docs: add debugging tips to DEVELOPMENT.md

Adds a 'Debugging the controller locally' section to DEVELOPMENT.md
covering the knobs that come up most often when reproducing issues:

- RUNTIME_NAMESPACE to scope the watch
- --concurrent=1 to serialize reconciles for a clean trace
- flux suspend for unrelated HelmReleases sharing the cluster
- Cross-controller setup with source-controller for paths that depend
  on HelmChart/HelmRepository artifacts

Mirrors the equivalent docs improvement on notification-controller
(fluxcd/notification-controller#362).

Fixes #462

Signed-off-by: alliasgher <alliasgher123@gmail.com>

Author: alliasgher

DEVELOPMENT.md

@@ -97,4 +97,50 @@ Deploy `helm-controller` into the cluster that is configured in the local kubeco
 
 ```sh
 make deploy
-```
+```
+
+## Debugging the controller locally
+
+When reproducing an issue or stepping through reconciliation logic, the
+following knobs make local runs cheaper and the resulting logs easier to
+read.
+
+### Limit the watched namespace
+
+The controller watches every namespace by default. To narrow it to a single
+namespace, set the `RUNTIME_NAMESPACE` environment variable before invoking
+`make run`:
+
+```sh
+RUNTIME_NAMESPACE=flux-system make run
+```
+
+### Reduce reconcile concurrency
+
+Each `HelmRelease` reconcile is processed concurrently (default
+`--concurrent=4`). When debugging it is almost always easier to follow a
+serial trace; pass `--concurrent=1` so reconciles run one at a time:
+
+```sh
+go run ./main.go --concurrent=1
+```
+
+### Suspend unrelated objects
+
+If the controller is sharing a cluster with other Flux objects, suspend
+anything not relevant to the test you're running so their reconciles don't
+interleave with yours:
+
+```sh
+flux suspend helmrelease <name>
+```
+
+Resume with `flux resume helmrelease <name>` when you're done.
+
+### Cross-controller interactions
+
+`helm-controller` reads `HelmChart` and `HelmRepository` artifacts produced
+by `source-controller`. When testing a path that needs a fresh chart, run
+`source-controller` locally too (or keep its in-cluster instance running)
+and point both at the same management cluster. Suspending unrelated
+controllers' objects per the previous step avoids racing reconciles.