terragrunt runner pool feature breaks run --all workflows #5998
Description
Terragrunt's new runner pool feature, which replaces the previous group-based scheduling, breaks existing workflows that rely on run --all, and may impact other workflows as well. Runner pool was made generally available in v0.89.0.
Why it breaks
When discovering units to run, terragrunt ignores hidden directories, not only under the working directory but also above it. Atlantis jobs happen to run inside ~/.atlantis, which triggers the issue.
I attempted to work around the limitation using TG_QUEUE_INCLUDE_DIR, but it didn’t behave consistently. In the end, the only reliable fix was to change the Atlantis data directory from ~/.atlantis to a non-hidden one (in our case ~/atlantis-data).
Not sure if it should be considered a bug in Atlantis, but definitely something that Atlantis users and developers should be aware of.