Dependencies CI on main is intermittently failing due to jobs being cancelled (not clear why)

[alamb]

Describe the bug

I have noticed recently that often CI checks fail on main (not PRs). For example, looking at: https://github.com/apache/datafusion/commits/main/

If you look at what is failing it is that these jobs are "cancelled"

To Reproduce

I am not sure why they don't pass or fail all the time. Maybe it is related to if we have merged another PR within a short time?

Here is an example failure:
https://github.com/apache/datafusion/actions/runs/25127768271/job/73645414258

Disk space before cleanup:
Filesystem      Size  Used Avail Use% Mounted on
overlay         145G   59G   86G  41% /
tmpfs            64M     0   64M   0% /dev
shm              64M     0   64M   0% /dev/shm
/dev/root       145G   59G   86G  41% /__t
tmpfs           3.2G  1.1M  3.2G   1% /run/docker.sock
tmpfs           7.9G     0  7.9G   0% /proc/acpi
tmpfs           7.9G     0  7.9G   0% /proc/scsi
tmpfs           7.9G     0  7.9G   0% /sys/firmware
context canceled
Error: The operation was canceled.

It seems to only affect "Dependencies" jobs

Expected behavior

Additional context

No response

[alamb]

@blaginin does this look familiar to you?

[blaginin]

feels like an easy fix, will do today

[alamb]

FWIW codex made this

• [codex] Avoid duplicate Dev and Security audit workflow runs #21939
  But I am not sure if it is the right fix (though the explanation sounds plausible)

[blaginin]

I believe the following is happening. When PR is in the Queue, GitHub spawns ALL the actions for that PR with the merge_group trigger. This includes circular dependency check:

But github doesn't wait for ALL of actions to finish - only for the required ones. The rest get cancelled. There are two ways to handle this:

• make the check required (so it'll be awaited)
• do not run the check at all