Socket write failure crashes program

[cosmastech]

Hello all!

We intermittently get this failure on our production environments.

ErrorException: socket_sendto(): Unable to write to socket [111]: Connection refused in /var/www/vendor/datadog/php-datadogstatsd/src/DogStatsd.php:502

We are running on a Kubernetes cluster.

This is incredibly frustrating, as we have to wrap every call to write metrics in try-catches to avoid having DB transactions rolled back or returning 500 errors to users.

Was wondering if there's any reason that this would be happening, and if there should be a configuration option to allow gracefully handling these? (See: League's statsd implementation https://github.com/thephpleague/statsd/blob/master/src/Client.php#L341-L354)

[StephenWakely]

The first thing to check would be the health of the Datadog agent. Is the agent pod regularly restarting, and does this coincide with the connection refused errors you see?

You might try changing the transport to using UDP, specify a host and port instead of socket_path. That would rule out any issues with the volume.

Happy to take a PR to handle these errors. Let me know when it's ready for review.

[cosmastech]

The first thing to check would be the health of the Datadog agent. Is the agent pod regularly restarting, and does this coincide with the connection refused errors you see?

You might try changing the transport to using UDP, specify a host and port instead of socket_path. That would rule out any issues with the volume.

Happy to take a PR to handle these errors. Let me know when it's ready for review.

Thanks for the reply! We do indeed have some investigating to do on our end as to what's causing it. From a cursory glance, it does indeed seem to coincide with scale up/down, so we're thinking that the agent is evicted.

I have a PR for review. It's pretty ugly, given the PHP version required in composer.json.

[StephenWakely]

Fixed by #169 .