Introduction

Monitors partition-level changes on a Kafka topic during an experiment. This check observes whether specific partition events occur — such as a leader election, replicas going offline, or in-sync replicas changing — and evaluates them against an expected change list.

This action targets topic partitions, not brokers or consumer groups. For broker-level monitoring (controller elections, downtime), use Check Brokers. For consumer group state monitoring (Stable, Dead, Rebalancing), use Check Consumer State.

Important: This action uses changeCheckMode and expectedChanges parameters — not stateCheckMode or expectedStateList (those belong to Check Consumer State).

Use Cases

• Verify that a leader election occurs on a topic's partition when the current leader broker is killed.
• Assert that in-sync replicas change when a broker hosting a replica is network-isolated.
• Confirm that no unexpected partition changes happen during a consumer-level fault injection.

Parameters

Parameter	Description	Default
Duration	How long the check runs. The check continuously polls partition metadata for this duration.	10s
Expected Changes	Which partition-level changes to watch for. Valid values: New Leader Elected (leader changed), Replicas Changed (replicas changed), Offline Replicas Changed (offline replicas changed), In-Sync Replicas Changed (in sync replicas changed). Multiple values can be selected. If left empty, any partition change triggers.
Change Check Mode	How the expected changes are evaluated. All the time (allTheTime): every poll must detect the change. At least once (atLeastOnce): the change must be observed at least once during the duration.	allTheTime