[libcxx-commits] [libcxx] [llvm] [CI][Github] Add linux premerge workflow (PR #119635)

[Nathan Gauër via libcxx-commits]

Keenuts wrote:

> > This just adds the workflow in for testing. There is at least one kink that needs to be worked out on the infrastructure side, namely that sometimes scaling down kills the runner/container pod.
> 
> There is no way to wait for the current job to finish? I implemented a downscaling of a Kubernetes cluster for running builbot workers where the downscale wouldn't just kill the pod but send a signal to terminate. That would instruct the buildbot master to not schedule new job on the worker running on this pod and the worker would exit at the end of the current job. The pod would then be garbage collected.

In our case we use GCP autoscale for scale up/down. The fact that a pod can be killed is surprising as the scale down trigger should be 10mn with almost 0 CPU activity on the node (some instance services are always running).

https://github.com/llvm/llvm-project/pull/119635