G8: bump ericomeehan.gitea for the Actions cache; document it

Picks up DevOps/ansible-role-gitea@96e5b92, which turns on the runner's
built-in cache server with a persistent volume. runner.md describes the
cache and how to troubleshoot or clear it, and corrects the DinD CPU limit
to the 750m now set for the current runner node. adding-ci.md shows how a
workflow uses actions/cache, pinned by SHA.

Refs #56

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This commit is contained in:
2026-11-17 09:24:30 -05:00
co-authored by Claude Opus 5.5
parent 8154b5f9fe
commit 78ae47bf11
3 changed files with 61 additions and 3 deletions
+26
View File
@@ -174,6 +174,32 @@ jobs:
| `dockerfiles` | every `Dockerfile*` and `*.Dockerfile` | Space-separated paths. The job fails if it finds none. |
| `failure-threshold` | `warning` | `error`, `warning`, `info` or `style`. Use a `.hadolint.yaml` for per-rule ignores. |
## Caching dependencies (G8)
The runner serves the Actions cache, so `actions/cache` works as on GitHub.
Use it for downloads that are slow and change rarely (npm, pip, Go modules,
Ansible collections). Don't cache build outputs that are cheap to rebuild.
```yaml
- name: restore npm cache
uses: actions/cache@caa296126883cff596d87d8935842f9db880ef25 # v5.1.0
with:
path: ~/.npm
key: npm-${{ hashFiles('package-lock.json') }}
restore-keys: npm-
```
- Pin by SHA, like every other action here (see
[actions-inventory.md](actions-inventory.md)). v4.3.0
(`0057852bfaa89a56745cba8c7296529d2fc39830`) also works.
- Put a hash of the lock file in the key, so a dependency change makes a new
entry instead of restoring a stale one.
- Entries are per repository. Unused entries expire after 7 days, and every
entry after 30. The first run on a new key is always a miss.
- Never cache anything that holds a secret. Anyone who can run a workflow in
the repo can restore its cache.
- If the cache is down, the step warns and the job carries on without it.
## How these were tested (2026-11-14)
Each workflow's `run:` steps were taken from the YAML as written and run inside
+34 -2
View File
@@ -18,7 +18,8 @@ gitea (Deployment, ns gitea) <── polls for jobs ── gitea-actions-runner-
├─ init: reach-gitea waits until Gitea answers
├─ dind Docker daemon; job containers run here
├─ docker-prune (sidecar) cleans Docker data every 15 min (S5)
└─ runner act_runner; talks to dind over a socket
└─ runner act_runner; talks to dind over a socket;
serves the Actions cache on port 8088 (G8)
```
- **One runner for everyone.** It is registered at instance level, so every
@@ -39,11 +40,21 @@ gitea (Deployment, ns gitea) <── polls for jobs ── gitea-actions-runner-
- **Registration token.** Kubernetes Secret `gitea-actions-runner-token` in
namespace `gitea`. The role creates it once with
`gitea actions generate-runner-token` and never prints it.
- **Actions cache** (G8, #56). The runner's built-in cache server backs
`actions/cache`. Its files are on a separate PVC, `gitea-actions-cache`
(`gitea_actions_cache_size`, 20Gi), mounted at `/cache` in the `runner`
container, so cached dependencies survive a runner restart even though
DinD's Docker data doesn't. Job containers reach the server at the pod's own
IP on port 8088 (`gitea_actions_cache_port`). Each job gets its own token,
and entries are scoped per repository. The runner removes entries unused for
7 days, and any entry older than 30 days. The storage is NFS, which doesn't
enforce the PVC size, so that expiry is what keeps it bounded.
- **Job images.** `ubuntu-latest`, `ubuntu-24.04` and `ubuntu-22.04` map to
dated `docker.gitea.com/runner-images` tags (`gitea_actions_runner_labels`).
`force_pull` is off, so they stay cached in DinD and the cleanup never
removes them.
- **Limits.** DinD gets up to 2 CPU and 8Gi memory, and holds every job
- **Limits.** DinD gets up to 750m CPU (sized for the current 1-CPU runner
node; go back to 2 on a dedicated node) and 8Gi memory, and holds every job
container. A job over the memory limit is OOM-killed and fails with a step
error. The runner fails a job after 2h; Gitea fails one after 2h30m
(`ENDLESS_TASK_TIMEOUT`), marks one whose runner stops reporting as failed
@@ -161,6 +172,27 @@ device`.
until Gitea's task timeout fails it; cancel it first. See
[soak-test.md](soak-test.md) for what this looks like in a job's log.
### Cache misses, or cache errors in jobs
Signs: `actions/cache` warns `Failed to restore` or `Failed to save`, or
always reports a miss for a key that a previous run saved.
1. Check the runner log for `artifactcache` lines:
`kubectl -n gitea logs gitea-actions-runner-0 -c runner | grep -i cache`.
2. Check the volume is mounted and has space:
`kubectl -n gitea exec gitea-actions-runner-0 -c runner -- df -h /cache`.
3. A miss is normal for the first run on a new key, and after the entry has
expired (see Architecture). Keys are per repository, so one repo never
restores another's cache.
4. To clear the whole cache, while no job is running, remove the files and
restart the pod so the runner starts a fresh index:
`kubectl -n gitea exec gitea-actions-runner-0 -c runner -- sh -c 'rm -rf /cache/*'`,
then `kubectl -n gitea delete pod gitea-actions-runner-0`. Jobs only lose
speed, never data.
5. To switch the cache off, set `gitea_actions_cache_enabled: false` and apply
with `--tags actions`. Workflows that use `actions/cache` still pass; the
step just warns and skips.
### kubectl itself fails
If `kubectl` errors with a TLS or certificate message, the problem is the