Merge branch 'bettercluster' into unstable

src/endianconv.h

@@ -61,4 +61,14 @@ uint64_t intrev64(uint64_t v);
 #define intrev64ifbe(v) intrev64(v)
 #endif
 
+/* The functions htonu64() and ntohu64() convert the specified value to
+ * network byte ordering and back. In big endian systems they are no-ops. */
+#if (BYTE_ORDER == BIG_ENDIAN)
+#define htonu64(v) (v)
+#define ntohu64(v) (v)
+#else
+#define htonu64(v) intrev64(v)
+#define ntohu64(v) intrev64(v)
+#endif
+
 #endif

src/redis-trib.rb

@@ -118,8 +118,8 @@ class ClusterNode
         nodes.each{|n|
             # name addr flags role ping_sent ping_recv link_status slots
             split = n.split
-            name,addr,flags,role,ping_sent,ping_recv,link_status = split[0..6]
-            slots = split[7..-1]
+            name,addr,flags,role,ping_sent,ping_recv,config_epoch,link_status = split[0..6]
+            slots = split[8..-1]
             info = {
                 :name => name,
                 :addr => addr,
@@ -230,7 +230,7 @@ class ClusterNode
         config = []
         @r.cluster("nodes").each_line{|l|
             s = l.split
-            slots = s[7..-1].select {|x| x[0..0] != "["}
+            slots = s[8..-1].select {|x| x[0..0] != "["}
             next if slots.length == 0
             config << s[0]+":"+(slots.sort.join(","))
         }

src/redis.c

@@ -1203,6 +1203,9 @@ void beforeSleep(struct aeEventLoop *eventLoop) {
 
     /* Write the AOF buffer on disk */
     flushAppendOnlyFile(0);
+
+    /* Call the Redis Cluster before sleep function. */
+    if (server.cluster_enabled) clusterBeforeSleep();
 }
 
 /* =========================== Server initialization ======================== */

src/redis.h

@@ -368,6 +368,8 @@
  * Data types
  *----------------------------------------------------------------------------*/
 
+typedef long long mstime_t; /* millisecond time type. */
+
 /* A redis object, that is a type able to hold a string / list / set */
 
 /* The actual Redis Object */
@@ -581,7 +583,7 @@ typedef struct redisOpArray {
 #define REDIS_CLUSTER_FAIL_UNDO_TIME_MULT 2 /* Undo fail if master is back. */
 #define REDIS_CLUSTER_FAIL_UNDO_TIME_ADD 10 /* Some additional time. */
 #define REDIS_CLUSTER_SLAVE_VALIDITY_MULT 10 /* Slave data validity. */
-#define REDIS_CLUSTER_FAILOVER_AUTH_RETRY_MULT 1 /* Auth request retry time. */
+#define REDIS_CLUSTER_FAILOVER_AUTH_RETRY_MULT 4 /* Auth request retry time. */
 #define REDIS_CLUSTER_FAILOVER_DELAY 5 /* Seconds */
 
 struct clusterNode;
@@ -617,6 +619,7 @@ struct clusterNode {
     time_t ctime;   /* Node object creation time. */
     char name[REDIS_CLUSTER_NAMELEN]; /* Node name, hex string, sha1-size */
     int flags;      /* REDIS_NODE_... */
+    uint64_t configEpoch; /* Last configEpoch observed for this node */
     unsigned char slots[REDIS_CLUSTER_SLOTS/8]; /* slots handled by this node */
     int numslots;   /* Number of slots handled by this node */
     int numslaves;  /* Number of slave nodes, if this is a master */
@@ -625,6 +628,7 @@ struct clusterNode {
     time_t ping_sent;       /* Unix time we sent latest ping */
     time_t pong_received;   /* Unix time we received the pong */
     time_t fail_time;       /* Unix time when FAIL flag was set */
+    time_t voted_time;      /* Last time we voted for a slave of this master */
     char ip[REDIS_IP_STR_LEN];  /* Latest known IP address of this node */
     int port;                   /* Latest known port of this node */
     clusterLink *link;          /* TCP/IP link with this node */
@@ -634,6 +638,7 @@ typedef struct clusterNode clusterNode;
 
 typedef struct {
     clusterNode *myself;  /* This node */
+    uint64_t currentEpoch;
     int state;            /* REDIS_CLUSTER_OK, REDIS_CLUSTER_FAIL, ... */
     int size;             /* Num of master nodes with at least one slot */
     dict *nodes;          /* Hash table of name -> clusterNode structures */
@@ -641,10 +646,24 @@ typedef struct {
     clusterNode *importing_slots_from[REDIS_CLUSTER_SLOTS];
     clusterNode *slots[REDIS_CLUSTER_SLOTS];
     zskiplist *slots_to_keys;
-    int failover_auth_time;     /* Time at which we sent the AUTH request. */
-    int failover_auth_count;    /* Number of authorizations received. */
+    /* The following fields are used to take the slave state on elections. */
+    mstime_t failover_auth_time;/* Time at which we'll try to get elected in ms*/
+    int failover_auth_count;    /* Number of votes received so far. */
+    int failover_auth_sent;     /* True if we already asked for votes. */
+    uint64_t failover_auth_epoch; /* Epoch of the current election. */
+    /* The followign fields are uesd by masters to take state on elections. */
+    uint64_t last_vote_epoch;   /* Epoch of the last vote granted. */
+    int todo_before_sleep; /* Things to do in clusterBeforeSleep(). */
+    long long stats_bus_messages_sent;  /* Num of msg sent via cluster bus. */
+    long long stats_bus_messages_received; /* Num of msg received via cluster bus. */
 } clusterState;
 
+/* clusterState todo_before_sleep flags. */
+#define CLUSTER_TODO_HANDLE_FAILOVER (1<<0)
+#define CLUSTER_TODO_UPDATE_STATE (1<<1)
+#define CLUSTER_TODO_SAVE_CONFIG (1<<2)
+#define CLUSTER_TODO_FSYNC_CONFIG (1<<3)
+
 /* Redis cluster messages header */
 
 /* Note that the PING, PONG and MEET messages are actually the same exact
@@ -704,9 +723,9 @@ typedef struct {
     uint32_t totlen;    /* Total length of this message */
     uint16_t type;      /* Message type */
     uint16_t count;     /* Only used for some kind of messages. */
-    uint64_t time;      /* Time at which this request was sent (in milliseconds),
-                           this field is copied in reply messages so that the
-                           original sender knows how old the reply is. */
+    uint64_t currentEpoch;  /* The epoch accordingly to the sending node. */
+    uint64_t configEpoch;   /* The config epoch if it's a master, or the last epoch
+                               advertised by its master if it is a slave. */
     char sender[REDIS_CLUSTER_NAMELEN]; /* Name of the sender node */
     unsigned char myslots[REDIS_CLUSTER_SLOTS/8];
     char slaveof[REDIS_CLUSTER_NAMELEN];
@@ -1367,6 +1386,7 @@ void clusterCron(void);
 clusterNode *getNodeByQuery(redisClient *c, struct redisCommand *cmd, robj **argv, int argc, int *hashslot, int *ask);
 void clusterPropagatePublish(robj *channel, robj *message);
 void migrateCloseTimedoutSockets(void);
+void clusterBeforeSleep(void);
 
 /* Sentinel */
 void initSentinelConfig(void);

src/replication.c

@@ -343,7 +343,7 @@ int masterTryPartialResynchronization(redisClient *c) {
         /* Run id "?" is used by slaves that want to force a full resync. */
         if (master_runid[0] != '?') {
             redisLog(REDIS_NOTICE,"Partial resynchronization not accepted: "
-                "Runid mismatch (Client asked for '%s', I'm '%s')",
+                "Runid mismatch (Client asked for runid '%s', my runid is '%s')",
                 master_runid, server.runid);
         } else {
             redisLog(REDIS_NOTICE,"Full resync requested by slave.");

src/sentinel.c

@@ -43,8 +43,6 @@ extern char **environ;
 
 /* ======================== Sentinel global state =========================== */
 
-typedef long long mstime_t; /* millisecond time type. */
-
 /* Address object, used to describe an ip:port pair. */
 typedef struct sentinelAddr {
     char *ip;