]> git.saurik.com Git - redis.git/blob - src/redis.c
Revert "Now maxmemory, VM, and everything else uses the fast RSS memory used estimati...
[redis.git] / src / redis.c
1 /*
2 * Copyright (c) 2009-2010, 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
32 #ifdef HAVE_BACKTRACE
33 #include <execinfo.h>
34 #include <ucontext.h>
35 #endif /* HAVE_BACKTRACE */
36
37 #include <time.h>
38 #include <signal.h>
39 #include <sys/wait.h>
40 #include <errno.h>
41 #include <assert.h>
42 #include <ctype.h>
43 #include <stdarg.h>
44 #include <arpa/inet.h>
45 #include <sys/stat.h>
46 #include <fcntl.h>
47 #include <sys/time.h>
48 #include <sys/resource.h>
49 #include <sys/uio.h>
50 #include <limits.h>
51 #include <float.h>
52 #include <math.h>
53 #include <pthread.h>
54 #include <sys/resource.h>
55
56 /* Our shared "common" objects */
57
58 struct sharedObjectsStruct shared;
59
60 /* Global vars that are actally used as constants. The following double
61 * values are used for double on-disk serialization, and are initialized
62 * at runtime to avoid strange compiler optimizations. */
63
64 double R_Zero, R_PosInf, R_NegInf, R_Nan;
65
66 /*================================= Globals ================================= */
67
68 /* Global vars */
69 struct redisServer server; /* server global state */
70 struct redisCommand *commandTable;
71 struct redisCommand readonlyCommandTable[] = {
72 {"get",getCommand,2,0,NULL,1,1,1},
73 {"set",setCommand,3,REDIS_CMD_DENYOOM,NULL,0,0,0},
74 {"setnx",setnxCommand,3,REDIS_CMD_DENYOOM,NULL,0,0,0},
75 {"setex",setexCommand,4,REDIS_CMD_DENYOOM,NULL,0,0,0},
76 {"append",appendCommand,3,REDIS_CMD_DENYOOM,NULL,1,1,1},
77 {"substr",substrCommand,4,0,NULL,1,1,1},
78 {"strlen",strlenCommand,2,0,NULL,1,1,1},
79 {"del",delCommand,-2,0,NULL,0,0,0},
80 {"exists",existsCommand,2,0,NULL,1,1,1},
81 {"incr",incrCommand,2,REDIS_CMD_DENYOOM,NULL,1,1,1},
82 {"decr",decrCommand,2,REDIS_CMD_DENYOOM,NULL,1,1,1},
83 {"mget",mgetCommand,-2,0,NULL,1,-1,1},
84 {"rpush",rpushCommand,3,REDIS_CMD_DENYOOM,NULL,1,1,1},
85 {"lpush",lpushCommand,3,REDIS_CMD_DENYOOM,NULL,1,1,1},
86 {"rpushx",rpushxCommand,3,REDIS_CMD_DENYOOM,NULL,1,1,1},
87 {"lpushx",lpushxCommand,3,REDIS_CMD_DENYOOM,NULL,1,1,1},
88 {"linsert",linsertCommand,5,REDIS_CMD_DENYOOM,NULL,1,1,1},
89 {"rpop",rpopCommand,2,0,NULL,1,1,1},
90 {"lpop",lpopCommand,2,0,NULL,1,1,1},
91 {"brpop",brpopCommand,-3,0,NULL,1,1,1},
92 {"blpop",blpopCommand,-3,0,NULL,1,1,1},
93 {"llen",llenCommand,2,0,NULL,1,1,1},
94 {"lindex",lindexCommand,3,0,NULL,1,1,1},
95 {"lset",lsetCommand,4,REDIS_CMD_DENYOOM,NULL,1,1,1},
96 {"lrange",lrangeCommand,4,0,NULL,1,1,1},
97 {"ltrim",ltrimCommand,4,0,NULL,1,1,1},
98 {"lrem",lremCommand,4,0,NULL,1,1,1},
99 {"rpoplpush",rpoplpushcommand,3,REDIS_CMD_DENYOOM,NULL,1,2,1},
100 {"sadd",saddCommand,3,REDIS_CMD_DENYOOM,NULL,1,1,1},
101 {"srem",sremCommand,3,0,NULL,1,1,1},
102 {"smove",smoveCommand,4,0,NULL,1,2,1},
103 {"sismember",sismemberCommand,3,0,NULL,1,1,1},
104 {"scard",scardCommand,2,0,NULL,1,1,1},
105 {"spop",spopCommand,2,0,NULL,1,1,1},
106 {"srandmember",srandmemberCommand,2,0,NULL,1,1,1},
107 {"sinter",sinterCommand,-2,REDIS_CMD_DENYOOM,NULL,1,-1,1},
108 {"sinterstore",sinterstoreCommand,-3,REDIS_CMD_DENYOOM,NULL,2,-1,1},
109 {"sunion",sunionCommand,-2,REDIS_CMD_DENYOOM,NULL,1,-1,1},
110 {"sunionstore",sunionstoreCommand,-3,REDIS_CMD_DENYOOM,NULL,2,-1,1},
111 {"sdiff",sdiffCommand,-2,REDIS_CMD_DENYOOM,NULL,1,-1,1},
112 {"sdiffstore",sdiffstoreCommand,-3,REDIS_CMD_DENYOOM,NULL,2,-1,1},
113 {"smembers",sinterCommand,2,0,NULL,1,1,1},
114 {"zadd",zaddCommand,4,REDIS_CMD_DENYOOM,NULL,1,1,1},
115 {"zincrby",zincrbyCommand,4,REDIS_CMD_DENYOOM,NULL,1,1,1},
116 {"zrem",zremCommand,3,0,NULL,1,1,1},
117 {"zremrangebyscore",zremrangebyscoreCommand,4,0,NULL,1,1,1},
118 {"zremrangebyrank",zremrangebyrankCommand,4,0,NULL,1,1,1},
119 {"zunionstore",zunionstoreCommand,-4,REDIS_CMD_DENYOOM,zunionInterBlockClientOnSwappedKeys,0,0,0},
120 {"zinterstore",zinterstoreCommand,-4,REDIS_CMD_DENYOOM,zunionInterBlockClientOnSwappedKeys,0,0,0},
121 {"zrange",zrangeCommand,-4,0,NULL,1,1,1},
122 {"zrangebyscore",zrangebyscoreCommand,-4,0,NULL,1,1,1},
123 {"zrevrangebyscore",zrevrangebyscoreCommand,-4,0,NULL,1,1,1},
124 {"zcount",zcountCommand,4,0,NULL,1,1,1},
125 {"zrevrange",zrevrangeCommand,-4,0,NULL,1,1,1},
126 {"zcard",zcardCommand,2,0,NULL,1,1,1},
127 {"zscore",zscoreCommand,3,REDIS_CMD_DENYOOM,NULL,1,1,1},
128 {"zrank",zrankCommand,3,0,NULL,1,1,1},
129 {"zrevrank",zrevrankCommand,3,0,NULL,1,1,1},
130 {"hset",hsetCommand,4,REDIS_CMD_DENYOOM,NULL,1,1,1},
131 {"hsetnx",hsetnxCommand,4,REDIS_CMD_DENYOOM,NULL,1,1,1},
132 {"hget",hgetCommand,3,0,NULL,1,1,1},
133 {"hmset",hmsetCommand,-4,REDIS_CMD_DENYOOM,NULL,1,1,1},
134 {"hmget",hmgetCommand,-3,0,NULL,1,1,1},
135 {"hincrby",hincrbyCommand,4,REDIS_CMD_DENYOOM,NULL,1,1,1},
136 {"hdel",hdelCommand,3,0,NULL,1,1,1},
137 {"hlen",hlenCommand,2,0,NULL,1,1,1},
138 {"hkeys",hkeysCommand,2,0,NULL,1,1,1},
139 {"hvals",hvalsCommand,2,0,NULL,1,1,1},
140 {"hgetall",hgetallCommand,2,0,NULL,1,1,1},
141 {"hexists",hexistsCommand,3,0,NULL,1,1,1},
142 {"incrby",incrbyCommand,3,REDIS_CMD_DENYOOM,NULL,1,1,1},
143 {"decrby",decrbyCommand,3,REDIS_CMD_DENYOOM,NULL,1,1,1},
144 {"getset",getsetCommand,3,REDIS_CMD_DENYOOM,NULL,1,1,1},
145 {"mset",msetCommand,-3,REDIS_CMD_DENYOOM,NULL,1,-1,2},
146 {"msetnx",msetnxCommand,-3,REDIS_CMD_DENYOOM,NULL,1,-1,2},
147 {"randomkey",randomkeyCommand,1,0,NULL,0,0,0},
148 {"select",selectCommand,2,0,NULL,0,0,0},
149 {"move",moveCommand,3,0,NULL,1,1,1},
150 {"rename",renameCommand,3,0,NULL,1,1,1},
151 {"renamenx",renamenxCommand,3,0,NULL,1,1,1},
152 {"expire",expireCommand,3,0,NULL,0,0,0},
153 {"expireat",expireatCommand,3,0,NULL,0,0,0},
154 {"keys",keysCommand,2,0,NULL,0,0,0},
155 {"dbsize",dbsizeCommand,1,0,NULL,0,0,0},
156 {"auth",authCommand,2,0,NULL,0,0,0},
157 {"ping",pingCommand,1,0,NULL,0,0,0},
158 {"echo",echoCommand,2,0,NULL,0,0,0},
159 {"save",saveCommand,1,0,NULL,0,0,0},
160 {"bgsave",bgsaveCommand,1,0,NULL,0,0,0},
161 {"bgrewriteaof",bgrewriteaofCommand,1,0,NULL,0,0,0},
162 {"shutdown",shutdownCommand,1,0,NULL,0,0,0},
163 {"lastsave",lastsaveCommand,1,0,NULL,0,0,0},
164 {"type",typeCommand,2,0,NULL,1,1,1},
165 {"multi",multiCommand,1,0,NULL,0,0,0},
166 {"exec",execCommand,1,REDIS_CMD_DENYOOM,execBlockClientOnSwappedKeys,0,0,0},
167 {"discard",discardCommand,1,0,NULL,0,0,0},
168 {"sync",syncCommand,1,0,NULL,0,0,0},
169 {"flushdb",flushdbCommand,1,0,NULL,0,0,0},
170 {"flushall",flushallCommand,1,0,NULL,0,0,0},
171 {"sort",sortCommand,-2,REDIS_CMD_DENYOOM,NULL,1,1,1},
172 {"info",infoCommand,1,0,NULL,0,0,0},
173 {"monitor",monitorCommand,1,0,NULL,0,0,0},
174 {"ttl",ttlCommand,2,0,NULL,1,1,1},
175 {"persist",persistCommand,2,0,NULL,1,1,1},
176 {"slaveof",slaveofCommand,3,0,NULL,0,0,0},
177 {"debug",debugCommand,-2,0,NULL,0,0,0},
178 {"config",configCommand,-2,0,NULL,0,0,0},
179 {"subscribe",subscribeCommand,-2,0,NULL,0,0,0},
180 {"unsubscribe",unsubscribeCommand,-1,0,NULL,0,0,0},
181 {"psubscribe",psubscribeCommand,-2,0,NULL,0,0,0},
182 {"punsubscribe",punsubscribeCommand,-1,0,NULL,0,0,0},
183 {"publish",publishCommand,3,REDIS_CMD_FORCE_REPLICATION,NULL,0,0,0},
184 {"watch",watchCommand,-2,0,NULL,0,0,0},
185 {"unwatch",unwatchCommand,1,0,NULL,0,0,0}
186 };
187
188 /*============================ Utility functions ============================ */
189
190 void redisLog(int level, const char *fmt, ...) {
191 va_list ap;
192 FILE *fp;
193 char *c = ".-*#";
194 char buf[64];
195 time_t now;
196
197 if (level < server.verbosity) return;
198
199 fp = (server.logfile == NULL) ? stdout : fopen(server.logfile,"a");
200 if (!fp) return;
201
202 va_start(ap, fmt);
203 now = time(NULL);
204 strftime(buf,64,"%d %b %H:%M:%S",localtime(&now));
205 fprintf(fp,"[%d] %s %c ",(int)getpid(),buf,c[level]);
206 vfprintf(fp, fmt, ap);
207 fprintf(fp,"\n");
208 fflush(fp);
209 va_end(ap);
210
211 if (server.logfile) fclose(fp);
212 }
213
214 /* Redis generally does not try to recover from out of memory conditions
215 * when allocating objects or strings, it is not clear if it will be possible
216 * to report this condition to the client since the networking layer itself
217 * is based on heap allocation for send buffers, so we simply abort.
218 * At least the code will be simpler to read... */
219 void oom(const char *msg) {
220 redisLog(REDIS_WARNING, "%s: Out of memory\n",msg);
221 sleep(1);
222 abort();
223 }
224
225 /*====================== Hash table type implementation ==================== */
226
227 /* This is an hash table type that uses the SDS dynamic strings libary as
228 * keys and radis objects as values (objects can hold SDS strings,
229 * lists, sets). */
230
231 void dictVanillaFree(void *privdata, void *val)
232 {
233 DICT_NOTUSED(privdata);
234 zfree(val);
235 }
236
237 void dictListDestructor(void *privdata, void *val)
238 {
239 DICT_NOTUSED(privdata);
240 listRelease((list*)val);
241 }
242
243 int dictSdsKeyCompare(void *privdata, const void *key1,
244 const void *key2)
245 {
246 int l1,l2;
247 DICT_NOTUSED(privdata);
248
249 l1 = sdslen((sds)key1);
250 l2 = sdslen((sds)key2);
251 if (l1 != l2) return 0;
252 return memcmp(key1, key2, l1) == 0;
253 }
254
255 void dictRedisObjectDestructor(void *privdata, void *val)
256 {
257 DICT_NOTUSED(privdata);
258
259 if (val == NULL) return; /* Values of swapped out keys as set to NULL */
260 decrRefCount(val);
261 }
262
263 void dictSdsDestructor(void *privdata, void *val)
264 {
265 DICT_NOTUSED(privdata);
266
267 sdsfree(val);
268 }
269
270 int dictObjKeyCompare(void *privdata, const void *key1,
271 const void *key2)
272 {
273 const robj *o1 = key1, *o2 = key2;
274 return dictSdsKeyCompare(privdata,o1->ptr,o2->ptr);
275 }
276
277 unsigned int dictObjHash(const void *key) {
278 const robj *o = key;
279 return dictGenHashFunction(o->ptr, sdslen((sds)o->ptr));
280 }
281
282 unsigned int dictSdsHash(const void *key) {
283 return dictGenHashFunction((unsigned char*)key, sdslen((char*)key));
284 }
285
286 int dictEncObjKeyCompare(void *privdata, const void *key1,
287 const void *key2)
288 {
289 robj *o1 = (robj*) key1, *o2 = (robj*) key2;
290 int cmp;
291
292 if (o1->encoding == REDIS_ENCODING_INT &&
293 o2->encoding == REDIS_ENCODING_INT)
294 return o1->ptr == o2->ptr;
295
296 o1 = getDecodedObject(o1);
297 o2 = getDecodedObject(o2);
298 cmp = dictSdsKeyCompare(privdata,o1->ptr,o2->ptr);
299 decrRefCount(o1);
300 decrRefCount(o2);
301 return cmp;
302 }
303
304 unsigned int dictEncObjHash(const void *key) {
305 robj *o = (robj*) key;
306
307 if (o->encoding == REDIS_ENCODING_RAW) {
308 return dictGenHashFunction(o->ptr, sdslen((sds)o->ptr));
309 } else {
310 if (o->encoding == REDIS_ENCODING_INT) {
311 char buf[32];
312 int len;
313
314 len = ll2string(buf,32,(long)o->ptr);
315 return dictGenHashFunction((unsigned char*)buf, len);
316 } else {
317 unsigned int hash;
318
319 o = getDecodedObject(o);
320 hash = dictGenHashFunction(o->ptr, sdslen((sds)o->ptr));
321 decrRefCount(o);
322 return hash;
323 }
324 }
325 }
326
327 /* Sets type */
328 dictType setDictType = {
329 dictEncObjHash, /* hash function */
330 NULL, /* key dup */
331 NULL, /* val dup */
332 dictEncObjKeyCompare, /* key compare */
333 dictRedisObjectDestructor, /* key destructor */
334 NULL /* val destructor */
335 };
336
337 /* Sorted sets hash (note: a skiplist is used in addition to the hash table) */
338 dictType zsetDictType = {
339 dictEncObjHash, /* hash function */
340 NULL, /* key dup */
341 NULL, /* val dup */
342 dictEncObjKeyCompare, /* key compare */
343 dictRedisObjectDestructor, /* key destructor */
344 NULL /* val destructor */
345 };
346
347 /* Db->dict, keys are sds strings, vals are Redis objects. */
348 dictType dbDictType = {
349 dictSdsHash, /* hash function */
350 NULL, /* key dup */
351 NULL, /* val dup */
352 dictSdsKeyCompare, /* key compare */
353 dictSdsDestructor, /* key destructor */
354 dictRedisObjectDestructor /* val destructor */
355 };
356
357 /* Db->expires */
358 dictType keyptrDictType = {
359 dictSdsHash, /* hash function */
360 NULL, /* key dup */
361 NULL, /* val dup */
362 dictSdsKeyCompare, /* key compare */
363 NULL, /* key destructor */
364 NULL /* val destructor */
365 };
366
367 /* Hash type hash table (note that small hashes are represented with zimpaps) */
368 dictType hashDictType = {
369 dictEncObjHash, /* hash function */
370 NULL, /* key dup */
371 NULL, /* val dup */
372 dictEncObjKeyCompare, /* key compare */
373 dictRedisObjectDestructor, /* key destructor */
374 dictRedisObjectDestructor /* val destructor */
375 };
376
377 /* Keylist hash table type has unencoded redis objects as keys and
378 * lists as values. It's used for blocking operations (BLPOP) and to
379 * map swapped keys to a list of clients waiting for this keys to be loaded. */
380 dictType keylistDictType = {
381 dictObjHash, /* hash function */
382 NULL, /* key dup */
383 NULL, /* val dup */
384 dictObjKeyCompare, /* key compare */
385 dictRedisObjectDestructor, /* key destructor */
386 dictListDestructor /* val destructor */
387 };
388
389 int htNeedsResize(dict *dict) {
390 long long size, used;
391
392 size = dictSlots(dict);
393 used = dictSize(dict);
394 return (size && used && size > DICT_HT_INITIAL_SIZE &&
395 (used*100/size < REDIS_HT_MINFILL));
396 }
397
398 /* If the percentage of used slots in the HT reaches REDIS_HT_MINFILL
399 * we resize the hash table to save memory */
400 void tryResizeHashTables(void) {
401 int j;
402
403 for (j = 0; j < server.dbnum; j++) {
404 if (htNeedsResize(server.db[j].dict))
405 dictResize(server.db[j].dict);
406 if (htNeedsResize(server.db[j].expires))
407 dictResize(server.db[j].expires);
408 }
409 }
410
411 /* Our hash table implementation performs rehashing incrementally while
412 * we write/read from the hash table. Still if the server is idle, the hash
413 * table will use two tables for a long time. So we try to use 1 millisecond
414 * of CPU time at every serverCron() loop in order to rehash some key. */
415 void incrementallyRehash(void) {
416 int j;
417
418 for (j = 0; j < server.dbnum; j++) {
419 if (dictIsRehashing(server.db[j].dict)) {
420 dictRehashMilliseconds(server.db[j].dict,1);
421 break; /* already used our millisecond for this loop... */
422 }
423 }
424 }
425
426 /* This function is called once a background process of some kind terminates,
427 * as we want to avoid resizing the hash tables when there is a child in order
428 * to play well with copy-on-write (otherwise when a resize happens lots of
429 * memory pages are copied). The goal of this function is to update the ability
430 * for dict.c to resize the hash tables accordingly to the fact we have o not
431 * running childs. */
432 void updateDictResizePolicy(void) {
433 if (server.bgsavechildpid == -1 && server.bgrewritechildpid == -1)
434 dictEnableResize();
435 else
436 dictDisableResize();
437 }
438
439 /* ======================= Cron: called every 100 ms ======================== */
440
441 /* Try to expire a few timed out keys. The algorithm used is adaptive and
442 * will use few CPU cycles if there are few expiring keys, otherwise
443 * it will get more aggressive to avoid that too much memory is used by
444 * keys that can be removed from the keyspace. */
445 void activeExpireCycle(void) {
446 int j;
447
448 for (j = 0; j < server.dbnum; j++) {
449 int expired;
450 redisDb *db = server.db+j;
451
452 /* Continue to expire if at the end of the cycle more than 25%
453 * of the keys were expired. */
454 do {
455 long num = dictSize(db->expires);
456 time_t now = time(NULL);
457
458 expired = 0;
459 if (num > REDIS_EXPIRELOOKUPS_PER_CRON)
460 num = REDIS_EXPIRELOOKUPS_PER_CRON;
461 while (num--) {
462 dictEntry *de;
463 time_t t;
464
465 if ((de = dictGetRandomKey(db->expires)) == NULL) break;
466 t = (time_t) dictGetEntryVal(de);
467 if (now > t) {
468 sds key = dictGetEntryKey(de);
469 robj *keyobj = createStringObject(key,sdslen(key));
470
471 propagateExpire(db,keyobj);
472 dbDelete(db,keyobj);
473 decrRefCount(keyobj);
474 expired++;
475 server.stat_expiredkeys++;
476 }
477 }
478 } while (expired > REDIS_EXPIRELOOKUPS_PER_CRON/4);
479 }
480 }
481
482 void updateLRUClock(void) {
483 server.lruclock = (time(NULL)/REDIS_LRU_CLOCK_RESOLUTION) &
484 REDIS_LRU_CLOCK_MAX;
485 }
486
487 int serverCron(struct aeEventLoop *eventLoop, long long id, void *clientData) {
488 int j, loops = server.cronloops++;
489 REDIS_NOTUSED(eventLoop);
490 REDIS_NOTUSED(id);
491 REDIS_NOTUSED(clientData);
492
493 /* We take a cached value of the unix time in the global state because
494 * with virtual memory and aging there is to store the current time
495 * in objects at every object access, and accuracy is not needed.
496 * To access a global var is faster than calling time(NULL) */
497 server.unixtime = time(NULL);
498 /* We have just 22 bits per object for LRU information.
499 * So we use an (eventually wrapping) LRU clock with 10 seconds resolution.
500 * 2^22 bits with 10 seconds resoluton is more or less 1.5 years.
501 *
502 * Note that even if this will wrap after 1.5 years it's not a problem,
503 * everything will still work but just some object will appear younger
504 * to Redis. But for this to happen a given object should never be touched
505 * for 1.5 years.
506 *
507 * Note that you can change the resolution altering the
508 * REDIS_LRU_CLOCK_RESOLUTION define.
509 */
510 updateLRUClock();
511
512 /* We received a SIGTERM, shutting down here in a safe way, as it is
513 * not ok doing so inside the signal handler. */
514 if (server.shutdown_asap) {
515 if (prepareForShutdown() == REDIS_OK) exit(0);
516 redisLog(REDIS_WARNING,"SIGTERM received but errors trying to shut down the server, check the logs for more information");
517 }
518
519 /* Show some info about non-empty databases */
520 for (j = 0; j < server.dbnum; j++) {
521 long long size, used, vkeys;
522
523 size = dictSlots(server.db[j].dict);
524 used = dictSize(server.db[j].dict);
525 vkeys = dictSize(server.db[j].expires);
526 if (!(loops % 50) && (used || vkeys)) {
527 redisLog(REDIS_VERBOSE,"DB %d: %lld keys (%lld volatile) in %lld slots HT.",j,used,vkeys,size);
528 /* dictPrintStats(server.dict); */
529 }
530 }
531
532 /* We don't want to resize the hash tables while a bacground saving
533 * is in progress: the saving child is created using fork() that is
534 * implemented with a copy-on-write semantic in most modern systems, so
535 * if we resize the HT while there is the saving child at work actually
536 * a lot of memory movements in the parent will cause a lot of pages
537 * copied. */
538 if (server.bgsavechildpid == -1 && server.bgrewritechildpid == -1) {
539 if (!(loops % 10)) tryResizeHashTables();
540 if (server.activerehashing) incrementallyRehash();
541 }
542
543 /* Show information about connected clients */
544 if (!(loops % 50)) {
545 redisLog(REDIS_VERBOSE,"%d clients connected (%d slaves), %zu bytes in use",
546 listLength(server.clients)-listLength(server.slaves),
547 listLength(server.slaves),
548 zmalloc_used_memory());
549 }
550
551 /* Close connections of timedout clients */
552 if ((server.maxidletime && !(loops % 100)) || server.blpop_blocked_clients)
553 closeTimedoutClients();
554
555 /* Check if a background saving or AOF rewrite in progress terminated */
556 if (server.bgsavechildpid != -1 || server.bgrewritechildpid != -1) {
557 int statloc;
558 pid_t pid;
559
560 if ((pid = wait3(&statloc,WNOHANG,NULL)) != 0) {
561 if (pid == server.bgsavechildpid) {
562 backgroundSaveDoneHandler(statloc);
563 } else {
564 backgroundRewriteDoneHandler(statloc);
565 }
566 updateDictResizePolicy();
567 }
568 } else {
569 /* If there is not a background saving in progress check if
570 * we have to save now */
571 time_t now = time(NULL);
572 for (j = 0; j < server.saveparamslen; j++) {
573 struct saveparam *sp = server.saveparams+j;
574
575 if (server.dirty >= sp->changes &&
576 now-server.lastsave > sp->seconds) {
577 redisLog(REDIS_NOTICE,"%d changes in %d seconds. Saving...",
578 sp->changes, sp->seconds);
579 rdbSaveBackground(server.dbfilename);
580 break;
581 }
582 }
583 }
584
585 /* Expire a few keys per cycle, only if this is a master.
586 * On slaves we wait for DEL operations synthesized by the master
587 * in order to guarantee a strict consistency. */
588 if (server.masterhost == NULL) activeExpireCycle();
589
590 /* Swap a few keys on disk if we are over the memory limit and VM
591 * is enbled. Try to free objects from the free list first. */
592 if (vmCanSwapOut()) {
593 while (server.vm_enabled && zmalloc_used_memory() >
594 server.vm_max_memory)
595 {
596 int retval;
597
598 if (tryFreeOneObjectFromFreelist() == REDIS_OK) continue;
599 retval = (server.vm_max_threads == 0) ?
600 vmSwapOneObjectBlocking() :
601 vmSwapOneObjectThreaded();
602 if (retval == REDIS_ERR && !(loops % 300) &&
603 zmalloc_used_memory() >
604 (server.vm_max_memory+server.vm_max_memory/10))
605 {
606 redisLog(REDIS_WARNING,"WARNING: vm-max-memory limit exceeded by more than 10%% but unable to swap more objects out!");
607 }
608 /* Note that when using threade I/O we free just one object,
609 * because anyway when the I/O thread in charge to swap this
610 * object out will finish, the handler of completed jobs
611 * will try to swap more objects if we are still out of memory. */
612 if (retval == REDIS_ERR || server.vm_max_threads > 0) break;
613 }
614 }
615
616 /* Check if we should connect to a MASTER */
617 if (server.replstate == REDIS_REPL_CONNECT && !(loops % 10)) {
618 redisLog(REDIS_NOTICE,"Connecting to MASTER...");
619 if (syncWithMaster() == REDIS_OK) {
620 redisLog(REDIS_NOTICE,"MASTER <-> SLAVE sync succeeded");
621 if (server.appendonly) rewriteAppendOnlyFileBackground();
622 }
623 }
624
625 /* Update fragmentation info, used for fast RSS estimation */
626 if (!(loops % 10)) server.fragmentation = zmalloc_get_fragmentation_ratio();
627 return 100;
628 }
629
630 /* This function gets called every time Redis is entering the
631 * main loop of the event driven library, that is, before to sleep
632 * for ready file descriptors. */
633 void beforeSleep(struct aeEventLoop *eventLoop) {
634 REDIS_NOTUSED(eventLoop);
635
636 /* Awake clients that got all the swapped keys they requested */
637 if (server.vm_enabled && listLength(server.io_ready_clients)) {
638 listIter li;
639 listNode *ln;
640
641 listRewind(server.io_ready_clients,&li);
642 while((ln = listNext(&li))) {
643 redisClient *c = ln->value;
644 struct redisCommand *cmd;
645
646 /* Resume the client. */
647 listDelNode(server.io_ready_clients,ln);
648 c->flags &= (~REDIS_IO_WAIT);
649 server.vm_blocked_clients--;
650 aeCreateFileEvent(server.el, c->fd, AE_READABLE,
651 readQueryFromClient, c);
652 cmd = lookupCommand(c->argv[0]->ptr);
653 redisAssert(cmd != NULL);
654 call(c,cmd);
655 resetClient(c);
656 /* There may be more data to process in the input buffer. */
657 if (c->querybuf && sdslen(c->querybuf) > 0)
658 processInputBuffer(c);
659 }
660 }
661 /* Write the AOF buffer on disk */
662 flushAppendOnlyFile();
663 }
664
665 /* =========================== Server initialization ======================== */
666
667 void createSharedObjects(void) {
668 int j;
669
670 shared.crlf = createObject(REDIS_STRING,sdsnew("\r\n"));
671 shared.ok = createObject(REDIS_STRING,sdsnew("+OK\r\n"));
672 shared.err = createObject(REDIS_STRING,sdsnew("-ERR\r\n"));
673 shared.emptybulk = createObject(REDIS_STRING,sdsnew("$0\r\n\r\n"));
674 shared.czero = createObject(REDIS_STRING,sdsnew(":0\r\n"));
675 shared.cone = createObject(REDIS_STRING,sdsnew(":1\r\n"));
676 shared.cnegone = createObject(REDIS_STRING,sdsnew(":-1\r\n"));
677 shared.nullbulk = createObject(REDIS_STRING,sdsnew("$-1\r\n"));
678 shared.nullmultibulk = createObject(REDIS_STRING,sdsnew("*-1\r\n"));
679 shared.emptymultibulk = createObject(REDIS_STRING,sdsnew("*0\r\n"));
680 shared.pong = createObject(REDIS_STRING,sdsnew("+PONG\r\n"));
681 shared.queued = createObject(REDIS_STRING,sdsnew("+QUEUED\r\n"));
682 shared.wrongtypeerr = createObject(REDIS_STRING,sdsnew(
683 "-ERR Operation against a key holding the wrong kind of value\r\n"));
684 shared.nokeyerr = createObject(REDIS_STRING,sdsnew(
685 "-ERR no such key\r\n"));
686 shared.syntaxerr = createObject(REDIS_STRING,sdsnew(
687 "-ERR syntax error\r\n"));
688 shared.sameobjecterr = createObject(REDIS_STRING,sdsnew(
689 "-ERR source and destination objects are the same\r\n"));
690 shared.outofrangeerr = createObject(REDIS_STRING,sdsnew(
691 "-ERR index out of range\r\n"));
692 shared.space = createObject(REDIS_STRING,sdsnew(" "));
693 shared.colon = createObject(REDIS_STRING,sdsnew(":"));
694 shared.plus = createObject(REDIS_STRING,sdsnew("+"));
695 shared.select0 = createStringObject("select 0\r\n",10);
696 shared.select1 = createStringObject("select 1\r\n",10);
697 shared.select2 = createStringObject("select 2\r\n",10);
698 shared.select3 = createStringObject("select 3\r\n",10);
699 shared.select4 = createStringObject("select 4\r\n",10);
700 shared.select5 = createStringObject("select 5\r\n",10);
701 shared.select6 = createStringObject("select 6\r\n",10);
702 shared.select7 = createStringObject("select 7\r\n",10);
703 shared.select8 = createStringObject("select 8\r\n",10);
704 shared.select9 = createStringObject("select 9\r\n",10);
705 shared.messagebulk = createStringObject("$7\r\nmessage\r\n",13);
706 shared.pmessagebulk = createStringObject("$8\r\npmessage\r\n",14);
707 shared.subscribebulk = createStringObject("$9\r\nsubscribe\r\n",15);
708 shared.unsubscribebulk = createStringObject("$11\r\nunsubscribe\r\n",18);
709 shared.psubscribebulk = createStringObject("$10\r\npsubscribe\r\n",17);
710 shared.punsubscribebulk = createStringObject("$12\r\npunsubscribe\r\n",19);
711 shared.mbulk3 = createStringObject("*3\r\n",4);
712 shared.mbulk4 = createStringObject("*4\r\n",4);
713 for (j = 0; j < REDIS_SHARED_INTEGERS; j++) {
714 shared.integers[j] = createObject(REDIS_STRING,(void*)(long)j);
715 shared.integers[j]->encoding = REDIS_ENCODING_INT;
716 }
717 }
718
719 void initServerConfig() {
720 server.dbnum = REDIS_DEFAULT_DBNUM;
721 server.port = REDIS_SERVERPORT;
722 server.verbosity = REDIS_VERBOSE;
723 server.maxidletime = REDIS_MAXIDLETIME;
724 server.saveparams = NULL;
725 server.logfile = NULL; /* NULL = log on standard output */
726 server.bindaddr = NULL;
727 server.glueoutputbuf = 1;
728 server.daemonize = 0;
729 server.appendonly = 0;
730 server.appendfsync = APPENDFSYNC_EVERYSEC;
731 server.no_appendfsync_on_rewrite = 0;
732 server.lastfsync = time(NULL);
733 server.appendfd = -1;
734 server.appendseldb = -1; /* Make sure the first time will not match */
735 server.pidfile = zstrdup("/var/run/redis.pid");
736 server.dbfilename = zstrdup("dump.rdb");
737 server.appendfilename = zstrdup("appendonly.aof");
738 server.requirepass = NULL;
739 server.rdbcompression = 1;
740 server.activerehashing = 1;
741 server.fragmentation = 1;
742 server.maxclients = 0;
743 server.blpop_blocked_clients = 0;
744 server.maxmemory = 0;
745 server.maxmemory_policy = REDIS_MAXMEMORY_VOLATILE_LRU;
746 server.maxmemory_samples = 3;
747 server.vm_enabled = 0;
748 server.vm_swap_file = zstrdup("/tmp/redis-%p.vm");
749 server.vm_page_size = 256; /* 256 bytes per page */
750 server.vm_pages = 1024*1024*100; /* 104 millions of pages */
751 server.vm_max_memory = 1024LL*1024*1024*1; /* 1 GB of RAM */
752 server.vm_max_threads = 4;
753 server.vm_blocked_clients = 0;
754 server.hash_max_zipmap_entries = REDIS_HASH_MAX_ZIPMAP_ENTRIES;
755 server.hash_max_zipmap_value = REDIS_HASH_MAX_ZIPMAP_VALUE;
756 server.list_max_ziplist_entries = REDIS_LIST_MAX_ZIPLIST_ENTRIES;
757 server.list_max_ziplist_value = REDIS_LIST_MAX_ZIPLIST_VALUE;
758 server.set_max_intset_entries = REDIS_SET_MAX_INTSET_ENTRIES;
759 server.shutdown_asap = 0;
760
761 updateLRUClock();
762 resetServerSaveParams();
763
764 appendServerSaveParams(60*60,1); /* save after 1 hour and 1 change */
765 appendServerSaveParams(300,100); /* save after 5 minutes and 100 changes */
766 appendServerSaveParams(60,10000); /* save after 1 minute and 10000 changes */
767 /* Replication related */
768 server.isslave = 0;
769 server.masterauth = NULL;
770 server.masterhost = NULL;
771 server.masterport = 6379;
772 server.master = NULL;
773 server.replstate = REDIS_REPL_NONE;
774
775 /* Double constants initialization */
776 R_Zero = 0.0;
777 R_PosInf = 1.0/R_Zero;
778 R_NegInf = -1.0/R_Zero;
779 R_Nan = R_Zero/R_Zero;
780 }
781
782 void initServer() {
783 int j;
784
785 signal(SIGHUP, SIG_IGN);
786 signal(SIGPIPE, SIG_IGN);
787 setupSigSegvAction();
788
789 server.mainthread = pthread_self();
790 server.devnull = fopen("/dev/null","w");
791 if (server.devnull == NULL) {
792 redisLog(REDIS_WARNING, "Can't open /dev/null: %s", server.neterr);
793 exit(1);
794 }
795 server.clients = listCreate();
796 server.slaves = listCreate();
797 server.monitors = listCreate();
798 server.objfreelist = listCreate();
799 createSharedObjects();
800 server.el = aeCreateEventLoop();
801 server.db = zmalloc(sizeof(redisDb)*server.dbnum);
802 server.fd = anetTcpServer(server.neterr, server.port, server.bindaddr);
803 if (server.fd == -1) {
804 redisLog(REDIS_WARNING, "Opening TCP port: %s", server.neterr);
805 exit(1);
806 }
807 for (j = 0; j < server.dbnum; j++) {
808 server.db[j].dict = dictCreate(&dbDictType,NULL);
809 server.db[j].expires = dictCreate(&keyptrDictType,NULL);
810 server.db[j].blocking_keys = dictCreate(&keylistDictType,NULL);
811 server.db[j].watched_keys = dictCreate(&keylistDictType,NULL);
812 if (server.vm_enabled)
813 server.db[j].io_keys = dictCreate(&keylistDictType,NULL);
814 server.db[j].id = j;
815 }
816 server.pubsub_channels = dictCreate(&keylistDictType,NULL);
817 server.pubsub_patterns = listCreate();
818 listSetFreeMethod(server.pubsub_patterns,freePubsubPattern);
819 listSetMatchMethod(server.pubsub_patterns,listMatchPubsubPattern);
820 server.cronloops = 0;
821 server.bgsavechildpid = -1;
822 server.bgrewritechildpid = -1;
823 server.bgrewritebuf = sdsempty();
824 server.aofbuf = sdsempty();
825 server.lastsave = time(NULL);
826 server.dirty = 0;
827 server.stat_numcommands = 0;
828 server.stat_numconnections = 0;
829 server.stat_expiredkeys = 0;
830 server.stat_starttime = time(NULL);
831 server.stat_keyspace_misses = 0;
832 server.stat_keyspace_hits = 0;
833 server.unixtime = time(NULL);
834 aeCreateTimeEvent(server.el, 1, serverCron, NULL, NULL);
835 if (aeCreateFileEvent(server.el, server.fd, AE_READABLE,
836 acceptHandler, NULL) == AE_ERR) oom("creating file event");
837
838 if (server.appendonly) {
839 server.appendfd = open(server.appendfilename,O_WRONLY|O_APPEND|O_CREAT,0644);
840 if (server.appendfd == -1) {
841 redisLog(REDIS_WARNING, "Can't open the append-only file: %s",
842 strerror(errno));
843 exit(1);
844 }
845 }
846
847 if (server.vm_enabled) vmInit();
848 }
849
850 int qsortRedisCommands(const void *r1, const void *r2) {
851 return strcasecmp(
852 ((struct redisCommand*)r1)->name,
853 ((struct redisCommand*)r2)->name);
854 }
855
856 void sortCommandTable() {
857 /* Copy and sort the read-only version of the command table */
858 commandTable = (struct redisCommand*)zmalloc(sizeof(readonlyCommandTable));
859 memcpy(commandTable,readonlyCommandTable,sizeof(readonlyCommandTable));
860 qsort(commandTable,
861 sizeof(readonlyCommandTable)/sizeof(struct redisCommand),
862 sizeof(struct redisCommand),qsortRedisCommands);
863 }
864
865 /* ====================== Commands lookup and execution ===================== */
866
867 struct redisCommand *lookupCommand(char *name) {
868 struct redisCommand tmp = {name,NULL,0,0,NULL,0,0,0};
869 return bsearch(
870 &tmp,
871 commandTable,
872 sizeof(readonlyCommandTable)/sizeof(struct redisCommand),
873 sizeof(struct redisCommand),
874 qsortRedisCommands);
875 }
876
877 /* Call() is the core of Redis execution of a command */
878 void call(redisClient *c, struct redisCommand *cmd) {
879 long long dirty;
880
881 dirty = server.dirty;
882 cmd->proc(c);
883 dirty = server.dirty-dirty;
884
885 if (server.appendonly && dirty)
886 feedAppendOnlyFile(cmd,c->db->id,c->argv,c->argc);
887 if ((dirty || cmd->flags & REDIS_CMD_FORCE_REPLICATION) &&
888 listLength(server.slaves))
889 replicationFeedSlaves(server.slaves,c->db->id,c->argv,c->argc);
890 if (listLength(server.monitors))
891 replicationFeedMonitors(server.monitors,c->db->id,c->argv,c->argc);
892 server.stat_numcommands++;
893 }
894
895 /* If this function gets called we already read a whole
896 * command, argments are in the client argv/argc fields.
897 * processCommand() execute the command or prepare the
898 * server for a bulk read from the client.
899 *
900 * If 1 is returned the client is still alive and valid and
901 * and other operations can be performed by the caller. Otherwise
902 * if 0 is returned the client was destroied (i.e. after QUIT). */
903 int processCommand(redisClient *c) {
904 struct redisCommand *cmd;
905
906 /* The QUIT command is handled separately. Normal command procs will
907 * go through checking for replication and QUIT will cause trouble
908 * when FORCE_REPLICATION is enabled and would be implemented in
909 * a regular command proc. */
910 if (!strcasecmp(c->argv[0]->ptr,"quit")) {
911 addReply(c,shared.ok);
912 c->flags |= REDIS_CLOSE_AFTER_REPLY;
913 return REDIS_ERR;
914 }
915
916 /* Now lookup the command and check ASAP about trivial error conditions
917 * such wrong arity, bad command name and so forth. */
918 cmd = lookupCommand(c->argv[0]->ptr);
919 if (!cmd) {
920 addReplyErrorFormat(c,"unknown command '%s'",
921 (char*)c->argv[0]->ptr);
922 return REDIS_OK;
923 } else if ((cmd->arity > 0 && cmd->arity != c->argc) ||
924 (c->argc < -cmd->arity)) {
925 addReplyErrorFormat(c,"wrong number of arguments for '%s' command",
926 cmd->name);
927 return REDIS_OK;
928 }
929
930 /* Check if the user is authenticated */
931 if (server.requirepass && !c->authenticated && cmd->proc != authCommand) {
932 addReplyError(c,"operation not permitted");
933 return REDIS_OK;
934 }
935
936 /* Handle the maxmemory directive.
937 *
938 * First we try to free some memory if possible (if there are volatile
939 * keys in the dataset). If there are not the only thing we can do
940 * is returning an error. */
941 if (server.maxmemory) freeMemoryIfNeeded();
942 if (server.maxmemory && (cmd->flags & REDIS_CMD_DENYOOM) &&
943 zmalloc_used_memory() > server.maxmemory)
944 {
945 addReplyError(c,"command not allowed when used memory > 'maxmemory'");
946 return REDIS_OK;
947 }
948
949 /* Only allow SUBSCRIBE and UNSUBSCRIBE in the context of Pub/Sub */
950 if ((dictSize(c->pubsub_channels) > 0 || listLength(c->pubsub_patterns) > 0)
951 &&
952 cmd->proc != subscribeCommand && cmd->proc != unsubscribeCommand &&
953 cmd->proc != psubscribeCommand && cmd->proc != punsubscribeCommand) {
954 addReplyError(c,"only (P)SUBSCRIBE / (P)UNSUBSCRIBE / QUIT allowed in this context");
955 return REDIS_OK;
956 }
957
958 /* Exec the command */
959 if (c->flags & REDIS_MULTI &&
960 cmd->proc != execCommand && cmd->proc != discardCommand &&
961 cmd->proc != multiCommand && cmd->proc != watchCommand)
962 {
963 queueMultiCommand(c,cmd);
964 addReply(c,shared.queued);
965 } else {
966 if (server.vm_enabled && server.vm_max_threads > 0 &&
967 blockClientOnSwappedKeys(c,cmd)) return REDIS_ERR;
968 call(c,cmd);
969 }
970 return REDIS_OK;
971 }
972
973 /*================================== Shutdown =============================== */
974
975 int prepareForShutdown() {
976 redisLog(REDIS_WARNING,"User requested shutdown, saving DB...");
977 /* Kill the saving child if there is a background saving in progress.
978 We want to avoid race conditions, for instance our saving child may
979 overwrite the synchronous saving did by SHUTDOWN. */
980 if (server.bgsavechildpid != -1) {
981 redisLog(REDIS_WARNING,"There is a live saving child. Killing it!");
982 kill(server.bgsavechildpid,SIGKILL);
983 rdbRemoveTempFile(server.bgsavechildpid);
984 }
985 if (server.appendonly) {
986 /* Append only file: fsync() the AOF and exit */
987 aof_fsync(server.appendfd);
988 if (server.vm_enabled) unlink(server.vm_swap_file);
989 } else if (server.saveparamslen > 0) {
990 /* Snapshotting. Perform a SYNC SAVE and exit */
991 if (rdbSave(server.dbfilename) != REDIS_OK) {
992 /* Ooops.. error saving! The best we can do is to continue
993 * operating. Note that if there was a background saving process,
994 * in the next cron() Redis will be notified that the background
995 * saving aborted, handling special stuff like slaves pending for
996 * synchronization... */
997 redisLog(REDIS_WARNING,"Error trying to save the DB, can't exit");
998 return REDIS_ERR;
999 }
1000 } else {
1001 redisLog(REDIS_WARNING,"Not saving DB.");
1002 }
1003 if (server.daemonize) unlink(server.pidfile);
1004 redisLog(REDIS_WARNING,"Server exit now, bye bye...");
1005 return REDIS_OK;
1006 }
1007
1008 /*================================== Commands =============================== */
1009
1010 void authCommand(redisClient *c) {
1011 if (!server.requirepass || !strcmp(c->argv[1]->ptr, server.requirepass)) {
1012 c->authenticated = 1;
1013 addReply(c,shared.ok);
1014 } else {
1015 c->authenticated = 0;
1016 addReplyError(c,"invalid password");
1017 }
1018 }
1019
1020 void pingCommand(redisClient *c) {
1021 addReply(c,shared.pong);
1022 }
1023
1024 void echoCommand(redisClient *c) {
1025 addReplyBulk(c,c->argv[1]);
1026 }
1027
1028 /* Convert an amount of bytes into a human readable string in the form
1029 * of 100B, 2G, 100M, 4K, and so forth. */
1030 void bytesToHuman(char *s, unsigned long long n) {
1031 double d;
1032
1033 if (n < 1024) {
1034 /* Bytes */
1035 sprintf(s,"%lluB",n);
1036 return;
1037 } else if (n < (1024*1024)) {
1038 d = (double)n/(1024);
1039 sprintf(s,"%.2fK",d);
1040 } else if (n < (1024LL*1024*1024)) {
1041 d = (double)n/(1024*1024);
1042 sprintf(s,"%.2fM",d);
1043 } else if (n < (1024LL*1024*1024*1024)) {
1044 d = (double)n/(1024LL*1024*1024);
1045 sprintf(s,"%.2fG",d);
1046 }
1047 }
1048
1049 /* Create the string returned by the INFO command. This is decoupled
1050 * by the INFO command itself as we need to report the same information
1051 * on memory corruption problems. */
1052 sds genRedisInfoString(void) {
1053 sds info;
1054 time_t uptime = time(NULL)-server.stat_starttime;
1055 int j;
1056 char hmem[64];
1057 struct rusage self_ru, c_ru;
1058
1059 getrusage(RUSAGE_SELF, &self_ru);
1060 getrusage(RUSAGE_CHILDREN, &c_ru);
1061
1062 bytesToHuman(hmem,zmalloc_used_memory());
1063 info = sdscatprintf(sdsempty(),
1064 "redis_version:%s\r\n"
1065 "redis_git_sha1:%s\r\n"
1066 "redis_git_dirty:%d\r\n"
1067 "arch_bits:%s\r\n"
1068 "multiplexing_api:%s\r\n"
1069 "process_id:%ld\r\n"
1070 "uptime_in_seconds:%ld\r\n"
1071 "uptime_in_days:%ld\r\n"
1072 "lru_clock:%ld\r\n"
1073 "used_cpu_sys:%.2f\r\n"
1074 "used_cpu_user:%.2f\r\n"
1075 "used_cpu_sys_childrens:%.2f\r\n"
1076 "used_cpu_user_childrens:%.2f\r\n"
1077 "connected_clients:%d\r\n"
1078 "connected_slaves:%d\r\n"
1079 "blocked_clients:%d\r\n"
1080 "used_memory:%zu\r\n"
1081 "used_memory_human:%s\r\n"
1082 "used_memory_rss:%zu\r\n"
1083 "used_memory_estimated_rss:%zu\r\n"
1084 "mem_fragmentation_ratio:%.2f\r\n"
1085 "use_tcmalloc:%d\r\n"
1086 "changes_since_last_save:%lld\r\n"
1087 "bgsave_in_progress:%d\r\n"
1088 "last_save_time:%ld\r\n"
1089 "bgrewriteaof_in_progress:%d\r\n"
1090 "total_connections_received:%lld\r\n"
1091 "total_commands_processed:%lld\r\n"
1092 "expired_keys:%lld\r\n"
1093 "keyspace_hits:%lld\r\n"
1094 "keyspace_misses:%lld\r\n"
1095 "hash_max_zipmap_entries:%zu\r\n"
1096 "hash_max_zipmap_value:%zu\r\n"
1097 "pubsub_channels:%ld\r\n"
1098 "pubsub_patterns:%u\r\n"
1099 "vm_enabled:%d\r\n"
1100 "role:%s\r\n"
1101 ,REDIS_VERSION,
1102 redisGitSHA1(),
1103 strtol(redisGitDirty(),NULL,10) > 0,
1104 (sizeof(long) == 8) ? "64" : "32",
1105 aeGetApiName(),
1106 (long) getpid(),
1107 uptime,
1108 uptime/(3600*24),
1109 (unsigned long) server.lruclock,
1110 (float)self_ru.ru_utime.tv_sec+(float)self_ru.ru_utime.tv_usec/1000000,
1111 (float)self_ru.ru_stime.tv_sec+(float)self_ru.ru_stime.tv_usec/1000000,
1112 (float)c_ru.ru_utime.tv_sec+(float)c_ru.ru_utime.tv_usec/1000000,
1113 (float)c_ru.ru_stime.tv_sec+(float)c_ru.ru_stime.tv_usec/1000000,
1114 listLength(server.clients)-listLength(server.slaves),
1115 listLength(server.slaves),
1116 server.blpop_blocked_clients,
1117 zmalloc_used_memory(),
1118 hmem,
1119 zmalloc_get_rss(),
1120 redisEstimateRSS(),
1121 zmalloc_get_fragmentation_ratio(),
1122 #ifdef USE_TCMALLOC
1123 1,
1124 #else
1125 0,
1126 #endif
1127 server.dirty,
1128 server.bgsavechildpid != -1,
1129 server.lastsave,
1130 server.bgrewritechildpid != -1,
1131 server.stat_numconnections,
1132 server.stat_numcommands,
1133 server.stat_expiredkeys,
1134 server.stat_keyspace_hits,
1135 server.stat_keyspace_misses,
1136 server.hash_max_zipmap_entries,
1137 server.hash_max_zipmap_value,
1138 dictSize(server.pubsub_channels),
1139 listLength(server.pubsub_patterns),
1140 server.vm_enabled != 0,
1141 server.masterhost == NULL ? "master" : "slave"
1142 );
1143 if (server.masterhost) {
1144 info = sdscatprintf(info,
1145 "master_host:%s\r\n"
1146 "master_port:%d\r\n"
1147 "master_link_status:%s\r\n"
1148 "master_last_io_seconds_ago:%d\r\n"
1149 ,server.masterhost,
1150 server.masterport,
1151 (server.replstate == REDIS_REPL_CONNECTED) ?
1152 "up" : "down",
1153 server.master ? ((int)(time(NULL)-server.master->lastinteraction)) : -1
1154 );
1155 }
1156 if (server.vm_enabled) {
1157 lockThreadedIO();
1158 info = sdscatprintf(info,
1159 "vm_conf_max_memory:%llu\r\n"
1160 "vm_conf_page_size:%llu\r\n"
1161 "vm_conf_pages:%llu\r\n"
1162 "vm_stats_used_pages:%llu\r\n"
1163 "vm_stats_swapped_objects:%llu\r\n"
1164 "vm_stats_swappin_count:%llu\r\n"
1165 "vm_stats_swappout_count:%llu\r\n"
1166 "vm_stats_io_newjobs_len:%lu\r\n"
1167 "vm_stats_io_processing_len:%lu\r\n"
1168 "vm_stats_io_processed_len:%lu\r\n"
1169 "vm_stats_io_active_threads:%lu\r\n"
1170 "vm_stats_blocked_clients:%lu\r\n"
1171 ,(unsigned long long) server.vm_max_memory,
1172 (unsigned long long) server.vm_page_size,
1173 (unsigned long long) server.vm_pages,
1174 (unsigned long long) server.vm_stats_used_pages,
1175 (unsigned long long) server.vm_stats_swapped_objects,
1176 (unsigned long long) server.vm_stats_swapins,
1177 (unsigned long long) server.vm_stats_swapouts,
1178 (unsigned long) listLength(server.io_newjobs),
1179 (unsigned long) listLength(server.io_processing),
1180 (unsigned long) listLength(server.io_processed),
1181 (unsigned long) server.io_active_threads,
1182 (unsigned long) server.vm_blocked_clients
1183 );
1184 unlockThreadedIO();
1185 }
1186 for (j = 0; j < server.dbnum; j++) {
1187 long long keys, vkeys;
1188
1189 keys = dictSize(server.db[j].dict);
1190 vkeys = dictSize(server.db[j].expires);
1191 if (keys || vkeys) {
1192 info = sdscatprintf(info, "db%d:keys=%lld,expires=%lld\r\n",
1193 j, keys, vkeys);
1194 }
1195 }
1196 return info;
1197 }
1198
1199 void infoCommand(redisClient *c) {
1200 sds info = genRedisInfoString();
1201 addReplySds(c,sdscatprintf(sdsempty(),"$%lu\r\n",
1202 (unsigned long)sdslen(info)));
1203 addReplySds(c,info);
1204 addReply(c,shared.crlf);
1205 }
1206
1207 void monitorCommand(redisClient *c) {
1208 /* ignore MONITOR if aleady slave or in monitor mode */
1209 if (c->flags & REDIS_SLAVE) return;
1210
1211 c->flags |= (REDIS_SLAVE|REDIS_MONITOR);
1212 c->slaveseldb = 0;
1213 listAddNodeTail(server.monitors,c);
1214 addReply(c,shared.ok);
1215 }
1216
1217 /* ============================ Maxmemory directive ======================== */
1218
1219 /* Try to free one object form the pre-allocated objects free list.
1220 * This is useful under low mem conditions as by default we take 1 million
1221 * free objects allocated. On success REDIS_OK is returned, otherwise
1222 * REDIS_ERR. */
1223 int tryFreeOneObjectFromFreelist(void) {
1224 robj *o;
1225
1226 if (server.vm_enabled) pthread_mutex_lock(&server.obj_freelist_mutex);
1227 if (listLength(server.objfreelist)) {
1228 listNode *head = listFirst(server.objfreelist);
1229 o = listNodeValue(head);
1230 listDelNode(server.objfreelist,head);
1231 if (server.vm_enabled) pthread_mutex_unlock(&server.obj_freelist_mutex);
1232 zfree(o);
1233 return REDIS_OK;
1234 } else {
1235 if (server.vm_enabled) pthread_mutex_unlock(&server.obj_freelist_mutex);
1236 return REDIS_ERR;
1237 }
1238 }
1239
1240 /* A fast RSS sampling function.
1241 *
1242 * The function is reasonably accurate while fast, since it uses the trick of
1243 * using the server.fragmentation ratio that is computed every second and
1244 * is the ratio between the RSS and our zmalloc() count of allocated bytes.
1245 *
1246 * So in order to compute the current RSS used we just need to multiply
1247 * the zmalloc() memory reporting, that is as fast as reading a counter,
1248 * for the latest estimation of fragmentation.
1249 *
1250 * The behavior of this function is also very desirable because it is
1251 * very responsive to memory changes: while the real RSS is actually measured
1252 * in pages, the RSS estimation will actually change even if just a few bytes
1253 * are freed, and this is a good property when the function is used in order
1254 * to evict keys for Virtual Memory of for 'maxmemory' directive.
1255 *
1256 * Note that when the memory reported by zmalloc is smaller than the RSS
1257 * (that is, fragmentation < 1) this means that something is odd (many pages
1258 * swapped since the Redis instance is idle) and we consider the fragmentation
1259 * ratio 1. */
1260 size_t redisEstimateRSS(void) {
1261 size_t used = zmalloc_used_memory();
1262 float maxfrag;
1263
1264 if (server.fragmentation < 1) return used;
1265 maxfrag = (float)SIZE_MAX / used;
1266
1267 /* If multiplying memory usage reported by zmalloc per fragmentation
1268 * ratio will lead to an overflow we just return SIZE_MAX. */
1269 if (maxfrag < server.fragmentation) return SIZE_MAX;
1270
1271 return (size_t)((used * server.fragmentation));
1272 }
1273
1274 /* This function gets called when 'maxmemory' is set on the config file to limit
1275 * the max memory used by the server, and we are out of memory.
1276 * This function will try to, in order:
1277 *
1278 * - Free objects from the free list
1279 * - Try to remove keys with an EXPIRE set
1280 *
1281 * It is not possible to free enough memory to reach used-memory < maxmemory
1282 * the server will start refusing commands that will enlarge even more the
1283 * memory usage.
1284 */
1285 void freeMemoryIfNeeded(void) {
1286 /* Remove keys accordingly to the active policy as long as we are
1287 * over the memory limit. */
1288 while (server.maxmemory && zmalloc_used_memory() > server.maxmemory) {
1289 int j, k, freed = 0;
1290
1291 /* Basic strategy -- remove objects from the free list. */
1292 if (tryFreeOneObjectFromFreelist() == REDIS_OK) continue;
1293
1294 for (j = 0; j < server.dbnum; j++) {
1295 long bestval = 0; /* just to prevent warning */
1296 sds bestkey = NULL;
1297 struct dictEntry *de;
1298 redisDb *db = server.db+j;
1299 dict *dict;
1300
1301 if (server.maxmemory_policy == REDIS_MAXMEMORY_ALLKEYS_LRU ||
1302 server.maxmemory_policy == REDIS_MAXMEMORY_ALLKEYS_RANDOM)
1303 {
1304 dict = server.db[j].dict;
1305 } else {
1306 dict = server.db[j].expires;
1307 }
1308 if (dictSize(dict) == 0) continue;
1309
1310 /* volatile-random and allkeys-random policy */
1311 if (server.maxmemory_policy == REDIS_MAXMEMORY_ALLKEYS_RANDOM ||
1312 server.maxmemory_policy == REDIS_MAXMEMORY_VOLATILE_RANDOM)
1313 {
1314 de = dictGetRandomKey(dict);
1315 bestkey = dictGetEntryKey(de);
1316 }
1317
1318 /* volatile-lru and allkeys-lru policy */
1319 else if (server.maxmemory_policy == REDIS_MAXMEMORY_ALLKEYS_LRU ||
1320 server.maxmemory_policy == REDIS_MAXMEMORY_VOLATILE_LRU)
1321 {
1322 for (k = 0; k < server.maxmemory_samples; k++) {
1323 sds thiskey;
1324 long thisval;
1325 robj *o;
1326
1327 de = dictGetRandomKey(dict);
1328 thiskey = dictGetEntryKey(de);
1329 o = dictGetEntryVal(de);
1330 thisval = estimateObjectIdleTime(o);
1331
1332 /* Higher idle time is better candidate for deletion */
1333 if (bestkey == NULL || thisval > bestval) {
1334 bestkey = thiskey;
1335 bestval = thisval;
1336 }
1337 }
1338 }
1339
1340 /* volatile-ttl */
1341 else if (server.maxmemory_policy == REDIS_MAXMEMORY_VOLATILE_TTL) {
1342 for (k = 0; k < server.maxmemory_samples; k++) {
1343 sds thiskey;
1344 long thisval;
1345
1346 de = dictGetRandomKey(dict);
1347 thiskey = dictGetEntryKey(de);
1348 thisval = (long) dictGetEntryVal(de);
1349
1350 /* Expire sooner (minor expire unix timestamp) is better
1351 * candidate for deletion */
1352 if (bestkey == NULL || thisval < bestval) {
1353 bestkey = thiskey;
1354 bestval = thisval;
1355 }
1356 }
1357 }
1358
1359 /* Finally remove the selected key. */
1360 if (bestkey) {
1361 robj *keyobj = createStringObject(bestkey,sdslen(bestkey));
1362 dbDelete(db,keyobj);
1363 server.stat_expiredkeys++;
1364 decrRefCount(keyobj);
1365 freed++;
1366 }
1367 }
1368 if (!freed) return; /* nothing to free... */
1369 }
1370
1371 while(0) {
1372 int j, k, freed = 0;
1373 for (j = 0; j < server.dbnum; j++) {
1374 int minttl = -1;
1375 sds minkey = NULL;
1376 robj *keyobj = NULL;
1377 struct dictEntry *de;
1378
1379 if (dictSize(server.db[j].expires)) {
1380 freed = 1;
1381 /* From a sample of three keys drop the one nearest to
1382 * the natural expire */
1383 for (k = 0; k < 3; k++) {
1384 time_t t;
1385
1386 de = dictGetRandomKey(server.db[j].expires);
1387 t = (time_t) dictGetEntryVal(de);
1388 if (minttl == -1 || t < minttl) {
1389 minkey = dictGetEntryKey(de);
1390 minttl = t;
1391 }
1392 }
1393 keyobj = createStringObject(minkey,sdslen(minkey));
1394 dbDelete(server.db+j,keyobj);
1395 server.stat_expiredkeys++;
1396 decrRefCount(keyobj);
1397 }
1398 }
1399 if (!freed) return; /* nothing to free... */
1400 }
1401 }
1402
1403 /* =================================== Main! ================================ */
1404
1405 #ifdef __linux__
1406 int linuxOvercommitMemoryValue(void) {
1407 FILE *fp = fopen("/proc/sys/vm/overcommit_memory","r");
1408 char buf[64];
1409
1410 if (!fp) return -1;
1411 if (fgets(buf,64,fp) == NULL) {
1412 fclose(fp);
1413 return -1;
1414 }
1415 fclose(fp);
1416
1417 return atoi(buf);
1418 }
1419
1420 void linuxOvercommitMemoryWarning(void) {
1421 if (linuxOvercommitMemoryValue() == 0) {
1422 redisLog(REDIS_WARNING,"WARNING overcommit_memory is set to 0! Background save may fail under low memory condition. To fix this issue add 'vm.overcommit_memory = 1' to /etc/sysctl.conf and then reboot or run the command 'sysctl vm.overcommit_memory=1' for this to take effect.");
1423 }
1424 }
1425 #endif /* __linux__ */
1426
1427 void createPidFile(void) {
1428 /* Try to write the pid file in a best-effort way. */
1429 FILE *fp = fopen(server.pidfile,"w");
1430 if (fp) {
1431 fprintf(fp,"%d\n",getpid());
1432 fclose(fp);
1433 }
1434 }
1435
1436 void daemonize(void) {
1437 int fd;
1438
1439 if (fork() != 0) exit(0); /* parent exits */
1440 setsid(); /* create a new session */
1441
1442 /* Every output goes to /dev/null. If Redis is daemonized but
1443 * the 'logfile' is set to 'stdout' in the configuration file
1444 * it will not log at all. */
1445 if ((fd = open("/dev/null", O_RDWR, 0)) != -1) {
1446 dup2(fd, STDIN_FILENO);
1447 dup2(fd, STDOUT_FILENO);
1448 dup2(fd, STDERR_FILENO);
1449 if (fd > STDERR_FILENO) close(fd);
1450 }
1451 }
1452
1453 void version() {
1454 printf("Redis server version %s (%s:%d)\n", REDIS_VERSION,
1455 redisGitSHA1(), atoi(redisGitDirty()) > 0);
1456 exit(0);
1457 }
1458
1459 void usage() {
1460 fprintf(stderr,"Usage: ./redis-server [/path/to/redis.conf]\n");
1461 fprintf(stderr," ./redis-server - (read config from stdin)\n");
1462 exit(1);
1463 }
1464
1465 int main(int argc, char **argv) {
1466 time_t start;
1467
1468 initServerConfig();
1469 sortCommandTable();
1470 if (argc == 2) {
1471 if (strcmp(argv[1], "-v") == 0 ||
1472 strcmp(argv[1], "--version") == 0) version();
1473 if (strcmp(argv[1], "--help") == 0) usage();
1474 resetServerSaveParams();
1475 loadServerConfig(argv[1]);
1476 } else if ((argc > 2)) {
1477 usage();
1478 } else {
1479 redisLog(REDIS_WARNING,"Warning: no config file specified, using the default config. In order to specify a config file use 'redis-server /path/to/redis.conf'");
1480 }
1481 if (server.daemonize) daemonize();
1482 initServer();
1483 if (server.daemonize) createPidFile();
1484 redisLog(REDIS_NOTICE,"Server started, Redis version " REDIS_VERSION);
1485 #ifdef __linux__
1486 linuxOvercommitMemoryWarning();
1487 #endif
1488 start = time(NULL);
1489 if (server.appendonly) {
1490 if (loadAppendOnlyFile(server.appendfilename) == REDIS_OK)
1491 redisLog(REDIS_NOTICE,"DB loaded from append only file: %ld seconds",time(NULL)-start);
1492 } else {
1493 if (rdbLoad(server.dbfilename) == REDIS_OK)
1494 redisLog(REDIS_NOTICE,"DB loaded from disk: %ld seconds",time(NULL)-start);
1495 }
1496 redisLog(REDIS_NOTICE,"The server is now ready to accept connections on port %d", server.port);
1497 aeSetBeforeSleepProc(server.el,beforeSleep);
1498 aeMain(server.el);
1499 aeDeleteEventLoop(server.el);
1500 return 0;
1501 }
1502
1503 /* ============================= Backtrace support ========================= */
1504
1505 #ifdef HAVE_BACKTRACE
1506 void *getMcontextEip(ucontext_t *uc) {
1507 #if defined(__FreeBSD__)
1508 return (void*) uc->uc_mcontext.mc_eip;
1509 #elif defined(__dietlibc__)
1510 return (void*) uc->uc_mcontext.eip;
1511 #elif defined(__APPLE__) && !defined(MAC_OS_X_VERSION_10_6)
1512 #if __x86_64__
1513 return (void*) uc->uc_mcontext->__ss.__rip;
1514 #else
1515 return (void*) uc->uc_mcontext->__ss.__eip;
1516 #endif
1517 #elif defined(__APPLE__) && defined(MAC_OS_X_VERSION_10_6)
1518 #if defined(_STRUCT_X86_THREAD_STATE64) && !defined(__i386__)
1519 return (void*) uc->uc_mcontext->__ss.__rip;
1520 #else
1521 return (void*) uc->uc_mcontext->__ss.__eip;
1522 #endif
1523 #elif defined(__i386__)
1524 return (void*) uc->uc_mcontext.gregs[14]; /* Linux 32 */
1525 #elif defined(__X86_64__) || defined(__x86_64__)
1526 return (void*) uc->uc_mcontext.gregs[16]; /* Linux 64 */
1527 #elif defined(__ia64__) /* Linux IA64 */
1528 return (void*) uc->uc_mcontext.sc_ip;
1529 #else
1530 return NULL;
1531 #endif
1532 }
1533
1534 void segvHandler(int sig, siginfo_t *info, void *secret) {
1535 void *trace[100];
1536 char **messages = NULL;
1537 int i, trace_size = 0;
1538 ucontext_t *uc = (ucontext_t*) secret;
1539 sds infostring;
1540 struct sigaction act;
1541 REDIS_NOTUSED(info);
1542
1543 redisLog(REDIS_WARNING,
1544 "======= Ooops! Redis %s got signal: -%d- =======", REDIS_VERSION, sig);
1545 infostring = genRedisInfoString();
1546 redisLog(REDIS_WARNING, "%s",infostring);
1547 /* It's not safe to sdsfree() the returned string under memory
1548 * corruption conditions. Let it leak as we are going to abort */
1549
1550 trace_size = backtrace(trace, 100);
1551 /* overwrite sigaction with caller's address */
1552 if (getMcontextEip(uc) != NULL) {
1553 trace[1] = getMcontextEip(uc);
1554 }
1555 messages = backtrace_symbols(trace, trace_size);
1556
1557 for (i=1; i<trace_size; ++i)
1558 redisLog(REDIS_WARNING,"%s", messages[i]);
1559
1560 /* free(messages); Don't call free() with possibly corrupted memory. */
1561 if (server.daemonize) unlink(server.pidfile);
1562
1563 /* Make sure we exit with the right signal at the end. So for instance
1564 * the core will be dumped if enabled. */
1565 sigemptyset (&act.sa_mask);
1566 /* When the SA_SIGINFO flag is set in sa_flags then sa_sigaction
1567 * is used. Otherwise, sa_handler is used */
1568 act.sa_flags = SA_NODEFER | SA_ONSTACK | SA_RESETHAND;
1569 act.sa_handler = SIG_DFL;
1570 sigaction (sig, &act, NULL);
1571 kill(getpid(),sig);
1572 }
1573
1574 void sigtermHandler(int sig) {
1575 REDIS_NOTUSED(sig);
1576
1577 redisLog(REDIS_WARNING,"SIGTERM received, scheduling shutting down...");
1578 server.shutdown_asap = 1;
1579 }
1580
1581 void setupSigSegvAction(void) {
1582 struct sigaction act;
1583
1584 sigemptyset (&act.sa_mask);
1585 /* When the SA_SIGINFO flag is set in sa_flags then sa_sigaction
1586 * is used. Otherwise, sa_handler is used */
1587 act.sa_flags = SA_NODEFER | SA_ONSTACK | SA_RESETHAND | SA_SIGINFO;
1588 act.sa_sigaction = segvHandler;
1589 sigaction (SIGSEGV, &act, NULL);
1590 sigaction (SIGBUS, &act, NULL);
1591 sigaction (SIGFPE, &act, NULL);
1592 sigaction (SIGILL, &act, NULL);
1593 sigaction (SIGBUS, &act, NULL);
1594
1595 act.sa_flags = SA_NODEFER | SA_ONSTACK | SA_RESETHAND;
1596 act.sa_handler = sigtermHandler;
1597 sigaction (SIGTERM, &act, NULL);
1598 return;
1599 }
1600
1601 #else /* HAVE_BACKTRACE */
1602 void setupSigSegvAction(void) {
1603 }
1604 #endif /* HAVE_BACKTRACE */
1605
1606 /* The End */