Skip to content
Use the agentsJava service

Deploy a Java service

How-to
On this page
  1. Supported topologies
  2. What the agent handles for you
  3. Conventions worth following
  4. When something fails

The Java agent deploys Spring Boot and Quarkus applications (Micronaut in the generative path) and is currently the platform’s most heavily verified agent: its topologies are the ones exercised by the published reliability campaigns on proof.hivedeploy.in.

Supported topologies

CloudTopologies
GCPCloud Run (single + multi-service) · GKE (standard, hardened, multi-zone HA) · 3-VM Compute Engine (api + workers) · Managed Instance Group + global HTTPS LB
AWSFargate HA · EKS · Auto Scaling Group + ALB · 3-VM EC2 · Lambda SnapStart

Ask for anything else and the generative path plans it — behind the same gates and health verification. Verified generative successes are promoted into the deterministic template library (how that works).

What the agent handles for you

  • Build: your source (repo or zip) is built inside your cloud — Cloud Build on GCP, CodeBuild on AWS — and images are pinned by digest, so a deploy can never pick up a stale image.
  • JVM sizing: instance/machine types are chosen for JVM memory behavior, and the runtime gets JAVA_TOOL_OPTIONS heap bounds plus a restart policy. Undersized asks get bumped rather than deployed as crash-loops.
  • Health contract: probes try /actuator/health and the standard candidate paths automatically. If your app exposes Spring Actuator with DB health indicators, backing-store connectivity shows up in the health verification’s connection measurement for free.
  • Workers: headless workers (queue consumers, SERVICE=worker processes) are routed to platforms that allow non-HTTP containers — never onto Cloud Run, which requires every service to bind $PORT.

Conventions worth following

  • Expose Spring Actuator (spring-boot-starter-actuator) — it gives the platform a health path and store-aware checks with zero custom code.
  • Externalize store config via environment (SPRING_DATASOURCE_URL, SPRING_DATA_MONGODB_URI, Redis host/port vars) — the platform injects credentials for the datasources you declared at Gate 3.
  • Keep the app booting in under ~3 minutes on your chosen tier; load balancers only start routing after first health pass.

When something fails

The per-cloud pages cover every recurring class and whether it self-heals: GCP · AWS.

Note
Two Java-specific ones to know: small-instance JVM OOM (auto-hardened — see the AWS page) and Cloud Run’s external /healthz 404 (Google front-end behavior, auto-handled — see the GCP page).
Was this page helpful?