Finally, after 2 years (!) of adopting pantsbuild and parallelizing our test suite, I resolved intermittent test hangs completely. See the story here: https://fosstodon.org/@achimnol/112738442561779050 TLDR: There is an internal race condition in the Docker to reassign TCP ports to different containers. I had to write up a filelock-based monotonic counter to assign TCP ports for all containers used in the test suite.

Thank you for the write up! Do you happen to recall what the error looked like before? I have a fair amount of Docker falkyness and it is hard to keep which flake is which straight.

The tests hung randomly, in a high probability.