[connection] do not open the connection on initialization #214
Conversation
Co-authored-by: Olivier Vielpeau <olivielpeau@users.noreply.github.com>
| logger.debug { "Statsd: #{payload}" } if logger | ||
|
|
There was a problem hiding this comment.
⬇️ GitHub doesn't let me comment on line 34, but it appears to me that if we always try to connect before send_message, then the Connection class no longer even needs to know about #connect: it effectively becomes a private implementation detail of the subclasses.
The remaining contract we have is that we can call close and then expect a subsequent send_message to still work.
There was a problem hiding this comment.
You are right, but, I like the clarity of https://github.com/DataDog/dogstatsd-ruby/pull/214/files#diff-01cb409d6d5a1fb2aebe52e78e877a91e3f6654758931899427a43656211a073R32-R35
You can explicitly read in the retry mechanism that it is closing the connection, re-opening it, and retrying.
I know the connect isn't needed since send_message will do it.
Not a strong opinion though, I can remove the connect here.
WDYT?
There was a problem hiding this comment.
I don't quite like having the mix of implicit (in send_message) and explicit (in the failure handler) but I can live with it :)
Co-authored-by: Ivo Anjo <ivo.anjo@datadoghq.com>
…sd-ruby into remeh/connection-on-send
A recent change clarifying the connection code introduced in v5.3.0 has changed when the socket are opened for both UDP and UDS. Because of that, resolution of an unreachable domain or IP or an non-existing UDS socket made the instantiation of the client throw an error in v5.3.0 since it was happening in the '#initialize' methods, where these errors are not caught.
This PR changes when the socket are opened (on write) where the connection errors are caught by the retry mechanism like it was before v5.3.0.
Should address #213