I have one question regarding Gitlab pods scheduling. We are using kops with 2 instance groups. This instance groups are scaling up dynamically using cluster autoscaler. We've set labels for this 2 instance groups. I have 1 question:

1) How to schedule gitlab runner "helper" pods (runner-concurrent) on different nodes and to not schedule this pods on the same worker. Because for now we have situation when we are running 2 jobs in parallel , CA spin ups 2 worker nodes , but this 2 jobs are scheduled on the same node. I've set pod limits to 3 (2 system pods and 1 space for gitlab-runner job , but in this case 1 job will fail sometime due to lack of resource (pod limits). But sometimes It will schedule properly pods on different nodes). I know about nodeAffinity etc , but Gitlab currently does not support this. Is there any possibility to avoid affinity in Gitlab scope?

If It's not clear from my side: when pods runner-gitlab-concurrent is already scheduled on 1 of the worker nodes , I want to schedule another pod to another worker node. In my case I see that when I am running 2 jobs or more in parallel , they see that worker nodes are free and can be scheduled on the same node. But other 2 nodes becomes free.