174 ssize_t aofRewriteBufferWrite(int fd) {
175 listNode *ln;
176 listIter li;
177 ssize_t count = 0;
178
179 listRewind(server.aof_rewrite_buf_blocks,&li);
180 while((ln = listNext(&li))) {
181 aofrwblock *block = listNodeValue(ln);
182 ssize_t nwritten;
183
184 if (block->used) {
185 nwritten = write(fd,block->buf,block->used);
186 if (nwritten != (ssize_t)block->used) {
187 if (nwritten == 0) errno = EIO;
188 return -1;
189 }
190 count += nwritten;
191 }
192 }
193 return count;
194 }
195
196 /* ----------------------------------------------------------------------------
197 * AOF file implementation
198 * ------------------------------------------------------------------------- */
199
200 /* Return true if an AOf fsync is currently already in progress in a
201 * BIO thread. */
aofFsyncInProgress(void)202 int aofFsyncInProgress(void) {
203 return bioPendingJobsOfType(BIO_AOF_FSYNC) != 0;
204 }
205
206 /* Starts a background task that performs fsync() against the specified
207 * file descriptor (the one of the AOF file) in another thread. */
aof_background_fsync(int fd)208 void aof_background_fsync(int fd) {
209 bioCreateBackgroundJob(BIO_AOF_FSYNC,(void*)(long)fd,NULL,NULL);
210 }
211
212 /* Kills an AOFRW child process if exists */
killAppendOnlyChild(void)213 static void killAppendOnlyChild(void) {
214 int statloc;
215 /* No AOFRW child? return. */
216 if (server.aof_child_pid == -1) return;
217 /* Kill AOFRW child, wait for child exit. */
218 serverLog(LL_NOTICE,"Killing running AOF rewrite child: %ld",
219 (long) server.aof_child_pid);
220 if (kill(server.aof_child_pid,SIGUSR1) != -1) {
221 while(wait3(&statloc,0,NULL) != server.aof_child_pid);
222 }
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 /* Called when the user switches from "appendonly yes" to "appendonly no"
233 * at runtime using the CONFIG command. */
stopAppendOnly(void)234 void stopAppendOnly(void) {
235 serverAssert(server.aof_state != AOF_OFF);
236 flushAppendOnlyFile(1);
237 redis_fsync(server.aof_fd);
238 close(server.aof_fd);
239
240 server.aof_fd = -1;
241 server.aof_selected_db = -1;
242 server.aof_state = AOF_OFF;
243 killAppendOnlyChild();
244 }
245
246 /* Called when the user switches from "appendonly no" to "appendonly yes"
247 * at runtime using the CONFIG command. */
startAppendOnly(void)248 int startAppendOnly(void) {
249 char cwd[MAXPATHLEN]; /* Current working dir path for error messages. */
250 int newfd;
251
252 newfd = open(server.aof_filename,O_WRONLY|O_APPEND|O_CREAT,0644);
253 serverAssert(server.aof_state == AOF_OFF);
254 if (newfd == -1) {
255 char *cwdp = getcwd(cwd,MAXPATHLEN);
256
257 serverLog(LL_WARNING,
258 "Redis needs to enable the AOF but can't open the "
259 "append only file %s (in server root dir %s): %s",
260 server.aof_filename,
261 cwdp ? cwdp : "unknown",
262 strerror(errno));
263 return C_ERR;
264 }
265 if (server.rdb_child_pid != -1) {
266 server.aof_rewrite_scheduled = 1;
267 serverLog(LL_WARNING,"AOF was enabled but there is already a child process saving an RDB file on disk. An AOF background was scheduled to start when possible.");
268 } else {
269 /* If there is a pending AOF rewrite, we need to switch it off and
270 * start a new one: the old one cannot be reused because it is not
271 * accumulating the AOF buffer. */
272 if (server.aof_child_pid != -1) {
273 serverLog(LL_WARNING,"AOF was enabled but there is already an AOF rewriting in background. Stopping background AOF and starting a rewrite now.");
274 killAppendOnlyChild();
275 }
276 if (rewriteAppendOnlyFileBackground() == C_ERR) {
277 close(newfd);
278 serverLog(LL_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.");
279 return C_ERR;
280 }
281 }
282 /* We correctly switched on AOF, now wait for the rewrite to be complete
283 * in order to append data on disk. */
284 server.aof_state = AOF_WAIT_REWRITE;
285 server.aof_last_fsync = server.unixtime;
286 server.aof_fd = newfd;
287 return C_OK;
288 }
289
290 /* This is a wrapper to the write syscall in order to retry on short writes
291 * or if the syscall gets interrupted. It could look strange that we retry
292 * on short writes given that we are writing to a block device: normally if
293 * the first call is short, there is a end-of-space condition, so the next
294 * is likely to fail. However apparently in modern systems this is no longer
295 * true, and in general it looks just more resilient to retry the write. If
296 * there is an actual error condition we'll get it at the next try. */
aofWrite(int fd,const char * buf,size_t len)297 ssize_t aofWrite(int fd, const char *buf, size_t len) {
298 ssize_t nwritten = 0, totwritten = 0;
299
300 while(len) {
301 nwritten = write(fd, buf, len);
302
303 if (nwritten < 0) {
304 if (errno == EINTR) {
305 continue;
306 }
307 return totwritten ? totwritten : -1;
308 }
309
310 len -= nwritten;
311 buf += nwritten;
312 totwritten += nwritten;
313 }
314
315 return totwritten;
316 }
317
318 /* Write the append only file buffer on disk.
319 *
320 * Since we are required to write the AOF before replying to the client,
321 * and the only way the client socket can get a write is entering when the
322 * the event loop, we accumulate all the AOF writes in a memory
323 * buffer and write it on disk using this function just before entering
324 * the event loop again.
325 *
326 * About the 'force' argument:
327 *
328 * When the fsync policy is set to 'everysec' we may delay the flush if there
329 * is still an fsync() going on in the background thread, since for instance
330 * on Linux write(2) will be blocked by the background fsync anyway.
331 * When this happens we remember that there is some aof buffer to be
332 * flushed ASAP, and will try to do that in the serverCron() function.
333 *
334 * However if force is set to 1 we'll write regardless of the background
335 * fsync. */
336 #define AOF_WRITE_LOG_ERROR_RATE 30 /* Seconds between errors logging. */
flushAppendOnlyFile(int force)337 void flushAppendOnlyFile(int force) {
338 ssize_t nwritten;
339 int sync_in_progress = 0;
340 mstime_t latency;
341
342 if (sdslen(server.aof_buf) == 0) {
343 /* Check if we need to do fsync even the aof buffer is empty,
344 * because previously in AOF_FSYNC_EVERYSEC mode, fsync is
345 * called only when aof buffer is not empty, so if users
346 * stop write commands before fsync called in one second,
347 * the data in page cache cannot be flushed in time. */
348 if (server.aof_fsync == AOF_FSYNC_EVERYSEC &&
349 server.aof_fsync_offset != server.aof_current_size &&
350 server.unixtime > server.aof_last_fsync &&
351 !(sync_in_progress = aofFsyncInProgress())) {
352 goto try_fsync;
353 } else {
354 return;
355 }
356 }
357
358 if (server.aof_fsync == AOF_FSYNC_EVERYSEC)
359 sync_in_progress = aofFsyncInProgress();
360
361 if (server.aof_fsync == AOF_FSYNC_EVERYSEC && !force) {
362 /* With this append fsync policy we do background fsyncing.
363 * If the fsync is still in progress we can try to delay
364 * the write for a couple of seconds. */
365 if (sync_in_progress) {
366 if (server.aof_flush_postponed_start == 0) {
367 /* No previous write postponing, remember that we are
368 * postponing the flush and return. */
369 server.aof_flush_postponed_start = server.unixtime;
370 return;
371 } else if (server.unixtime - server.aof_flush_postponed_start < 2) {
372 /* We were already waiting for fsync to finish, but for less
373 * than two seconds this is still ok. Postpone again. */
374 return;
375 }
376 /* Otherwise fall trough, and go write since we can't wait
377 * over two seconds. */
378 server.aof_delayed_fsync++;
379 serverLog(LL_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.");
380 }
381 }
382 /* We want to perform a single write. This should be guaranteed atomic
383 * at least if the filesystem we are writing is a real physical one.
384 * While this will save us against the server being killed I don't think
385 * there is much to do about the whole server stopping for power problems
386 * or alike */
387
388 latencyStartMonitor(latency);
389 nwritten = aofWrite(server.aof_fd,server.aof_buf,sdslen(server.aof_buf));
390 latencyEndMonitor(latency);
391 /* We want to capture different events for delayed writes:
392 * when the delay happens with a pending fsync, or with a saving child
393 * active, and when the above two conditions are missing.
394 * We also use an additional event name to save all samples which is
395 * useful for graphing / monitoring purposes. */
396 if (sync_in_progress) {
397 latencyAddSampleIfNeeded("aof-write-pending-fsync",latency);
398 } else if (server.aof_child_pid != -1 || server.rdb_child_pid != -1) {
399 latencyAddSampleIfNeeded("aof-write-active-child",latency);
400 } else {
401 latencyAddSampleIfNeeded("aof-write-alone",latency);
402 }
403 latencyAddSampleIfNeeded("aof-write",latency);
404
405 /* We performed the write so reset the postponed flush sentinel to zero. */
406 server.aof_flush_postponed_start = 0;
407
408 if (nwritten != (ssize_t)sdslen(server.aof_buf)) {
409 static time_t last_write_error_log = 0;
410 int can_log = 0;
411
412 /* Limit logging rate to 1 line per AOF_WRITE_LOG_ERROR_RATE seconds. */
413 if ((server.unixtime - last_write_error_log) > AOF_WRITE_LOG_ERROR_RATE) {
414 can_log = 1;
415 last_write_error_log = server.unixtime;
416 }
417
418 /* Log the AOF write error and record the error code. */
419 if (nwritten == -1) {
420 if (can_log) {
421 serverLog(LL_WARNING,"Error writing to the AOF file: %s",
422 strerror(errno));
423 server.aof_last_write_errno = errno;
424 }
425 } else {
426 if (can_log) {
427 serverLog(LL_WARNING,"Short write while writing to "
428 "the AOF file: (nwritten=%lld, "
429 "expected=%lld)",
430 (long long)nwritten,
431 (long long)sdslen(server.aof_buf));
432 }
433
434 if (ftruncate(server.aof_fd, server.aof_current_size) == -1) {
435 if (can_log) {
436 serverLog(LL_WARNING, "Could not remove short write "
437 "from the append-only file. Redis may refuse "
438 "to load the AOF the next time it starts. "
439 "ftruncate: %s", strerror(errno));
440 }
441 } else {
442 /* If the ftruncate() succeeded we can set nwritten to
443 * -1 since there is no longer partial data into the AOF. */
444 nwritten = -1;
445 }
446 server.aof_last_write_errno = ENOSPC;
447 }
448
449 /* Handle the AOF write error. */
450 if (server.aof_fsync == AOF_FSYNC_ALWAYS) {
451 /* We can't recover when the fsync policy is ALWAYS since the
452 * reply for the client is already in the output buffers, and we
453 * have the contract with the user that on acknowledged write data
454 * is synced on disk. */
455 serverLog(LL_WARNING,"Can't recover from AOF write error when the AOF fsync policy is 'always'. Exiting...");
456 exit(1);
457 } else {
458 /* Recover from failed write leaving data into the buffer. However
459 * set an error to stop accepting writes as long as the error
460 * condition is not cleared. */
461 server.aof_last_write_status = C_ERR;
462
463 /* Trim the sds buffer if there was a partial write, and there
464 * was no way to undo it with ftruncate(2). */
465 if (nwritten > 0) {
466 server.aof_current_size += nwritten;
467 sdsrange(server.aof_buf,nwritten,-1);
468 }
469 return; /* We'll try again on the next call... */
470 }
471 } else {
472 /* Successful write(2). If AOF was in error state, restore the
473 * OK state and log the event. */
474 if (server.aof_last_write_status == C_ERR) {
475 serverLog(LL_WARNING,
476 "AOF write error looks solved, Redis can write again.");
477 server.aof_last_write_status = C_OK;
478 }
479 }
480 server.aof_current_size += nwritten;
481
482 /* Re-use AOF buffer when it is small enough. The maximum comes from the
483 * arena size of 4k minus some overhead (but is otherwise arbitrary). */
484 if ((sdslen(server.aof_buf)+sdsavail(server.aof_buf)) < 4000) {
485 sdsclear(server.aof_buf);
486 } else {
487 sdsfree(server.aof_buf);
488 server.aof_buf = sdsempty();
489 }
490
491 try_fsync:
492 /* Don't fsync if no-appendfsync-on-rewrite is set to yes and there are
493 * children doing I/O in the background. */
494 if (server.aof_no_fsync_on_rewrite &&
495 (server.aof_child_pid != -1 || server.rdb_child_pid != -1))
496 return;
497
498 /* Perform the fsync if needed. */
499 if (server.aof_fsync == AOF_FSYNC_ALWAYS) {
500 /* redis_fsync is defined as fdatasync() for Linux in order to avoid
501 * flushing metadata. */
502 latencyStartMonitor(latency);
503 redis_fsync(server.aof_fd); /* Let's try to get this data on the disk */
504 latencyEndMonitor(latency);
505 latencyAddSampleIfNeeded("aof-fsync-always",latency);
506 server.aof_fsync_offset = server.aof_current_size;
507 server.aof_last_fsync = server.unixtime;
508 } else if ((server.aof_fsync == AOF_FSYNC_EVERYSEC &&
509 server.unixtime > server.aof_last_fsync)) {
510 if (!sync_in_progress) {
511 aof_background_fsync(server.aof_fd);
512 server.aof_fsync_offset = server.aof_current_size;
513 }
514 server.aof_last_fsync = server.unixtime;
515 }
516 }
517
catAppendOnlyGenericCommand(sds dst,int argc,robj ** argv)518 sds catAppendOnlyGenericCommand(sds dst, int argc, robj **argv) {
519 char buf[32];
520 int len, j;
521 robj *o;
522
523 buf[0] = '*';
524 len = 1+ll2string(buf+1,sizeof(buf)-1,argc);
525 buf[len++] = '\r';
526 buf[len++] = '\n';
527 dst = sdscatlen(dst,buf,len);
528
529 for (j = 0; j < argc; j++) {
530 o = getDecodedObject(argv[j]);
531 buf[0] = '$';
532 len = 1+ll2string(buf+1,sizeof(buf)-1,sdslen(o->ptr));
533 buf[len++] = '\r';
534 buf[len++] = '\n';
535 dst = sdscatlen(dst,buf,len);
536 dst = sdscatlen(dst,o->ptr,sdslen(o->ptr));
537 dst = sdscatlen(dst,"\r\n",2);
538 decrRefCount(o);
539 }
540 return dst;
541 }
542
543 /* Create the sds representation of an PEXPIREAT command, using
544 * 'seconds' as time to live and 'cmd' to understand what command
545 * we are translating into a PEXPIREAT.
546 *
547 * This command is used in order to translate EXPIRE and PEXPIRE commands
548 * into PEXPIREAT command so that we retain precision in the append only
549 * file, and the time is always absolute and not relative. */
catAppendOnlyExpireAtCommand(sds buf,struct redisCommand * cmd,robj * key,robj * seconds)550 sds catAppendOnlyExpireAtCommand(sds buf, struct redisCommand *cmd, robj *key, robj *seconds) {
551 long long when;
552 robj *argv[3];
553
554 /* Make sure we can use strtoll */
555 seconds = getDecodedObject(seconds);
556 when = strtoll(seconds->ptr,NULL,10);
557 /* Convert argument into milliseconds for EXPIRE, SETEX, EXPIREAT */
558 if (cmd->proc == expireCommand || cmd->proc == setexCommand ||
559 cmd->proc == expireatCommand)
560 {
561 when *= 1000;
562 }
563 /* Convert into absolute time for EXPIRE, PEXPIRE, SETEX, PSETEX */
564 if (cmd->proc == expireCommand || cmd->proc == pexpireCommand ||
565 cmd->proc == setexCommand || cmd->proc == psetexCommand)
566 {
567 when += mstime();
568 }
569 decrRefCount(seconds);
570
571 argv[0] = createStringObject("PEXPIREAT",9);
572 argv[1] = key;
573 argv[2] = createStringObjectFromLongLong(when);
574 buf = catAppendOnlyGenericCommand(buf, 3, argv);
575 decrRefCount(argv[0]);
576 decrRefCount(argv[2]);
577 return buf;
578 }
579
feedAppendOnlyFile(struct redisCommand * cmd,int dictid,robj ** argv,int argc)580 void feedAppendOnlyFile(struct redisCommand *cmd, int dictid, robj **argv, int argc) {
581 sds buf = sdsempty();
582 robj *tmpargv[3];
583
584 /* The DB this command was targeting is not the same as the last command
585 * we appended. To issue a SELECT command is needed. */
586 if (dictid != server.aof_selected_db) {
587 char seldb[64];
588
589 snprintf(seldb,sizeof(seldb),"%d",dictid);
590 buf = sdscatprintf(buf,"*2\r\n$6\r\nSELECT\r\n$%lu\r\n%s\r\n",
591 (unsigned long)strlen(seldb),seldb);
592 server.aof_selected_db = dictid;
593 }
594
595 if (cmd->proc == expireCommand || cmd->proc == pexpireCommand ||
596 cmd->proc == expireatCommand) {
597 /* Translate EXPIRE/PEXPIRE/EXPIREAT into PEXPIREAT */
598 buf = catAppendOnlyExpireAtCommand(buf,cmd,argv[1],argv[2]);
599 } else if (cmd->proc == setexCommand || cmd->proc == psetexCommand) {
600 /* Translate SETEX/PSETEX to SET and PEXPIREAT */
601 tmpargv[0] = createStringObject("SET",3);
602 tmpargv[1] = argv[1];
603 tmpargv[2] = argv[3];
604 buf = catAppendOnlyGenericCommand(buf,3,tmpargv);
605 decrRefCount(tmpargv[0]);
606 buf = catAppendOnlyExpireAtCommand(buf,cmd,argv[1],argv[2]);
607 } else if (cmd->proc == setCommand && argc > 3) {
608 int i;
609 robj *exarg = NULL, *pxarg = NULL;
610 /* Translate SET [EX seconds][PX milliseconds] to SET and PEXPIREAT */
611 buf = catAppendOnlyGenericCommand(buf,3,argv);
612 for (i = 3; i < argc; i ++) {
613 if (!strcasecmp(argv[i]->ptr, "ex")) exarg = argv[i+1];
614 if (!strcasecmp(argv[i]->ptr, "px")) pxarg = argv[i+1];
615 }
616 serverAssert(!(exarg && pxarg));
617 if (exarg)
618 buf = catAppendOnlyExpireAtCommand(buf,server.expireCommand,argv[1],
619 exarg);
620 if (pxarg)
621 buf = catAppendOnlyExpireAtCommand(buf,server.pexpireCommand,argv[1],
622 pxarg);
623 } else {
624 /* All the other commands don't need translation or need the
625 * same translation already operated in the command vector
626 * for the replication itself. */
627 buf = catAppendOnlyGenericCommand(buf,argc,argv);
628 }
629
630 /* Append to the AOF buffer. This will be flushed on disk just before
631 * of re-entering the event loop, so before the client will get a
632 * positive reply about the operation performed. */
633 if (server.aof_state == AOF_ON)
634 server.aof_buf = sdscatlen(server.aof_buf,buf,sdslen(buf));
635
636 /* If a background append only file rewriting is in progress we want to
637 * accumulate the differences between the child DB and the current one
638 * in a buffer, so that when the child process will do its work we
639 * can append the differences to the new append only file. */
640 if (server.aof_child_pid != -1)
641 aofRewriteBufferAppend((unsigned char*)buf,sdslen(buf));
642
643 sdsfree(buf);
644 }
645
646 /* ----------------------------------------------------------------------------
647 * AOF loading
648 * ------------------------------------------------------------------------- */
649
650 /* In Redis commands are always executed in the context of a client, so in
651 * order to load the append only file we need to create a fake client. */
createFakeClient(void)652 struct client *createFakeClient(void) {
653 struct client *c = zmalloc(sizeof(*c));
654
655 selectDb(c,0);
656 c->fd = -1;
657 c->name = NULL;
658 c->querybuf = sdsempty();
659 c->querybuf_peak = 0;
660 c->argc = 0;
661 c->argv = NULL;
662 c->bufpos = 0;
663 c->flags = 0;
664 c->btype = BLOCKED_NONE;
665 /* We set the fake client as a slave waiting for the synchronization
666 * so that Redis will not try to send replies to this client. */
667 c->replstate = SLAVE_STATE_WAIT_BGSAVE_START;
668 c->reply = listCreate();
669 c->reply_bytes = 0;
670 c->obuf_soft_limit_reached_time = 0;
671 c->watched_keys = listCreate();
672 c->peerid = NULL;
673 listSetFreeMethod(c->reply,freeClientReplyValue);
674 listSetDupMethod(c->reply,dupClientReplyValue);
675 initClientMultiState(c);
676 return c;
677 }
678
freeFakeClientArgv(struct client * c)679 void freeFakeClientArgv(struct client *c) {
680 int j;
681
682 for (j = 0; j < c->argc; j++)
683 decrRefCount(c->argv[j]);
684 zfree(c->argv);
685 }
686
freeFakeClient(struct client * c)687 void freeFakeClient(struct client *c) {
688 sdsfree(c->querybuf);
689 listRelease(c->reply);
690 listRelease(c->watched_keys);
691 freeClientMultiState(c);
692 zfree(c);
693 }
694
695 /* Replay the append log file. On success C_OK is returned. On non fatal
696 * error (the append only file is zero-length) C_ERR is returned. On
697 * fatal error an error message is logged and the program exists. */
loadAppendOnlyFile(char * filename)698 int loadAppendOnlyFile(char *filename) {
699 struct client *fakeClient;
700 FILE *fp = fopen(filename,"r");
701 struct redis_stat sb;
702 int old_aof_state = server.aof_state;
703 long loops = 0;
704 off_t valid_up_to = 0; /* Offset of latest well-formed command loaded. */
705 off_t valid_before_multi = 0; /* Offset before MULTI command loaded. */
706
707 if (fp == NULL) {
708 serverLog(LL_WARNING,"Fatal error: can't open the append log file for reading: %s",strerror(errno));
709 exit(1);
710 }
711
712 /* Handle a zero-length AOF file as a special case. An empty AOF file
713 * is a valid AOF because an empty server with AOF enabled will create
714 * a zero length file at startup, that will remain like that if no write
715 * operation is received. */
716 if (fp && redis_fstat(fileno(fp),&sb) != -1 && sb.st_size == 0) {
717 server.aof_current_size = 0;
718 server.aof_fsync_offset = server.aof_current_size;
719 fclose(fp);
720 return C_ERR;
721 }
722
723 /* Temporarily disable AOF, to prevent EXEC from feeding a MULTI
724 * to the same file we're about to read. */
725 server.aof_state = AOF_OFF;
726
727 fakeClient = createFakeClient();
728 startLoading(fp);
729
730 /* Check if this AOF file has an RDB preamble. In that case we need to
731 * load the RDB file and later continue loading the AOF tail. */
732 char sig[5]; /* "REDIS" */
733 if (fread(sig,1,5,fp) != 5 || memcmp(sig,"REDIS",5) != 0) {
734 /* No RDB preamble, seek back at 0 offset. */
735 if (fseek(fp,0,SEEK_SET) == -1) goto readerr;
736 } else {
737 /* RDB preamble. Pass loading the RDB functions. */
738 rio rdb;
739
740 serverLog(LL_NOTICE,"Reading RDB preamble from AOF file...");
741 if (fseek(fp,0,SEEK_SET) == -1) goto readerr;
742 rioInitWithFile(&rdb,fp);
743 if (rdbLoadRio(&rdb,NULL,1) != C_OK) {
744 serverLog(LL_WARNING,"Error reading the RDB preamble of the AOF file, AOF loading aborted");
745 goto readerr;
746 } else {
747 serverLog(LL_NOTICE,"Reading the remaining AOF tail...");
748 }
749 }
750
751 /* Read the actual AOF file, in REPL format, command by command. */
752 while(1) {
753 int argc, j;
754 unsigned long len;
755 robj **argv;
756 char buf[128];
757 sds argsds;
758 struct redisCommand *cmd;
759
760 /* Serve the clients from time to time */
761 if (!(loops++ % 1000)) {
762 loadingProgress(ftello(fp));
763 processEventsWhileBlocked();
764 }
765
766 if (fgets(buf,sizeof(buf),fp) == NULL) {
767 if (feof(fp))
768 break;
769 else
770 goto readerr;
771 }
772 if (buf[0] != '*') goto fmterr;
773 if (buf[1] == '\0') goto readerr;
774 argc = atoi(buf+1);
775 if (argc < 1) goto fmterr;
776
777 argv = zmalloc(sizeof(robj*)*argc);
778 fakeClient->argc = argc;
779 fakeClient->argv = argv;
780
781 for (j = 0; j < argc; j++) {
782 if (fgets(buf,sizeof(buf),fp) == NULL) {
783 fakeClient->argc = j; /* Free up to j-1. */
784 freeFakeClientArgv(fakeClient);
785 goto readerr;
786 }
787 if (buf[0] != '$') goto fmterr;
788 len = strtol(buf+1,NULL,10);
789 argsds = sdsnewlen(SDS_NOINIT,len);
790 if (len && fread(argsds,len,1,fp) == 0) {
791 sdsfree(argsds);
792 fakeClient->argc = j; /* Free up to j-1. */
793 freeFakeClientArgv(fakeClient);
794 goto readerr;
795 }
796 argv[j] = createObject(OBJ_STRING,argsds);
797 if (fread(buf,2,1,fp) == 0) {
798 fakeClient->argc = j+1; /* Free up to j. */
799 freeFakeClientArgv(fakeClient);
800 goto readerr; /* discard CRLF */
801 }
802 }
803
804 /* Command lookup */
805 cmd = lookupCommand(argv[0]->ptr);
806 if (!cmd) {
807 serverLog(LL_WARNING,
808 "Unknown command '%s' reading the append only file",
809 (char*)argv[0]->ptr);
810 exit(1);
811 }
812
813 if (cmd == server.multiCommand) valid_before_multi = valid_up_to;
814
815 /* Run the command in the context of a fake client */
816 fakeClient->cmd = cmd;
817 if (fakeClient->flags & CLIENT_MULTI &&
818 fakeClient->cmd->proc != execCommand)
819 {
820 queueMultiCommand(fakeClient);
821 } else {
822 cmd->proc(fakeClient);
823 }
824
825 /* The fake client should not have a reply */
826 serverAssert(fakeClient->bufpos == 0 &&
827 listLength(fakeClient->reply) == 0);
828
829 /* The fake client should never get blocked */
830 serverAssert((fakeClient->flags & CLIENT_BLOCKED) == 0);
831
832 /* Clean up. Command code may have changed argv/argc so we use the
833 * argv/argc of the client instead of the local variables. */
834 freeFakeClientArgv(fakeClient);
835 fakeClient->cmd = NULL;
836 if (server.aof_load_truncated) valid_up_to = ftello(fp);
837 }
838
839 /* This point can only be reached when EOF is reached without errors.
840 * If the client is in the middle of a MULTI/EXEC, handle it as it was
841 * a short read, even if technically the protocol is correct: we want
842 * to remove the unprocessed tail and continue. */
843 if (fakeClient->flags & CLIENT_MULTI) {
844 serverLog(LL_WARNING,
845 "Revert incomplete MULTI/EXEC transaction in AOF file");
846 valid_up_to = valid_before_multi;
847 goto uxeof;
848 }
849
850 loaded_ok: /* DB loaded, cleanup and return C_OK to the caller. */
851 fclose(fp);
852 freeFakeClient(fakeClient);
853 server.aof_state = old_aof_state;
854 stopLoading();
855 aofUpdateCurrentSize();
856 server.aof_rewrite_base_size = server.aof_current_size;
857 server.aof_fsync_offset = server.aof_current_size;
858 return C_OK;
859
860 readerr: /* Read error. If feof(fp) is true, fall through to unexpected EOF. */
861 if (!feof(fp)) {
862 if (fakeClient) freeFakeClient(fakeClient); /* avoid valgrind warning */
863 serverLog(LL_WARNING,"Unrecoverable error reading the append only file: %s", strerror(errno));
864 exit(1);
865 }
866
867 uxeof: /* Unexpected AOF end of file. */
868 if (server.aof_load_truncated) {
869 serverLog(LL_WARNING,"!!! Warning: short read while loading the AOF file !!!");
870 serverLog(LL_WARNING,"!!! Truncating the AOF at offset %llu !!!",
871 (unsigned long long) valid_up_to);
872 if (valid_up_to == -1 || truncate(filename,valid_up_to) == -1) {
873 if (valid_up_to == -1) {
874 serverLog(LL_WARNING,"Last valid command offset is invalid");
875 } else {
876 serverLog(LL_WARNING,"Error truncating the AOF file: %s",
877 strerror(errno));
878 }
879 } else {
880 /* Make sure the AOF file descriptor points to the end of the
881 * file after the truncate call. */
882 if (server.aof_fd != -1 && lseek(server.aof_fd,0,SEEK_END) == -1) {
883 serverLog(LL_WARNING,"Can't seek the end of the AOF file: %s",
884 strerror(errno));
885 } else {
886 serverLog(LL_WARNING,
887 "AOF loaded anyway because aof-load-truncated is enabled");
888 goto loaded_ok;
889 }
890 }
891 }
892 if (fakeClient) freeFakeClient(fakeClient); /* avoid valgrind warning */
893 serverLog(LL_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.");
894 exit(1);
895
896 fmterr: /* Format error. */
897 if (fakeClient) freeFakeClient(fakeClient); /* avoid valgrind warning */
898 serverLog(LL_WARNING,"Bad file format reading the append only file: make a backup of your AOF file, then use ./redis-check-aof --fix <filename>");
899 exit(1);
900 }
901
902 /* ----------------------------------------------------------------------------
903 * AOF rewrite
904 * ------------------------------------------------------------------------- */
905
906 /* Delegate writing an object to writing a bulk string or bulk long long.
907 * This is not placed in rio.c since that adds the server.h dependency. */
rioWriteBulkObject(rio * r,robj * obj)908 int rioWriteBulkObject(rio *r, robj *obj) {
909 /* Avoid using getDecodedObject to help copy-on-write (we are often
910 * in a child process when this function is called). */
911 if (obj->encoding == OBJ_ENCODING_INT) {
912 return rioWriteBulkLongLong(r,(long)obj->ptr);
913 } else if (sdsEncodedObject(obj)) {
914 return rioWriteBulkString(r,obj->ptr,sdslen(obj->ptr));
915 } else {
916 serverPanic("Unknown string encoding");
917 }
918 }
919
920 /* Emit the commands needed to rebuild a list object.
921 * The function returns 0 on error, 1 on success. */
rewriteListObject(rio * r,robj * key,robj * o)922 int rewriteListObject(rio *r, robj *key, robj *o) {
923 long long count = 0, items = listTypeLength(o);
924
925 if (o->encoding == OBJ_ENCODING_QUICKLIST) {
926 quicklist *list = o->ptr;
927 quicklistIter *li = quicklistGetIterator(list, AL_START_HEAD);
928 quicklistEntry entry;
929
930 while (quicklistNext(li,&entry)) {
931 if (count == 0) {
932 int cmd_items = (items > AOF_REWRITE_ITEMS_PER_CMD) ?
933 AOF_REWRITE_ITEMS_PER_CMD : items;
934 if (rioWriteBulkCount(r,'*',2+cmd_items) == 0) return 0;
935 if (rioWriteBulkString(r,"RPUSH",5) == 0) return 0;
936 if (rioWriteBulkObject(r,key) == 0) return 0;
937 }
938
939 if (entry.value) {
940 if (rioWriteBulkString(r,(char*)entry.value,entry.sz) == 0) return 0;
941 } else {
942 if (rioWriteBulkLongLong(r,entry.longval) == 0) return 0;
943 }
944 if (++count == AOF_REWRITE_ITEMS_PER_CMD) count = 0;
945 items--;
946 }
947 quicklistReleaseIterator(li);
948 } else {
949 serverPanic("Unknown list encoding");
950 }
951 return 1;
952 }
953
954 /* Emit the commands needed to rebuild a set object.
955 * The function returns 0 on error, 1 on success. */
rewriteSetObject(rio * r,robj * key,robj * o)956 int rewriteSetObject(rio *r, robj *key, robj *o) {
957 long long count = 0, items = setTypeSize(o);
958
959 if (o->encoding == OBJ_ENCODING_INTSET) {
960 int ii = 0;
961 int64_t llval;
962
963 while(intsetGet(o->ptr,ii++,&llval)) {
964 if (count == 0) {
965 int cmd_items = (items > AOF_REWRITE_ITEMS_PER_CMD) ?
966 AOF_REWRITE_ITEMS_PER_CMD : items;
967
968 if (rioWriteBulkCount(r,'*',2+cmd_items) == 0) return 0;
969 if (rioWriteBulkString(r,"SADD",4) == 0) return 0;
970 if (rioWriteBulkObject(r,key) == 0) return 0;
971 }
972 if (rioWriteBulkLongLong(r,llval) == 0) return 0;
973 if (++count == AOF_REWRITE_ITEMS_PER_CMD) count = 0;
974 items--;
975 }
976 } else if (o->encoding == OBJ_ENCODING_HT) {
977 dictIterator *di = dictGetIterator(o->ptr);
978 dictEntry *de;
979
980 while((de = dictNext(di)) != NULL) {
981 sds ele = dictGetKey(de);
982 if (count == 0) {
983 int cmd_items = (items > AOF_REWRITE_ITEMS_PER_CMD) ?
984 AOF_REWRITE_ITEMS_PER_CMD : items;
985
986 if (rioWriteBulkCount(r,'*',2+cmd_items) == 0) return 0;
987 if (rioWriteBulkString(r,"SADD",4) == 0) return 0;
988 if (rioWriteBulkObject(r,key) == 0) return 0;
989 }
990 if (rioWriteBulkString(r,ele,sdslen(ele)) == 0) return 0;
991 if (++count == AOF_REWRITE_ITEMS_PER_CMD) count = 0;
992 items--;
993 }
994 dictReleaseIterator(di);
995 } else {
996 serverPanic("Unknown set encoding");
997 }
998 return 1;
999 }
1000
1001 /* Emit the commands needed to rebuild a sorted set object.
1002 * The function returns 0 on error, 1 on success. */
rewriteSortedSetObject(rio * r,robj * key,robj * o)1003 int rewriteSortedSetObject(rio *r, robj *key, robj *o) {
1004 long long count = 0, items = zsetLength(o);
1005
1006 if (o->encoding == OBJ_ENCODING_ZIPLIST) {
1007 unsigned char *zl = o->ptr;
1008 unsigned char *eptr, *sptr;
1009 unsigned char *vstr;
1010 unsigned int vlen;
1011 long long vll;
1012 double score;
1013
1014 eptr = ziplistIndex(zl,0);
1015 serverAssert(eptr != NULL);
1016 sptr = ziplistNext(zl,eptr);
1017 serverAssert(sptr != NULL);
1018
1019 while (eptr != NULL) {
1020 serverAssert(ziplistGet(eptr,&vstr,&vlen,&vll));
1021 score = zzlGetScore(sptr);
1022
1023 if (count == 0) {
1024 int cmd_items = (items > AOF_REWRITE_ITEMS_PER_CMD) ?
1025 AOF_REWRITE_ITEMS_PER_CMD : items;
1026
1027 if (rioWriteBulkCount(r,'*',2+cmd_items*2) == 0) return 0;
1028 if (rioWriteBulkString(r,"ZADD",4) == 0) return 0;
1029 if (rioWriteBulkObject(r,key) == 0) return 0;
1030 }
1031 if (rioWriteBulkDouble(r,score) == 0) return 0;
1032 if (vstr != NULL) {
1033 if (rioWriteBulkString(r,(char*)vstr,vlen) == 0) return 0;
1034 } else {
1035 if (rioWriteBulkLongLong(r,vll) == 0) return 0;
1036 }
1037 zzlNext(zl,&eptr,&sptr);
1038 if (++count == AOF_REWRITE_ITEMS_PER_CMD) count = 0;
1039 items--;
1040 }
1041 } else if (o->encoding == OBJ_ENCODING_SKIPLIST) {
1042 zset *zs = o->ptr;
1043 dictIterator *di = dictGetIterator(zs->dict);
1044 dictEntry *de;
1045
1046 while((de = dictNext(di)) != NULL) {
1047 sds ele = dictGetKey(de);
1048 double *score = dictGetVal(de);
1049
1050 if (count == 0) {
1051 int cmd_items = (items > AOF_REWRITE_ITEMS_PER_CMD) ?
1052 AOF_REWRITE_ITEMS_PER_CMD : items;
1053
1054 if (rioWriteBulkCount(r,'*',2+cmd_items*2) == 0) return 0;
1055 if (rioWriteBulkString(r,"ZADD",4) == 0) return 0;
1056 if (rioWriteBulkObject(r,key) == 0) return 0;
1057 }
1058 if (rioWriteBulkDouble(r,*score) == 0) return 0;
1059 if (rioWriteBulkString(r,ele,sdslen(ele)) == 0) return 0;
1060 if (++count == AOF_REWRITE_ITEMS_PER_CMD) count = 0;
1061 items--;
1062 }
1063 dictReleaseIterator(di);
1064 } else {
1065 serverPanic("Unknown sorted zset encoding");
1066 }
1067 return 1;
1068 }
1069
1070 /* Write either the key or the value of the currently selected item of a hash.
1071 * The 'hi' argument passes a valid Redis hash iterator.
1072 * The 'what' filed specifies if to write a key or a value and can be
1073 * either OBJ_HASH_KEY or OBJ_HASH_VALUE.
1074 *
1075 * The function returns 0 on error, non-zero on success. */
rioWriteHashIteratorCursor(rio * r,hashTypeIterator * hi,int what)1076 static int rioWriteHashIteratorCursor(rio *r, hashTypeIterator *hi, int what) {
1077 if (hi->encoding == OBJ_ENCODING_ZIPLIST) {
1078 unsigned char *vstr = NULL;
1079 unsigned int vlen = UINT_MAX;
1080 long long vll = LLONG_MAX;
1081
1082 hashTypeCurrentFromZiplist(hi, what, &vstr, &vlen, &vll);
1083 if (vstr)
1084 return rioWriteBulkString(r, (char*)vstr, vlen);
1085 else
1086 return rioWriteBulkLongLong(r, vll);
1087 } else if (hi->encoding == OBJ_ENCODING_HT) {
1088 sds value = hashTypeCurrentFromHashTable(hi, what);
1089 return rioWriteBulkString(r, value, sdslen(value));
1090 }
1091
1092 serverPanic("Unknown hash encoding");
1093 return 0;
1094 }
1095
1096 /* Emit the commands needed to rebuild a hash object.
1097 * The function returns 0 on error, 1 on success. */
rewriteHashObject(rio * r,robj * key,robj * o)1098 int rewriteHashObject(rio *r, robj *key, robj *o) {
1099 hashTypeIterator *hi;
1100 long long count = 0, items = hashTypeLength(o);
1101
1102 hi = hashTypeInitIterator(o);
1103 while (hashTypeNext(hi) != C_ERR) {
1104 if (count == 0) {
1105 int cmd_items = (items > AOF_REWRITE_ITEMS_PER_CMD) ?
1106 AOF_REWRITE_ITEMS_PER_CMD : items;
1107
1108 if (rioWriteBulkCount(r,'*',2+cmd_items*2) == 0) return 0;
1109 if (rioWriteBulkString(r,"HMSET",5) == 0) return 0;
1110 if (rioWriteBulkObject(r,key) == 0) return 0;
1111 }
1112
1113 if (rioWriteHashIteratorCursor(r, hi, OBJ_HASH_KEY) == 0) return 0;
1114 if (rioWriteHashIteratorCursor(r, hi, OBJ_HASH_VALUE) == 0) return 0;
1115 if (++count == AOF_REWRITE_ITEMS_PER_CMD) count = 0;
1116 items--;
1117 }
1118
1119 hashTypeReleaseIterator(hi);
1120
1121 return 1;
1122 }
1123
1124 /* Helper for rewriteStreamObject() that generates a bulk string into the
1125 * AOF representing the ID 'id'. */
rioWriteBulkStreamID(rio * r,streamID * id)1126 int rioWriteBulkStreamID(rio *r,streamID *id) {
1127 int retval;
1128
1129 sds replyid = sdscatfmt(sdsempty(),"%U-%U",id->ms,id->seq);
1130 if ((retval = rioWriteBulkString(r,replyid,sdslen(replyid))) == 0) return 0;
1131 sdsfree(replyid);
1132 return retval;
1133 }
1134
1135 /* Helper for rewriteStreamObject(): emit the XCLAIM needed in order to
1136 * add the message described by 'nack' having the id 'rawid', into the pending
1137 * list of the specified consumer. All this in the context of the specified
1138 * key and group. */
rioWriteStreamPendingEntry(rio * r,robj * key,const char * groupname,size_t groupname_len,streamConsumer * consumer,unsigned char * rawid,streamNACK * nack)1139 int rioWriteStreamPendingEntry(rio *r, robj *key, const char *groupname, size_t groupname_len, streamConsumer *consumer, unsigned char *rawid, streamNACK *nack) {
1140 /* XCLAIM <key> <group> <consumer> 0 <id> TIME <milliseconds-unix-time>
1141 RETRYCOUNT <count> JUSTID FORCE. */
1142 streamID id;
1143 streamDecodeID(rawid,&id);
1144 if (rioWriteBulkCount(r,'*',12) == 0) return 0;
1145 if (rioWriteBulkString(r,"XCLAIM",6) == 0) return 0;
1146 if (rioWriteBulkObject(r,key) == 0) return 0;
1147 if (rioWriteBulkString(r,groupname,groupname_len) == 0) return 0;
1148 if (rioWriteBulkString(r,consumer->name,sdslen(consumer->name)) == 0) return 0;
1149 if (rioWriteBulkString(r,"0",1) == 0) return 0;
1150 if (rioWriteBulkStreamID(r,&id) == 0) return 0;
1151 if (rioWriteBulkString(r,"TIME",4) == 0) return 0;
1152 if (rioWriteBulkLongLong(r,nack->delivery_time) == 0) return 0;
1153 if (rioWriteBulkString(r,"RETRYCOUNT",10) == 0) return 0;
1154 if (rioWriteBulkLongLong(r,nack->delivery_count) == 0) return 0;
1155 if (rioWriteBulkString(r,"JUSTID",6) == 0) return 0;
1156 if (rioWriteBulkString(r,"FORCE",5) == 0) return 0;
1157 return 1;
1158 }
1159
1160 /* Emit the commands needed to rebuild a stream object.
1161 * The function returns 0 on error, 1 on success. */
rewriteStreamObject(rio * r,robj * key,robj * o)1162 int rewriteStreamObject(rio *r, robj *key, robj *o) {
1163 stream *s = o->ptr;
1164 streamIterator si;
1165 streamIteratorStart(&si,s,NULL,NULL,0);
1166 streamID id;
1167 int64_t numfields;
1168
1169 if (s->length) {
1170 /* Reconstruct the stream data using XADD commands. */
1171 while(streamIteratorGetID(&si,&id,&numfields)) {
1172 /* Emit a two elements array for each item. The first is
1173 * the ID, the second is an array of field-value pairs. */
1174
1175 /* Emit the XADD <key> <id> ...fields... command. */
1176 if (rioWriteBulkCount(r,'*',3+numfields*2) == 0) return 0;
1177 if (rioWriteBulkString(r,"XADD",4) == 0) return 0;
1178 if (rioWriteBulkObject(r,key) == 0) return 0;
1179 if (rioWriteBulkStreamID(r,&id) == 0) return 0;
1180 while(numfields--) {
1181 unsigned char *field, *value;
1182 int64_t field_len, value_len;
1183 streamIteratorGetField(&si,&field,&value,&field_len,&value_len);
1184 if (rioWriteBulkString(r,(char*)field,field_len) == 0) return 0;
1185 if (rioWriteBulkString(r,(char*)value,value_len) == 0) return 0;
1186 }
1187 }
1188 } else {
1189 /* Use the XADD MAXLEN 0 trick to generate an empty stream if
1190 * the key we are serializing is an empty string, which is possible
1191 * for the Stream type. */
1192 if (rioWriteBulkCount(r,'*',7) == 0) return 0;
1193 if (rioWriteBulkString(r,"XADD",4) == 0) return 0;
1194 if (rioWriteBulkObject(r,key) == 0) return 0;
1195 if (rioWriteBulkString(r,"MAXLEN",6) == 0) return 0;
1196 if (rioWriteBulkString(r,"0",1) == 0) return 0;
1197 if (rioWriteBulkStreamID(r,&s->last_id) == 0) return 0;
1198 if (rioWriteBulkString(r,"x",1) == 0) return 0;
1199 if (rioWriteBulkString(r,"y",1) == 0) return 0;
1200 }
1201
1202 /* Append XSETID after XADD, make sure lastid is correct,
1203 * in case of XDEL lastid. */
1204 if (rioWriteBulkCount(r,'*',3) == 0) return 0;
1205 if (rioWriteBulkString(r,"XSETID",6) == 0) return 0;
1206 if (rioWriteBulkObject(r,key) == 0) return 0;
1207 if (rioWriteBulkStreamID(r,&s->last_id) == 0) return 0;
1208
1209
1210 /* Create all the stream consumer groups. */
1211 if (s->cgroups) {
1212 raxIterator ri;
1213 raxStart(&ri,s->cgroups);
1214 raxSeek(&ri,"^",NULL,0);
1215 while(raxNext(&ri)) {
1216 streamCG *group = ri.data;
1217 /* Emit the XGROUP CREATE in order to create the group. */
1218 if (rioWriteBulkCount(r,'*',5) == 0) return 0;
1219 if (rioWriteBulkString(r,"XGROUP",6) == 0) return 0;
1220 if (rioWriteBulkString(r,"CREATE",6) == 0) return 0;
1221 if (rioWriteBulkObject(r,key) == 0) return 0;
1222 if (rioWriteBulkString(r,(char*)ri.key,ri.key_len) == 0) return 0;
1223 if (rioWriteBulkStreamID(r,&group->last_id) == 0) return 0;
1224
1225 /* Generate XCLAIMs for each consumer that happens to
1226 * have pending entries. Empty consumers have no semantical
1227 * value so they are discarded. */
1228 raxIterator ri_cons;
1229 raxStart(&ri_cons,group->consumers);
1230 raxSeek(&ri_cons,"^",NULL,0);
1231 while(raxNext(&ri_cons)) {
1232 streamConsumer *consumer = ri_cons.data;
1233 /* For the current consumer, iterate all the PEL entries
1234 * to emit the XCLAIM protocol. */
1235 raxIterator ri_pel;
1236 raxStart(&ri_pel,consumer->pel);
1237 raxSeek(&ri_pel,"^",NULL,0);
1238 while(raxNext(&ri_pel)) {
1239 streamNACK *nack = ri_pel.data;
1240 if (rioWriteStreamPendingEntry(r,key,(char*)ri.key,
1241 ri.key_len,consumer,
1242 ri_pel.key,nack) == 0)
1243 {
1244 return 0;
1245 }
1246 }
1247 raxStop(&ri_pel);
1248 }
1249 raxStop(&ri_cons);
1250 }
1251 raxStop(&ri);
1252 }
1253
1254 streamIteratorStop(&si);
1255 return 1;
1256 }
1257
1258 /* Call the module type callback in order to rewrite a data type
1259 * that is exported by a module and is not handled by Redis itself.
1260 * The function returns 0 on error, 1 on success. */
rewriteModuleObject(rio * r,robj * key,robj * o)1261 int rewriteModuleObject(rio *r, robj *key, robj *o) {
1262 RedisModuleIO io;
1263 moduleValue *mv = o->ptr;
1264 moduleType *mt = mv->type;
1265 moduleInitIOContext(io,mt,r,key);
1266 mt->aof_rewrite(&io,key,mv->value);
1267 if (io.ctx) {
1268 moduleFreeContext(io.ctx);
1269 zfree(io.ctx);
1270 }
1271 return io.error ? 0 : 1;
1272 }
1273
1274 /* This function is called by the child rewriting the AOF file to read
1275 * the difference accumulated from the parent into a buffer, that is
1276 * concatenated at the end of the rewrite. */
aofReadDiffFromParent(void)1277 ssize_t aofReadDiffFromParent(void) {
1278 char buf[65536]; /* Default pipe buffer size on most Linux systems. */
1279 ssize_t nread, total = 0;
1280
1281 while ((nread =
1282 read(server.aof_pipe_read_data_from_parent,buf,sizeof(buf))) > 0) {
1283 server.aof_child_diff = sdscatlen(server.aof_child_diff,buf,nread);
1284 total += nread;
1285 }
1286 return total;
1287 }
1288
rewriteAppendOnlyFileRio(rio * aof)1289 int rewriteAppendOnlyFileRio(rio *aof) {
1290 dictIterator *di = NULL;
1291 dictEntry *de;
1292 size_t processed = 0;
1293 int j;
1294
1295 for (j = 0; j < server.dbnum; j++) {
1296 char selectcmd[] = "*2\r\n$6\r\nSELECT\r\n";
1297 redisDb *db = server.db+j;
1298 dict *d = db->dict;
1299 if (dictSize(d) == 0) continue;
1300 di = dictGetSafeIterator(d);
1301
1302 /* SELECT the new DB */
1303 if (rioWrite(aof,selectcmd,sizeof(selectcmd)-1) == 0) goto werr;
1304 if (rioWriteBulkLongLong(aof,j) == 0) goto werr;
1305
1306 /* Iterate this DB writing every entry */
1307 while((de = dictNext(di)) != NULL) {
1308 sds keystr;
1309 robj key, *o;
1310 long long expiretime;
1311
1312 keystr = dictGetKey(de);
1313 o = dictGetVal(de);
1314 initStaticStringObject(key,keystr);
1315
1316 expiretime = getExpire(db,&key);
1317
1318 /* Save the key and associated value */
1319 if (o->type == OBJ_STRING) {
1320 /* Emit a SET command */
1321 char cmd[]="*3\r\n$3\r\nSET\r\n";
1322 if (rioWrite(aof,cmd,sizeof(cmd)-1) == 0) goto werr;
1323 /* Key and value */
1324 if (rioWriteBulkObject(aof,&key) == 0) goto werr;
1325 if (rioWriteBulkObject(aof,o) == 0) goto werr;
1326 } else if (o->type == OBJ_LIST) {
1327 if (rewriteListObject(aof,&key,o) == 0) goto werr;
1328 } else if (o->type == OBJ_SET) {
1329 if (rewriteSetObject(aof,&key,o) == 0) goto werr;
1330 } else if (o->type == OBJ_ZSET) {
1331 if (rewriteSortedSetObject(aof,&key,o) == 0) goto werr;
1332 } else if (o->type == OBJ_HASH) {
1333 if (rewriteHashObject(aof,&key,o) == 0) goto werr;
1334 } else if (o->type == OBJ_STREAM) {
1335 if (rewriteStreamObject(aof,&key,o) == 0) goto werr;
1336 } else if (o->type == OBJ_MODULE) {
1337 if (rewriteModuleObject(aof,&key,o) == 0) goto werr;
1338 } else {
1339 serverPanic("Unknown object type");
1340 }
1341 /* Save the expire time */
1342 if (expiretime != -1) {
1343 char cmd[]="*3\r\n$9\r\nPEXPIREAT\r\n";
1344 if (rioWrite(aof,cmd,sizeof(cmd)-1) == 0) goto werr;
1345 if (rioWriteBulkObject(aof,&key) == 0) goto werr;
1346 if (rioWriteBulkLongLong(aof,expiretime) == 0) goto werr;
1347 }
1348 /* Read some diff from the parent process from time to time. */
1349 if (aof->processed_bytes > processed+AOF_READ_DIFF_INTERVAL_BYTES) {
1350 processed = aof->processed_bytes;
1351 aofReadDiffFromParent();
1352 }
1353 }
1354 dictReleaseIterator(di);
1355 di = NULL;
1356 }
1357 return C_OK;
1358
1359 werr:
1360 if (di) dictReleaseIterator(di);
1361 return C_ERR;
1362 }
1363
1364 /* Write a sequence of commands able to fully rebuild the dataset into
1365 * "filename". Used both by REWRITEAOF and BGREWRITEAOF.
1366 *
1367 * In order to minimize the number of commands needed in the rewritten
1368 * log Redis uses variadic commands when possible, such as RPUSH, SADD
1369 * and ZADD. However at max AOF_REWRITE_ITEMS_PER_CMD items per time
1370 * are inserted using a single command. */
rewriteAppendOnlyFile(char * filename)1371 int rewriteAppendOnlyFile(char *filename) {
1372 rio aof;
1373 FILE *fp;
1374 char tmpfile[256];
1375 char byte;
1376
1377 /* Note that we have to use a different temp name here compared to the
1378 * one used by rewriteAppendOnlyFileBackground() function. */
1379 snprintf(tmpfile,256,"temp-rewriteaof-%d.aof", (int) getpid());
1380 fp = fopen(tmpfile,"w");
1381 if (!fp) {
1382 serverLog(LL_WARNING, "Opening the temp file for AOF rewrite in rewriteAppendOnlyFile(): %s", strerror(errno));
1383 return C_ERR;
1384 }
1385
1386 server.aof_child_diff = sdsempty();
1387 rioInitWithFile(&aof,fp);
1388
1389 if (server.aof_rewrite_incremental_fsync)
1390 rioSetAutoSync(&aof,REDIS_AUTOSYNC_BYTES);
1391
1392 if (server.aof_use_rdb_preamble) {
1393 int error;
1394 if (rdbSaveRio(&aof,&error,RDB_SAVE_AOF_PREAMBLE,NULL) == C_ERR) {
1395 errno = error;
1396 goto werr;
1397 }
1398 } else {
1399 if (rewriteAppendOnlyFileRio(&aof) == C_ERR) goto werr;
1400 }
1401
1402 /* Do an initial slow fsync here while the parent is still sending
1403 * data, in order to make the next final fsync faster. */
1404 if (fflush(fp) == EOF) goto werr;
1405 if (fsync(fileno(fp)) == -1) goto werr;
1406
1407 /* Read again a few times to get more data from the parent.
1408 * We can't read forever (the server may receive data from clients
1409 * faster than it is able to send data to the child), so we try to read
1410 * some more data in a loop as soon as there is a good chance more data
1411 * will come. If it looks like we are wasting time, we abort (this
1412 * happens after 20 ms without new data). */
1413 int nodata = 0;
1414 mstime_t start = mstime();
1415 while(mstime()-start < 1000 && nodata < 20) {
1416 if (aeWait(server.aof_pipe_read_data_from_parent, AE_READABLE, 1) <= 0)
1417 {
1418 nodata++;
1419 continue;
1420 }
1421 nodata = 0; /* Start counting from zero, we stop on N *contiguous*
1422 timeouts. */
1423 aofReadDiffFromParent();
1424 }
1425
1426 /* Ask the master to stop sending diffs. */
1427 if (write(server.aof_pipe_write_ack_to_parent,"!",1) != 1) goto werr;
1428 if (anetNonBlock(NULL,server.aof_pipe_read_ack_from_parent) != ANET_OK)
1429 goto werr;
1430 /* We read the ACK from the server using a 10 seconds timeout. Normally
1431 * it should reply ASAP, but just in case we lose its reply, we are sure
1432 * the child will eventually get terminated. */
1433 if (syncRead(server.aof_pipe_read_ack_from_parent,&byte,1,5000) != 1 ||
1434 byte != '!') goto werr;
1435 serverLog(LL_NOTICE,"Parent agreed to stop sending diffs. Finalizing AOF...");
1436
1437 /* Read the final diff if any. */
1438 aofReadDiffFromParent();
1439
1440 /* Write the received diff to the file. */
1441 serverLog(LL_NOTICE,
1442 "Concatenating %.2f MB of AOF diff received from parent.",
1443 (double) sdslen(server.aof_child_diff) / (1024*1024));
1444 if (rioWrite(&aof,server.aof_child_diff,sdslen(server.aof_child_diff)) == 0)
1445 goto werr;
1446
1447 /* Make sure data will not remain on the OS's output buffers */
1448 if (fflush(fp) == EOF) goto werr;
1449 if (fsync(fileno(fp)) == -1) goto werr;
1450 if (fclose(fp) == EOF) goto werr;
1451
1452 /* Use RENAME to make sure the DB file is changed atomically only
1453 * if the generate DB file is ok. */
1454 if (rename(tmpfile,filename) == -1) {
1455 serverLog(LL_WARNING,"Error moving temp append only file on the final destination: %s", strerror(errno));
1456 unlink(tmpfile);
1457 return C_ERR;
1458 }
1459 serverLog(LL_NOTICE,"SYNC append only file rewrite performed");
1460 return C_OK;
1461
1462 werr:
1463 serverLog(LL_WARNING,"Write error writing append only file on disk: %s", strerror(errno));
1464 fclose(fp);
1465 unlink(tmpfile);
1466 return C_ERR;
1467 }
1468
1469 /* ----------------------------------------------------------------------------
1470 * AOF rewrite pipes for IPC
1471 * -------------------------------------------------------------------------- */
1472
1473 /* This event handler is called when the AOF rewriting child sends us a
1474 * single '!' char to signal we should stop sending buffer diffs. The
1475 * parent sends a '!' as well to acknowledge. */
aofChildPipeReadable(aeEventLoop * el,int fd,void * privdata,int mask)1476 void aofChildPipeReadable(aeEventLoop *el, int fd, void *privdata, int mask) {
1477 char byte;
1478 UNUSED(el);
1479 UNUSED(privdata);
1480 UNUSED(mask);
1481
1482 if (read(fd,&byte,1) == 1 && byte == '!') {
1483 serverLog(LL_NOTICE,"AOF rewrite child asks to stop sending diffs.");
1484 server.aof_stop_sending_diff = 1;
1485 if (write(server.aof_pipe_write_ack_to_child,"!",1) != 1) {
1486 /* If we can't send the ack, inform the user, but don't try again
1487 * since in the other side the children will use a timeout if the
1488 * kernel can't buffer our write, or, the children was
1489 * terminated. */
1490 serverLog(LL_WARNING,"Can't send ACK to AOF child: %s",
1491 strerror(errno));
1492 }
1493 }
1494 /* Remove the handler since this can be called only one time during a
1495 * rewrite. */
1496 aeDeleteFileEvent(server.el,server.aof_pipe_read_ack_from_child,AE_READABLE);
1497 }
1498
1499 /* Create the pipes used for parent - child process IPC during rewrite.
1500 * We have a data pipe used to send AOF incremental diffs to the child,
1501 * and two other pipes used by the children to signal it finished with
1502 * the rewrite so no more data should be written, and another for the
1503 * parent to acknowledge it understood this new condition. */
aofCreatePipes(void)1504 int aofCreatePipes(void) {
1505 int fds[6] = {-1, -1, -1, -1, -1, -1};
1506 int j;
1507
1508 if (pipe(fds) == -1) goto error; /* parent -> children data. */
1509 if (pipe(fds+2) == -1) goto error; /* children -> parent ack. */
1510 if (pipe(fds+4) == -1) goto error; /* parent -> children ack. */
1511 /* Parent -> children data is non blocking. */
1512 if (anetNonBlock(NULL,fds[0]) != ANET_OK) goto error;
1513 if (anetNonBlock(NULL,fds[1]) != ANET_OK) goto error;
1514 if (aeCreateFileEvent(server.el, fds[2], AE_READABLE, aofChildPipeReadable, NULL) == AE_ERR) goto error;
1515
1516 server.aof_pipe_write_data_to_child = fds[1];
1517 server.aof_pipe_read_data_from_parent = fds[0];
1518 server.aof_pipe_write_ack_to_parent = fds[3];
1519 server.aof_pipe_read_ack_from_child = fds[2];
1520 server.aof_pipe_write_ack_to_child = fds[5];
1521 server.aof_pipe_read_ack_from_parent = fds[4];
1522 server.aof_stop_sending_diff = 0;
1523 return C_OK;
1524
1525 error:
1526 serverLog(LL_WARNING,"Error opening /setting AOF rewrite IPC pipes: %s",
1527 strerror(errno));
1528 for (j = 0; j < 6; j++) if(fds[j] != -1) close(fds[j]);
1529 return C_ERR;
1530 }
1531
aofClosePipes(void)1532 void aofClosePipes(void) {
1533 aeDeleteFileEvent(server.el,server.aof_pipe_read_ack_from_child,AE_READABLE);
1534 aeDeleteFileEvent(server.el,server.aof_pipe_write_data_to_child,AE_WRITABLE);
1535 close(server.aof_pipe_write_data_to_child);
1536 close(server.aof_pipe_read_data_from_parent);
1537 close(server.aof_pipe_write_ack_to_parent);
1538 close(server.aof_pipe_read_ack_from_child);
1539 close(server.aof_pipe_write_ack_to_child);
1540 close(server.aof_pipe_read_ack_from_parent);
1541 }
1542
1543 /* ----------------------------------------------------------------------------
1544 * AOF background rewrite
1545 * ------------------------------------------------------------------------- */
1546
1547 /* This is how rewriting of the append only file in background works:
1548 *
1549 * 1) The user calls BGREWRITEAOF
1550 * 2) Redis calls this function, that forks():
1551 * 2a) the child rewrite the append only file in a temp file.
1552 * 2b) the parent accumulates differences in server.aof_rewrite_buf.
1553 * 3) When the child finished '2a' exists.
1554 * 4) The parent will trap the exit code, if it's OK, will append the
1555 * data accumulated into server.aof_rewrite_buf into the temp file, and
1556 * finally will rename(2) the temp file in the actual file name.
1557 * The the new file is reopened as the new append only file. Profit!
1558 */
rewriteAppendOnlyFileBackground(void)1559 int rewriteAppendOnlyFileBackground(void) {
1560 pid_t childpid;
1561 long long start;
1562
1563 if (server.aof_child_pid != -1 || server.rdb_child_pid != -1) return C_ERR;
1564 if (aofCreatePipes() != C_OK) return C_ERR;
1565 openChildInfoPipe();
1566 start = ustime();
1567 if ((childpid = fork()) == 0) {
1568 char tmpfile[256];
1569
1570 /* Child */
1571 closeListeningSockets(0);
1572 resetCpuAffinity("aof-rewrite");
1573 redisSetProcTitle("redis-aof-rewrite");
1574 snprintf(tmpfile,256,"temp-rewriteaof-bg-%d.aof", (int) getpid());
1575 if (rewriteAppendOnlyFile(tmpfile) == C_OK) {
1576 size_t private_dirty = zmalloc_get_private_dirty(-1);
1577
1578 if (private_dirty) {
1579 serverLog(LL_NOTICE,
1580 "AOF rewrite: %zu MB of memory used by copy-on-write",
1581 private_dirty/(1024*1024));
1582 }
1583
1584 server.child_info_data.cow_size = private_dirty;
1585 sendChildInfo(CHILD_INFO_TYPE_AOF);
1586 exitFromChild(0);
1587 } else {
1588 exitFromChild(1);
1589 }
1590 } else {
1591 /* Parent */
1592 server.stat_fork_time = ustime()-start;
1593 server.stat_fork_rate = (double) zmalloc_used_memory() * 1000000 / server.stat_fork_time / (1024*1024*1024); /* GB per second. */
1594 latencyAddSampleIfNeeded("fork",server.stat_fork_time/1000);
1595 if (childpid == -1) {
1596 closeChildInfoPipe();
1597 serverLog(LL_WARNING,
1598 "Can't rewrite append only file in background: fork: %s",
1599 strerror(errno));
1600 aofClosePipes();
1601 return C_ERR;
1602 }
1603 serverLog(LL_NOTICE,
1604 "Background append only file rewriting started by pid %d",childpid);
1605 server.aof_rewrite_scheduled = 0;
1606 server.aof_rewrite_time_start = time(NULL);
1607 server.aof_child_pid = childpid;
1608 updateDictResizePolicy();
1609 /* We set appendseldb to -1 in order to force the next call to the
1610 * feedAppendOnlyFile() to issue a SELECT command, so the differences
1611 * accumulated by the parent into server.aof_rewrite_buf will start
1612 * with a SELECT statement and it will be safe to merge. */
1613 server.aof_selected_db = -1;
1614 replicationScriptCacheFlush();
1615 return C_OK;
1616 }
1617 return C_OK; /* unreached */
1618 }
1619
bgrewriteaofCommand(client * c)1620 void bgrewriteaofCommand(client *c) {
1621 if (server.aof_child_pid != -1) {
1622 addReplyError(c,"Background append only file rewriting already in progress");
1623 } else if (server.rdb_child_pid != -1) {
1624 server.aof_rewrite_scheduled = 1;
1625 addReplyStatus(c,"Background append only file rewriting scheduled");
1626 } else if (rewriteAppendOnlyFileBackground() == C_OK) {
1627 addReplyStatus(c,"Background append only file rewriting started");
1628 } else {
1629 addReply(c,shared.err);
1630 }
1631 }
1632
aofRemoveTempFile(pid_t childpid)1633 void aofRemoveTempFile(pid_t childpid) {
1634 char tmpfile[256];
1635
1636 snprintf(tmpfile,256,"temp-rewriteaof-bg-%d.aof", (int) childpid);
1637 unlink(tmpfile);
1638 }
1639
1640 /* Update the server.aof_current_size field explicitly using stat(2)
1641 * to check the size of the file. This is useful after a rewrite or after
1642 * a restart, normally the size is updated just adding the write length
1643 * to the current length, that is much faster. */
aofUpdateCurrentSize(void)1644 void aofUpdateCurrentSize(void) {
1645 struct redis_stat sb;
1646 mstime_t latency;
1647
1648 latencyStartMonitor(latency);
1649 if (redis_fstat(server.aof_fd,&sb) == -1) {
1650 serverLog(LL_WARNING,"Unable to obtain the AOF file length. stat: %s",
1651 strerror(errno));
1652 } else {
1653 server.aof_current_size = sb.st_size;
1654 }
1655 latencyEndMonitor(latency);
1656 latencyAddSampleIfNeeded("aof-fstat",latency);
1657 }
1658
1659 /* A background append only file rewriting (BGREWRITEAOF) terminated its work.
1660 * Handle this. */
backgroundRewriteDoneHandler(int exitcode,int bysignal)1661 void backgroundRewriteDoneHandler(int exitcode, int bysignal) {
1662 if (!bysignal && exitcode == 0) {
1663 int newfd, oldfd;
1664 char tmpfile[256];
1665 long long now = ustime();
1666 mstime_t latency;
1667
1668 serverLog(LL_NOTICE,
1669 "Background AOF rewrite terminated with success");
1670
1671 /* Flush the differences accumulated by the parent to the
1672 * rewritten AOF. */
1673 latencyStartMonitor(latency);
1674 snprintf(tmpfile,256,"temp-rewriteaof-bg-%d.aof",
1675 (int)server.aof_child_pid);
1676 newfd = open(tmpfile,O_WRONLY|O_APPEND);
1677 if (newfd == -1) {
1678 serverLog(LL_WARNING,
1679 "Unable to open the temporary AOF produced by the child: %s", strerror(errno));
1680 goto cleanup;
1681 }
1682
1683 if (aofRewriteBufferWrite(newfd) == -1) {
1684 serverLog(LL_WARNING,
1685 "Error trying to flush the parent diff to the rewritten AOF: %s", strerror(errno));
1686 close(newfd);
1687 goto cleanup;
1688 }
1689 latencyEndMonitor(latency);
1690 latencyAddSampleIfNeeded("aof-rewrite-diff-write",latency);
1691
1692 serverLog(LL_NOTICE,
1693 "Residual parent diff successfully flushed to the rewritten AOF (%.2f MB)", (double) aofRewriteBufferSize() / (1024*1024));
1694
1695 /* The only remaining thing to do is to rename the temporary file to
1696 * the configured file and switch the file descriptor used to do AOF
1697 * writes. We don't want close(2) or rename(2) calls to block the
1698 * server on old file deletion.
1699 *
1700 * There are two possible scenarios:
1701 *
1702 * 1) AOF is DISABLED and this was a one time rewrite. The temporary
1703 * file will be renamed to the configured file. When this file already
1704 * exists, it will be unlinked, which may block the server.
1705 *
1706 * 2) AOF is ENABLED and the rewritten AOF will immediately start
1707 * receiving writes. After the temporary file is renamed to the
1708 * configured file, the original AOF file descriptor will be closed.
1709 * Since this will be the last reference to that file, closing it
1710 * causes the underlying file to be unlinked, which may block the
1711 * server.
1712 *
1713 * To mitigate the blocking effect of the unlink operation (either
1714 * caused by rename(2) in scenario 1, or by close(2) in scenario 2), we
1715 * use a background thread to take care of this. First, we
1716 * make scenario 1 identical to scenario 2 by opening the target file
1717 * when it exists. The unlink operation after the rename(2) will then
1718 * be executed upon calling close(2) for its descriptor. Everything to
1719 * guarantee atomicity for this switch has already happened by then, so
1720 * we don't care what the outcome or duration of that close operation
1721 * is, as long as the file descriptor is released again. */
1722 if (server.aof_fd == -1) {
1723 /* AOF disabled */
1724
1725 /* Don't care if this fails: oldfd will be -1 and we handle that.
1726 * One notable case of -1 return is if the old file does
1727 * not exist. */
1728 oldfd = open(server.aof_filename,O_RDONLY|O_NONBLOCK);
1729 } else {
1730 /* AOF enabled */
1731 oldfd = -1; /* We'll set this to the current AOF filedes later. */
1732 }
1733
1734 /* Rename the temporary file. This will not unlink the target file if
1735 * it exists, because we reference it with "oldfd". */
1736 latencyStartMonitor(latency);
1737 if (rename(tmpfile,server.aof_filename) == -1) {
1738 serverLog(LL_WARNING,
1739 "Error trying to rename the temporary AOF file %s into %s: %s",
1740 tmpfile,
1741 server.aof_filename,
1742 strerror(errno));
1743 close(newfd);
1744 if (oldfd != -1) close(oldfd);
1745 goto cleanup;
1746 }
1747 latencyEndMonitor(latency);
1748 latencyAddSampleIfNeeded("aof-rename",latency);
1749
1750 if (server.aof_fd == -1) {
1751 /* AOF disabled, we don't need to set the AOF file descriptor
1752 * to this new file, so we can close it. */
1753 close(newfd);
1754 } else {
1755 /* AOF enabled, replace the old fd with the new one. */
1756 oldfd = server.aof_fd;
1757 server.aof_fd = newfd;
1758 if (server.aof_fsync == AOF_FSYNC_ALWAYS)
1759 redis_fsync(newfd);
1760 else if (server.aof_fsync == AOF_FSYNC_EVERYSEC)
1761 aof_background_fsync(newfd);
1762 server.aof_selected_db = -1; /* Make sure SELECT is re-issued */
1763 aofUpdateCurrentSize();
1764 server.aof_rewrite_base_size = server.aof_current_size;
1765 server.aof_current_size = server.aof_current_size;
1766
1767 /* Clear regular AOF buffer since its contents was just written to
1768 * the new AOF from the background rewrite buffer. */
1769 sdsfree(server.aof_buf);
1770 server.aof_buf = sdsempty();
1771 }
1772
1773 server.aof_lastbgrewrite_status = C_OK;
1774
1775 serverLog(LL_NOTICE, "Background AOF rewrite finished successfully");
1776 /* Change state from WAIT_REWRITE to ON if needed */
1777 if (server.aof_state == AOF_WAIT_REWRITE)
1778 server.aof_state = AOF_ON;
1779
1780 /* Asynchronously close the overwritten AOF. */
1781 if (oldfd != -1) bioCreateBackgroundJob(BIO_CLOSE_FILE,(void*)(long)oldfd,NULL,NULL);
1782
1783 serverLog(LL_VERBOSE,
1784 "Background AOF rewrite signal handler took %lldus", ustime()-now);
1785 } else if (!bysignal && exitcode != 0) {
1786 /* SIGUSR1 is whitelisted, so we have a way to kill a child without
1787 * tirggering an error condition. */
1788 if (bysignal != SIGUSR1)
1789 server.aof_lastbgrewrite_status = C_ERR;
1790 serverLog(LL_WARNING,
1791 "Background AOF rewrite terminated with error");
1792 } else {
1793 server.aof_lastbgrewrite_status = C_ERR;
1794
1795 serverLog(LL_WARNING,
1796 "Background AOF rewrite terminated by signal %d", bysignal);
1797 }
1798
1799 cleanup:
1800 aofClosePipes();
1801 aofRewriteBufferReset();
1802 aofRemoveTempFile(server.aof_child_pid);
1803 server.aof_child_pid = -1;
1804 server.aof_rewrite_time_last = time(NULL)-server.aof_rewrite_time_start;
1805 server.aof_rewrite_time_start = -1;
1806 /* Schedule a new rewrite if we are waiting for it to switch the AOF ON. */
1807 if (server.aof_state == AOF_WAIT_REWRITE)
1808 server.aof_rewrite_scheduled = 1;
1809 }
1810