Best practices

Use randomly generated cluster names to avoid clashes between instances meant to belong to different clusters, especially when running tests concurrently.

Use randomly assigned ports for the member endpoints, in case a network stack is spawned.

Always clean up resources after each test using factory.shutdownAll() or Hazelcast.shutdownAll() in a teardown hook.

Use mock networking for tests by default: it avoids the need for real sockets and speeds up execution. In case a real network is required use -Dhazelcast.test.use.network=true.

Reserve real networking setups for network integration or system tests only.

Keep test clusters small (2–3 nodes) to validate partitioning, replication, and quorum logic efficiently.

Assert cluster size explicitly using assertClusterSizeEventually(…​) before testing distributed operations relying on fully formed clusters.

Assert nodes state explicitly with waitAllForSafeState(…​) to make sure that all nodes are ready to operate (partition state is SAFE)

Use a small number of partitions (for example 11) in your cluster configuration to expedite the cluster creation and test execution.

Use @Repeat to detect flaky tests that might otherwise pass unreliably.

Enable thread dump logging on failure by setting hazelcast.test.threadDumpOnFailure=true.

hazelcast.test.use.network=false: This is the default value to use mock networking. Set to true to use real networking.

hazelcast.test.threadDumpOnFailure=true: Print thread dumps when a test fails to aid debugging.

hazelcast.phone.home.enabled=false: Disable phone home feature for clean test output.

hazelcast.wait.seconds.before.join=1: Reduce wait time before a member attempts to join the cluster.

hazelcast.local.localAddress=127.0.0.1: Bind test members to the loopback address to avoid conflicts.

java.net.preferIPv4Stack=true: Ensure consistent networking on dual-stack systems.

hazelcast.logging.type=log4j2: Configures logging with the specified log library (supported: jdk, log4j, log4j2, `none)

hazelcast.logging.details.enabled=true: Logs name, IP address and version of the cluster

Test execution durations are printed automatically when using HazelcastParallelClassRunner or HazelcastSerialClassRunner.

When using HazelcastParallelClassRunner and tests fail with an error in initialization or instances aren’t shut down, some (other) tests may not properly shut down the clients/instances created. Make sure your tests do so.