X-Git-Url: https://git.saurik.com/redis.git/blobdiff_plain/7eac2a75a43bace72947fb534db9016707af1a02..b1ee7da75a4191bf1785f598c158b191104fdc38:/src/networking.c?ds=sidebyside diff --git a/src/networking.c b/src/networking.c index 6c3e7d81..f922e297 100644 --- a/src/networking.c +++ b/src/networking.c @@ -3,6 +3,14 @@ static void setProtocolError(redisClient *c, int pos); +/* To evaluate the output buffer size of a client we need to get size of + * allocated objects, however we can't used zmalloc_size() directly on sds + * strings because of the trick they use to work (the header is before the + * returned pointer), so we use this helper function. */ +size_t zmalloc_size_sds(sds s) { + return zmalloc_size(s-sizeof(struct sdshdr)); +} + void *dupClientReplyValue(void *o) { incrRefCount((robj*)o); return o; @@ -14,7 +22,6 @@ int listMatchObjects(void *a, void *b) { redisClient *createClient(int fd) { redisClient *c = zmalloc(sizeof(redisClient)); - c->bufpos = 0; /* passing -1 as fd it is possible to create a non connected client. * This is useful since all the Redis commands needs to be executed @@ -34,7 +41,9 @@ redisClient *createClient(int fd) { selectDb(c,0); c->fd = fd; + c->bufpos = 0; c->querybuf = sdsempty(); + c->querybuf_peak = 0; c->reqtype = 0; c->argc = 0; c->argv = NULL; @@ -43,7 +52,7 @@ redisClient *createClient(int fd) { c->bulklen = -1; c->sentlen = 0; c->flags = 0; - c->lastinteraction = time(NULL); + c->ctime = c->lastinteraction = server.unixtime; c->authenticated = 0; c->replstate = REDIS_REPL_NONE; c->reply = listCreate(); @@ -67,11 +76,23 @@ redisClient *createClient(int fd) { return c; } -/* Set the event loop to listen for write events on the client's socket. - * Typically gets called every time a reply is built. */ -int _installWriteEvent(redisClient *c) { +/* This function is called every time we are going to transmit new data + * to the client. The behavior is the following: + * + * If the client should receive new data (normal clients will) the function + * returns REDIS_OK, and make sure to install the write handler in our event + * loop so that when the socket is writable new data gets written. + * + * If the client should not receive new data, because it is a fake client + * or a slave, or because the setup of the write handler failed, the function + * returns REDIS_ERR. + * + * Typically gets called every time a reply is built, before adding more + * data to the clients output buffers. If the function returns REDIS_ERR no + * data should be appended to the output buffers. */ +int prepareClientToWrite(redisClient *c) { if (c->flags & REDIS_LUA_CLIENT) return REDIS_OK; - if (c->fd <= 0) return REDIS_ERR; + if (c->fd <= 0) return REDIS_ERR; /* Fake client */ if (c->bufpos == 0 && listLength(c->reply) == 0 && (c->replstate == REDIS_REPL_NONE || c->replstate == REDIS_REPL_ONLINE) && @@ -125,6 +146,7 @@ void _addReplyObjectToList(redisClient *c, robj *o) { if (listLength(c->reply) == 0) { incrRefCount(o); listAddNodeTail(c->reply,o); + c->reply_bytes += zmalloc_size_sds(o->ptr); } else { tail = listNodeValue(listLast(c->reply)); @@ -132,14 +154,16 @@ void _addReplyObjectToList(redisClient *c, robj *o) { if (tail->ptr != NULL && sdslen(tail->ptr)+sdslen(o->ptr) <= REDIS_REPLY_CHUNK_BYTES) { + c->reply_bytes -= zmalloc_size_sds(tail->ptr); tail = dupLastObjectIfNeeded(c->reply); tail->ptr = sdscatlen(tail->ptr,o->ptr,sdslen(o->ptr)); + c->reply_bytes += zmalloc_size_sds(tail->ptr); } else { incrRefCount(o); listAddNodeTail(c->reply,o); + c->reply_bytes += zmalloc_size_sds(o->ptr); } } - c->reply_bytes += sdslen(o->ptr); asyncCloseClientOnOutputBufferLimitReached(c); } @@ -153,9 +177,9 @@ void _addReplySdsToList(redisClient *c, sds s) { return; } - c->reply_bytes += sdslen(s); if (listLength(c->reply) == 0) { listAddNodeTail(c->reply,createObject(REDIS_STRING,s)); + c->reply_bytes += zmalloc_size_sds(s); } else { tail = listNodeValue(listLast(c->reply)); @@ -163,11 +187,14 @@ void _addReplySdsToList(redisClient *c, sds s) { if (tail->ptr != NULL && sdslen(tail->ptr)+sdslen(s) <= REDIS_REPLY_CHUNK_BYTES) { + c->reply_bytes -= zmalloc_size_sds(tail->ptr); tail = dupLastObjectIfNeeded(c->reply); tail->ptr = sdscatlen(tail->ptr,s,sdslen(s)); + c->reply_bytes += zmalloc_size_sds(tail->ptr); sdsfree(s); } else { listAddNodeTail(c->reply,createObject(REDIS_STRING,s)); + c->reply_bytes += zmalloc_size_sds(s); } } asyncCloseClientOnOutputBufferLimitReached(c); @@ -179,7 +206,10 @@ void _addReplyStringToList(redisClient *c, char *s, size_t len) { if (c->flags & REDIS_CLOSE_AFTER_REPLY) return; if (listLength(c->reply) == 0) { - listAddNodeTail(c->reply,createStringObject(s,len)); + robj *o = createStringObject(s,len); + + listAddNodeTail(c->reply,o); + c->reply_bytes += zmalloc_size_sds(o->ptr); } else { tail = listNodeValue(listLast(c->reply)); @@ -187,13 +217,17 @@ void _addReplyStringToList(redisClient *c, char *s, size_t len) { if (tail->ptr != NULL && sdslen(tail->ptr)+len <= REDIS_REPLY_CHUNK_BYTES) { + c->reply_bytes -= zmalloc_size_sds(tail->ptr); tail = dupLastObjectIfNeeded(c->reply); tail->ptr = sdscatlen(tail->ptr,s,len); + c->reply_bytes += zmalloc_size_sds(tail->ptr); } else { - listAddNodeTail(c->reply,createStringObject(s,len)); + robj *o = createStringObject(s,len); + + listAddNodeTail(c->reply,o); + c->reply_bytes += zmalloc_size_sds(o->ptr); } } - c->reply_bytes += len; asyncCloseClientOnOutputBufferLimitReached(c); } @@ -203,7 +237,7 @@ void _addReplyStringToList(redisClient *c, char *s, size_t len) { * -------------------------------------------------------------------------- */ void addReply(redisClient *c, robj *obj) { - if (_installWriteEvent(c) != REDIS_OK) return; + if (prepareClientToWrite(c) != REDIS_OK) return; /* This is an important place where we can avoid copy-on-write * when there is a saving child running, avoiding touching the @@ -215,19 +249,31 @@ void addReply(redisClient *c, robj *obj) { if (obj->encoding == REDIS_ENCODING_RAW) { if (_addReplyToBuffer(c,obj->ptr,sdslen(obj->ptr)) != REDIS_OK) _addReplyObjectToList(c,obj); - } else { - /* FIXME: convert the long into string and use _addReplyToBuffer() - * instead of calling getDecodedObject. As this place in the - * code is too performance critical. */ + } else if (obj->encoding == REDIS_ENCODING_INT) { + /* Optimization: if there is room in the static buffer for 32 bytes + * (more than the max chars a 64 bit integer can take as string) we + * avoid decoding the object and go for the lower level approach. */ + if (listLength(c->reply) == 0 && (sizeof(c->buf) - c->bufpos) >= 32) { + char buf[32]; + int len; + + len = ll2string(buf,sizeof(buf),(long)obj->ptr); + if (_addReplyToBuffer(c,buf,len) == REDIS_OK) + return; + /* else... continue with the normal code path, but should never + * happen actually since we verified there is room. */ + } obj = getDecodedObject(obj); if (_addReplyToBuffer(c,obj->ptr,sdslen(obj->ptr)) != REDIS_OK) _addReplyObjectToList(c,obj); decrRefCount(obj); + } else { + redisPanic("Wrong obj->encoding in addReply()"); } } void addReplySds(redisClient *c, sds s) { - if (_installWriteEvent(c) != REDIS_OK) { + if (prepareClientToWrite(c) != REDIS_OK) { /* The caller expects the sds to be free'd. */ sdsfree(s); return; @@ -241,19 +287,19 @@ void addReplySds(redisClient *c, sds s) { } void addReplyString(redisClient *c, char *s, size_t len) { - if (_installWriteEvent(c) != REDIS_OK) return; + if (prepareClientToWrite(c) != REDIS_OK) return; if (_addReplyToBuffer(c,s,len) != REDIS_OK) _addReplyStringToList(c,s,len); } -void _addReplyError(redisClient *c, char *s, size_t len) { +void addReplyErrorLength(redisClient *c, char *s, size_t len) { addReplyString(c,"-ERR ",5); addReplyString(c,s,len); addReplyString(c,"\r\n",2); } void addReplyError(redisClient *c, char *err) { - _addReplyError(c,err,strlen(err)); + addReplyErrorLength(c,err,strlen(err)); } void addReplyErrorFormat(redisClient *c, const char *fmt, ...) { @@ -268,18 +314,18 @@ void addReplyErrorFormat(redisClient *c, const char *fmt, ...) { for (j = 0; j < l; j++) { if (s[j] == '\r' || s[j] == '\n') s[j] = ' '; } - _addReplyError(c,s,sdslen(s)); + addReplyErrorLength(c,s,sdslen(s)); sdsfree(s); } -void _addReplyStatus(redisClient *c, char *s, size_t len) { +void addReplyStatusLength(redisClient *c, char *s, size_t len) { addReplyString(c,"+",1); addReplyString(c,s,len); addReplyString(c,"\r\n",2); } void addReplyStatus(redisClient *c, char *status) { - _addReplyStatus(c,status,strlen(status)); + addReplyStatusLength(c,status,strlen(status)); } void addReplyStatusFormat(redisClient *c, const char *fmt, ...) { @@ -287,7 +333,7 @@ void addReplyStatusFormat(redisClient *c, const char *fmt, ...) { va_start(ap,fmt); sds s = sdscatvprintf(sdsempty(),fmt,ap); va_end(ap); - _addReplyStatus(c,s,sdslen(s)); + addReplyStatusLength(c,s,sdslen(s)); sdsfree(s); } @@ -297,7 +343,7 @@ void *addDeferredMultiBulkLength(redisClient *c) { /* Note that we install the write event here even if the object is not * ready to be sent, since we are sure that before returning to the * event loop setDeferredMultiBulkLength() will be called. */ - if (_installWriteEvent(c) != REDIS_OK) return NULL; + if (prepareClientToWrite(c) != REDIS_OK) return NULL; listAddNodeTail(c->reply,createObject(REDIS_STRING,NULL)); return listLast(c->reply); } @@ -312,7 +358,7 @@ void setDeferredMultiBulkLength(redisClient *c, void *node, long length) { len = listNodeValue(ln); len->ptr = sdscatprintf(sdsempty(),"*%ld\r\n",length); - c->reply_bytes += sdslen(len->ptr); + c->reply_bytes += zmalloc_size_sds(len->ptr); if (ln->next != NULL) { next = listNodeValue(ln->next); @@ -336,9 +382,21 @@ void addReplyDouble(redisClient *c, double d) { /* Add a long long as integer reply or bulk len / multi bulk count. * Basically this is used to output . */ -void _addReplyLongLong(redisClient *c, long long ll, char prefix) { +void addReplyLongLongWithPrefix(redisClient *c, long long ll, char prefix) { char buf[128]; int len; + + /* Things like $3\r\n or *2\r\n are emitted very often by the protocol + * so we have a few shared objects to use if the integer is small + * like it is most of the times. */ + if (prefix == '*' && ll < REDIS_SHARED_BULKHDR_LEN) { + addReply(c,shared.mbulkhdr[ll]); + return; + } else if (prefix == '$' && ll < REDIS_SHARED_BULKHDR_LEN) { + addReply(c,shared.bulkhdr[ll]); + return; + } + buf[0] = prefix; len = ll2string(buf+1,sizeof(buf)-1,ll); buf[len+1] = '\r'; @@ -352,11 +410,11 @@ void addReplyLongLong(redisClient *c, long long ll) { else if (ll == 1) addReply(c,shared.cone); else - _addReplyLongLong(c,ll,':'); + addReplyLongLongWithPrefix(c,ll,':'); } void addReplyMultiBulkLen(redisClient *c, long length) { - _addReplyLongLong(c,length,'*'); + addReplyLongLongWithPrefix(c,length,'*'); } /* Create the length prefix of a bulk reply, example: $2234 */ @@ -378,7 +436,7 @@ void addReplyBulkLen(redisClient *c, robj *obj) { len++; } } - _addReplyLongLong(c,len,'$'); + addReplyLongLongWithPrefix(c,len,'$'); } /* Add a Redis Object as a bulk reply */ @@ -390,7 +448,7 @@ void addReplyBulk(redisClient *c, robj *obj) { /* Add a C buffer as bulk reply */ void addReplyBulkCBuffer(redisClient *c, void *p, size_t len) { - _addReplyLongLong(c,len,'$'); + addReplyLongLongWithPrefix(c,len,'$'); addReplyString(c,p,len); addReply(c,shared.crlf); } @@ -489,6 +547,16 @@ static void freeClientArgv(redisClient *c) { c->cmd = NULL; } +/* Close all the slaves connections. This is useful in chained replication + * when we resync with our own master and want to force all our slaves to + * resync with us as well. */ +void disconnectSlaves(void) { + while (listLength(server.slaves)) { + listNode *ln = listFirst(server.slaves); + freeClient((redisClient*)ln->value); + } +} + void freeClient(redisClient *c) { listNode *ln; @@ -546,22 +614,13 @@ void freeClient(redisClient *c) { if (c->flags & REDIS_MASTER) { server.master = NULL; server.repl_state = REDIS_REPL_CONNECT; - server.repl_down_since = time(NULL); - /* Since we lost the connection with the master, we should also - * close the connection with all our slaves if we have any, so - * when we'll resync with the master the other slaves will sync again - * with us as well. Note that also when the slave is not connected - * to the master it will keep refusing connections by other slaves. + server.repl_down_since = server.unixtime; + /* We lost connection with our master, force our slaves to resync + * with us as well to load the new data set. * - * We do this only if server.masterhost != NULL. If it is NULL this - * means the user called SLAVEOF NO ONE and we are freeing our - * link with the master, so no need to close link with slaves. */ - if (server.masterhost != NULL) { - while (listLength(server.slaves)) { - ln = listFirst(server.slaves); - freeClient((redisClient*)ln->value); - } - } + * If server.masterhost is NULL the user called SLAVEOF NO ONE so + * slave resync is not needed. */ + if (server.masterhost != NULL) disconnectSlaves(); } /* If this client was scheduled for async freeing we need to remove it @@ -602,6 +661,7 @@ void freeClientsInAsyncFreeQueue(void) { void sendReplyToClient(aeEventLoop *el, int fd, void *privdata, int mask) { redisClient *c = privdata; int nwritten = 0, totwritten = 0, objlen; + size_t objmem; robj *o; REDIS_NOTUSED(el); REDIS_NOTUSED(mask); @@ -627,6 +687,7 @@ void sendReplyToClient(aeEventLoop *el, int fd, void *privdata, int mask) { } else { o = listNodeValue(listFirst(c->reply)); objlen = sdslen(o->ptr); + objmem = zmalloc_size_sds(o->ptr); if (objlen == 0) { listDelNode(c->reply,listFirst(c->reply)); @@ -647,15 +708,20 @@ void sendReplyToClient(aeEventLoop *el, int fd, void *privdata, int mask) { if (c->sentlen == objlen) { listDelNode(c->reply,listFirst(c->reply)); c->sentlen = 0; - c->reply_bytes -= objlen; + c->reply_bytes -= objmem; } } - /* Note that we avoid to send more thank REDIS_MAX_WRITE_PER_EVENT + /* Note that we avoid to send more than REDIS_MAX_WRITE_PER_EVENT * bytes, in a single threaded server it's a good idea to serve * other clients as well, even if a very large request comes from * super fast link that is always able to accept data (in real world - * scenario think about 'KEYS *' against the loopback interfae) */ - if (totwritten > REDIS_MAX_WRITE_PER_EVENT) break; + * scenario think about 'KEYS *' against the loopback interface). + * + * However if we are over the maxmemory limit we ignore that and + * just deliver as much data as it is possible to deliver. */ + if (totwritten > REDIS_MAX_WRITE_PER_EVENT && + (server.maxmemory == 0 || + zmalloc_used_memory() < server.maxmemory)) break; } if (nwritten == -1) { if (errno == EAGAIN) { @@ -667,7 +733,7 @@ void sendReplyToClient(aeEventLoop *el, int fd, void *privdata, int mask) { return; } } - if (totwritten > 0) c->lastinteraction = time(NULL); + if (totwritten > 0) c->lastinteraction = server.unixtime; if (c->bufpos == 0 && listLength(c->reply) == 0) { c->sentlen = 0; aeDeleteFileEvent(server.el,c->fd,AE_WRITABLE); @@ -687,34 +753,6 @@ void resetClient(redisClient *c) { if (!(c->flags & REDIS_MULTI)) c->flags &= (~REDIS_ASKING); } -void closeTimedoutClients(void) { - redisClient *c; - listNode *ln; - time_t now = time(NULL); - listIter li; - - listRewind(server.clients,&li); - while ((ln = listNext(&li)) != NULL) { - c = listNodeValue(ln); - if (server.maxidletime && - !(c->flags & REDIS_SLAVE) && /* no timeout for slaves */ - !(c->flags & REDIS_MASTER) && /* no timeout for masters */ - !(c->flags & REDIS_BLOCKED) && /* no timeout for BLPOP */ - dictSize(c->pubsub_channels) == 0 && /* no timeout for pubsub */ - listLength(c->pubsub_patterns) == 0 && - (now - c->lastinteraction > server.maxidletime)) - { - redisLog(REDIS_VERBOSE,"Closing idle client"); - freeClient(c); - } else if (c->flags & REDIS_BLOCKED) { - if (c->bpop.timeout != 0 && c->bpop.timeout < now) { - addReply(c,shared.nullmultibulk); - unblockClientWaitingData(c); - } - } - } -} - int processInlineBuffer(redisClient *c) { char *newline = strstr(c->querybuf,"\r\n"); int argc, j; @@ -962,6 +1000,7 @@ void readQueryFromClient(aeEventLoop *el, int fd, void *privdata, int mask) { } qblen = sdslen(c->querybuf); + if (c->querybuf_peak < qblen) c->querybuf_peak = qblen; c->querybuf = sdsMakeRoomFor(c->querybuf, readlen); nread = read(fd, c->querybuf+qblen, readlen); if (nread == -1) { @@ -979,7 +1018,7 @@ void readQueryFromClient(aeEventLoop *el, int fd, void *privdata, int mask) { } if (nread) { sdsIncrLen(c->querybuf,nread); - c->lastinteraction = time(NULL); + c->lastinteraction = server.unixtime; } else { server.current_client = NULL; return; @@ -1020,14 +1059,9 @@ void getClientsMaxBuffers(unsigned long *longest_output_list, sds getClientInfoString(redisClient *client) { char ip[32], flags[16], events[3], *p; int port; - time_t now = time(NULL); int emask; - if (anetPeerToString(client->fd,ip,&port) == -1) { - ip[0] = '?'; - ip[1] = '\0'; - port = 0; - } + anetPeerToString(client->fd,ip,&port); p = flags; if (client->flags & REDIS_SLAVE) { if (client->flags & REDIS_MONITOR) @@ -1051,14 +1085,17 @@ sds getClientInfoString(redisClient *client) { if (emask & AE_WRITABLE) *p++ = 'w'; *p = '\0'; return sdscatprintf(sdsempty(), - "addr=%s:%d fd=%d idle=%ld flags=%s db=%d sub=%d psub=%d qbuf=%lu obl=%lu oll=%lu omem=%lu events=%s cmd=%s", + "addr=%s:%d fd=%d age=%ld idle=%ld flags=%s db=%d sub=%d psub=%d multi=%d qbuf=%lu qbuf-free=%lu obl=%lu oll=%lu omem=%lu events=%s cmd=%s", ip,port,client->fd, - (long)(now - client->lastinteraction), + (long)(server.unixtime - client->ctime), + (long)(server.unixtime - client->lastinteraction), flags, client->db->id, (int) dictSize(client->pubsub_channels), (int) listLength(client->pubsub_patterns), + (client->flags & REDIS_MULTI) ? client->mstate.count : -1, (unsigned long) sdslen(client->querybuf), + (unsigned long) sdsavail(client->querybuf), (unsigned long) client->bufpos, (unsigned long) listLength(client->reply), getClientOutputBufferMemoryUsage(client), @@ -1179,9 +1216,9 @@ void rewriteClientCommandArgument(redisClient *c, int i, robj *newval) { * * Note: this function is very fast so can be called as many time as * the caller wishes. The main usage of this function currently is - * enforcing the client output lenght limits. */ + * enforcing the client output length limits. */ unsigned long getClientOutputBufferMemoryUsage(redisClient *c) { - unsigned long list_item_size = sizeof(listNode); + unsigned long list_item_size = sizeof(listNode)+sizeof(robj); return c->reply_bytes + (list_item_size*listLength(c->reply)); } @@ -1201,6 +1238,22 @@ int getClientLimitClass(redisClient *c) { return REDIS_CLIENT_LIMIT_CLASS_NORMAL; } +int getClientLimitClassByName(char *name) { + if (!strcasecmp(name,"normal")) return REDIS_CLIENT_LIMIT_CLASS_NORMAL; + else if (!strcasecmp(name,"slave")) return REDIS_CLIENT_LIMIT_CLASS_SLAVE; + else if (!strcasecmp(name,"pubsub")) return REDIS_CLIENT_LIMIT_CLASS_PUBSUB; + else return -1; +} + +char *getClientLimitClassName(int class) { + switch(class) { + case REDIS_CLIENT_LIMIT_CLASS_NORMAL: return "normal"; + case REDIS_CLIENT_LIMIT_CLASS_SLAVE: return "slave"; + case REDIS_CLIENT_LIMIT_CLASS_PUBSUB: return "pubsub"; + default: return NULL; + } +} + /* The function checks if the client reached output buffer soft or hard * limit, and also update the state needed to check the soft limit as * a side effect. @@ -1242,21 +1295,40 @@ int checkClientOutputBufferLimits(redisClient *c) { } /* Asynchronously close a client if soft or hard limit is reached on the - * output buffer size. If the client will be closed 1 is returend, otherwise 0 - * is returned. + * output buffer size. The caller can check if the client will be closed + * checking if the client REDIS_CLOSE_ASAP flag is set. * * Note: we need to close the client asynchronously because this function is * called from contexts where the client can't be freed safely, i.e. from the * lower level functions pushing data inside the client output buffers. */ -int asyncCloseClientOnOutputBufferLimitReached(redisClient *c) { +void asyncCloseClientOnOutputBufferLimitReached(redisClient *c) { + if (c->reply_bytes == 0 || c->flags & REDIS_CLOSE_ASAP) return; if (checkClientOutputBufferLimits(c)) { sds client = getClientInfoString(c); freeClientAsync(c); - redisLog(REDIS_NOTICE,"Client %s scheduled to be closed ASAP for overcoming of output buffer limits."); + redisLog(REDIS_WARNING,"Client %s scheduled to be closed ASAP for overcoming of output buffer limits.", client); sdsfree(client); - return 1; - } else { - return 0; + } +} + +/* Helper function used by freeMemoryIfNeeded() in order to flush slaves + * output buffers without returning control to the event loop. */ +void flushSlavesOutputBuffers(void) { + listIter li; + listNode *ln; + + listRewind(server.slaves,&li); + while((ln = listNext(&li))) { + redisClient *slave = listNodeValue(ln); + int events; + + events = aeGetFileEvents(server.el,slave->fd); + if (events & AE_WRITABLE && + slave->replstate == REDIS_REPL_ONLINE && + listLength(slave->reply)) + { + sendReplyToClient(server.el,slave->fd,slave,0); + } } }