#include <sys/time.h>
#include <sys/resource.h>
#include <limits.h>
+#include <math.h>
+
+#if defined(__sun)
+#include "solarisfixes.h"
+#endif
#include "redis.h"
#include "ae.h" /* Event driven programming library */
/* Anti-warning macro... */
#define REDIS_NOTUSED(V) ((void) V)
+#define ZSKIPLIST_MAXLEVEL 32 /* Should be enough for 2^32 elements */
+#define ZSKIPLIST_P 0.25 /* Skiplist P = 1/4 */
+
+/* Append only defines */
+#define APPENDFSYNC_NO 0
+#define APPENDFSYNC_ALWAYS 1
+#define APPENDFSYNC_EVERYSEC 2
/*================================= Data types ============================== */
int maxidletime;
int dbnum;
int daemonize;
+ int appendonly;
+ int appendfsync;
+ time_t lastfsync;
+ int appendfd;
+ int appendseldb;
char *pidfile;
int bgsaveinprogress;
pid_t bgsavechildpid;
char *logfile;
char *bindaddr;
char *dbfilename;
+ char *appendfilename;
char *requirepass;
int shareobjects;
/* Replication related */
robj *pattern;
} redisSortOperation;
+/* ZSETs use a specialized version of Skiplists */
+
+typedef struct zskiplistNode {
+ struct zskiplistNode **forward;
+ struct zskiplistNode *backward;
+ double score;
+ robj *obj;
+} zskiplistNode;
+
+typedef struct zskiplist {
+ struct zskiplistNode *header, *tail;
+ unsigned long length;
+ int level;
+} zskiplist;
+
typedef struct zset {
dict *dict;
- tree *tree;
+ zskiplist *zsl;
} zset;
+/* Our shared "common" objects */
+
struct sharedObjectsStruct {
robj *crlf, *ok, *err, *emptybulk, *czero, *cone, *pong, *space,
*colon, *nullbulk, *nullmultibulk,
*select5, *select6, *select7, *select8, *select9;
} shared;
+/* Global vars that are actally used as constants. The following double
+ * values are used for double on-disk serialization, and are initialized
+ * at runtime to avoid strange compiler optimizations. */
+
+static double R_Zero, R_PosInf, R_NegInf, R_Nan;
+
/*================================ Prototypes =============================== */
static void freeStringObject(robj *o);
static int rdbSaveBackground(char *filename);
static robj *createStringObject(char *ptr, size_t len);
static void replicationFeedSlaves(list *slaves, struct redisCommand *cmd, int dictid, robj **argv, int argc);
+static void feedAppendOnlyFile(struct redisCommand *cmd, int dictid, robj **argv, int argc);
static int syncWithMaster(void);
static robj *tryObjectSharing(robj *o);
static int tryObjectEncoding(robj *o);
static void rdbRemoveTempFile(pid_t childpid);
static size_t stringObjectLen(robj *o);
static void processInputBuffer(redisClient *c);
+static zskiplist *zslCreate(void);
+static void zslFree(zskiplist *zsl);
+static void zslInsert(zskiplist *zsl, double score, robj *obj);
static void authCommand(redisClient *c);
static void pingCommand(redisClient *c);
static void mgetCommand(redisClient *c);
static void monitorCommand(redisClient *c);
static void expireCommand(redisClient *c);
+static void expireatCommand(redisClient *c);
static void getsetCommand(redisClient *c);
static void ttlCommand(redisClient *c);
static void slaveofCommand(redisClient *c);
static void debugCommand(redisClient *c);
static void msetCommand(redisClient *c);
static void msetnxCommand(redisClient *c);
+static void zaddCommand(redisClient *c);
+static void zrangeCommand(redisClient *c);
+static void zrangebyscoreCommand(redisClient *c);
+static void zrevrangeCommand(redisClient *c);
+static void zcardCommand(redisClient *c);
+static void zremCommand(redisClient *c);
+static void zscoreCommand(redisClient *c);
+static void zremrangebyscoreCommand(redisClient *c);
/*================================= Globals ================================= */
{"sdiff",sdiffCommand,-2,REDIS_CMD_INLINE|REDIS_CMD_DENYOOM},
{"sdiffstore",sdiffstoreCommand,-3,REDIS_CMD_INLINE|REDIS_CMD_DENYOOM},
{"smembers",sinterCommand,2,REDIS_CMD_INLINE},
+ {"zadd",zaddCommand,4,REDIS_CMD_BULK|REDIS_CMD_DENYOOM},
+ {"zrem",zremCommand,3,REDIS_CMD_BULK},
+ {"zremrangebyscore",zremrangebyscoreCommand,4,REDIS_CMD_INLINE},
+ {"zrange",zrangeCommand,4,REDIS_CMD_INLINE},
+ {"zrangebyscore",zrangebyscoreCommand,4,REDIS_CMD_INLINE},
+ {"zrevrange",zrevrangeCommand,4,REDIS_CMD_INLINE},
+ {"zcard",zcardCommand,2,REDIS_CMD_INLINE},
+ {"zscore",zscoreCommand,3,REDIS_CMD_BULK|REDIS_CMD_DENYOOM},
{"incrby",incrbyCommand,3,REDIS_CMD_INLINE|REDIS_CMD_DENYOOM},
{"decrby",decrbyCommand,3,REDIS_CMD_INLINE|REDIS_CMD_DENYOOM},
{"getset",getsetCommand,3,REDIS_CMD_BULK|REDIS_CMD_DENYOOM},
{"rename",renameCommand,3,REDIS_CMD_INLINE},
{"renamenx",renamenxCommand,3,REDIS_CMD_INLINE},
{"expire",expireCommand,3,REDIS_CMD_INLINE},
+ {"expireat",expireatCommand,3,REDIS_CMD_INLINE},
{"keys",keysCommand,2,REDIS_CMD_INLINE},
{"dbsize",dbsizeCommand,1,REDIS_CMD_INLINE},
{"auth",authCommand,2,REDIS_CMD_INLINE},
time_t now;
now = time(NULL);
- strftime(buf,64,"%d %b %H:%M:%S",gmtime(&now));
+ strftime(buf,64,"%d %b %H:%M:%S",localtime(&now));
fprintf(fp,"%s %c ",buf,c[level]);
vfprintf(fp, fmt, ap);
fprintf(fp,"\n");
static void appendServerSaveParams(time_t seconds, int changes) {
server.saveparams = zrealloc(server.saveparams,sizeof(struct saveparam)*(server.saveparamslen+1));
- if (server.saveparams == NULL) oom("appendServerSaveParams");
server.saveparams[server.saveparamslen].seconds = seconds;
server.saveparams[server.saveparamslen].changes = changes;
server.saveparamslen++;
server.bindaddr = NULL;
server.glueoutputbuf = 1;
server.daemonize = 0;
+ server.appendonly = 0;
+ server.appendfsync = APPENDFSYNC_ALWAYS;
+ server.lastfsync = time(NULL);
+ server.appendfd = -1;
+ server.appendseldb = -1; /* Make sure the first time will not match */
server.pidfile = "/var/run/redis.pid";
server.dbfilename = "dump.rdb";
+ server.appendfilename = "appendonly.log";
server.requirepass = NULL;
server.shareobjects = 0;
server.sharingpoolsize = 1024;
server.masterport = 6379;
server.master = NULL;
server.replstate = REDIS_REPL_NONE;
+
+ /* Double constants initialization */
+ R_Zero = 0.0;
+ R_PosInf = 1.0/R_Zero;
+ R_NegInf = -1.0/R_Zero;
+ R_Nan = R_Zero/R_Zero;
}
static void initServer() {
server.el = aeCreateEventLoop();
server.db = zmalloc(sizeof(redisDb)*server.dbnum);
server.sharingpool = dictCreate(&setDictType,NULL);
- if (!server.db || !server.clients || !server.slaves || !server.monitors || !server.el || !server.objfreelist)
- oom("server initialization"); /* Fatal OOM */
server.fd = anetTcpServer(server.neterr, server.port, server.bindaddr);
if (server.fd == -1) {
redisLog(REDIS_WARNING, "Opening TCP port: %s", server.neterr);
server.stat_numconnections = 0;
server.stat_starttime = time(NULL);
aeCreateTimeEvent(server.el, 1000, serverCron, NULL, NULL);
+
+ if (server.appendonly) {
+ server.appendfd = open(server.appendfilename,O_WRONLY|O_APPEND|O_CREAT,0644);
+ if (server.appendfd == -1) {
+ redisLog(REDIS_WARNING, "Can't open the append-only file: %s",
+ strerror(errno));
+ exit(1);
+ }
+ }
}
/* Empty the whole database */
if ((server.daemonize = yesnotoi(argv[1])) == -1) {
err = "argument must be 'yes' or 'no'"; goto loaderr;
}
+ } else if (!strcasecmp(argv[0],"appendonly") && argc == 2) {
+ if ((server.appendonly = yesnotoi(argv[1])) == -1) {
+ err = "argument must be 'yes' or 'no'"; goto loaderr;
+ }
+ } else if (!strcasecmp(argv[0],"appendfsync") && argc == 2) {
+ if (strcasecmp(argv[1],"no")) {
+ server.appendfsync = APPENDFSYNC_NO;
+ } else if (strcasecmp(argv[1],"always")) {
+ server.appendfsync = APPENDFSYNC_ALWAYS;
+ } else if (strcasecmp(argv[1],"everysec")) {
+ server.appendfsync = APPENDFSYNC_EVERYSEC;
+ } else {
+ err = "argument must be 'no', 'always' or 'everysec'";
+ goto loaderr;
+ }
} else if (!strcasecmp(argv[0],"requirepass") && argc == 2) {
server.requirepass = zstrdup(argv[1]);
} else if (!strcasecmp(argv[0],"pidfile") && argc == 2) {
}
/* Now the output buffer is empty, add the new single element */
o = createObject(REDIS_STRING,sdsnewlen(buf,totlen));
- if (!listAddNodeTail(c->reply,o)) oom("listAddNodeTail");
+ listAddNodeTail(c->reply,o);
}
}
/* Exec the command */
dirty = server.dirty;
cmd->proc(c);
+ if (server.appendonly != 0)
+ feedAppendOnlyFile(cmd,c->db->id,c->argv,c->argc);
if (server.dirty-dirty != 0 && listLength(server.slaves))
replicationFeedSlaves(server.slaves,cmd,c->db->id,c->argv,c->argc);
if (listLength(server.monitors))
outv = static_outv;
} else {
outv = zmalloc(sizeof(robj*)*(argc*2+1));
- if (!outv) oom("replicationFeedSlaves");
}
for (j = 0; j < argc; j++) {
if (outv != static_outv) zfree(outv);
}
+/* TODO: translate EXPIREs into EXPIRETOs */
+static void feedAppendOnlyFile(struct redisCommand *cmd, int dictid, robj **argv, int argc) {
+ sds buf = sdsempty();
+ int j;
+ ssize_t nwritten;
+ time_t now;
+ robj *tmpargv[3];
+
+ /* The DB this command was targetting is not the same as the last command
+ * we appendend. To issue a SELECT command is needed. */
+ if (dictid != server.appendseldb) {
+ char seldb[64];
+
+ snprintf(seldb,sizeof(seldb),"%d",dictid);
+ buf = sdscatprintf(buf,"*2\r\n$6\r\nSELECT\r\n$%d\r\n%s\r\n",
+ strlen(seldb),seldb);
+ server.appendseldb = dictid;
+ }
+
+ /* "Fix" the argv vector if the command is EXPIRE. We want to translate
+ * EXPIREs into EXPIREATs calls */
+ if (cmd->proc == expireCommand) {
+ long when;
+
+ tmpargv[0] = createStringObject("EXPIREAT",8);
+ tmpargv[1] = argv[1];
+ incrRefCount(argv[1]);
+ when = time(NULL)+strtol(argv[2]->ptr,NULL,10);
+ tmpargv[2] = createObject(REDIS_STRING,
+ sdscatprintf(sdsempty(),"%ld",when));
+ argv = tmpargv;
+ }
+
+ /* Append the actual command */
+ buf = sdscatprintf(buf,"*%d\r\n",argc);
+ for (j = 0; j < argc; j++) {
+ robj *o = argv[j];
+
+ if (o->encoding != REDIS_ENCODING_RAW)
+ o = getDecodedObject(o);
+ buf = sdscatprintf(buf,"$%d\r\n",sdslen(o->ptr));
+ buf = sdscatlen(buf,o->ptr,sdslen(o->ptr));
+ buf = sdscatlen(buf,"\r\n",2);
+ if (o != argv[j])
+ decrRefCount(o);
+ }
+
+ /* Free the objects from the modified argv for EXPIREAT */
+ if (cmd->proc == expireCommand) {
+ for (j = 0; j < 3; j++)
+ decrRefCount(argv[j]);
+ }
+
+ /* We want to perform a single write. This should be guaranteed atomic
+ * at least if the filesystem we are writing is a real physical one.
+ * While this will save us against the server being killed I don't think
+ * there is much to do about the whole server stopping for power problems
+ * or alike */
+ nwritten = write(server.appendfd,buf,sdslen(buf));
+ if (nwritten != (signed)sdslen(buf)) {
+ /* Ooops, we are in troubles. The best thing to do for now is
+ * to simply exit instead to give the illusion that everything is
+ * working as expected. */
+ if (nwritten == -1) {
+ redisLog(REDIS_WARNING,"Aborting on error writing to the append-only file: %s",strerror(errno));
+ } else {
+ redisLog(REDIS_WARNING,"Aborting on short write while writing to the append-only file: %s",strerror(errno));
+ }
+ abort();
+ }
+ now = time(NULL);
+ if (server.appendfsync == APPENDFSYNC_ALWAYS ||
+ (server.appendfsync == APPENDFSYNC_EVERYSEC &&
+ now-server.lastfsync > 1))
+ {
+ fsync(server.appendfd); /* Let's try to get this data on the disk */
+ server.lastfsync = now;
+ }
+}
+
static void processInputBuffer(redisClient *c) {
again:
if (c->bulklen == -1) {
return;
}
argv = sdssplitlen(query,sdslen(query)," ",1,&argc);
- if (argv == NULL) oom("sdssplitlen");
sdsfree(query);
if (c->argv) zfree(c->argv);
c->argv = zmalloc(sizeof(robj*)*argc);
- if (c->argv == NULL) oom("allocating arguments list for client");
for (j = 0; j < argc; j++) {
if (sdslen(argv[j])) {
c->lastinteraction = time(NULL);
c->authenticated = 0;
c->replstate = REDIS_REPL_NONE;
- if ((c->reply = listCreate()) == NULL) oom("listCreate");
+ c->reply = listCreate();
listSetFreeMethod(c->reply,decrRefCount);
listSetDupMethod(c->reply,dupClientReplyValue);
if (aeCreateFileEvent(server.el, c->fd, AE_READABLE,
freeClient(c);
return NULL;
}
- if (!listAddNodeTail(server.clients,c)) oom("listAddNodeTail");
+ listAddNodeTail(server.clients,c);
return c;
}
} else {
incrRefCount(obj);
}
- if (!listAddNodeTail(c->reply,obj)) oom("listAddNodeTail");
+ listAddNodeTail(c->reply,obj);
}
static void addReplySds(redisClient *c, sds s) {
} else {
o = zmalloc(sizeof(*o));
}
- if (!o) oom("createObject");
o->type = type;
o->encoding = REDIS_ENCODING_RAW;
o->ptr = ptr;
static robj *createListObject(void) {
list *l = listCreate();
- if (!l) oom("listCreate");
listSetFreeMethod(l,decrRefCount);
return createObject(REDIS_LIST,l);
}
static robj *createSetObject(void) {
dict *d = dictCreate(&setDictType,NULL);
- if (!d) oom("dictCreate");
return createObject(REDIS_SET,d);
}
static robj *createZsetObject(void) {
- dict *d = dictCreate(&zsetDictType,NULL);
- if (!d) oom("dictCreate");
- return createObject(REDIS_ZSET,d);
+ zset *zs = zmalloc(sizeof(*zs));
+
+ zs->dict = dictCreate(&zsetDictType,NULL);
+ zs->zsl = zslCreate();
+ return createObject(REDIS_ZSET,zs);
}
static void freeStringObject(robj *o) {
dictRelease((dict*) o->ptr);
}
+static void freeZsetObject(robj *o) {
+ zset *zs = o->ptr;
+
+ dictRelease(zs->dict);
+ zslFree(zs->zsl);
+ zfree(zs);
+}
+
static void freeHashObject(robj *o) {
dictRelease((dict*) o->ptr);
}
case REDIS_STRING: freeStringObject(o); break;
case REDIS_LIST: freeListObject(o); break;
case REDIS_SET: freeSetObject(o); break;
+ case REDIS_ZSET: freeZsetObject(o); break;
case REDIS_HASH: freeHashObject(o); break;
default: assert(0 != 0); break;
}
}
}
+/* Compare two string objects via strcmp() or alike.
+ * Note that the objects may be integer-encoded. In such a case we
+ * use snprintf() to get a string representation of the numbers on the stack
+ * and compare the strings, it's much faster than calling getDecodedObject(). */
static int compareStringObjects(robj *a, robj *b) {
assert(a->type == REDIS_STRING && b->type == REDIS_STRING);
-
- if (a->encoding == REDIS_ENCODING_INT && b->encoding == REDIS_ENCODING_INT){
- return (long)a->ptr - (long)b->ptr;
+ char bufa[128], bufb[128], *astr, *bstr;
+ int bothsds = 1;
+
+ if (a == b) return 0;
+ if (a->encoding != REDIS_ENCODING_RAW) {
+ snprintf(bufa,sizeof(bufa),"%ld",(long) a->ptr);
+ astr = bufa;
+ bothsds = 0;
} else {
- int retval;
-
- incrRefCount(a);
- incrRefCount(b);
- if (a->encoding != REDIS_ENCODING_RAW) a = getDecodedObject(a);
- if (b->encoding != REDIS_ENCODING_RAW) b = getDecodedObject(a);
- retval = sdscmp(a->ptr,b->ptr);
- decrRefCount(a);
- decrRefCount(b);
- return retval;
+ astr = a->ptr;
+ }
+ if (b->encoding != REDIS_ENCODING_RAW) {
+ snprintf(bufb,sizeof(bufb),"%ld",(long) b->ptr);
+ bstr = bufb;
+ bothsds = 0;
+ } else {
+ bstr = b->ptr;
}
+ return bothsds ? sdscmp(astr,bstr) : strcmp(astr,bstr);
}
static size_t stringObjectLen(robj *o) {
}
}
+/* Save a double value. Doubles are saved as strings prefixed by an unsigned
+ * 8 bit integer specifing the length of the representation.
+ * This 8 bit integer has special values in order to specify the following
+ * conditions:
+ * 253: not a number
+ * 254: + inf
+ * 255: - inf
+ */
+static int rdbSaveDoubleValue(FILE *fp, double val) {
+ unsigned char buf[128];
+ int len;
+
+ if (isnan(val)) {
+ buf[0] = 253;
+ len = 1;
+ } else if (!isfinite(val)) {
+ len = 1;
+ buf[0] = (val < 0) ? 255 : 254;
+ } else {
+ snprintf((char*)buf+1,sizeof(buf)-1,"%.16g",val);
+ buf[0] = strlen((char*)buf);
+ len = buf[0]+1;
+ }
+ if (fwrite(buf,len,1,fp) == 0) return -1;
+ return 0;
+}
+
/* Save the DB on disk. Return REDIS_ERR on error, REDIS_OK on success */
static int rdbSave(char *filename) {
dictIterator *di = NULL;
dictIterator *di = dictGetIterator(set);
dictEntry *de;
- if (!set) oom("dictGetIteraotr");
if (rdbSaveLen(fp,dictSize(set)) == -1) goto werr;
while((de = dictNext(di)) != NULL) {
robj *eleobj = dictGetEntryKey(de);
if (rdbSaveStringObject(fp,eleobj) == -1) goto werr;
}
dictReleaseIterator(di);
+ } else if (o->type == REDIS_ZSET) {
+ /* Save a set value */
+ zset *zs = o->ptr;
+ dictIterator *di = dictGetIterator(zs->dict);
+ dictEntry *de;
+
+ if (rdbSaveLen(fp,dictSize(zs->dict)) == -1) goto werr;
+ while((de = dictNext(di)) != NULL) {
+ robj *eleobj = dictGetEntryKey(de);
+ double *score = dictGetEntryVal(de);
+
+ if (rdbSaveStringObject(fp,eleobj) == -1) goto werr;
+ if (rdbSaveDoubleValue(fp,*score) == -1) goto werr;
+ }
+ dictReleaseIterator(di);
} else {
assert(0 != 0);
}
/* Use RENAME to make sure the DB file is changed atomically only
* if the generate DB file is ok. */
if (rename(tmpfile,filename) == -1) {
- redisLog(REDIS_WARNING,"Error moving temp DB file on the final destionation: %s", strerror(errno));
+ redisLog(REDIS_WARNING,"Error moving temp DB file on the final destination: %s", strerror(errno));
unlink(tmpfile);
return REDIS_ERR;
}
return tryObjectSharing(createObject(REDIS_STRING,val));
}
+/* For information about double serialization check rdbSaveDoubleValue() */
+static int rdbLoadDoubleValue(FILE *fp, double *val) {
+ char buf[128];
+ unsigned char len;
+
+ if (fread(&len,1,1,fp) == 0) return -1;
+ switch(len) {
+ case 255: *val = R_NegInf; return 0;
+ case 254: *val = R_PosInf; return 0;
+ case 253: *val = R_Nan; return 0;
+ default:
+ if (fread(buf,len,1,fp) == 0) return -1;
+ sscanf(buf, "%lg", val);
+ return 0;
+ }
+}
+
static int rdbLoad(char *filename) {
FILE *fp;
robj *keyobj = NULL;
if ((ele = rdbLoadStringObject(fp,rdbver)) == NULL) goto eoferr;
tryObjectEncoding(ele);
if (type == REDIS_LIST) {
- if (!listAddNodeTail((list*)o->ptr,ele))
- oom("listAddNodeTail");
+ listAddNodeTail((list*)o->ptr,ele);
} else {
- if (dictAdd((dict*)o->ptr,ele,NULL) == DICT_ERR)
- oom("dictAdd");
+ dictAdd((dict*)o->ptr,ele,NULL);
}
}
+ } else if (type == REDIS_ZSET) {
+ /* Read list/set value */
+ uint32_t zsetlen;
+ zset *zs;
+
+ if ((zsetlen = rdbLoadLen(fp,rdbver,NULL)) == REDIS_RDB_LENERR)
+ goto eoferr;
+ o = createZsetObject();
+ zs = o->ptr;
+ /* Load every single element of the list/set */
+ while(zsetlen--) {
+ robj *ele;
+ double *score = zmalloc(sizeof(double));
+
+ if ((ele = rdbLoadStringObject(fp,rdbver)) == NULL) goto eoferr;
+ tryObjectEncoding(ele);
+ if (rdbLoadDoubleValue(fp,score) == -1) goto eoferr;
+ dictAdd(zs->dict,ele,score);
+ zslInsert(zs->zsl,*score,ele);
+ incrRefCount(ele); /* added to skiplist */
+ }
} else {
assert(0 != 0);
}
addReply(c,shared.ok);
} else {
c->authenticated = 0;
- addReply(c,shared.err);
+ addReplySds(c,sdscatprintf(sdsempty(),"-ERR invalid password\r\n"));
}
}
robj *lenobj = createObject(REDIS_STRING,NULL);
di = dictGetIterator(c->db->dict);
- if (!di) oom("dictGetIterator");
addReply(c,lenobj);
decrRefCount(lenobj);
while((de = dictNext(di)) != NULL) {
lobj = createListObject();
list = lobj->ptr;
if (where == REDIS_HEAD) {
- if (!listAddNodeHead(list,c->argv[2])) oom("listAddNodeHead");
+ listAddNodeHead(list,c->argv[2]);
} else {
- if (!listAddNodeTail(list,c->argv[2])) oom("listAddNodeTail");
+ listAddNodeTail(list,c->argv[2]);
}
dictAdd(c->db->dict,c->argv[1],lobj);
incrRefCount(c->argv[1]);
}
list = lobj->ptr;
if (where == REDIS_HEAD) {
- if (!listAddNodeHead(list,c->argv[2])) oom("listAddNodeHead");
+ listAddNodeHead(list,c->argv[2]);
} else {
- if (!listAddNodeTail(list,c->argv[2])) oom("listAddNodeTail");
+ listAddNodeTail(list,c->argv[2]);
}
incrRefCount(c->argv[2]);
}
robj *lenobj = NULL, *dstset = NULL;
int j, cardinality = 0;
- if (!dv) oom("sinterGenericCommand");
for (j = 0; j < setsnum; j++) {
robj *setobj;
* the element against all the other sets, if at least one set does
* not include the element it is discarded */
di = dictGetIterator(dv[0]);
- if (!di) oom("dictGetIterator");
while((de = dictNext(di)) != NULL) {
robj *ele;
robj *dstset = NULL;
int j, cardinality = 0;
- if (!dv) oom("sunionDiffGenericCommand");
for (j = 0; j < setsnum; j++) {
robj *setobj;
if (!dv[j]) continue; /* non existing keys are like empty sets */
di = dictGetIterator(dv[j]);
- if (!di) oom("dictGetIterator");
while((de = dictNext(di)) != NULL) {
robj *ele;
if (!dstkey) {
addReplySds(c,sdscatprintf(sdsempty(),"*%d\r\n",cardinality));
di = dictGetIterator(dstset->ptr);
- if (!di) oom("dictGetIterator");
while((de = dictNext(di)) != NULL) {
robj *ele;
sunionDiffGenericCommand(c,c->argv+2,c->argc-2,c->argv[1],REDIS_OP_DIFF);
}
+/* ==================================== ZSets =============================== */
+
+/* ZSETs are ordered sets using two data structures to hold the same elements
+ * in order to get O(log(N)) INSERT and REMOVE operations into a sorted
+ * data structure.
+ *
+ * The elements are added to an hash table mapping Redis objects to scores.
+ * At the same time the elements are added to a skip list mapping scores
+ * to Redis objects (so objects are sorted by scores in this "view"). */
+
+/* This skiplist implementation is almost a C translation of the original
+ * algorithm described by William Pugh in "Skip Lists: A Probabilistic
+ * Alternative to Balanced Trees", modified in three ways:
+ * a) this implementation allows for repeated values.
+ * b) the comparison is not just by key (our 'score') but by satellite data.
+ * c) there is a back pointer, so it's a doubly linked list with the back
+ * pointers being only at "level 1". This allows to traverse the list
+ * from tail to head, useful for ZREVRANGE. */
+
+static zskiplistNode *zslCreateNode(int level, double score, robj *obj) {
+ zskiplistNode *zn = zmalloc(sizeof(*zn));
+
+ zn->forward = zmalloc(sizeof(zskiplistNode*) * level);
+ zn->score = score;
+ zn->obj = obj;
+ return zn;
+}
+
+static zskiplist *zslCreate(void) {
+ int j;
+ zskiplist *zsl;
+
+ zsl = zmalloc(sizeof(*zsl));
+ zsl->level = 1;
+ zsl->length = 0;
+ zsl->header = zslCreateNode(ZSKIPLIST_MAXLEVEL,0,NULL);
+ for (j = 0; j < ZSKIPLIST_MAXLEVEL; j++)
+ zsl->header->forward[j] = NULL;
+ zsl->header->backward = NULL;
+ zsl->tail = NULL;
+ return zsl;
+}
+
+static void zslFreeNode(zskiplistNode *node) {
+ decrRefCount(node->obj);
+ zfree(node->forward);
+ zfree(node);
+}
+
+static void zslFree(zskiplist *zsl) {
+ zskiplistNode *node = zsl->header->forward[0], *next;
+
+ zfree(zsl->header->forward);
+ zfree(zsl->header);
+ while(node) {
+ next = node->forward[0];
+ zslFreeNode(node);
+ node = next;
+ }
+ zfree(zsl);
+}
+
+static int zslRandomLevel(void) {
+ int level = 1;
+ while ((random()&0xFFFF) < (ZSKIPLIST_P * 0xFFFF))
+ level += 1;
+ return level;
+}
+
+static void zslInsert(zskiplist *zsl, double score, robj *obj) {
+ zskiplistNode *update[ZSKIPLIST_MAXLEVEL], *x;
+ int i, level;
+
+ x = zsl->header;
+ for (i = zsl->level-1; i >= 0; i--) {
+ while (x->forward[i] &&
+ (x->forward[i]->score < score ||
+ (x->forward[i]->score == score &&
+ compareStringObjects(x->forward[i]->obj,obj) < 0)))
+ x = x->forward[i];
+ update[i] = x;
+ }
+ /* we assume the key is not already inside, since we allow duplicated
+ * scores, and the re-insertion of score and redis object should never
+ * happpen since the caller of zslInsert() should test in the hash table
+ * if the element is already inside or not. */
+ level = zslRandomLevel();
+ if (level > zsl->level) {
+ for (i = zsl->level; i < level; i++)
+ update[i] = zsl->header;
+ zsl->level = level;
+ }
+ x = zslCreateNode(level,score,obj);
+ for (i = 0; i < level; i++) {
+ x->forward[i] = update[i]->forward[i];
+ update[i]->forward[i] = x;
+ }
+ x->backward = (update[0] == zsl->header) ? NULL : update[0];
+ if (x->forward[0])
+ x->forward[0]->backward = x;
+ else
+ zsl->tail = x;
+ zsl->length++;
+}
+
+/* Delete an element with matching score/object from the skiplist. */
+static int zslDelete(zskiplist *zsl, double score, robj *obj) {
+ zskiplistNode *update[ZSKIPLIST_MAXLEVEL], *x;
+ int i;
+
+ x = zsl->header;
+ for (i = zsl->level-1; i >= 0; i--) {
+ while (x->forward[i] &&
+ (x->forward[i]->score < score ||
+ (x->forward[i]->score == score &&
+ compareStringObjects(x->forward[i]->obj,obj) < 0)))
+ x = x->forward[i];
+ update[i] = x;
+ }
+ /* We may have multiple elements with the same score, what we need
+ * is to find the element with both the right score and object. */
+ x = x->forward[0];
+ if (x && score == x->score && compareStringObjects(x->obj,obj) == 0) {
+ for (i = 0; i < zsl->level; i++) {
+ if (update[i]->forward[i] != x) break;
+ update[i]->forward[i] = x->forward[i];
+ }
+ if (x->forward[0]) {
+ x->forward[0]->backward = (x->backward == zsl->header) ?
+ NULL : x->backward;
+ } else {
+ zsl->tail = x->backward;
+ }
+ zslFreeNode(x);
+ while(zsl->level > 1 && zsl->header->forward[zsl->level-1] == NULL)
+ zsl->level--;
+ zsl->length--;
+ return 1;
+ } else {
+ return 0; /* not found */
+ }
+ return 0; /* not found */
+}
+
+/* Delete all the elements with score between min and max from the skiplist.
+ * Min and mx are inclusive, so a score >= min || score <= max is deleted.
+ * Note that this function takes the reference to the hash table view of the
+ * sorted set, in order to remove the elements from the hash table too. */
+static unsigned long zslDeleteRange(zskiplist *zsl, double min, double max, dict *dict) {
+ zskiplistNode *update[ZSKIPLIST_MAXLEVEL], *x;
+ unsigned long removed = 0;
+ int i;
+
+ x = zsl->header;
+ for (i = zsl->level-1; i >= 0; i--) {
+ while (x->forward[i] && x->forward[i]->score < min)
+ x = x->forward[i];
+ update[i] = x;
+ }
+ /* We may have multiple elements with the same score, what we need
+ * is to find the element with both the right score and object. */
+ x = x->forward[0];
+ while (x && x->score <= max) {
+ zskiplistNode *next;
+
+ for (i = 0; i < zsl->level; i++) {
+ if (update[i]->forward[i] != x) break;
+ update[i]->forward[i] = x->forward[i];
+ }
+ if (x->forward[0]) {
+ x->forward[0]->backward = (x->backward == zsl->header) ?
+ NULL : x->backward;
+ } else {
+ zsl->tail = x->backward;
+ }
+ next = x->forward[0];
+ dictDelete(dict,x->obj);
+ zslFreeNode(x);
+ while(zsl->level > 1 && zsl->header->forward[zsl->level-1] == NULL)
+ zsl->level--;
+ zsl->length--;
+ removed++;
+ x = next;
+ }
+ return removed; /* not found */
+}
+
+/* Find the first node having a score equal or greater than the specified one.
+ * Returns NULL if there is no match. */
+static zskiplistNode *zslFirstWithScore(zskiplist *zsl, double score) {
+ zskiplistNode *x;
+ int i;
+
+ x = zsl->header;
+ for (i = zsl->level-1; i >= 0; i--) {
+ while (x->forward[i] && x->forward[i]->score < score)
+ x = x->forward[i];
+ }
+ /* We may have multiple elements with the same score, what we need
+ * is to find the element with both the right score and object. */
+ return x->forward[0];
+}
+
+/* The actual Z-commands implementations */
+
+static void zaddCommand(redisClient *c) {
+ robj *zsetobj;
+ zset *zs;
+ double *score;
+
+ zsetobj = lookupKeyWrite(c->db,c->argv[1]);
+ if (zsetobj == NULL) {
+ zsetobj = createZsetObject();
+ dictAdd(c->db->dict,c->argv[1],zsetobj);
+ incrRefCount(c->argv[1]);
+ } else {
+ if (zsetobj->type != REDIS_ZSET) {
+ addReply(c,shared.wrongtypeerr);
+ return;
+ }
+ }
+ score = zmalloc(sizeof(double));
+ *score = strtod(c->argv[2]->ptr,NULL);
+ zs = zsetobj->ptr;
+ if (dictAdd(zs->dict,c->argv[3],score) == DICT_OK) {
+ /* case 1: New element */
+ incrRefCount(c->argv[3]); /* added to hash */
+ zslInsert(zs->zsl,*score,c->argv[3]);
+ incrRefCount(c->argv[3]); /* added to skiplist */
+ server.dirty++;
+ addReply(c,shared.cone);
+ } else {
+ dictEntry *de;
+ double *oldscore;
+
+ /* case 2: Score update operation */
+ de = dictFind(zs->dict,c->argv[3]);
+ assert(de != NULL);
+ oldscore = dictGetEntryVal(de);
+ if (*score != *oldscore) {
+ int deleted;
+
+ deleted = zslDelete(zs->zsl,*oldscore,c->argv[3]);
+ assert(deleted != 0);
+ zslInsert(zs->zsl,*score,c->argv[3]);
+ incrRefCount(c->argv[3]);
+ dictReplace(zs->dict,c->argv[3],score);
+ server.dirty++;
+ } else {
+ zfree(score);
+ }
+ addReply(c,shared.czero);
+ }
+}
+
+static void zremCommand(redisClient *c) {
+ robj *zsetobj;
+ zset *zs;
+
+ zsetobj = lookupKeyWrite(c->db,c->argv[1]);
+ if (zsetobj == NULL) {
+ addReply(c,shared.czero);
+ } else {
+ dictEntry *de;
+ double *oldscore;
+ int deleted;
+
+ if (zsetobj->type != REDIS_ZSET) {
+ addReply(c,shared.wrongtypeerr);
+ return;
+ }
+ zs = zsetobj->ptr;
+ de = dictFind(zs->dict,c->argv[2]);
+ if (de == NULL) {
+ addReply(c,shared.czero);
+ return;
+ }
+ /* Delete from the skiplist */
+ oldscore = dictGetEntryVal(de);
+ deleted = zslDelete(zs->zsl,*oldscore,c->argv[2]);
+ assert(deleted != 0);
+
+ /* Delete from the hash table */
+ dictDelete(zs->dict,c->argv[2]);
+ if (htNeedsResize(zs->dict)) dictResize(zs->dict);
+ server.dirty++;
+ addReply(c,shared.cone);
+ }
+}
+
+static void zremrangebyscoreCommand(redisClient *c) {
+ double min = strtod(c->argv[2]->ptr,NULL);
+ double max = strtod(c->argv[3]->ptr,NULL);
+ robj *zsetobj;
+ zset *zs;
+
+ zsetobj = lookupKeyWrite(c->db,c->argv[1]);
+ if (zsetobj == NULL) {
+ addReply(c,shared.czero);
+ } else {
+ long deleted;
+
+ if (zsetobj->type != REDIS_ZSET) {
+ addReply(c,shared.wrongtypeerr);
+ return;
+ }
+ zs = zsetobj->ptr;
+ deleted = zslDeleteRange(zs->zsl,min,max,zs->dict);
+ if (htNeedsResize(zs->dict)) dictResize(zs->dict);
+ server.dirty += deleted;
+ addReplySds(c,sdscatprintf(sdsempty(),":%lu\r\n",deleted));
+ }
+}
+
+static void zrangeGenericCommand(redisClient *c, int reverse) {
+ robj *o;
+ int start = atoi(c->argv[2]->ptr);
+ int end = atoi(c->argv[3]->ptr);
+
+ o = lookupKeyRead(c->db,c->argv[1]);
+ if (o == NULL) {
+ addReply(c,shared.nullmultibulk);
+ } else {
+ if (o->type != REDIS_ZSET) {
+ addReply(c,shared.wrongtypeerr);
+ } else {
+ zset *zsetobj = o->ptr;
+ zskiplist *zsl = zsetobj->zsl;
+ zskiplistNode *ln;
+
+ int llen = zsl->length;
+ int rangelen, j;
+ robj *ele;
+
+ /* convert negative indexes */
+ if (start < 0) start = llen+start;
+ if (end < 0) end = llen+end;
+ if (start < 0) start = 0;
+ if (end < 0) end = 0;
+
+ /* indexes sanity checks */
+ if (start > end || start >= llen) {
+ /* Out of range start or start > end result in empty list */
+ addReply(c,shared.emptymultibulk);
+ return;
+ }
+ if (end >= llen) end = llen-1;
+ rangelen = (end-start)+1;
+
+ /* Return the result in form of a multi-bulk reply */
+ if (reverse) {
+ ln = zsl->tail;
+ while (start--)
+ ln = ln->backward;
+ } else {
+ ln = zsl->header->forward[0];
+ while (start--)
+ ln = ln->forward[0];
+ }
+
+ addReplySds(c,sdscatprintf(sdsempty(),"*%d\r\n",rangelen));
+ for (j = 0; j < rangelen; j++) {
+ ele = ln->obj;
+ addReplyBulkLen(c,ele);
+ addReply(c,ele);
+ addReply(c,shared.crlf);
+ ln = reverse ? ln->backward : ln->forward[0];
+ }
+ }
+ }
+}
+
+static void zrangeCommand(redisClient *c) {
+ zrangeGenericCommand(c,0);
+}
+
+static void zrevrangeCommand(redisClient *c) {
+ zrangeGenericCommand(c,1);
+}
+
+static void zrangebyscoreCommand(redisClient *c) {
+ robj *o;
+ double min = strtod(c->argv[2]->ptr,NULL);
+ double max = strtod(c->argv[3]->ptr,NULL);
+
+ o = lookupKeyRead(c->db,c->argv[1]);
+ if (o == NULL) {
+ addReply(c,shared.nullmultibulk);
+ } else {
+ if (o->type != REDIS_ZSET) {
+ addReply(c,shared.wrongtypeerr);
+ } else {
+ zset *zsetobj = o->ptr;
+ zskiplist *zsl = zsetobj->zsl;
+ zskiplistNode *ln;
+ robj *ele, *lenobj;
+ unsigned int rangelen = 0;
+
+ /* Get the first node with the score >= min */
+ ln = zslFirstWithScore(zsl,min);
+ if (ln == NULL) {
+ /* No element matching the speciifed interval */
+ addReply(c,shared.emptymultibulk);
+ return;
+ }
+
+ /* We don't know in advance how many matching elements there
+ * are in the list, so we push this object that will represent
+ * the multi-bulk length in the output buffer, and will "fix"
+ * it later */
+ lenobj = createObject(REDIS_STRING,NULL);
+ addReply(c,lenobj);
+
+ while(ln && ln->score <= max) {
+ ele = ln->obj;
+ addReplyBulkLen(c,ele);
+ addReply(c,ele);
+ addReply(c,shared.crlf);
+ ln = ln->forward[0];
+ rangelen++;
+ }
+ lenobj->ptr = sdscatprintf(sdsempty(),"*%d\r\n",rangelen);
+ }
+ }
+}
+
+static void zcardCommand(redisClient *c) {
+ robj *o;
+ zset *zs;
+
+ o = lookupKeyRead(c->db,c->argv[1]);
+ if (o == NULL) {
+ addReply(c,shared.czero);
+ return;
+ } else {
+ if (o->type != REDIS_ZSET) {
+ addReply(c,shared.wrongtypeerr);
+ } else {
+ zs = o->ptr;
+ addReplySds(c,sdscatprintf(sdsempty(),":%d\r\n",zs->zsl->length));
+ }
+ }
+}
+
+static void zscoreCommand(redisClient *c) {
+ robj *o;
+ zset *zs;
+
+ o = lookupKeyRead(c->db,c->argv[1]);
+ if (o == NULL) {
+ addReply(c,shared.czero);
+ return;
+ } else {
+ if (o->type != REDIS_ZSET) {
+ addReply(c,shared.wrongtypeerr);
+ } else {
+ dictEntry *de;
+
+ zs = o->ptr;
+ de = dictFind(zs->dict,c->argv[2]);
+ if (!de) {
+ addReply(c,shared.nullbulk);
+ } else {
+ char buf[128];
+ double *score = dictGetEntryVal(de);
+
+ snprintf(buf,sizeof(buf),"%.16g",*score);
+ addReplySds(c,sdscatprintf(sdsempty(),"$%d\r\n%s\r\n",
+ strlen(buf),buf));
+ }
+ }
+ }
+}
+
+/* ========================= Non type-specific commands ==================== */
+
static void flushdbCommand(redisClient *c) {
server.dirty += dictSize(c->db->dict);
dictEmpty(c->db->dict);
static redisSortOperation *createSortOperation(int type, robj *pattern) {
redisSortOperation *so = zmalloc(sizeof(*so));
- if (!so) oom("createSortOperation");
so->type = type;
so->pattern = pattern;
return so;
listLength((list*)sortval->ptr) :
dictSize((dict*)sortval->ptr);
vector = zmalloc(sizeof(redisSortObject)*vectorlen);
- if (!vector) oom("allocating objects vector for SORT");
j = 0;
if (sortval->type == REDIS_LIST) {
list *list = sortval->ptr;
dictEntry *setele;
di = dictGetIterator(set);
- if (!di) oom("dictGetIterator");
while((setele = dictNext(di)) != NULL) {
vector[j].obj = dictGetEntryKey(setele);
vector[j].u.score = 0;
server.masterport,
(server.replstate == REDIS_REPL_CONNECTED) ?
"up" : "down",
- (int)(time(NULL)-server.master->lastinteraction)
+ server.master ? ((int)(time(NULL)-server.master->lastinteraction)) : -1
);
}
for (j = 0; j < server.dbnum; j++) {
c->flags |= (REDIS_SLAVE|REDIS_MONITOR);
c->slaveseldb = 0;
- if (!listAddNodeTail(server.monitors,c)) oom("listAddNodeTail");
+ listAddNodeTail(server.monitors,c);
addReply(c,shared.ok);
}
return dictDelete(db->dict,key) == DICT_OK;
}
-static void expireCommand(redisClient *c) {
+static void expireGenericCommand(redisClient *c, robj *key, time_t seconds) {
dictEntry *de;
- int seconds = atoi(c->argv[2]->ptr);
- de = dictFind(c->db->dict,c->argv[1]);
+ de = dictFind(c->db->dict,key);
if (de == NULL) {
addReply(c,shared.czero);
return;
}
- if (seconds <= 0) {
- addReply(c, shared.czero);
+ if (seconds < 0) {
+ if (deleteKey(c->db,key)) server.dirty++;
+ addReply(c, shared.cone);
return;
} else {
time_t when = time(NULL)+seconds;
- if (setExpire(c->db,c->argv[1],when)) {
+ if (setExpire(c->db,key,when)) {
addReply(c,shared.cone);
server.dirty++;
} else {
}
}
+static void expireCommand(redisClient *c) {
+ expireGenericCommand(c,c->argv[1],strtol(c->argv[2]->ptr,NULL,10));
+}
+
+static void expireatCommand(redisClient *c) {
+ expireGenericCommand(c,c->argv[1],strtol(c->argv[2]->ptr,NULL,10)-time(NULL));
+}
+
static void ttlCommand(redisClient *c) {
time_t expire;
int ttl = -1;
* another slave. Set the right state, and copy the buffer. */
listRelease(c->reply);
c->reply = listDup(slave->reply);
- if (!c->reply) oom("listDup copying slave reply list");
c->replstate = REDIS_REPL_WAIT_BGSAVE_END;
redisLog(REDIS_NOTICE,"Waiting for end of BGSAVE for SYNC");
} else {
c->repldbfd = -1;
c->flags |= REDIS_SLAVE;
c->slaveseldb = 0;
- if (!listAddNodeTail(server.slaves,c)) oom("listAddNodeTail");
+ listAddNodeTail(server.slaves,c);
return;
}
strerror(errno));
return REDIS_ERR;
}
+ if (buf[0] != '$') {
+ close(fd);
+ redisLog(REDIS_WARNING,"Bad protocol from MASTER, the first byte is not '$', are you sure the host and port are right?");
+ return REDIS_ERR;
+ }
dumpsize = atoi(buf+1);
redisLog(REDIS_NOTICE,"Receiving %d bytes data dump from MASTER",dumpsize);
/* Read the bulk write data on a temp file */
{"mgetCommand", (unsigned long)mgetCommand},
{"monitorCommand", (unsigned long)monitorCommand},
{"expireCommand", (unsigned long)expireCommand},
+{"expireatCommand", (unsigned long)expireatCommand},
{"getsetCommand", (unsigned long)getsetCommand},
{"ttlCommand", (unsigned long)ttlCommand},
{"slaveofCommand", (unsigned long)slaveofCommand},
{"msetGenericCommand", (unsigned long)msetGenericCommand},
{"msetCommand", (unsigned long)msetCommand},
{"msetnxCommand", (unsigned long)msetnxCommand},
+{"zslCreateNode", (unsigned long)zslCreateNode},
+{"zslCreate", (unsigned long)zslCreate},
+{"zslFreeNode",(unsigned long)zslFreeNode},
+{"zslFree",(unsigned long)zslFree},
+{"zslRandomLevel",(unsigned long)zslRandomLevel},
+{"zslInsert",(unsigned long)zslInsert},
+{"zslDelete",(unsigned long)zslDelete},
+{"createZsetObject",(unsigned long)createZsetObject},
+{"zaddCommand",(unsigned long)zaddCommand},
+{"zrangeGenericCommand",(unsigned long)zrangeGenericCommand},
+{"zrangeCommand",(unsigned long)zrangeCommand},
+{"zrevrangeCommand",(unsigned long)zrevrangeCommand},
+{"zremCommand",(unsigned long)zremCommand},
+{"rdbSaveDoubleValue",(unsigned long)rdbSaveDoubleValue},
+{"rdbLoadDoubleValue",(unsigned long)rdbLoadDoubleValue},
+{"feedAppendOnlyFile",(unsigned long)feedAppendOnlyFile},
{NULL,0}
};
#elif defined(__APPLE__) && !defined(MAC_OS_X_VERSION_10_6)
return (void*) uc->uc_mcontext->__ss.__eip;
#elif defined(__APPLE__) && defined(MAC_OS_X_VERSION_10_6)
- #ifdef _STRUCT_X86_THREAD_STATE64
+ #if defined(_STRUCT_X86_THREAD_STATE64) && !defined(__i386__)
return (void*) uc->uc_mcontext->__ss.__rip;
#else
return (void*) uc->uc_mcontext->__ss.__eip;
projects / redis.git / blobdiff