Java Profiler Deployment and Operations

This page is the English entry point for platform administrators, SREs, and security administrators. The Chinese operational manual contains the longer installation and troubleshooting procedures.

Product boundary

The first version targets HotSpot-compatible Java services on Kubernetes:

• The collector runs as a node-local DaemonSet.
• Profiling is opt-in through Kubernetes annotations or labels.
• CPU, allocation, and lock profiling use async-profiler.
• Profiles, thread diagnostics, target status, and ingestion evidence are stored in ClickHouse.
• Collected data is retained for no more than seven days; optional raw artifacts are limited to 24 hours.
• Prometheus-compatible metrics are exposed by collector/backend exporters; this product does not store Prometheus time series.

The platform does not expand into general logging, tracing, service maps, non-Java profiling, or a long-term metrics platform.

Administrator responsibilities

Administrators own installation, configuration, permissions, upgrades, rollback, ClickHouse retention, Web-to-backend routing, collector discovery, and operational health. Service owners own the interpretation of CPU, allocation, lock, deadlock, and thread evidence.

Use the Profiling Contracts for field names and status values. Use the Real Profiling Acceptance Standard when a release changes the profiling path.

Required operational checks

Before handing a deployment to users, verify:

1. Collector, backend, Web, and ClickHouse are healthy.
2. Kubernetes RBAC allows node-local JVM discovery and attach operations.
3. The Web /api/* proxy reaches the backend rather than the static asset server.
4. A target workload is accepted only when its metadata enables profiling.
5. ClickHouse TTLs are bounded and cleanup health is observable.
6. Real acceptance produces non-empty profile data, ingestion evidence, browser evidence, and no unexpected workload restart.

Release and troubleshooting

The release scripts build and publish versioned artifacts. Treat a release as ready only after the deployment, backend, ClickHouse, collector, and UI checks pass. For detailed Helm, Secret, RBAC, proxy, retention, upgrade, rollback, and failure-case procedures, use the full Chinese operational manual.