Merge pull request #878 from fluxcd/pprof-guide

docs/dev-guides/debugging.md

@@ -0,0 +1,42 @@
+# Advanced debugging
+
+This guide covers more advanced debugging topics such as collecting
+runtime profiling data from GitOps Toolkit components.
+
+As a user, this page normally should be a last resort, but you may
+be asked by a maintainer to share a [collected profile](#collecting-a-profile)
+to debug e.g. performance issues.
+
+## Pprof
+
+The [GitOps Toolkit components](../components/index.md) serve [`pprof`](https://golang.org/pkg/net/http/pprof/)
+runtime profiling data on their metrics HTTP server (default `:8080`).
+
+### Endpoints
+
+| Endpoint    | Path                   |
+|-------------|------------------------|
+| Index       | `/debug/pprof/`        |
+| CPU profile | `/debug/pprof/profile` |
+| Symbol      | `/debug/pprof/symbol`  |
+| Trace       | `/debug/pprof/trace`   |
+
+### Collecting a profile
+
+To collect a profile, port-forward to the component's metrics endpoint
+and collect the data from the [endpoint](#endpoints) of choice:
+
+```console
+$ kubectl port-forward -n <namespace> deploy/<component> 8080
+$ curl -Sk -v http://localhost:8080/debug/pprof/heap > heap.out
+```
+
+The collected profile [can be analyzed using `go`](https://blog.golang.org/pprof),
+or shared with one of the maintainers.
+
+## Resource usage
+
+As `kubectl top` gives a limited (and at times inaccurate) overview of
+resource usage, it is often better to make use of the Grafana metrics
+to gather insights. See [monitoring](../guides/monitoring.md) for a
+guide on how to visualize this data with a Grafana dashboard.

mkdocs.yml

@@ -196,6 +196,7 @@ nav:
     - Uninstall: cmd/flux_uninstall.md
   - Dev Guides:
       - Watching for source changes: dev-guides/source-watcher.md
+      - Advanced debugging: dev-guides/debugging.md
   - Roadmap: roadmap/index.md
   - Contributing: contributing/index.md
   - FAQ: faq/index.md