First implementation of the ASKING command. Semantics still to verify.

[redis.git] / src / cluster.c

diff --git a/src/cluster.c b/src/cluster.c
index 1c0b1c23e54a3b7b7401fb9a9f99fc48d4aaf8a4..ccb230c91777f338062842ef64c18e5580c7807b 100644 (file)
--- a/src/cluster.c
+++ b/src/cluster.c
@@ -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) {
 /* 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
 }
 
 /* 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;
 
     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 ||
     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;
     }
         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;
     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();
         }
             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 {
     } else {

-        redisLog(REDIS_NOTICE,"Received unknown packet type: %d", type);
+        redisLog(REDIS_WARNING,"Received unknown packet type: %d", type);

     }
     return 1;
 }
     }
     return 1;
 }


@@ -758,6 +786,22 @@ void clusterSendMessage(clusterLink *link, unsigned char *msg, size_t msglen) {
     link->sndbuf = sdscatlen(link->sndbuf, msg, 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 = dictGetEntryVal(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;
 /* Build the message header */
 void clusterBuildMessageHdr(clusterMsg *hdr, int type) {
     int totlen;


@@ -842,20 +886,48 @@ void clusterSendPing(clusterLink *link, int type) {
     clusterSendMessage(link,buf,totlen);
 }
 
     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.
 }
 
 /* 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));
 }
 
     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
  * -------------------------------------------------------------------------- */
 /* -----------------------------------------------------------------------------
  * CLUSTER cron job
  * -------------------------------------------------------------------------- */


@@ -1099,7 +1182,8 @@ sds clusterGenNodesDescription(void) {
         ci = sdscatprintf(ci,"%ld %ld %s",
             (long) node->ping_sent,
             (long) node->pong_received,
         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;
 
         /* Slots served by this instance */
         start = -1;


@@ -1191,7 +1275,10 @@ void clusterCommand(redisClient *c) {
         addReplyBulk(c,o);
         decrRefCount(o);
     } else if ((!strcasecmp(c->argv[1]->ptr,"addslots") ||
         addReplyBulk(c,o);
         decrRefCount(o);
     } else if ((!strcasecmp(c->argv[1]->ptr,"addslots") ||

-               !strcasecmp(c->argv[1]->ptr,"delslots")) && c->argc >= 3) {
+               !strcasecmp(c->argv[1]->ptr,"delslots")) && c->argc >= 3)
+    {
+        /* CLUSTER ADDSLOTS <slot> [slot] ... */
+        /* CLUSTER DELSLOTS <slot> [slot] ... */

         int j, slot;
         unsigned char *slots = zmalloc(REDIS_CLUSTER_SLOTS);
         int del = !strcasecmp(c->argv[1]->ptr,"delslots");
         int j, slot;
         unsigned char *slots = zmalloc(REDIS_CLUSTER_SLOTS);
         int del = !strcasecmp(c->argv[1]->ptr,"delslots");


@@ -1231,7 +1318,7 @@ void clusterCommand(redisClient *c) {
 
                 retval = del ? clusterDelSlot(j) :
                                clusterAddSlot(server.cluster.myself,j);
 
                 retval = del ? clusterDelSlot(j) :
                                clusterAddSlot(server.cluster.myself,j);

-                redisAssert(retval == REDIS_OK);
+                redisAssertWithInfo(c,NULL,retval == REDIS_OK);

             }
         }
         zfree(slots);
             }
         }
         zfree(slots);


@@ -1239,9 +1326,10 @@ void clusterCommand(redisClient *c) {
         clusterSaveConfigOrDie();
         addReply(c,shared.ok);
     } else if (!strcasecmp(c->argv[1]->ptr,"setslot") && c->argc >= 4) {
         clusterSaveConfigOrDie();
         addReply(c,shared.ok);
     } else if (!strcasecmp(c->argv[1]->ptr,"setslot") && c->argc >= 4) {

-        /* SETSLOT 10 MIGRATING <instance ID> */
-        /* SETSLOT 10 IMPORTING <instance ID> */
+        /* SETSLOT 10 MIGRATING <node ID> */
+        /* SETSLOT 10 IMPORTING <node ID> */

         /* SETSLOT 10 STABLE */
         /* SETSLOT 10 STABLE */

+        /* SETSLOT 10 NODE <node ID> */

         int slot;
         clusterNode *n;
 
         int slot;
         clusterNode *n;
 


@@ -1274,7 +1362,7 @@ void clusterCommand(redisClient *c) {
             /* CLUSTER SETSLOT <SLOT> STABLE */
             server.cluster.importing_slots_from[slot] = NULL;
             server.cluster.migrating_slots_to[slot] = NULL;
             /* CLUSTER SETSLOT <SLOT> 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 == 4) {
+        } else if (!strcasecmp(c->argv[3]->ptr,"node") && c->argc == 5) {

             /* CLUSTER SETSLOT <SLOT> NODE <NODE ID> */
             clusterNode *n = clusterLookupNode(c->argv[4]->ptr);
 
             /* CLUSTER SETSLOT <SLOT> NODE <NODE ID> */
             clusterNode *n = clusterLookupNode(c->argv[4]->ptr);
 


@@ -1291,7 +1379,7 @@ void clusterCommand(redisClient *c) {
                 keys = zmalloc(sizeof(robj*)*1);
                 numkeys = GetKeysInSlot(slot, keys, 1);
                 zfree(keys);
                 keys = zmalloc(sizeof(robj*)*1);
                 numkeys = GetKeysInSlot(slot, keys, 1);
                 zfree(keys);

-                if (numkeys == 0) {
+                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;
                 }
                     addReplyErrorFormat(c, "Can't assign hashslot %d to a different node while I still hold keys for this hash slot.", slot);
                     return;
                 }


@@ -1303,6 +1391,11 @@ void clusterCommand(redisClient *c) {
                 server.cluster.migrating_slots_to[slot])
                 server.cluster.migrating_slots_to[slot] = NULL;
 
                 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 {
             clusterDelSlot(slot);
             clusterAddSlot(n,slot);
         } else {


@@ -1389,7 +1482,7 @@ void restoreCommand(redisClient *c) {
     robj *obj;
 
     /* Make sure this key does not already exist here... */
     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;
     }
         addReplyError(c,"Target key name is busy.");
         return;
     }


@@ -1402,7 +1495,7 @@ void restoreCommand(redisClient *c) {
         return;
     }
 
         return;
     }
 

-    payload = rioInitWithBuffer(c->argv[3]->ptr);
+    rioInitWithBuffer(&payload,c->argv[3]->ptr);

     if (((type = rdbLoadObjectType(&payload)) == -1) ||
         ((obj = rdbLoadObject(type,&payload)) == NULL))
     {
     if (((type = rdbLoadObjectType(&payload)) == -1) ||
         ((obj = rdbLoadObject(type,&payload)) == NULL))
     {


@@ -1414,6 +1507,7 @@ void restoreCommand(redisClient *c) {
     dbAdd(c->db,c->argv[1],obj);
     if (ttl) setExpire(c->db,c->argv[1],time(NULL)+ttl);
     addReply(c,shared.ok);
     dbAdd(c->db,c->argv[1],obj);
     if (ttl) setExpire(c->db,c->argv[1],time(NULL)+ttl);
     addReply(c,shared.ok);

+    server.dirty++;

 }
 
 /* MIGRATE host port key dbid timeout */
 }
 
 /* MIGRATE host port key dbid timeout */


@@ -1436,7 +1530,7 @@ 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) {
      * 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;
     }
     
         return;
     }
     


@@ -1453,24 +1547,24 @@ void migrateCommand(redisClient *c) {
         return;
     }
 
         return;
     }
 

-    cmd = rioInitWithBuffer(sdsempty());
-    redisAssert(rioWriteBulkCount(&cmd,'*',2));
-    redisAssert(rioWriteBulkString(&cmd,"SELECT",6));
-    redisAssert(rioWriteBulkLongLong(&cmd,dbid));
+    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]);
 
     ttl = getExpire(c->db,c->argv[3]);

-    redisAssert(rioWriteBulkCount(&cmd,'*',4));
-    redisAssert(rioWriteBulkString(&cmd,"RESTORE",7));
-    redisAssert(c->argv[3]->encoding == REDIS_ENCODING_RAW);
-    redisAssert(rioWriteBulkString(&cmd,c->argv[3]->ptr,sdslen(c->argv[3]->ptr)));
-    redisAssert(rioWriteBulkLongLong(&cmd,(ttl == -1) ? 0 : ttl));
+    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: <type><rdb-serialized-object>. */
 
     /* Finally the last argument that is the serailized object payload
      * in the form: <type><rdb-serialized-object>. */

-    payload = rioInitWithBuffer(sdsempty());
-    redisAssert(rdbSaveObjectType(&payload,o));
-    redisAssert(rdbSaveObject(&payload,o) != -1);
-    redisAssert(rioWriteBulkString(&cmd,payload.io.buffer.ptr,sdslen(payload.io.buffer.ptr)));
+    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. */
     sdsfree(payload.io.buffer.ptr);
 
     /* Tranfer the query to the other node in 64K chunks. */


@@ -1501,8 +1595,16 @@ void migrateCommand(redisClient *c) {
             addReplyErrorFormat(c,"Target instance replied with error: %s",
                 (buf1[0] == '-') ? buf1+1 : buf2+1);
         } else {
             addReplyErrorFormat(c,"Target instance replied with error: %s",
                 (buf1[0] == '-') ? buf1+1 : buf2+1);
         } else {

+            robj *aux;
+

             dbDelete(c->db,c->argv[3]);
             addReply(c,shared.ok);
             dbDelete(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);

         }
     }
 
         }
     }
 


@@ -1544,9 +1646,9 @@ void dumpCommand(redisClient *c) {
 
     /* 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. */
 
     /* 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. */

-    payload = rioInitWithBuffer(sdsempty());
-    redisAssert(rdbSaveObjectType(&payload,o));
-    redisAssert(rdbSaveObject(&payload,o));
+    rioInitWithBuffer(&payload,sdsempty());
+    redisAssertWithInfo(c,NULL,rdbSaveObjectType(&payload,o));
+    redisAssertWithInfo(c,NULL,rdbSaveObject(&payload,o));

 
     /* Transfer to the client */
     dumpobj = createObject(REDIS_STRING,payload.io.buffer.ptr);
 
     /* Transfer to the client */
     dumpobj = createObject(REDIS_STRING,payload.io.buffer.ptr);


@@ -1555,6 +1657,19 @@ void dumpCommand(redisClient *c) {
     return;
 }
 
     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);
+}
+

 /* -----------------------------------------------------------------------------
  * Cluster functions related to serving / redirecting clients
  * -------------------------------------------------------------------------- */
 /* -----------------------------------------------------------------------------
  * Cluster functions related to serving / redirecting clients
  * -------------------------------------------------------------------------- */


@@ -1616,7 +1731,7 @@ clusterNode *getNodeByQuery(redisClient *c, struct redisCommand *cmd, robj **arg
 
                 slot = keyHashSlot((char*)firstkey->ptr, sdslen(firstkey->ptr));
                 n = server.cluster.slots[slot];
 
                 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. */
             } else {
                 /* If it is not the first key, make sure it is exactly
                  * the same key as the first we saw. */


@@ -1648,9 +1763,12 @@ 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
     }
     /* 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;
         return server.cluster.myself;

+    }

     /* It's not a -ASK case. Base case: just return the right node. */
     return n;
 }
     /* It's not a -ASK case. Base case: just return the right node. */
     return n;
 }