Document persistent offline reuse of pinned Runner Actions [skip ci]

This commit is contained in:
Pigway
2026-10-11 07:57:21 +08:00
parent b682bdd5b0
commit 7c29cc00cb
+26 -6
View File
@@ -293,12 +293,32 @@ server. Forks download Actions through their own Runner/network; no public actio
mirror is served by this project. The former Tea auxiliary repository is private
and is no longer required by current workflows.
Keep the NAS act_runner's existing action cache and Python toolcache persistent.
A new action URL or commit may require an initial download; an old mirror cache
is not automatically interchangeable with an upstream Git repository. Do not
prune Runner caches between jobs. Python toolcache misses still download Python
from its upstream release source. Network failures remain failures, never reasons
to skip validation or silently fall back to the owner's Tea server.
Keep the Runner's action cache and Python toolcache persistent. For Gitea Runner
versions supporting `cache.offline_mode`, use a private Runner configuration:
```yaml
runner:
file: /data/.runner
cache:
offline_mode: true
```
In the Runner's own Compose service (not the application), set `HOME=/data` and
`CONFIG_FILE=/data/config.yaml`, with `/data` already persisted. Before recreating
an idle Runner, back up its configuration and copy the existing action cache from
`/root/.cache/act` to `/data/.cache/act`. Preserve registration, labels, image,
mounts and networks; do not interrupt active jobs or recreate other services.
Verify cached actions against the pinned upstream commit and tree before reuse.
This is administrator configuration, not a public workflow dependency.
Offline reuse avoids refreshing cached actions from GitHub on each job. Moving
tags/branches may remain at the cached version; this workflow deliberately pins
immutable SHAs. A new action URL or commit requires an initial download or verified
administrator prewarming; an old mirror subtree is not automatically equivalent
to an upstream Git repository. Do not prune Runner caches between jobs. Python
toolcache misses still download Python from its upstream release source. Network
failures remain failures, never reasons to skip validation or silently fall back
to the owner's Tea server.
Administrators needing a mainland-specific download route can configure their
own Runner/network cache privately. Such settings must not introduce public