Public: anyone can read this; the crew can change it.
A change is live in two minutes, proven on a copy before anyone sees it, and… #11889
This pull covers only part of the issue: item 5 (a hotfix does not wait) and a first step on timings (item 1) and releasekeeper (item 6). The rest is still open: the faster build, the proof on a copy, the switch that disturbs nothing, one-step rollback, and the /flow display.
stopped closed without shipping deploy c41d8bb
2 earlier comments
This pull covers only part of the issue: item 5 (a hotfix does not wait) and a first step on timings (item 1) and release_keeper (item 6). The rest is still open: the faster build, the proof on a copy, the switch that disturbs nothing, one-step rollback, and the /flow display. bun scripts/gate.ts and bun x tsc --noEmit came back clean. 269 keeper, MCP and docs tests pass. I did not run the whole suite, because scripts/fast.ts called the diff too broad to name.
What changed
release_now { reason }: a person who may roll back (never an agent) with an admin-scope token asks the keeper's next tick to cut outside the half-hour spacing. A newkeeper_nowtable records the reason and who asked. The request is spent by the cut it triggers, and it lapses after an hour unused.- What still stops a hotfix: only the interval is skipped. A release in flight, a standing hold, the merge queue or a disk over its limit still refuse.
release_keepershowsblocked, so a refused hotfix says why. - Timings on
release_keeper: it now returnscurrent(the release in flight, else the newest) andreleaseNow(the pending request). Each release carriestimings: cut→live, verify and total seconds, and which stage was slowest (null for a stage not reached). - Docs and tests:
docs/MCP.mdanddocs/REFERENCE.mdare updated.test/release-now.test.tscovers the cadence rule, the once-only request, the agent and empty-reason refusals, a hold still stopping a hotfix, and the timings.
Not done
- Timings: build, boot proof and switch are one stage, "cut→live", because the deployer doesn't stamp them separately yet. The release card and
/flowdon't show timings or holds yet. - The rest of the issue: nothing for the faster build, the proof on a copy (status probes, stranger journey, covered-page check), runs surviving a release, interrupted-run counts, one-step rollback or a warm previous process. Each of those needs the deployer reworked and is a separate pull. There are no production timings to paste, since nothing here changes how releases are built or switched.
Try it: as an admin, call release_now with a reason, then release_keeper. You should see releaseNow set, and after the next tick current.timings filled in.
The push hook asked for a "Try:" commit before it would accept the push, so there is an empty commit saying how to try it.
The two failing tests came from release_now, which writes and needs an admin-scope token but wasn't on the list of tools that ask first (ASKING in src/crew/conversation_confirm.ts). I added it there: "cut a release now, outside the half-hour spacing". An agent now has to get a person's confirmation before it can cut a release outside the spacing.
test/conversation-confirm.test.ts, test/release-now.test.ts and test/conversation-holds.test.ts pass locally (51 tests). The gate and tsc --noEmit printed nothing. I haven't seen the forge's full check on this push yet.
Earlier in this pull, a hotfix can skip the keeper's half-hour spacing with release_now and a reason. release_keeper shows the current release with stage timings. The migration pin is in too.
Try it by calling release_now with a reason using an admin-scope token, then reading release_keeper for releaseNow and current.timings.
Declined: already done on main in fd4d15b83. Issue #11839 was landed on main by #11841, which adds release_now, src/keeper/now.ts, the keeper stages migration, the /releases card and the tests. All eight conflicting files were a second implementation of the same feature, so I took main's version of each. Run log: https://forge.open-platform.sh/trevato/forge/runs/01m797kecz3mrjw0dsn3zm8zrh
This pull covers only part of the issue: item 5 (a hotfix does not wait) and a first step on timings (item 1) and
release_keeper(item 6). The rest is still open: the faster build, the proof on a copy, the switch that disturbs nothing, one-step rollback, and the/flowdisplay.bun scripts/gate.tsandbun x tsc --noEmitcame back clean. 269 keeper, MCP and docs tests pass. I did not run the whole suite, becausescripts/fast.tscalled the diff too broad to name.What changed
release_now { reason }: a person who may roll back (never an agent) with an admin-scope token asks the keeper's next tick to cut outside the half-hour spacing. A newkeeper_nowtable records the reason and who asked. The request is spent by the cut it triggers, and it lapses after an hour unused.release_keepershowsblocked, so a refused hotfix says why.release_keeper: it now returnscurrent(the release in flight, else the newest) andreleaseNow(the pending request). Each release carriestimings: cut→live, verify and total seconds, and which stage was slowest (null for a stage not reached).docs/MCP.mdanddocs/REFERENCE.mdare updated.test/release-now.test.tscovers the cadence rule, the once-only request, the agent and empty-reason refusals, a hold still stopping a hotfix, and the timings.Not done
/flowdon't show timings or holds yet.Try it: as an admin, call
release_nowwith a reason, thenrelease_keeper. You should seereleaseNowset, and after the next tickcurrent.timingsfilled in.The push hook asked for a "Try:" commit before it would accept the push, so there is an empty commit saying how to try it.
Closes #11839