Communication between non-primary containers becomes unavailable



We’re using 3 containers, which are [a] primary container (Ubuntu), [b] MySQL and [c] private container of related private app, and [c] is used to initialize MySQL database schema/data by running rake db:migrate or similar commands. Of course, it requires connection between [c] and [b]. After initialization, tests are executed on primary container.

Two week ago (when we started to use this configuration), this configuration works well.
But 9 hours ago, the connection between [b] and [c] seems to become unavailable.

Is this a kind of bug? Or results of intentional CircleCI update?

This is strongly required for us to use different container images from testing repository to test on schema/data defined by different repository.
It is impossible as documented to use remote docker environment, so that the current configuration is only a hack to do it for us.


It should be fine, I think. CircleCI merges the networking stacks of all primary and secondary containers to localhost for ease of use. I would start off by checking you’re using the right address/port in [c] when connecting to [b].

However, I imagine most people’s use-case is to call secondary containers only from the primary [a], so it is possible you’ve bumped into a bug.

Can you expand on why that is impossible? I use a single primary container, pull ~12 images from a private registry, and then start them on Docker Compose. It works flawlessly. Could you do the same?