The checkpointer thread runs on both primary and standby nodes, so AdaptiveCheckpointPolicy that was added in SERVER-120876 (#5dc8bbc) is currently active on all node types and its metrics (e.g. oplogBytesSinceLastCheckpoint) are emitted on all of them. The metric tracks closely on primary and standby since both process the same oplog stream.

Grafana

Checkpoints on standbys are largely a no-op for what the adaptive policy is optimizing, and there's a question of whether the policy could delay other work the MongoDB checkpoint thread performs on those nodes.

Ask: explore whether it makes sense to scope the adaptive policy to the primary (and nodes transitioning to secondary), with standbys staying on FixedIntervalPolicy (5s) — or whether running it everywhere is fine/preferable.