[Mohith26]

A Pull Request should be associated with an Issue.

Related issue: #1685

Description

This PR makes a pending unpostpone visible in status and lets operators withdraw it.

After unpostpone, cut-over still waits (by design, #921) until the changelog heartbeat lag is below both --max-lag-millis and --cut-over-lock-timeout-seconds. Until now status showed the same postponing cut-over whether a request was pending or a previous attempt had failed and reset the flag, and nothing could cancel a request, so tooling that drives cut-over from a UI had no way to observe or give up on it.

• Status: while a request is pending, State reads postponing cut-over, unpostpone requested, or postponing cut-over, unpostpone requested: waiting for heartbeat lag (4.20s) to drop below 1.50s when lag is what cut-over is waiting on (same state, detail shape as throttled, ...). With no request pending the output is unchanged.
• cancel-unpostpone: new interactive command that withdraws a pending request; optional table name, like no-throttle. Requests, cancellations and consumption go through MigrationContext helpers behind one mutex, and consuming a request ends postponement in the same critical section, so a request is either consumed by cut-over or cancelled, never both, and a repeated unpostpone is refused once cut-over has begun.
• The postpone closure in cutOver() is moved into shouldPostponeCutOver() / cutOverHeartbeatLag() with the same logic (lag > min(maxLag, lockTimeout) is equivalent to the old two comparisons), so it can be unit tested.
• Docs: doc/interactive-commands.md and the help command.

Tests (all unit level, no MySQL needed):

• TestServerUnpostponeCommands: unpostpone/cancel round trip, cancel with nothing pending, table-name check, help text.
• TestMigratorShouldPostponeCutOver: a pending request survives high heartbeat lag and is consumed once lag recovers; a cancelled request keeps postponing; state strings.
• TestMigratorUnpostponeCancelRace: 500 concurrent consume-vs-cancel rounds, exactly one side wins each time.
• TestMigratorRepeatedUnpostponeAfterConsume / TestMigratorRepeatedUnpostponeRace: a repeated unpostpone after (or racing) consumption is refused and a later cancel reports nothing pending (300 rounds under -race).
• TestMigratorCutOverPostponeLoop: drives the real sleepWhileTrue loop through unpostpone under lag, cancel, lag recovery (still postponed), and unpostpone again (proceeds).

Ran locally (Apple Silicon): go test -race ./go/... (all pass except the testcontainers suites TestApplier/TestMigrator/TestEventsStreamer, which need Docker and fail the same way on master here), golangci-lint run ./go/... v2.11.4: 0 issues on the first commit; go vet and gofmt -s -l go/ clean on the follow-up. I could not run the localtests replica harness without Docker, so this has not been exercised against a live MySQL.

• contributed code is using same conventions as original code
• script/cibuild returns with no formatting errors, build errors or unit test errors (ran the equivalent steps above; Docker-backed suites excepted)

[Copilot]

🟡 Changes recommended

A critical race in repeated unpostpone requests can cause cancellation to report success even though cut-over is already proceeding with the postpone file present.

1 open issue

• Synchronize repeated unpostpone requests with cut-over state · New

What changed in this PR

Extends gh-ost’s interactive cut-over controls so operators can observe and withdraw pending unpostpone requests, addressing #1685 while preserving heartbeat-lag gating.

Changes:

• Shows pending requests and heartbeat-lag waits in migration status.
• Adds cancel-unpostpone with optional table-name validation, help text, and documentation.
• Extracts postpone checks and adds unit tests for commands, lag handling, cancellation races, and polling behavior.

File	Description
go/​logic/​server.go	Adds the cancellation command and help entry.
go/​logic/​server_test.go	Tests command responses, table validation, and help text.
go/​logic/​migrator.go	Extracts postpone checks, atomically consumes requests, and expands status output.
go/​logic/​migrator_test.go	Tests lag gating, cancellation races, and the postpone loop.
doc/​interactive-commands.md	Documents pending-request status and cancellation behavior.

_{Review effort: Max (auto)}

[Mohith26]

Good catch, fixed in b37800e. Requests, cancellations and consumption now go through MigrationContext helpers behind one mutex; consuming a request clears IsPostponingCutOver in the same critical section, and leaving the postpone loop drops any late request. Added TestMigratorRepeatedUnpostponeAfterConsume (fails on the previous commit) and a 300-round TestMigratorRepeatedUnpostponeRace under -race.