A recovery action should restore the business guarantee, not just remove the visible error. If acks=all with min.insync.replicas=2 rejects writes while only one replica is in sync, the rejection is protecting the two-copy requirement. Lowering the minimum changes that requirement. Adding retries cannot create a second copy while the replica stays unavailable. These tasks ask for a recommendation based on the observed ISR and the stated durability boundary. Restore an eligible replica and verify catch-up before retrying the workload. The local controls make those states immediate; a real incident needs progress monitoring, communication and a measured recovery check. Record the decision in terms that explain what remains protected.
Inspect the topics, partition logs and group offsets before changing the cluster.