fix: ignore one-off container exit events in monitor

[waterWang]

Description

Fixes #14020

Since 2.39.0, the new monitor.go refactoring introduced a regression where docker compose up --abort-on-container-exit tears down the project when an unrelated one-off container created by docker compose run exits.

Root cause: The ContainerList call in monitor.Start() correctly excludes one-off containers via oneOffFilter(false), but the event subscription (Events()) does not. One-off containers carry com.docker.compose.oneoff=True and pass through the service guard, causing the monitor to see their exit and abort the project.

Fix: Add oneOffFilter(false) to the Events() filter args, matching the ContainerList filter. This is consistent with docker/compose-cli#1987 which established this exclusion deliberately.

Verification

• docker compose up --abort-on-container-exit main + docker compose run --rm main true no longer triggers premature abort
• Existing regression: docker compose up --abort-on-container-exit still aborts when a declared service exits (verified independent of this fix)

Related

• docker compose run brings down the cluster after one-off container finishes execution. docker-archive/compose-cli#1955, ignore one-off container events docker-archive/compose-cli#1987 (same fix for the old codebase)

[brano-osif]

Thanks for the fast turnaround. I'm the reporter — built this branch (cross-compiled darwin/arm64, tested against Engine 29.4.0 on macOS) and ran the repro from the issue plus two adjacent cases:

• one-off docker compose run --rm main true while up --abort-on-container-exit main is attached: no longer aborts, main keeps running
• up --abort-on-container-exit with no service arguments, then a one-off of a different service: also no longer aborts (this was the second trigger, where len(c.services) == 0 meant nothing filtered the event)
• a declared service exiting during up --abort-on-container-exit main other: still aborts, so the documented behaviour is unchanged

Confirmed fixed for me. Worth adding a regression test in monitor_test.go if you're up for it — heads up that #13990 is currently touching that file, so there may be a conflict to sort out.

[ndeloof]

LGTM, but please signoff your commit (this is a legal requirement)

[brano-osif]

Hi @ndeloof
DCO is still failing, and I doubt the author will come back to it (100 PRs across unrelated repos in the last two days, no force-pushes on any). I have the fix signed off on my side, with a regression test. Happy to open a PR to supersede this one.

[ndeloof]

Closing in favor of #14038, which carries this exact same one-liner (credit to @waterWang for identifying the fix) plus a regression test and a signed-off commit. The DCO sign-off is a legal requirement we cannot add on the author's behalf, and this PR can't merge without it — if you come back and amend, thanks again for the contribution!