Hello, I suspect that since this PR in introduction connection pooling to the metadata service, we’ve been getting

ValueError: Metaflow service [<http://metaflow/metadata/service.com>] unreachable.

Do you guys know why this is happening? I’m guessing it’s because the service is closing the connection for some reason? Are there any fixes that we can do?

does the error go away if you use an older version of metaflow that doesn't have that change?

yes, also it goes away if I change the

max_retries=0

to

max_retries=2

. I just randomly tested with something > 1.

it might be useful to add explicit retries here. any interest in contributing a PR?

Sure I can create a PR

because we have custom retry behavior

what’s the best way to create a PR? I’m getting an error:

ERROR: Permission to Netflix/metaflow.git denied to tlee14-upstart.

You can create from your fork

here is the PR: https://github.com/Netflix/metaflow/pull/2433