Handle ENOTCONN

[blaines]

Unset socket under the condition ENOTCONN this could happen in the event the datadog agent is restarted and using a unix socket file.

• Unset socket in the condition a connection is lost
• Added test to handle ENOTCONN

[Shadow6363]

It might actually be better to not add this and instead relax the retry conditions on this line

dogstatsd-ruby/lib/datadog/statsd.rb

Line 64 in 570f3c0

if retries <= 1 && boom.is_a?(IOError) && boom.message =~ /closed stream/i

to just

        if retries <= 1

In our case as well as I'm sure in many others, we'd much rather immediately retry establishing the socket and resending the message rather than losing the message and having to wait for the next one for the socket to re-establish itself.

[blaines]

Good idea, I changed the logic to retry the same message upon ENOTCONN

[gmmeyer]

this looks good to me! thanks a lot for doing this! 😄

[Shadow6363]

Hey @gmmeyer, thanks for merging this! Is there a standard release schedule you follow? Just wondering when I should update our agents to pull this in.

[Shadow6363]

Also, just realized @blaines, you never removed the original version of your code. There's a return block after that if so your change to retry isn't actually ever going to get reached. :/