Forgejo Actions CI never dispatches on Mark/transcriber — #133 fix and #134 never built (last success 2026-06-14) #136

New issue

Closed

opened 2026-08-14 06:06:05 +00:00 by Mark · 1 comment

Mark commented

2026-08-14 06:06:05 +00:00

Owner

Forgejo Actions CI is not running for Mark/transcriber. The push runs for the #133 fix commit (39af1b3, run 5629) and the #134 commit (1b1c329, run 5658) were never dispatched: run 5629 sat waiting then was cancelled without ever starting (no started_at, no conclusion); run 5658 is still waiting as of 2026-08-14 03:00Z. Last CI success on this repo is run 4121 (2026-06-14, commit 68c6c85).

Consequence: the Hilt/WorkManager code in #133 is compile-unverified — the kilo-gsd verify gate for #133 was grep-based, not a real build.

Actual: push events create runs that sit waiting indefinitely and get cancelled.
Expected: runs dispatch to an online runner within minutes (pattern: run 4121 in June).

Likely cause (from prior diagnosis pattern): workflow uses runs-on: ubuntu-latest (.forgejo/workflows/ci.yml) but no online runner is registered with that label for Mark/transcriber — repo-scoped runners on other repos silently ignore jobs from this repo. Check runner registration / last_online for Mark/transcriber; register a runner for this repo (or use the linux label that the blackhole runner registers) so runs dispatch.

Reproduction:

Push any commit to origin/main.
Watch the Actions run: status stays waiting, never gets a started_at.
Compare: run 4121 (2026-06-14) is the last run with a conclusion.

Filed from kanban card t_a946c33c (post-deploy probe of Mark/transcriber#133).

Forgejo Actions CI is not running for Mark/transcriber. The push runs for the #133 fix commit (39af1b3, run 5629) and the #134 commit (1b1c329, run 5658) were never dispatched: run 5629 sat `waiting` then was `cancelled` without ever starting (no started_at, no conclusion); run 5658 is still `waiting` as of 2026-08-14 03:00Z. Last CI success on this repo is run 4121 (2026-06-14, commit 68c6c85). Consequence: the Hilt/WorkManager code in #133 is compile-unverified — the kilo-gsd verify gate for #133 was grep-based, not a real build. Actual: push events create runs that sit `waiting` indefinitely and get cancelled. Expected: runs dispatch to an online runner within minutes (pattern: run 4121 in June). Likely cause (from prior diagnosis pattern): workflow uses `runs-on: ubuntu-latest` (.forgejo/workflows/ci.yml) but no online runner is registered with that label for Mark/transcriber — repo-scoped runners on other repos silently ignore jobs from this repo. Check runner registration / last_online for Mark/transcriber; register a runner for this repo (or use the `linux` label that the blackhole runner registers) so runs dispatch. Reproduction: - Push any commit to origin/main. - Watch the Actions run: status stays `waiting`, never gets a started_at. - Compare: run 4121 (2026-06-14) is the last run with a conclusion. Filed from kanban card t_a946c33c (post-deploy probe of Mark/transcriber#133).

Mark added the

bug

triage-approved

labels

2026-08-14 06:06:05 +00:00

Mark commented

2026-08-14 06:18:16 +00:00

Author

Owner

Re-probe confirmation (kanban card t_51a41bbe, post-deploy probe of #133) — 2026-08-14 ~07:00Z:

Root cause CONFIRMED, fix recipe now concrete:

Mark/infrastructure#78 (closed 2026-08-14T03:42Z): the generic localhost-runner registration (action_runner id 23) is repo-scoped to Mark/runner-test (repo_id 158), NOT user-level. Per-repo runners exist for kimi + nodedesk (named localhost-<repo>-runner, self-hosted label), registered via scripts at ~/forgejo-runner-kimi/{register,start}.sh.
No equivalent runner exists for Mark/transcriber — its CI can never dispatch until one is registered. Live check 2026-08-14 07:00Z: run 5658 (current HEAD 1b1c329) still waiting; run 5629 (39af1b3, the #133 fix) cancelled without ever starting. No localhost-transcriber registration found anywhere.

Action required (operator):

Register localhost-transcriber-runner (repo-scoped, self-hosted label) using the infra#78 pattern / scripts template.
Re-trigger run 5658 after the runner comes online (and optionally re-run 5629 commit for CI evidence on the #133 fix).

Note: the #133 fix itself is verified clean (issue closed, commit 39af1b3 on origin/main, assembleDebug + 72/72 unit tests pass) — this issue is solely about CI infrastructure, not the fix quality.

Re-probe confirmation (kanban card t_51a41bbe, post-deploy probe of #133) — 2026-08-14 ~07:00Z: Root cause CONFIRMED, fix recipe now concrete: - **Mark/infrastructure#78** (closed 2026-08-14T03:42Z): the generic `localhost-runner` registration (action_runner id 23) is repo-scoped to Mark/runner-test (repo_id 158), NOT user-level. Per-repo runners exist for kimi + nodedesk (named `localhost-<repo>-runner`, `self-hosted` label), registered via scripts at `~/forgejo-runner-kimi/{register,start}.sh`. - **No equivalent runner exists for Mark/transcriber** — its CI can never dispatch until one is registered. Live check 2026-08-14 07:00Z: run 5658 (current HEAD 1b1c329) still `waiting`; run 5629 (39af1b3, the #133 fix) `cancelled` without ever starting. No `localhost-transcriber` registration found anywhere. Action required (operator): 1. Register `localhost-transcriber-runner` (repo-scoped, `self-hosted` label) using the infra#78 pattern / scripts template. 2. Re-trigger run 5658 after the runner comes online (and optionally re-run 5629 commit for CI evidence on the #133 fix). Note: the #133 fix itself is verified clean (issue closed, commit 39af1b3 on origin/main, assembleDebug + 72/72 unit tests pass) — this issue is solely about CI infrastructure, not the fix quality.

Mark

2026-08-14 07:36:11 +00:00