X-Git-Url: https://git.saurik.com/redis.git/blobdiff_plain/46834808fe31e7cbf08d6c9f7638a54c5c8bff8a..6710ff24d067d56b6b8d768532a6a7d3c4bc7a80:/src/cluster.c diff --git a/src/cluster.c b/src/cluster.c index e8ca51cf..2b3ebe4d 100644 --- a/src/cluster.c +++ b/src/cluster.c @@ -377,7 +377,7 @@ clusterNode *clusterLookupNode(char *name) { de = dictFind(server.cluster.nodes,s); sdsfree(s); if (de == NULL) return NULL; - return dictGetEntryVal(de); + return dictGetVal(de); } /* This is only used after the handshake. When we connect a given IP/PORT @@ -439,7 +439,7 @@ void clusterProcessGossipSection(clusterMsg *hdr, clusterLink *link) { * time PONG figure if it is newer than our figure. * Note that it's not a problem if we have a PING already * in progress against this node. */ - if (node->pong_received < ntohl(g->pong_received)) { + if (node->pong_received < (signed) ntohl(g->pong_received)) { redisLog(REDIS_DEBUG,"Node pong_received updated by gossip"); node->pong_received = ntohl(g->pong_received); } @@ -493,6 +493,7 @@ void nodeIp2String(char *buf, clusterLink *link) { /* Update the node address to the IP address that can be extracted * from link->fd, and at the specified port. */ void nodeUpdateAddress(clusterNode *node, clusterLink *link, int port) { + /* TODO */ } /* When this function is called, there is a packet to process starting @@ -510,8 +511,10 @@ int clusterProcessPacket(clusterLink *link) { uint16_t type = ntohs(hdr->type); clusterNode *sender; - redisLog(REDIS_DEBUG,"--- packet to process %lu bytes (%lu) ---", - (unsigned long) totlen, sdslen(link->rcvbuf)); + redisLog(REDIS_DEBUG,"--- Processing packet of type %d, %lu bytes", + type, (unsigned long) totlen); + + /* Perform sanity checks */ if (totlen < 8) return 1; if (totlen > sdslen(link->rcvbuf)) return 1; if (type == CLUSTERMSG_TYPE_PING || type == CLUSTERMSG_TYPE_PONG || @@ -530,7 +533,16 @@ int clusterProcessPacket(clusterLink *link) { explen += sizeof(clusterMsgDataFail); if (totlen != explen) return 1; } + if (type == CLUSTERMSG_TYPE_PUBLISH) { + uint32_t explen = sizeof(clusterMsg)-sizeof(union clusterMsgData); + + explen += sizeof(clusterMsgDataPublish) + + ntohl(hdr->data.publish.msg.channel_len) + + ntohl(hdr->data.publish.msg.message_len); + if (totlen != explen) return 1; + } + /* Ready to process the packet. Dispatch by type. */ sender = clusterLookupNode(hdr->sender); if (type == CLUSTERMSG_TYPE_PING || type == CLUSTERMSG_TYPE_MEET) { int update_config = 0; @@ -664,8 +676,24 @@ int clusterProcessPacket(clusterLink *link) { clusterUpdateState(); clusterSaveConfigOrDie(); } + } else if (type == CLUSTERMSG_TYPE_PUBLISH) { + robj *channel, *message; + uint32_t channel_len, message_len; + + /* Don't bother creating useless objects if there are no Pub/Sub subscribers. */ + if (dictSize(server.pubsub_channels) || listLength(server.pubsub_patterns)) { + channel_len = ntohl(hdr->data.publish.msg.channel_len); + message_len = ntohl(hdr->data.publish.msg.message_len); + channel = createStringObject( + (char*)hdr->data.publish.msg.bulk_data,channel_len); + message = createStringObject( + (char*)hdr->data.publish.msg.bulk_data+channel_len, message_len); + pubsubPublishMessage(channel,message); + decrRefCount(channel); + decrRefCount(message); + } } else { - redisLog(REDIS_NOTICE,"Received unknown packet type: %d", type); + redisLog(REDIS_WARNING,"Received unknown packet type: %d", type); } return 1; } @@ -758,9 +786,25 @@ void clusterSendMessage(clusterLink *link, unsigned char *msg, size_t msglen) { link->sndbuf = sdscatlen(link->sndbuf, msg, msglen); } +/* Send a message to all the nodes with a reliable link */ +void clusterBroadcastMessage(void *buf, size_t len) { + dictIterator *di; + dictEntry *de; + + di = dictGetIterator(server.cluster.nodes); + while((de = dictNext(di)) != NULL) { + clusterNode *node = dictGetVal(de); + + if (!node->link) continue; + if (node->flags & (REDIS_NODE_MYSELF|REDIS_NODE_NOADDR)) continue; + clusterSendMessage(node->link,buf,len); + } + dictReleaseIterator(di); +} + /* Build the message header */ void clusterBuildMessageHdr(clusterMsg *hdr, int type) { - int totlen; + int totlen = 0; memset(hdr,0,sizeof(*hdr)); hdr->type = htons(type); @@ -805,7 +849,7 @@ void clusterSendPing(clusterLink *link, int type) { /* Populate the gossip fields */ while(freshnodes > 0 && gossipcount < 3) { struct dictEntry *de = dictGetRandomKey(server.cluster.nodes); - clusterNode *this = dictGetEntryVal(de); + clusterNode *this = dictGetVal(de); clusterMsgDataGossip *gossip; int j; @@ -842,20 +886,48 @@ void clusterSendPing(clusterLink *link, int type) { clusterSendMessage(link,buf,totlen); } -/* Send a message to all the nodes with a reliable link */ -void clusterBroadcastMessage(void *buf, size_t len) { - dictIterator *di; - dictEntry *de; +/* Send a PUBLISH message. + * + * If link is NULL, then the message is broadcasted to the whole cluster. */ +void clusterSendPublish(clusterLink *link, robj *channel, robj *message) { + unsigned char buf[4096], *payload; + clusterMsg *hdr = (clusterMsg*) buf; + uint32_t totlen; + uint32_t channel_len, message_len; - di = dictGetIterator(server.cluster.nodes); - while((de = dictNext(di)) != NULL) { - clusterNode *node = dictGetEntryVal(de); + channel = getDecodedObject(channel); + message = getDecodedObject(message); + channel_len = sdslen(channel->ptr); + message_len = sdslen(message->ptr); - if (!node->link) continue; - if (node->flags & (REDIS_NODE_MYSELF|REDIS_NODE_NOADDR)) continue; - clusterSendMessage(node->link,buf,len); + clusterBuildMessageHdr(hdr,CLUSTERMSG_TYPE_PUBLISH); + totlen = sizeof(clusterMsg)-sizeof(union clusterMsgData); + totlen += sizeof(clusterMsgDataPublish) + channel_len + message_len; + + hdr->data.publish.msg.channel_len = htonl(channel_len); + hdr->data.publish.msg.message_len = htonl(message_len); + hdr->totlen = htonl(totlen); + + /* Try to use the local buffer if possible */ + if (totlen < sizeof(buf)) { + payload = buf; + } else { + payload = zmalloc(totlen); + hdr = (clusterMsg*) payload; + memcpy(payload,hdr,sizeof(hdr)); } - dictReleaseIterator(di); + memcpy(hdr->data.publish.msg.bulk_data,channel->ptr,sdslen(channel->ptr)); + memcpy(hdr->data.publish.msg.bulk_data+sdslen(channel->ptr), + message->ptr,sdslen(message->ptr)); + + if (link) + clusterSendMessage(link,payload,totlen); + else + clusterBroadcastMessage(payload,totlen); + + decrRefCount(channel); + decrRefCount(message); + if (payload != buf) zfree(payload); } /* Send a FAIL message to all the nodes we are able to contact. @@ -872,6 +944,17 @@ void clusterSendFail(char *nodename) { clusterBroadcastMessage(buf,ntohl(hdr->totlen)); } +/* ----------------------------------------------------------------------------- + * CLUSTER Pub/Sub support + * + * For now we do very little, just propagating PUBLISH messages across the whole + * cluster. In the future we'll try to get smarter and avoiding propagating those + * messages to hosts without receives for a given channel. + * -------------------------------------------------------------------------- */ +void clusterPropagatePublish(robj *channel, robj *message) { + clusterSendPublish(NULL, channel, message); +} + /* ----------------------------------------------------------------------------- * CLUSTER cron job * -------------------------------------------------------------------------- */ @@ -887,7 +970,7 @@ void clusterCron(void) { /* Check if we have disconnected nodes and reestablish the connection. */ di = dictGetIterator(server.cluster.nodes); while((de = dictNext(di)) != NULL) { - clusterNode *node = dictGetEntryVal(de); + clusterNode *node = dictGetVal(de); if (node->flags & (REDIS_NODE_MYSELF|REDIS_NODE_NOADDR)) continue; if (node->link == NULL) { @@ -922,7 +1005,7 @@ void clusterCron(void) { * the oldest ping_sent time */ for (j = 0; j < 5; j++) { de = dictGetRandomKey(server.cluster.nodes); - clusterNode *this = dictGetEntryVal(de); + clusterNode *this = dictGetVal(de); if (this->link == NULL) continue; if (this->flags & (REDIS_NODE_MYSELF|REDIS_NODE_HANDSHAKE)) continue; @@ -939,7 +1022,7 @@ void clusterCron(void) { /* Iterate nodes to check if we need to flag something as failing */ di = dictGetIterator(server.cluster.nodes); while((de = dictNext(di)) != NULL) { - clusterNode *node = dictGetEntryVal(de); + clusterNode *node = dictGetVal(de); int delay; if (node->flags & @@ -1014,11 +1097,24 @@ int clusterNodeGetSlotBit(clusterNode *n, int slot) { * If the slot is already assigned to another instance this is considered * an error and REDIS_ERR is returned. */ int clusterAddSlot(clusterNode *n, int slot) { - redisAssert(clusterNodeSetSlotBit(n,slot) == 0); + if (clusterNodeSetSlotBit(n,slot) != 0) + return REDIS_ERR; server.cluster.slots[slot] = n; return REDIS_OK; } +/* Delete the specified slot marking it as unassigned. + * Returns REDIS_OK if the slot was assigned, otherwise if the slot was + * already unassigned REDIS_ERR is returned. */ +int clusterDelSlot(int slot) { + clusterNode *n = server.cluster.slots[slot]; + + if (!n) return REDIS_ERR; + redisAssert(clusterNodeClearSlotBit(n,slot) == 1); + server.cluster.slots[slot] = NULL; + return REDIS_OK; +} + /* ----------------------------------------------------------------------------- * Cluster state evaluation function * -------------------------------------------------------------------------- */ @@ -1057,7 +1153,7 @@ sds clusterGenNodesDescription(void) { di = dictGetIterator(server.cluster.nodes); while((de = dictNext(di)) != NULL) { - clusterNode *node = dictGetEntryVal(de); + clusterNode *node = dictGetVal(de); /* Node coordinates */ ci = sdscatprintf(ci,"%.40s %s:%d ", @@ -1086,7 +1182,8 @@ sds clusterGenNodesDescription(void) { ci = sdscatprintf(ci,"%ld %ld %s", (long) node->ping_sent, (long) node->pong_received, - node->link ? "connected" : "disconnected"); + (node->link || node->flags & REDIS_NODE_MYSELF) ? + "connected" : "disconnected"); /* Slots served by this instance */ start = -1; @@ -1128,6 +1225,18 @@ sds clusterGenNodesDescription(void) { return ci; } +int getSlotOrReply(redisClient *c, robj *o) { + long long slot; + + if (getLongLongFromObject(o,&slot) != REDIS_OK || + slot < 0 || slot > REDIS_CLUSTER_SLOTS) + { + addReplyError(c,"Invalid or out of range slot"); + return -1; + } + return (int) slot; +} + void clusterCommand(redisClient *c) { if (server.cluster_enabled == 0) { addReplyError(c,"This instance has cluster support disabled"); @@ -1165,24 +1274,29 @@ void clusterCommand(redisClient *c) { o = createObject(REDIS_STRING,ci); addReplyBulk(c,o); decrRefCount(o); - } else if (!strcasecmp(c->argv[1]->ptr,"addslots") && c->argc >= 3) { - int j; - long long slot; + } else if ((!strcasecmp(c->argv[1]->ptr,"addslots") || + !strcasecmp(c->argv[1]->ptr,"delslots")) && c->argc >= 3) + { + /* CLUSTER ADDSLOTS [slot] ... */ + /* CLUSTER DELSLOTS [slot] ... */ + int j, slot; unsigned char *slots = zmalloc(REDIS_CLUSTER_SLOTS); + int del = !strcasecmp(c->argv[1]->ptr,"delslots"); memset(slots,0,REDIS_CLUSTER_SLOTS); /* Check that all the arguments are parsable and that all the * slots are not already busy. */ for (j = 2; j < c->argc; j++) { - if (getLongLongFromObject(c->argv[j],&slot) != REDIS_OK || - slot < 0 || slot > REDIS_CLUSTER_SLOTS) - { - addReplyError(c,"Invalid or out of range slot index"); + if ((slot = getSlotOrReply(c,c->argv[j])) == -1) { zfree(slots); return; } - if (server.cluster.slots[slot]) { - addReplyErrorFormat(c,"Slot %lld is already busy", slot); + if (del && server.cluster.slots[slot] == NULL) { + addReplyErrorFormat(c,"Slot %d is already unassigned", slot); + zfree(slots); + return; + } else if (!del && server.cluster.slots[slot]) { + addReplyErrorFormat(c,"Slot %d is already busy", slot); zfree(slots); return; } @@ -1195,9 +1309,16 @@ void clusterCommand(redisClient *c) { } for (j = 0; j < REDIS_CLUSTER_SLOTS; j++) { if (slots[j]) { - int retval = clusterAddSlot(server.cluster.myself,j); - - redisAssert(retval == REDIS_OK); + int retval; + + /* If this slot was set as importing we can clear this + * state as now we are the real owner of the slot. */ + if (server.cluster.importing_slots_from[j]) + server.cluster.importing_slots_from[j] = NULL; + + retval = del ? clusterDelSlot(j) : + clusterAddSlot(server.cluster.myself,j); + redisAssertWithInfo(c,NULL,retval == REDIS_OK); } } zfree(slots); @@ -1205,25 +1326,20 @@ void clusterCommand(redisClient *c) { clusterSaveConfigOrDie(); addReply(c,shared.ok); } else if (!strcasecmp(c->argv[1]->ptr,"setslot") && c->argc >= 4) { - /* SETSLOT 10 MIGRATING */ - /* SETSLOT 10 IMPORTING */ + /* SETSLOT 10 MIGRATING */ + /* SETSLOT 10 IMPORTING */ /* SETSLOT 10 STABLE */ - long long aux; - unsigned int slot; + /* SETSLOT 10 NODE */ + int slot; clusterNode *n; - if (getLongLongFromObjectOrReply(c,c->argv[2],&aux,NULL) != REDIS_OK) - return; - if (aux < 0 || aux >= REDIS_CLUSTER_SLOTS) { - addReplyError(c,"Slot out of range"); - return; - } - slot = (unsigned int) aux; - if (server.cluster.slots[slot] != server.cluster.myself) { - addReplyErrorFormat(c,"I'm not the owner of hash slot %u",slot); - return; - } + if ((slot = getSlotOrReply(c,c->argv[2])) == -1) return; + if (!strcasecmp(c->argv[3]->ptr,"migrating") && c->argc == 5) { + if (server.cluster.slots[slot] != server.cluster.myself) { + addReplyErrorFormat(c,"I'm not the owner of hash slot %u",slot); + return; + } if ((n = clusterLookupNode(c->argv[4]->ptr)) == NULL) { addReplyErrorFormat(c,"I don't know about node %s", (char*)c->argv[4]->ptr); @@ -1231,6 +1347,11 @@ void clusterCommand(redisClient *c) { } server.cluster.migrating_slots_to[slot] = n; } else if (!strcasecmp(c->argv[3]->ptr,"importing") && c->argc == 5) { + if (server.cluster.slots[slot] == server.cluster.myself) { + addReplyErrorFormat(c, + "I'm already the owner of hash slot %u",slot); + return; + } if ((n = clusterLookupNode(c->argv[4]->ptr)) == NULL) { addReplyErrorFormat(c,"I don't know about node %s", (char*)c->argv[3]->ptr); @@ -1238,8 +1359,45 @@ void clusterCommand(redisClient *c) { } server.cluster.importing_slots_from[slot] = n; } else if (!strcasecmp(c->argv[3]->ptr,"stable") && c->argc == 4) { + /* CLUSTER SETSLOT STABLE */ server.cluster.importing_slots_from[slot] = NULL; server.cluster.migrating_slots_to[slot] = NULL; + } else if (!strcasecmp(c->argv[3]->ptr,"node") && c->argc == 5) { + /* CLUSTER SETSLOT NODE */ + clusterNode *n = clusterLookupNode(c->argv[4]->ptr); + + if (!n) addReplyErrorFormat(c,"Unknown node %s", + (char*)c->argv[4]->ptr); + /* If this hash slot was served by 'myself' before to switch + * make sure there are no longer local keys for this hash slot. */ + if (server.cluster.slots[slot] == server.cluster.myself && + n != server.cluster.myself) + { + int numkeys; + robj **keys; + + keys = zmalloc(sizeof(robj*)*1); + numkeys = GetKeysInSlot(slot, keys, 1); + zfree(keys); + if (numkeys != 0) { + addReplyErrorFormat(c, "Can't assign hashslot %d to a different node while I still hold keys for this hash slot.", slot); + return; + } + } + /* If this node was the slot owner and the slot was marked as + * migrating, assigning the slot to another node will clear + * the migratig status. */ + if (server.cluster.slots[slot] == server.cluster.myself && + server.cluster.migrating_slots_to[slot]) + server.cluster.migrating_slots_to[slot] = NULL; + + /* If this node was importing this slot, assigning the slot to + * itself also clears the importing status. */ + if (n == server.cluster.myself && server.cluster.importing_slots_from[slot]) + server.cluster.importing_slots_from[slot] = NULL; + + clusterDelSlot(slot); + clusterAddSlot(n,slot); } else { addReplyError(c,"Invalid CLUSTER SETSLOT action or number of arguments"); return; @@ -1318,14 +1476,13 @@ void clusterCommand(redisClient *c) { /* RESTORE key ttl serialized-value */ void restoreCommand(redisClient *c) { - FILE *fp; - char buf[64]; - robj *o; - unsigned char *data; long ttl; + rio payload; + int type; + robj *obj; /* Make sure this key does not already exist here... */ - if (dbExists(c->db,c->argv[1])) { + if (lookupKeyWrite(c->db,c->argv[1]) != NULL) { addReplyError(c,"Target key name is busy."); return; } @@ -1338,46 +1495,20 @@ void restoreCommand(redisClient *c) { return; } - /* rdbLoadObject() only works against file descriptors so we need to - * dump the serialized object into a file and reload. */ - snprintf(buf,sizeof(buf),"redis-restore-%d.tmp",getpid()); - fp = fopen(buf,"w+"); - if (!fp) { - redisLog(REDIS_WARNING,"Can't open tmp file for RESTORE: %s", - strerror(errno)); - addReplyErrorFormat(c,"RESTORE failed, tmp file creation error: %s", - strerror(errno)); - return; - } - unlink(buf); - - /* Write the actual data and rewind the file */ - data = (unsigned char*) c->argv[3]->ptr; - if (fwrite(data+1,sdslen((sds)data)-1,1,fp) != 1) { - redisLog(REDIS_WARNING,"Can't write against tmp file for RESTORE: %s", - strerror(errno)); - addReplyError(c,"RESTORE failed, tmp file I/O error."); - fclose(fp); - return; - } - rewind(fp); - - /* Finally create the object from the serialized dump and - * store it at the specified key. */ - if ((data[0] > 4 && data[0] < 9) || - data[0] > 11 || - (o = rdbLoadObject(data[0],fp)) == NULL) + rioInitWithBuffer(&payload,c->argv[3]->ptr); + if (((type = rdbLoadObjectType(&payload)) == -1) || + ((obj = rdbLoadObject(type,&payload)) == NULL)) { - addReplyError(c,"Bad data format."); - fclose(fp); + addReplyError(c,"Bad data format"); return; } - fclose(fp); /* Create the key and set the TTL if any */ - dbAdd(c->db,c->argv[1],o); + dbAdd(c->db,c->argv[1],obj); if (ttl) setExpire(c->db,c->argv[1],time(NULL)+ttl); + signalModifiedKey(c->db,c->argv[1]); addReply(c,shared.ok); + server.dirty++; } /* MIGRATE host port key dbid timeout */ @@ -1385,12 +1516,9 @@ void migrateCommand(redisClient *c) { int fd; long timeout; long dbid; - char buf[64]; - FILE *fp; time_t ttl; robj *o; - unsigned char type; - off_t payload_len; + rio cmd, payload; /* Sanity check */ if (getLongFromObjectOrReply(c,c->argv[5],&timeout,NULL) != REDIS_OK) @@ -1403,7 +1531,7 @@ void migrateCommand(redisClient *c) { * nothing to migrate (for instance the key expired in the meantime), but * we include such information in the reply string. */ if ((o = lookupKeyRead(c->db,c->argv[3])) == NULL) { - addReplySds(c,sdsnew("+NOKEY")); + addReplySds(c,sdsnew("+NOKEY\r\n")); return; } @@ -1420,54 +1548,41 @@ void migrateCommand(redisClient *c) { return; } - /* Create temp file */ - snprintf(buf,sizeof(buf),"redis-migrate-%d.tmp",getpid()); - fp = fopen(buf,"w+"); - if (!fp) { - redisLog(REDIS_WARNING,"Can't open tmp file for MIGRATE: %s", - strerror(errno)); - addReplyErrorFormat(c,"MIGRATE failed, tmp file creation error: %s.", - strerror(errno)); - return; - } - unlink(buf); - - /* Build the SELECT + RESTORE query writing it in our temp file. */ - if (fwriteBulkCount(fp,'*',2) == 0) goto file_wr_err; - if (fwriteBulkString(fp,"SELECT",6) == 0) goto file_wr_err; - if (fwriteBulkLongLong(fp,dbid) == 0) goto file_wr_err; + rioInitWithBuffer(&cmd,sdsempty()); + redisAssertWithInfo(c,NULL,rioWriteBulkCount(&cmd,'*',2)); + redisAssertWithInfo(c,NULL,rioWriteBulkString(&cmd,"SELECT",6)); + redisAssertWithInfo(c,NULL,rioWriteBulkLongLong(&cmd,dbid)); ttl = getExpire(c->db,c->argv[3]); - type = o->type; - if (fwriteBulkCount(fp,'*',4) == 0) goto file_wr_err; - if (fwriteBulkString(fp,"RESTORE",7) == 0) goto file_wr_err; - if (fwriteBulkObject(fp,c->argv[3]) == 0) goto file_wr_err; - if (fwriteBulkLongLong(fp, (ttl == -1) ? 0 : ttl) == 0) goto file_wr_err; + redisAssertWithInfo(c,NULL,rioWriteBulkCount(&cmd,'*',4)); + redisAssertWithInfo(c,NULL,rioWriteBulkString(&cmd,"RESTORE",7)); + redisAssertWithInfo(c,NULL,c->argv[3]->encoding == REDIS_ENCODING_RAW); + redisAssertWithInfo(c,NULL,rioWriteBulkString(&cmd,c->argv[3]->ptr,sdslen(c->argv[3]->ptr))); + redisAssertWithInfo(c,NULL,rioWriteBulkLongLong(&cmd,(ttl == -1) ? 0 : ttl)); /* Finally the last argument that is the serailized object payload - * in the form: . */ - payload_len = rdbSavedObjectLen(o); - if (fwriteBulkCount(fp,'$',payload_len+1) == 0) goto file_wr_err; - if (fwrite(&type,1,1,fp) == 0) goto file_wr_err; - if (rdbSaveObject(fp,o) == -1) goto file_wr_err; - if (fwrite("\r\n",2,1,fp) == 0) goto file_wr_err; - - /* Tranfer the query to the other node */ - rewind(fp); + * in the form: . */ + rioInitWithBuffer(&payload,sdsempty()); + redisAssertWithInfo(c,NULL,rdbSaveObjectType(&payload,o)); + redisAssertWithInfo(c,NULL,rdbSaveObject(&payload,o) != -1); + redisAssertWithInfo(c,NULL,rioWriteBulkString(&cmd,payload.io.buffer.ptr,sdslen(payload.io.buffer.ptr))); + sdsfree(payload.io.buffer.ptr); + + /* Tranfer the query to the other node in 64K chunks. */ { - char buf[4096]; - size_t nread; - - while ((nread = fread(buf,1,sizeof(buf),fp)) != 0) { - int nwritten; - - nwritten = syncWrite(fd,buf,nread,timeout); - if (nwritten != (signed)nread) goto socket_wr_err; + sds buf = cmd.io.buffer.ptr; + size_t pos = 0, towrite; + int nwritten = 0; + + while ((towrite = sdslen(buf)-pos) > 0) { + towrite = (towrite > (64*1024) ? (64*1024) : towrite); + nwritten = syncWrite(fd,buf+nwritten,towrite,timeout); + if (nwritten != (signed)towrite) goto socket_wr_err; + pos += nwritten; } - if (ferror(fp)) goto file_rd_err; } - /* Read back the reply */ + /* Read back the reply. */ { char buf1[1024]; char buf2[1024]; @@ -1476,34 +1591,26 @@ void migrateCommand(redisClient *c) { if (syncReadLine(fd, buf1, sizeof(buf1), timeout) <= 0) goto socket_rd_err; if (syncReadLine(fd, buf2, sizeof(buf2), timeout) <= 0) - goto socket_rd_err; + goto socket_rd_err; if (buf1[0] == '-' || buf2[0] == '-') { addReplyErrorFormat(c,"Target instance replied with error: %s", (buf1[0] == '-') ? buf1+1 : buf2+1); } else { + robj *aux; + dbDelete(c->db,c->argv[3]); + signalModifiedKey(c->db,c->argv[3]); addReply(c,shared.ok); + server.dirty++; + + /* Translate MIGRATE as DEL for replication/AOF. */ + aux = createStringObject("DEL",3); + rewriteClientCommandVector(c,2,aux,c->argv[3]); + decrRefCount(aux); } } - fclose(fp); - close(fd); - return; -file_wr_err: - redisLog(REDIS_WARNING,"Can't write on tmp file for MIGRATE: %s", - strerror(errno)); - addReplyErrorFormat(c,"MIGRATE failed, tmp file write error: %s.", - strerror(errno)); - fclose(fp); - close(fd); - return; - -file_rd_err: - redisLog(REDIS_WARNING,"Can't read from tmp file for MIGRATE: %s", - strerror(errno)); - addReplyErrorFormat(c,"MIGRATE failed, tmp file read error: %s.", - strerror(errno)); - fclose(fp); + sdsfree(cmd.io.buffer.ptr); close(fd); return; @@ -1512,7 +1619,7 @@ socket_wr_err: strerror(errno)); addReplyErrorFormat(c,"MIGRATE failed, writing to target node: %s.", strerror(errno)); - fclose(fp); + sdsfree(cmd.io.buffer.ptr); close(fd); return; @@ -1521,7 +1628,7 @@ socket_rd_err: strerror(errno)); addReplyErrorFormat(c,"MIGRATE failed, reading from target node: %s.", strerror(errno)); - fclose(fp); + sdsfree(cmd.io.buffer.ptr); close(fd); return; } @@ -1530,74 +1637,39 @@ socket_rd_err: * DUMP is actually not used by Redis Cluster but it is the obvious * complement of RESTORE and can be useful for different applications. */ void dumpCommand(redisClient *c) { - char buf[64]; - FILE *fp; robj *o, *dumpobj; - sds dump = NULL; - off_t payload_len; - unsigned int type; + rio payload; /* Check if the key is here. */ if ((o = lookupKeyRead(c->db,c->argv[1])) == NULL) { addReply(c,shared.nullbulk); return; } - - /* Create temp file */ - snprintf(buf,sizeof(buf),"redis-dump-%d.tmp",getpid()); - fp = fopen(buf,"w+"); - if (!fp) { - redisLog(REDIS_WARNING,"Can't open tmp file for MIGRATE: %s", - strerror(errno)); - addReplyErrorFormat(c,"DUMP failed, tmp file creation error: %s.", - strerror(errno)); - return; - } - unlink(buf); - - /* Dump the serailized object and read it back in memory. - * We prefix it with a one byte containing the type ID. - * This is the serialization format understood by RESTORE. */ - if (rdbSaveObject(fp,o) == -1) goto file_wr_err; - payload_len = ftello(fp); - if (fseeko(fp,0,SEEK_SET) == -1) goto file_rd_err; - dump = sdsnewlen(NULL,payload_len+1); - if (payload_len && fread(dump+1,payload_len,1,fp) != 1) goto file_rd_err; - fclose(fp); - type = o->type; - if (type == REDIS_LIST && o->encoding == REDIS_ENCODING_ZIPLIST) - type = REDIS_LIST_ZIPLIST; - else if (type == REDIS_HASH && o->encoding == REDIS_ENCODING_ZIPMAP) - type = REDIS_HASH_ZIPMAP; - else if (type == REDIS_SET && o->encoding == REDIS_ENCODING_INTSET) - type = REDIS_SET_INTSET; - else - type = o->type; - dump[0] = type; + + /* Serialize the object in a RDB-like format. It consist of an object type + * byte followed by the serialized object. This is understood by RESTORE. */ + rioInitWithBuffer(&payload,sdsempty()); + redisAssertWithInfo(c,NULL,rdbSaveObjectType(&payload,o)); + redisAssertWithInfo(c,NULL,rdbSaveObject(&payload,o)); /* Transfer to the client */ - dumpobj = createObject(REDIS_STRING,dump); + dumpobj = createObject(REDIS_STRING,payload.io.buffer.ptr); addReplyBulk(c,dumpobj); decrRefCount(dumpobj); return; +} -file_wr_err: - redisLog(REDIS_WARNING,"Can't write on tmp file for DUMP: %s", - strerror(errno)); - addReplyErrorFormat(c,"DUMP failed, tmp file write error: %s.", - strerror(errno)); - sdsfree(dump); - fclose(fp); - return; - -file_rd_err: - redisLog(REDIS_WARNING,"Can't read from tmp file for DUMP: %s", - strerror(errno)); - addReplyErrorFormat(c,"DUMP failed, tmp file read error: %s.", - strerror(errno)); - sdsfree(dump); - fclose(fp); - return; +/* The ASKING command is required after a -ASK redirection. + * The client should issue ASKING before to actualy send the command to + * the target instance. See the Redis Cluster specification for more + * information. */ +void askingCommand(redisClient *c) { + if (server.cluster_enabled == 0) { + addReplyError(c,"This instance has cluster support disabled"); + return; + } + c->flags |= REDIS_ASKING; + addReply(c,shared.ok); } /* ----------------------------------------------------------------------------- @@ -1661,7 +1733,7 @@ clusterNode *getNodeByQuery(redisClient *c, struct redisCommand *cmd, robj **arg slot = keyHashSlot((char*)firstkey->ptr, sdslen(firstkey->ptr)); n = server.cluster.slots[slot]; - redisAssert(n != NULL); + redisAssertWithInfo(c,firstkey,n != NULL); } else { /* If it is not the first key, make sure it is exactly * the same key as the first we saw. */ @@ -1693,9 +1765,12 @@ clusterNode *getNodeByQuery(redisClient *c, struct redisCommand *cmd, robj **arg } /* Handle the case in which we are receiving this hash slot from * another instance, so we'll accept the query even if in the table - * it is assigned to a different node. */ - if (server.cluster.importing_slots_from[slot] != NULL) + * it is assigned to a different node, but only if the client + * issued an ASKING command before. */ + if (server.cluster.importing_slots_from[slot] != NULL && + c->flags & REDIS_ASKING) { return server.cluster.myself; + } /* It's not a -ASK case. Base case: just return the right node. */ return n; }