问题
Since clustered Redis is still in the works, are there mechanisms in Resque that automatically will failover to a Redis slave should the master ever go down?
回答1:
I don't think so. However, you can implement the master election mechanism yourself quite easily using Apache Zookeeper:
require "rubygems"
require "zookeeper"
def log(msg)
puts "[#{Process.pid}] #{msg}"
end
def debug(obj)
log(obj.inspect)
end
def on_master_changed(&block)
loop do
wcb = Zookeeper::WatcherCallback.new
resp = @zookeeper.get_children(:path => @base_path, :watcher => wcb, :watcher_context => @base_path)
children = resp[:children].map{|name| "#{@base_path}/#{name}"}
new_master = children.sort.first
block.call(new_master)
while !wcb.completed?
sleep(0.1)
end
end
end
@zookeeper = Zookeeper.new("localhost:2181")
if @zookeeper.state != Zookeeper::ZOO_CONNECTED_STATE
log 'Unable to connect to Zookeeper!'
exit(1)
end
@base_path = "/nodes"
@zookeeper.create(:path => @base_path)
resp = @zookeeper.create(:path => "#{@base_path}/node-", :ephemeral => true, :sequence => true, :data => Process.pid.to_s)
my_node = resp[:path]
is_master = false
log "My node is: #{my_node}"
on_master_changed do |new_master|
if new_master == my_node
if is_master
log "I am still the master. Bow before me or die!"
else
log "I am the new master. Behold!"
end
is_master = true
else
pid = @zookeeper.get(:path => new_master)[:data]
log "New master is process #{pid}"
end
end
You could modify the script above to:
- Use IP/port of the redis server instead of PID of the process
- Use redis-cli along with the SLAVEOF command to handle "became master", "master changed" and "no longer master" scenarions.
回答2:
I've released a redis_failover gem that provides Redis failover for Ruby with ZooKeeper:
https://github.com/ryanlecompte/redis_failover
来源:https://stackoverflow.com/questions/7759572/solutions-for-resque-failover-redis