Repository navigation
v2.5.0
Changes
- Recover active mocks from surviving WireMock pods after all embedded Hazelcast members restart. Recovered mocks retain their pod identity and runtime version; last-access timestamps restart at recovery time. Replacement API pods joining a surviving cluster preserve its existing state. (#143)
- Default compatible API upgrades to RollingUpdate with zero unavailable replicas and one surge replica, graceful HTTP draining, and preferred node separation. (#143)
- Preserve the dashboard mock-ID filter after adding or selecting a mock. (#146, fixes #144)
- Report the application release version in OpenAPI and Swagger UI instead of a hard-coded version. (#147, fixes #145)
Upgrade notes
- Use
fleet.api.updateStrategy.type=Recreatefor incompatible Hazelcast or shared-state upgrades. Rolling updates require spare capacity for the surge replica. - RollingUpdate keeps replicas available but does not guarantee uninterrupted requests: local tests observed mock-list requests exceeding five seconds during Hazelcast membership changes. Full API restarts interrupt API access until recovery completes.
Full changelog: v2.4.3...v2.5.0