CLUSTER

   1 CLUSTER README
   2 ==============
   3
   4 Redis Cluster is currently a work in progress, however there are a few things
   5 that you can do already with it to see how it works.
   6
   7 The following guide show you how to setup a three nodes cluster and issue some
   8 basic command against it.
   9
  10 ... WORK IN PROGRESS ...
  11
  12 1) Show MIGRATE
  13 2) Show CLUSTER MEET
  14 3) Show link status detection with CLUSTER NODES
  15 4) Show how to add slots with CLUSTER ADDSLOTS
  16 5) Show redirection
  17 6) Show cluster down
  18
  19 ... WORK IN PROGRESS ...
  20
  21 TODO
  22 ====
  23
  24 *** WARNING: all the following probably has some meaning only for
  25 *** me (antirez), most info are not updated, so please consider this file
  26 *** as a private TODO list / brainstorming.
  27
  28 - disconnect FAIL clients after some pong idle time.
  29
  30 ---------------------------------
  31
  32 * Majority rule: the cluster con continue when there are all the hash slots covered AND when there are the majority of masters.
  33 * Shutdown on request rule: when a node sees many connections closed or even a timeout longer than usual on almost all the other nodes, it will usually wait for the normal timeout before to change the state, unless it receives a query from a client: in such a case it will put itself into error status.
  34
  35 --------------------------------
  36
  37 * When asked for a key that is not in a node's business it will reply:
  38
  39     -ASK 1.2.3.4:6379   (in case we want the client to ask just one time)
  40     -MOVED <slotid> 1.2.3.4:6379  (in case the hash slot is permanently moved)
  41
  42 So with -ASK a client should just retry the query against this new node, a single time.
  43
  44 With -MOVED the client should update its hash slots table to reflect the fact that now the specified node is the one to contact for the specified hash slot.
  45
  46 * Nodes communicate using a binary protocol.
  47
  48 * Node failure detection.
  49
  50     1) Every node contains information about all the other nodes:
  51         - If this node is believed to work ok or not
  52         - The hash slots for which this node is responsible
  53         - If the node is a master or a slave
  54         - If it is a slave, the slave of which node
  55         - if it is a master, the list of slave nodes
  56         - The slaves are ordered for "<ip>:<port>" string from lower to higher
  57           ordered lexicographically. When a master is down, the cluster will
  58           try to elect the first slave in the list.
  59
  60     2) Every node also contains the unix time where every other node was
  61        reported to work properly (that is, it replied to a ping or any other
  62        protocol request correctly). For every node we also store the timestamp
  63        at which we sent the latest ping, so we can easily compute the current
  64        lag.
  65
  66     3) From time to time a node pings a random node, selected among the nodes
  67        with the least recent "alive" time stamp. Three random nodes are selected
  68        and the one with lower alive time stamp is pinged.
  69
  70     4) The ping packet contains also information about a few random nodes
  71        alive time stamp. So that the receiver of the ping will update the
  72        alive table if the received alive timestamp is more recent the
  73        one present in the node local table.
  74
  75        In the ping packet every node "gossip" information is something like
  76        this:
  77
  78        <ip>:<port>:<status>:<pingsent_timestamp>:<pongreceived_timestamp>
  79
  80        status is OK, POSSIBLE_FAILURE, FAILURE.
  81
  82     5) The node replies to ping with a pong packet, that also contains a random
  83        selections of nodes timestamps.
  84
  85 A given node thinks another node may be in a failure state once there is a
  86 ping timeout bigger than 30 seconds (configurable).
  87
  88 When a possible failure is detected the node performs the following action:
  89
  90     1) Is the average between all the other nodes big? For instance bigger
  91        than 30 seconds / 2 = 15 seconds? Probably *we* are disconnected.
  92        In such a case we don't trust our lag data, and reset all the
  93        timestamps of sent ping to zero. This way when we'll reconnect there
  94        is no risk that we'll claim many nodes are down, taking inappropriate
  95        actions.
  96
  97     2) Messages from nodes marked as failed are *always* ignored by the other
  98        nodes. A new node needs to be "introduced" by a good online node.
  99
 100     3) If we are well connected (that is, condition "1" is not true) and a
 101        node timeout is > 30 seconds, we mark the node as POSSIBLE_FAILURE
 102        (a flat in the cluster node structure). Every time we sent a ping
 103        to another node we inform this other nodes that we detected this
 104        condition, as already stated.
 105
 106     4) Once a node receives a POSSIBLE_FAILURE status for a node that is
 107        already marked as POSSIBLE_FAILURE locally, it sends a message
 108        to all the other nodes of type NODE_FAILURE_DETECTED, communicating the
 109        ip/port of the specified node.
 110
 111        All the nodes need to update the status of this node setting it into
 112        FAILURE.
 113
 114     5) If the computer in FAILURE state is a master node, what is needed is
 115        to perform a Slave Election.
 116
 117 SLAVE ELECTION
 118
 119     1) The slave election is performed by the first slave (with slaves ordered
 120        lexicographically). Actually it is the first functioning slave, so if
 121        the first slave is marked as failing the next slave will perform the
 122        election and so forth. Such a slave is called the "Successor".
 123
 124     2) The Successor starts checking that all the nodes in the cluster already
 125        marked the master in FAILURE state. If at least one node does not agree
 126        no action is performed.
 127
 128     3) If all the nodes agree that the master is failing, the Successor does
 129        the following:
 130
 131        a) It will send a SUCCESSION message to all the other nodes, that will
 132           upgrade the hash slot tables accordingly. It will make sure that all
 133           the nodes are updated and if some node did not received the message
 134           it will keep trying.
 135        b) Once all nodes not marked as FAILURE accepted the SUCCESSION message
 136           it will update his own table and will start acting as a master
 137           accepting write queries.
 138        c) Every node receiving the succession message, if not already informed
 139           of the change will broadcast the same message to other three random
 140           nodes. No action is performed if the specified host was already marked
 141           as the master node.
 142        d) A node that was a slave of the original master that failed will
 143           switch master to the new one once the SUCCESSION message is received.
 144
 145 RANDOM
 146
 147     1) When selecting a slave, the system will try to pick one with an IP different than the master and other slaves, if possible.
 148
 149     2) The PING packet also contains information about the local configuration checksum. This is the SHA1 of the current configuration, without the bits that normally change form one node to another (like latest ping reply, failure status of nodes, and so forth). From time to time the local config SHA1 is checked against the list of the other nodes, and if there is a mismatch between our configuration and the most common one that lasts for more than N seconds, the most common configuration is asked and retrieved from another node. The event is logged.
 150
 151     3) Every time a node updates its internal cluster configuration, it dumps such a config in the cluster.conf file. On startup the configuration is reloaded.
 152     Nodes can share the cluster configuration when needed (for instance if SHA1 does not match) using this exact same format.
 153
 154 CLIENTS
 155
 156     - Clients may be configured to use slaves to perform reads, when read-after-write consistency is not required.