X-Git-Url: https://git.saurik.com/redis.git/blobdiff_plain/fd7a584f7ab9a796338e3bd6cacf01ce5476bb91..6fdc635447b1a5dd0a4f7b13c15fdd6c108dabee:/src/cluster.c diff --git a/src/cluster.c b/src/cluster.c index e2b82044..57243132 100644 --- a/src/cluster.c +++ b/src/cluster.c @@ -1,8 +1,10 @@ #include "redis.h" +#include "endianconv.h" #include #include #include +#include void clusterAcceptHandler(aeEventLoop *el, int fd, void *privdata, int mask); void clusterReadHandler(aeEventLoop *el, int fd, void *privdata, int mask); @@ -19,22 +21,6 @@ int clusterAddSlot(clusterNode *n, int slot); * Initialization * -------------------------------------------------------------------------- */ -void clusterGetRandomName(char *p) { - FILE *fp = fopen("/dev/urandom","r"); - char *charset = "0123456789abcdef"; - int j; - - if (!fp) { - redisLog(REDIS_WARNING, - "Unrecovarable error: can't open /dev/urandom:%s" ,strerror(errno)); - exit(1); - } - fread(p,REDIS_CLUSTER_NAMELEN,1,fp); - for (j = 0; j < REDIS_CLUSTER_NAMELEN; j++) - p[j] = charset[p[j] & 0x0F]; - fclose(fp); -} - int clusterLoadConfig(char *filename) { FILE *fp = fopen(filename,"r"); char *line; @@ -115,7 +101,30 @@ int clusterLoadConfig(char *filename) { for (j = 7; j < argc; j++) { int start, stop; - if ((p = strchr(argv[j],'-')) != NULL) { + if (argv[j][0] == '[') { + /* Here we handle migrating / importing slots */ + int slot; + char direction; + clusterNode *cn; + + p = strchr(argv[j],'-'); + redisAssert(p != NULL); + *p = '\0'; + direction = p[1]; /* Either '>' or '<' */ + slot = atoi(argv[j]+1); + p += 3; + cn = clusterLookupNode(p); + if (!cn) { + cn = createClusterNode(p,0); + clusterAddNode(cn); + } + if (direction == '>') { + server.cluster.migrating_slots_to[slot] = cn; + } else { + server.cluster.importing_slots_from[slot] = cn; + } + continue; + } else if ((p = strchr(argv[j],'-')) != NULL) { *p = '\0'; start = atoi(argv[j]); stop = atoi(p+1); @@ -201,7 +210,8 @@ void clusterInit(void) { exit(1); } if (aeCreateFileEvent(server.el, server.cfd, AE_READABLE, - clusterAcceptHandler, NULL) == AE_ERR) oom("creating file event"); + clusterAcceptHandler, NULL) == AE_ERR) redisPanic("Unrecoverable error creating Redis Cluster file event."); + server.cluster.slots_to_keys = zslCreate(); } /* ----------------------------------------------------------------------------- @@ -282,7 +292,7 @@ clusterNode *createClusterNode(char *nodename, int flags) { if (nodename) memcpy(node->name, nodename, REDIS_CLUSTER_NAMELEN); else - clusterGetRandomName(node->name); + getRandomHexChars(node->name, REDIS_CLUSTER_NAMELEN); node->flags = flags; memset(node->slots,0,sizeof(node->slots)); node->numslaves = 0; @@ -355,7 +365,7 @@ 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 @@ -417,7 +427,7 @@ 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); } @@ -471,6 +481,7 @@ 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 @@ -488,8 +499,10 @@ 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 || @@ -508,7 +521,16 @@ 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; @@ -579,7 +601,7 @@ int clusterProcessPacket(clusterLink *link) { } } /* Update our info about the node */ - link->node->pong_received = time(NULL); + if (link->node) link->node->pong_received = time(NULL); /* Update master/slave info */ if (sender) { @@ -642,8 +664,24 @@ 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; } @@ -736,9 +774,25 @@ 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); @@ -783,7 +837,7 @@ 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; @@ -820,20 +874,48 @@ 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. @@ -850,6 +932,17 @@ 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 * -------------------------------------------------------------------------- */ @@ -865,7 +958,7 @@ 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) { @@ -900,7 +993,7 @@ 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; @@ -917,7 +1010,7 @@ 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 & @@ -943,6 +1036,7 @@ void clusterCron(void) { node->flags &= ~REDIS_NODE_PFAIL; } else if (node->flags & REDIS_NODE_FAIL && !node->numslaves) { node->flags &= ~REDIS_NODE_FAIL; + clusterUpdateState(); } } else { /* Timeout reached. Set the noad se possibly failing if it is @@ -991,9 +1085,21 @@ 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; - printf("SLOT %d added to %.40s\n", slot, n->name); + 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; } @@ -1035,7 +1141,7 @@ 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 ", @@ -1064,7 +1170,8 @@ 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; @@ -1085,12 +1192,39 @@ sds clusterGenNodesDescription(void) { start = -1; } } + + /* Just for MYSELF node we also dump info about slots that + * we are migrating to other instances or importing from other + * instances. */ + if (node->flags & REDIS_NODE_MYSELF) { + for (j = 0; j < REDIS_CLUSTER_SLOTS; j++) { + if (server.cluster.migrating_slots_to[j]) { + ci = sdscatprintf(ci," [%d->-%.40s]",j, + server.cluster.migrating_slots_to[j]->name); + } else if (server.cluster.importing_slots_from[j]) { + ci = sdscatprintf(ci," [%d-<-%.40s]",j, + server.cluster.importing_slots_from[j]->name); + } + } + } ci = sdscatlen(ci,"\n",1); } dictReleaseIterator(di); 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"); @@ -1128,24 +1262,29 @@ 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; } @@ -1158,15 +1297,101 @@ 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); clusterUpdateState(); clusterSaveConfigOrDie(); addReply(c,shared.ok); + } else if (!strcasecmp(c->argv[1]->ptr,"setslot") && c->argc >= 4) { + /* SETSLOT 10 MIGRATING */ + /* SETSLOT 10 IMPORTING */ + /* SETSLOT 10 STABLE */ + /* SETSLOT 10 NODE */ + int slot; + clusterNode *n; + + 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); + return; + } + 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); + return; + } + 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; + } + clusterSaveConfigOrDie(); + addReply(c,shared.ok); } else if (!strcasecmp(c->argv[1]->ptr,"info") && c->argc == 2) { char *statestr[] = {"ok","fail","needhelp"}; int slots_assigned = 0, slots_ok = 0, slots_pfail = 0, slots_fail = 0; @@ -1192,35 +1417,137 @@ void clusterCommand(redisClient *c) { "cluster_slots_ok:%d\r\n" "cluster_slots_pfail:%d\r\n" "cluster_slots_fail:%d\r\n" + "cluster_known_nodes:%lu\r\n" , statestr[server.cluster.state], slots_assigned, slots_ok, slots_pfail, - slots_fail + slots_fail, + dictSize(server.cluster.nodes) ); addReplySds(c,sdscatprintf(sdsempty(),"$%lu\r\n", (unsigned long)sdslen(info))); addReplySds(c,info); addReply(c,shared.crlf); + } else if (!strcasecmp(c->argv[1]->ptr,"keyslot") && c->argc == 3) { + sds key = c->argv[2]->ptr; + + addReplyLongLong(c,keyHashSlot(key,sdslen(key))); + } else if (!strcasecmp(c->argv[1]->ptr,"getkeysinslot") && c->argc == 4) { + long long maxkeys, slot; + unsigned int numkeys, j; + robj **keys; + + if (getLongLongFromObjectOrReply(c,c->argv[2],&slot,NULL) != REDIS_OK) + return; + if (getLongLongFromObjectOrReply(c,c->argv[3],&maxkeys,NULL) != REDIS_OK) + return; + if (slot < 0 || slot >= REDIS_CLUSTER_SLOTS || maxkeys < 0 || + maxkeys > 1024*1024) { + addReplyError(c,"Invalid slot or number of keys"); + return; + } + + keys = zmalloc(sizeof(robj*)*maxkeys); + numkeys = GetKeysInSlot(slot, keys, maxkeys); + addReplyMultiBulkLen(c,numkeys); + for (j = 0; j < numkeys; j++) addReplyBulk(c,keys[j]); + zfree(keys); } else { addReplyError(c,"Wrong CLUSTER subcommand or number of arguments"); } } /* ----------------------------------------------------------------------------- - * RESTORE and MIGRATE commands + * DUMP, RESTORE and MIGRATE commands * -------------------------------------------------------------------------- */ +/* Generates a DUMP-format representation of the object 'o', adding it to the + * io stream pointed by 'rio'. This function can't fail. */ +void createDumpPayload(rio *payload, robj *o) { + unsigned char buf[2]; + uint64_t crc; + + /* 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()); + redisAssert(rdbSaveObjectType(payload,o)); + redisAssert(rdbSaveObject(payload,o)); + + /* Write the footer, this is how it looks like: + * ----------------+---------------------+---------------+ + * ... RDB payload | 2 bytes RDB version | 8 bytes CRC64 | + * ----------------+---------------------+---------------+ + * RDB version and CRC are both in little endian. + */ + + /* RDB version */ + buf[0] = REDIS_RDB_VERSION & 0xff; + buf[1] = (REDIS_RDB_VERSION >> 8) & 0xff; + payload->io.buffer.ptr = sdscatlen(payload->io.buffer.ptr,buf,2); + + /* CRC64 */ + crc = crc64(0,(unsigned char*)payload->io.buffer.ptr, + sdslen(payload->io.buffer.ptr)); + memrev64ifbe(&crc); + payload->io.buffer.ptr = sdscatlen(payload->io.buffer.ptr,&crc,8); +} + +/* Verify that the RDB version of the dump payload matches the one of this Redis + * instance and that the checksum is ok. + * If the DUMP payload looks valid REDIS_OK is returned, otherwise REDIS_ERR + * is returned. */ +int verifyDumpPayload(unsigned char *p, size_t len) { + unsigned char *footer; + uint16_t rdbver; + uint64_t crc; + + /* At least 2 bytes of RDB version and 8 of CRC64 should be present. */ + if (len < 10) return REDIS_ERR; + footer = p+(len-10); + + /* Verify RDB version */ + rdbver = (footer[1] << 8) | footer[0]; + if (rdbver != REDIS_RDB_VERSION) return REDIS_ERR; + + /* Verify CRC64 */ + crc = crc64(0,p,len-8); + memrev64ifbe(&crc); + return (memcmp(&crc,footer+2,8) == 0) ? REDIS_OK : REDIS_ERR; +} + +/* DUMP keyname + * 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) { + robj *o, *dumpobj; + rio payload; + + /* Check if the key is here. */ + if ((o = lookupKeyRead(c->db,c->argv[1])) == NULL) { + addReply(c,shared.nullbulk); + return; + } + + /* Create the DUMP encoded representation. */ + createDumpPayload(&payload,o); + + /* Transfer to the client */ + dumpobj = createObject(REDIS_STRING,payload.io.buffer.ptr); + addReplyBulk(c,dumpobj); + decrRefCount(dumpobj); + return; +} + /* 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; } @@ -1233,46 +1560,26 @@ 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); + /* Verify RDB version and data checksum. */ + if (verifyDumpPayload(c->argv[3]->ptr,sdslen(c->argv[3]->ptr)) == REDIS_ERR) { + addReplyError(c,"DUMP payload version or checksum are wrong"); 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); - if (ttl) setExpire(c->db,c->argv[1],time(NULL)+ttl); + dbAdd(c->db,c->argv[1],obj); + if (ttl) setExpire(c->db,c->argv[1],mstime()+ttl); + signalModifiedKey(c->db,c->argv[1]); addReply(c,shared.ok); + server.dirty++; } /* MIGRATE host port key dbid timeout */ @@ -1280,12 +1587,9 @@ void migrateCommand(redisClient *c) { int fd; long timeout; long dbid; - char buf[64]; - FILE *fp; - time_t ttl; + long long ttl = 0, expireat; robj *o; - unsigned char type; - off_t payload_len; + rio cmd, payload; /* Sanity check */ if (getLongFromObjectOrReply(c,c->argv[5],&timeout,NULL) != REDIS_OK) @@ -1298,7 +1602,7 @@ 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; } @@ -1311,58 +1615,49 @@ void migrateCommand(redisClient *c) { return; } if ((aeWait(fd,AE_WRITABLE,timeout*1000) & AE_WRITABLE) == 0) { - addReplyError(c,"Timeout connecting to the client"); - 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)); + addReplySds(c,sdsnew("-IOERR error or timeout connecting to the client\r\n")); 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; + /* Create RESTORE payload and generate the protocol to call the command. */ + 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; + expireat = getExpire(c->db,c->argv[3]); + if (expireat != -1) { + ttl = expireat-mstime(); + if (ttl < 1) ttl = 1; + } + 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)); /* 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); - { - char buf[4096]; - size_t nread; - - while ((nread = fread(buf,1,sizeof(buf),fp)) != 0) { - int nwritten; + * in the DUMP format. */ + createDumpPayload(&payload,o); + redisAssertWithInfo(c,NULL,rioWriteBulkString(&cmd,payload.io.buffer.ptr, + sdslen(payload.io.buffer.ptr))); + sdsfree(payload.io.buffer.ptr); - nwritten = syncWrite(fd,buf,nread,timeout); - if (nwritten != (signed)nread) goto socket_wr_err; + /* Tranfer the query to the other node in 64K chunks. */ + { + 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+pos,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]; @@ -1371,128 +1666,53 @@ 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; socket_wr_err: - redisLog(REDIS_NOTICE,"Can't write to target node for MIGRATE: %s", - strerror(errno)); - addReplyErrorFormat(c,"MIGRATE failed, writing to target node: %s.", - strerror(errno)); - fclose(fp); + addReplySds(c,sdsnew("-IOERR error or timeout writing to target instance\r\n")); + sdsfree(cmd.io.buffer.ptr); close(fd); return; socket_rd_err: - redisLog(REDIS_NOTICE,"Can't read from target node for MIGRATE: %s", - strerror(errno)); - addReplyErrorFormat(c,"MIGRATE failed, reading from target node: %s.", - strerror(errno)); - fclose(fp); + addReplySds(c,sdsnew("-IOERR error or timeout reading from target node\r\n")); + sdsfree(cmd.io.buffer.ptr); close(fd); return; } -/* DUMP keyname - * 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; - - /* 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)); +/* 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; } - 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; - - /* Transfer to the client */ - dumpobj = createObject(REDIS_STRING,dump); - 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; + c->flags |= REDIS_ASKING; + addReply(c,shared.ok); } /* ----------------------------------------------------------------------------- @@ -1502,12 +1722,19 @@ file_rd_err: /* Return the pointer to the cluster node that is able to serve the query * as all the keys belong to hash slots for which the node is in charge. * - * If keys in query spawn multiple nodes NULL is returned. */ -clusterNode *getNodeByQuery(redisClient *c, struct redisCommand *cmd, robj **argv, int argc, int *hashslot) { + * If the returned node should be used only for this request, the *ask + * integer is set to '1', otherwise to '0'. This is used in order to + * let the caller know if we should reply with -MOVED or with -ASK. + * + * If the request contains more than a single key NULL is returned, + * however a request with more then a key argument where the key is always + * the same is valid, like in: RPOPLPUSH mylist mylist.*/ +clusterNode *getNodeByQuery(redisClient *c, struct redisCommand *cmd, robj **argv, int argc, int *hashslot, int *ask) { clusterNode *n = NULL; + robj *firstkey = NULL; multiState *ms, _ms; multiCmd mc; - int i; + int i, slot = 0; /* We handle all the cases as if they were EXEC commands, so we have * a common code path for everything */ @@ -1517,7 +1744,9 @@ clusterNode *getNodeByQuery(redisClient *c, struct redisCommand *cmd, robj **arg if (!(c->flags & REDIS_MULTI)) return server.cluster.myself; ms = &c->mstate; } else { - /* Create a fake Multi State structure, with just one command */ + /* In order to have a single codepath create a fake Multi State + * structure if the client is not in MULTI/EXEC state, this way + * we have a single codepath below. */ ms = &_ms; _ms.commands = &mc; _ms.count = 1; @@ -1526,6 +1755,8 @@ clusterNode *getNodeByQuery(redisClient *c, struct redisCommand *cmd, robj **arg mc.cmd = cmd; } + /* Check that all the keys are the same key, and get the slot and + * node for this key. */ for (i = 0; i < ms->count; i++) { struct redisCommand *mcmd; robj **margv; @@ -1536,26 +1767,53 @@ clusterNode *getNodeByQuery(redisClient *c, struct redisCommand *cmd, robj **arg margv = ms->commands[i].argv; keyindex = getKeysFromCommand(mcmd,margv,margc,&numkeys, - REDIS_GETKEYS_PRELOAD); + REDIS_GETKEYS_ALL); for (j = 0; j < numkeys; j++) { - int slot = keyHashSlot((char*)margv[keyindex[j]]->ptr, - sdslen(margv[keyindex[j]]->ptr)); - struct clusterNode *slotnode; - - slotnode = server.cluster.slots[slot]; - if (hashslot) *hashslot = slot; - /* Node not assigned? (Should never happen actually - * if we reached this function). - * Different node than the previous one? - * Return NULL, the cluster can't serve multi-node requests */ - if (slotnode == NULL || (n && slotnode != n)) { - getKeysFreeResult(keyindex); - return NULL; + if (firstkey == NULL) { + /* This is the first key we see. Check what is the slot + * and node. */ + firstkey = margv[keyindex[j]]; + + slot = keyHashSlot((char*)firstkey->ptr, sdslen(firstkey->ptr)); + n = server.cluster.slots[slot]; + redisAssertWithInfo(c,firstkey,n != NULL); } else { - n = slotnode; + /* If it is not the first key, make sure it is exactly + * the same key as the first we saw. */ + if (!equalStringObjects(firstkey,margv[keyindex[j]])) { + decrRefCount(firstkey); + getKeysFreeResult(keyindex); + return NULL; + } } } getKeysFreeResult(keyindex); } - return (n == NULL) ? server.cluster.myself : n; + if (ask) *ask = 0; /* This is the default. Set to 1 if needed later. */ + /* No key at all in command? then we can serve the request + * without redirections. */ + if (n == NULL) return server.cluster.myself; + if (hashslot) *hashslot = slot; + /* This request is about a slot we are migrating into another instance? + * Then we need to check if we have the key. If we have it we can reply. + * If instead is a new key, we pass the request to the node that is + * receiving the slot. */ + if (n == server.cluster.myself && + server.cluster.migrating_slots_to[slot] != NULL) + { + if (lookupKeyRead(&server.db[0],firstkey) == NULL) { + if (ask) *ask = 1; + return server.cluster.migrating_slots_to[slot]; + } + } + /* 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, 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; }