From 545dda7c283cac7a5b3da70bbd56a9acbac22be4 Mon Sep 17 00:00:00 2001 From: Joris Wouter Jonkers Date: Wed, 26 Aug 2026 15:29:02 +0200 Subject: [PATCH] fix(deploy): point agent-gateway's liveness probe at the liveness group agent-gateway declares only health.path, so the probe rendering introduced in schema 0.21 falls back to the readiness path for livenessProbe as well. The two are different actuator groups: readiness reports readinessState, liveness reports livenessState and ping. Sharing one path means a failing dependency fails liveness too, and the kubelet restarts the pod instead of removing it from endpoints - turning a transient outage into a restart loop. The endpoint is configured, not assumed: services/agent-gateway/src/main/resources/application.yml sets management.endpoint.health.probes.enabled: true, management.endpoint.health.group.liveness.include: livenessState,ping and management.endpoints.web.base-path: /api/actuator, on server.port 8090 which is the port the manifest already probes. agents-login is unchanged. It is a Fastify service exposing a single /healthz with no separate groups, so one path for both probes is correct there. Verified by rendering against schema 0.22.0 and the published context: agent-gateway readiness=/api/actuator/health/readiness, liveness=/api/actuator/health/liveness; agents-login /healthz for both. --- platform/deployment.yml | 1 + 1 file changed, 1 insertion(+) diff --git a/platform/deployment.yml b/platform/deployment.yml index 387eb74..7977dcb 100644 --- a/platform/deployment.yml +++ b/platform/deployment.yml @@ -12,6 +12,7 @@ spec: image: agent-gateway health: path: /api/actuator/health/readiness + livenessPath: /api/actuator/health/liveness port: 8090 timeoutClass: stateless mandatory: true