[LRUG] Puma, CLOSE_WAIT. Arg.

[Jason Lee]

A long time ago I was using heroku, MRI, puma, and ActionController::Live.
Exceptions would cause database connections to leak, this was a problem for
me because the heroku free tier only gives you 20 database connections and
I was using a db connection on every streaming request in order to use
postresql's listen/notify feature ( instead of using, say, redis pub/sub).

This code https://gist.github.com/jlsync/ea499fa90a9abfe04fa9  manages
database connections by using

  ActiveRecord::Base.connection_pool.checkout

to get a connection and then a ruby "ensure" block at the end to "checkin"
the connection back to the pool.

The code also uses

  Thread.list.count

to monitor the active ruby/puma thread count,  and also

  ActiveRecord::Base.connection_pool.size
  ActiveRecord::Base.connection_pool.connections.size

to monitor database connections.

You can see these monitoring counts being logged directly to the UI here
https://jlquiz.herokuapp.com/ (after logging in).

my puma config included

  on_worker_boot do
    # worker specific setup
    ActiveSupport.on_load(:active_record) do
      ActiveRecord::Base.establish_connection
    end
  end


Using some of these ideas might help you.

Jason.

On Thu, Feb 18, 2016 at 4:01 PM, Simon Morley <simon at polkaspots.com> wrote:

> As an interim update, adding the recommended
> ActiveRecord::Base.connection_pool.disconnect! doesn't seem to help :(
>
> We're seeing far fewer timeouts but they're still happening on that db
> connection. And when they happen, it's carnage for 15s.
>
>
>
> Simon Morley
>
> Big Chief | PolkaSpots Supafly Wi-Fi
> Bigger Chief | Cucumber Tony
>
> Got an unlicensed Meraki? Set it free with Cucumber
> cucumberwifi.io/meraki
>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.lrug.org/pipermail/chat-lrug.org/attachments/20160218/b9ff82f9/attachment-0002.html>