pip install redis-trib
easy_install redis-trib
WARNING: The following console commands or APIs not support simultaneous operations on one cluster.
Start a cluster in a single redis node (the node should have cluster enabled, and not in a cluster)
redis-trib.py start NODE_HOST:PORT
Start a cluster in some redis nodes (the nodes should have cluster enabled, and none of them in any cluster)
redis-trib.py start_multi NODE_HOST_a:PORT_a NODE_HOST_b:PORT_b ...
Add another master node to a cluster
redis-trib.py join CLUSTER_HOST:PORT NEW_NODE_HOST:PORT
The above APIs balance slots automatically and not configurable.
Add another node to a cluster, but neither set as slave nor migrating slots to it
redis-trib.py join_no_load CLUSTER_HOST:PORT NEW_NODE_HOST:PORT
Add a slave node to a master (slave should not in any cluster)
redis-trib.py replicate MASTER_HOST:PORT SLAVE_HOST:PORT
Remove a node from its cluster
redis-trib.py quit NODE_HOST:PORT
Shutdown an empty cluster (there is only one node left and no keys in the node)
redis-trib.py shutdown NODE_HOST:PORT
Fix a migrating slot in a node
redis-trib.py fix HOST_HOST:PORT
Migrate slots (require source node holding all the migrating slots, and the two nodes are in the same cluster)
redis-trib.py migrate_slots SRC_HOST:PORT DST_HOST:PORT SLOT SLOT_BEGIN-SLOT_END
each of "slot" argument tuple could be an integer (indicating a single slot number) or a range (begin and end, both inclusive). For example
redis-trib.py migrate_slots 127.0.0.1:7000 127.0.0.1:7001 0 2 4-7
means migrate slot #0 #2 #4 #5 #6 #7 from 127.0.0.1:7000 to 127.0.0.1:7001.
Rescue a failed cluster, specify host, port of one node in the cluster, and a free node
redis-trib.py rescue 127.0.0.1:7000 127.0.0.1:8000
The program would check which slots are failed in the cluster which contains 127.0.0.1:7000, and add them to 127.0.0.1:8000.
import redistrib.command
# start cluster at node 127.0.0.1:7000
redistrib.command.start_cluster('127.0.0.1', 7000)
# start cluster on multiple nodes, all the slots will be shared among them
# the argument is a list of (HOST, PORT) tuples
# for example, the following call will start a cluster on 127.0.0.1:7000 and 127.0.0.1:7001
redistrib.command.start_cluster_on_multi([('127.0.0.1', 7000), ('127.0.0.1', 7001)])
# add node 127.0.0.1:7001 to the cluster as a master
redistrib.command.join_cluster('127.0.0.1', 7000, '127.0.0.1', 7001)
# add node 127.0.0.1:7002 to the cluster as a slave to 127.0.0.1:7000
redistrib.command.replicate('127.0.0.1', 7000, '127.0.0.1', 7002)
# just add node 127.0.0.1:7001 to the cluster, not specifying its role
# could call migrate_slot(s) on it later, so that it becomes a master
redistrib.command.join_no_load('127.0.0.1', 7000, '127.0.0.1', 7001)
# remove node 127.0.0.7000 from the cluster
redistrib.command.quit_cluster('127.0.0.1', 7000)
# shut down the cluster
redistrib.command.shutdown_cluster('127.0.0.1', 7001)
# fix a migrating slot in a node
redistrib.command.fix_migrating('127.0.0.1', 7001)
# migrate slots; require source node holding the slots
# migrate slots #1, #2, #3 from 127.0.0.1:7001 to 127.0.0.1:7002
redistrib.command.migrate_slots('127.0.0.1', 7001, '127.0.0.1', 7002, [1, 2, 3])
# rescue a failed cluster
# 127.0.0.1:7000 is one of the nodes that is still alive in the cluster
# and 127.0.0.1:8000 is the node that would take care of all failed slots
redistrib.command.rescue_cluster('127.0.0.1', 7000, '127.0.0.1', 8000)
See also https://github.com/antirez/redis/blob/3.0/src/redis-trib.rb
The join_cluster function takes 2 optional arguments balancer and balance_plan. The former is an object for calculating the weights of cluster nodes, and the latter is a function that calculates how the slots migrate to balance the load between nodes.
As crude examples, you could refer to redistrib.clusternode.BaseBalancer and redistrib.clusternode.base_balance_plan. An instance of BaseBalancer should implement weight method that returns the weight of a specified node, and a function like base_balance_plan should return a list of migration tuples (source node, destination node, slots count).
import redistrib.command
# list all cluster nodes (attributes of which shown in the next section)
# args
# - host: host of specified node
# - port: port of specified node
# - default_host: default host string if the specified node doesn't know its host
# returns
# - nodes: all cluster nodes
# - myself: the specified node itself, also contained by nodes
nodes, myself = redistrib.command.list_nodes('127.0.0.1', 7000, default_host='127.0.0.1')
# list all master nodes
# args same as list_nodes
# returns
# - nodes: all master nodes
# - myself: the specified node itself, contained by nodes if it's a master; won't be None even if it's a slave
nodes, myself = redistrib.command.list_masters('127.0.0.1', 7000, default_host='127.0.0.1')
redistrib.cluster.ClusterNode: cluster node, attributes:
node_id: node idhost: known host, this value could be empty string if the node is newly launchedport: listening portrole_in_cluster:"master"or"slave"master_id: master'snode_idif it's a slaveassigned_slots: a list of assigned slots if it's a master; it won't contain slots being migratedslots_migrating: boolean value for whether there are any slot(s) migrating or importing on this node