Generic connection pooling for Ruby.

MongoDB has its own connection pool. ActiveRecord has its own connection pool. This is a generic connection pool that can be used with anything, e.g. Redis, Dalli and other Ruby network clients.

Create a pool of objects to share amongst the fibers or threads in your Ruby application:

$memcached = ConnectionPool.new(size: 5, timeout: 5) { Dalli::Client.new }

Then use the pool in your application:

$memcached.with do |conn|
  conn.get('some-count')
end

If all the objects in the connection pool are in use, with will block until one becomes available. If no object is available within :timeout seconds, with will raise a ConnectionPool::TimeoutError (a subclass of Timeout::Error).

You can also use ConnectionPool#then to support both a connection pool and a raw client.

# Compatible with a raw Redis::Client, and ConnectionPool Redis
$redis.then { |r| r.set 'foo' 'bar' }

Optionally, you can specify a timeout override using the with-block semantics:

$memcached.with(timeout: 2.0) do |conn|
  conn.get('some-count')
end

This will only modify the resource-get timeout for this particular invocation. This is useful if you want to fail-fast on certain non-critical sections when a resource is not available, or conversely if you are comfortable blocking longer on a particular resource. This is not implemented in the ConnectionPool::Wrapper class.

You can use ConnectionPool::Wrapper to wrap a single global connection, making it easier to migrate existing connection code over time:

$redis = ConnectionPool::Wrapper.new(size: 5, timeout: 3) { Redis.new }
$redis.sadd('foo', 1)
$redis.smembers('foo')

The wrapper uses method_missing to checkout a connection, run the requested method and then immediately check the connection back into the pool. It's not high-performance so you'll want to port your performance sensitive code to use with as soon as possible.

$redis.with do |conn|
  conn.sadd('foo', 1)
  conn.smembers('foo')
end

Once you've ported your entire system to use with, you can simply remove Wrapper and use the simpler and faster ConnectionPool.

You can shut down a ConnectionPool instance once it should no longer be used. Further checkout attempts will immediately raise an error but existing checkouts will work.

cp = ConnectionPool.new { Redis.new }
cp.shutdown { |c| c.close }

Shutting down a connection pool will block until all connections are checked in and closed. Note that shutting down is completely optional; Ruby's garbage collector will reclaim unreferenced pools under normal circumstances.

You can reload a ConnectionPool instance in the case it is desired to close all connections to the pool and, unlike shutdown, afterwards recreate connections so the pool may continue to be used. Reloading may be useful after forking the process.

cp = ConnectionPool.new { Redis.new }
cp.reload { |conn| conn.quit }
cp.with { |conn| conn.get('some-count') }

Like shutdown, this will block until all connections are checked in and closed.

You can reap idle connections in the ConnectionPool instance to close connections that were created but have not been used for a certain amount of time. This can be useful to run periodically in a separate thread especially if keeping the connection open is resource intensive.

You can specify how many seconds the connections have to be idle for them to be reaped. Defaults to 60 seconds.

cp = ConnectionPool.new { Redis.new }
cp.reap(300) { |conn| conn.close } # Reaps connections that have been idle for 300 seconds (5 minutes).

You can start your own reaper thread to reap idle connections in the ConnectionPool instance on a regular interval.

cp = ConnectionPool.new { Redis.new }

# Start a reaper thread to reap connections that have been idle for 300 seconds (5 minutes).
Thread.new do
  loop do
    cp.reap(300) { |conn| conn.close }
    sleep 300
  end
end

There are several methods that return information about a pool.

cp = ConnectionPool.new(size: 10) { Redis.new }
cp.size # => 10
cp.available # => 10
cp.idle # => 0

cp.with do |conn|
  cp.size # => 10
  cp.available # => 9
  cp.idle # => 0
end

cp.idle # => 1

• Connections are lazily created as needed.
• There is no provision for repairing or checking the health of a connection; connections should be self-repairing. This is true of the Dalli and Redis clients.
• WARNING: Don't ever use Timeout.timeout in your Ruby code or you will see occasional silent corruption and mysterious errors. The Timeout API is unsafe and cannot be used correctly, ever. Use proper socket timeout options as exposed by Net::HTTP, Redis, Dalli, etc.

Mike Perham, @getajobmike, https://www.mikeperham.com