A Thrift client wrapper that encapsulates some common failover behavior.
Transparent connection management
Configurable failover and retry backoff
Ruby 1.9 compatibility
ThriftClient::Simple class, for working without generated bindings.
The Github source repository is here. Patches and contributions are very welcome.
Instantiate a client:
client = ThriftClient.new(CassandraRb::Client, '127.0.0.1:9160', :retries => 2)
You can then make calls to the server via the client instance as if was your internal Thrift client. The connection will be opened lazily and methods will be proxied through.
On failures, the client will try the remaining servers in the list before giving up. See ThriftClient for more.
Timeouts are enforced per-try, so if you have a timeout of n and do m retries, the total time it could take is n*m.
The library will shuffle the host list then work its way down this list, only moving to the next host if it received an error or you've doing more than server_max_requests requests with that host (defaults to 0 which means there's no limit).
Servers that throw an error get marked as dead and will only be retried every server_retry_period seconds (at that time all dead servers are retried, no matter long they've been marked as dead).
You need Ruby 1.8 or 1.9. If you have those, just run:
sudo gem install thrift_client
To contribute changes:
Fork the project
make your change, adding tests
send a pull request to twitter and ryanking
The Github issue tracker is here.
Copyright 2009-2012 Twitter, Inc. See included LICENSE file.
The public certificate for this gem is here.