grafana / rollout-operator

Kubernetes Rollout Operator
Apache License 2.0
132 stars 17 forks source link

Do not exit reconciliation if scaling errors #90

Closed jhalterman closed 11 months ago

jhalterman commented 11 months ago

Following #82, downscaling attempts are prevented in a zone if some other zone has non-ready pods. One situation where this could happen is if zone-C is rolling around the same time that zone-A starts experiencing evictions. In this case, zone C may also want to downscale some nodes, but is unable to because zone A has non-ready nodes, and since reconciliation exits when downscaling fails, zone C can't finish rolling its pods either.

This change allows reconciliation to continue even if scaling fails, so that updates may proceed.