feat(instanceset): allow updating crashed/unavailable pods while pres…

[ljluestc]

Title

feat(instanceset): allow updating crashed/unavailable pods while preserving rolling guarantees

Description

• Problem: ITS only updated Ready/Available pods, blocking recovery for crashed pods (e.g., OOMKilled) when users wanted to increase resources.
• Solution: Relax per-pod gating so NotReady/Unavailable pods can be updated. Preserve rollout safety by consuming maxUnavailable only when taking a currently-Available pod offline; already-unavailable pods don’t consume it.

Changes

• pkg/controller/instanceset/reconciler_update.go
  • Don’t block updates on NotReady/Unavailable/role-not-ready; keep image-mismatch guard.
  • Track unavailability consumption per pod; only Available→update consumes maxUnavailable.
• pkg/controller/instanceset/reconciler_update_test.go
  • Add test: updates a crashed/unavailable pod with outdated revision.

Risks and mitigations

• More permissive updates when degraded; still bounded by:
  • Image consistency checks.
  • Rolling quotas (replicas, maxUnavailable).
  • Member update rules for roleful sets.

How to verify

# From repo root
go test ./pkg/controller/instanceset -count=1 -v

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}

[apecloud-bot]

Auto Cherry-pick Instructions

Usage:
  - /nopick: Not auto cherry-pick when PR merged.
  - /pick: release-x.x [release-x.x]: Auto cherry-pick to the specified branch when PR merged.

Example:
  - /nopick
  - /pick release-1.0