[Discussion] Upgrading chart fails if requested resources exceed ~50%

[zelima]

It seems If the requested resources (CPU or Memory) exceed half of the allocatable resources we need to manually delete chart via helm and redeploy it, otherwise, deployment will fail with unschedulable error.

In our case for specstore 2.8 CPU is requested out of 3.92 total. If we make a change in specstore and let the auto-redeploy do its job it will fail. I left a not in README to do it manually instead.

Solution/Problem

It seems like the solution here would be to just decrease requested CPU for specstore <50% and let auto-redeploy do its job, but problem is that k8s is allocating all services on the biggest node in that case, leaving specstore with lack of CPU.

[zelima]

Manual deploy does not really work and it's time-consuming. I did the following:

• pushed all of them and let k8s allocate (deployed all of them on 1 node)
• purged specstore release and pushed with max available CPU and memory in the same node
  • helm --purge delete release-name
  • bash helm_upgrade_external_chart.sh specstore
• one by one re-deployed other services --> they moved to another node as no place was available on that node
• purged specstore again and now pushed with resources 50% of the total.

This way none of the resources need manual redeploy. (Hope other resources will not move back when redeployed)

[zelima]

closing as FIXED for now.