Frequent thread creation failures on aarch64 CI
The aarch64 CI runners for Calls have been subject to resource exhaustion issues since April of this year, with many errors like the following (job link):
libgomp: Thread creation failed: Resource temporarily unavailable
There is a summary of my investigation HERE, but the short version of it is:
- @julianfairfax attempted to use a known-good Flatpak manifest, but we continued to see errors
- I discovered that modifying the flatpak-builder CI template to not spawn additional threads could resolve the issue
- @alatiera let me know that my workaround is not acceptable, and that there are underlying infrastructure issues that should be fixed first.
@alatiera's diagnosis is that:
the real issue is that there are leftover zombie processes that are eating up the resources on the runner side.
Could somebody please investigate and fix the underlying failures?
Edited by Anton Lazarev