diff --git a/.gitea/workflows/main.yml b/.gitea/workflows/main.yml index a878e56a..457d83cf 100644 --- a/.gitea/workflows/main.yml +++ b/.gitea/workflows/main.yml @@ -344,9 +344,10 @@ jobs: # run (63), and on every push, where the nightly jobs it waited for are skipped (79). A throwaway # workflow started by hand reproduced it with nothing else in it (2026-11-11, three variants): # every job with both `needs` and `if` stayed Blocked, whatever the `if` said and whether or not - # the jobs it needed had passed. A job without `needs` is queued when the run starts, and the one - # runner (capacity 1) takes a run's jobs in the order they were created, which is their order in - # this file. A second runner would let this job overlap the others; it would still run. + # the jobs it needed had passed. A job without `needs` is runnable when the run starts, and the one + # runner (capacity 1) takes the job that became runnable first, so a run's jobs go in the order they + # were created, which is their order in this file. A second runner would let this job overlap the + # others; it would still run. performance: if: ${{ github.event_name != 'push' }} runs-on: ubuntu-latest diff --git a/.gitea/workflows/release.yml b/.gitea/workflows/release.yml index 8022ae99..05ace22e 100644 --- a/.gitea/workflows/release.yml +++ b/.gitea/workflows/release.yml @@ -142,6 +142,16 @@ jobs: # Build, inventory and scan the image; on a tag also push it. It waits on every evidence job: # `needs` does not reach across workflows, so a gate in another file could still be running. + # + # What `needs` costs on this runner. Gitea gives the runner the job that became runnable first, + # across every repository the runner serves. A job without `needs` is runnable when its run is + # created; this one only when the last gate finishes, so it waits behind every job created before + # that moment, other repositories' jobs included. Manual run 157 (2026-11-12): the gates finished at + # 17:50 UTC and this job started at 20:52, after four frontend pull-request runs created between + # 17:03 and 17:47; app-image-signing waited two hours more the same way. Their job time is what a + # release costs; the wait is the queue's. On an idle runner the wait is zero, so cut a release tag + # when no other run is queued. Kept as it is: without `needs` an image could be published before a + # gate failed, and folding the signing into this job would put the key in a job that builds. app-image-release: needs: - release