syntaqxWhether a repository's Actions are on is read once per request, however many runs it starts. actions_on read repo_settings in D1 in each trigger path and again in create_run for every run, so one push starting five workflows read it six times. The service now keeps what it read in ActionsOn (protection.rs), a per-request map of repository id to on or off: the service is made afresh for each fetch and for the minute's sweep, Actions::on_event clears it for each queued event, and saving a repository's Actions settings forgets that repository's entry. create_run stays the gate for every trigger (push, pull_request including a fork's, merge_group, schedule, workflow_run, dispatch and the rest), and record_failed_run, which now takes the repository, is gated the same way, so an unreadable workflow file in a repository with Actions off records nothing. on_event, run_schedules and merge_group still ask before reading workflows from git or a schedule's head, now from the same read; dispatch, repository_dispatch and re-run still refuse with the same message. Behaviour is unchanged.0c9ad8cHistory
- ..
- backup.test.ts
- backup.ts
- broker-egress.ts
- broker-relay.test.ts
- broker-relay.ts
- broker.test.ts
- broker.ts
- bump.test.ts
- bump.ts
- credentials.test.ts
- credentials.ts
- egress.test.ts
- egress.ts
- forward.test.ts
- forward.ts
- gate.test.ts
- guard.ts
- index.ts
- lifecycle.test.ts
- lifecycle.ts
- machines.test.ts
- self-hosted.test.ts
- self-hosted.ts