Skip to content

batches: use Docker CPU count as default parallelism, not GOMAXPROCS

Warren Gifford requested to merge aharvey/cpu-count into main

Created by: LawnGnome

This PR changes the default parallelism of src batch preview|apply runs to be based on the CPU cores reported by docker info.

Practically, this means that Docker Desktop users will now have their parallelism based on the VM Docker runs within, not their host machine. The trade off here is that Docker Desktop users will likely have their default parallelism cut significantly, but this is also likely to make issues caused by the Docker VM having its resources exhausted considerably less likely.

Implementation notes

  • I've migrated the previous docker version preflight to use docker info instead, which means we can get the CPU count and preflight Docker at the same time.
  • Since docker info should always be quick, this reuses the same timeout behaviour as #783, which has been renamed from being image inspect specific to being used for any "fast commands" issued to Docker. The behaviour is otherwise unchanged.
  • In testing this, I found an issue where os/exec is inconsistent in terms of the error returned when a deadline is exceeded; I've updated both places we use the fast command context to look both at the returned error and the context error.

Screenshot of the timeout

image

Test plan

Added unit tests, and tested this in various scenarios locally (Docker not running; Docker running but unresponsive; Docker running normally).

Merge request reports