fix: close events left open by a capture daemon that died

[nabbi]

Automated PR, tested in a zm-holodeck sandbox. An AI agent (Claude Opus 5.5) reproduced the bug on a disposable ZoneMinder built from upstream master 821030e17, wrote this fix and ran the checks below on the same lab. Minimal human review before opening: please review it as you would any contributor's PR.

TL;DR

• Fixes: An event open when zmc is killed is never closed on a default install #5200: an event that was open when its zmc was killed is never closed on a default install.
• Change: once per zmc start, zmwatch closes that monitor's events left open by the previous, dead zmc. Event::Close() (also behind zmeventtool.pl close) now works when the event has no Frames rows yet. 3 files, +179/-9 (120 of them a Perl test).
• Risk: one indexed query per monitor each time a zmc starts. Only events that started before the live zmc connected are touched, so it can't close an event a running zmc still owns. Re-analysis events are skipped.
• Tested: reproduced before, closed 13 s after the new zmc came up; a real lost-stop KILL (zmwatch restarts a zmc that is still starting, and zmdc loses the stop and KILLs it 30 s later, leaving its event open #5198) during the check run was recovered too. Script tests, smoke and event checks green apart from one known XFAIL.

Details: change, behaviour changes, checks

Before

Upstream master 821030e17: kill -9 a Recording=Always zmc mid-event. The event keeps EndDateTime NULL indefinitely:

1  2026-10-10 19:19:05  NULL  19.93  300  incomplete.h264.mp4  New Event

Change

• scripts/zmwatch.pl.in: when a monitor's heartbeat is fresh and its shared-memory startup_time differs from the last one seen, close its events with EndDateTime IS NULL, StartDateTime < FROM_UNIXTIME(startup_time) and Cause other than Re-analysis, through Event::Close(). A running zmc closes its event before reconnecting and sets startup_time after connecting, so it never owns such an event. Only this server's monitors.
• scripts/ZoneMinder/lib/ZoneMinder/Event.pm Close(): Length, Frames, AlarmFrames, TotScore and MaxScore = the larger of the event row and the Frames rows (zmc writes both in batches, so either can be ahead); EndDateTime follows the Length used; works with no Frames rows; refuses an event without a start. Name and Notes as before ((r) and Recovered., zmaudit's existing tags).
• Test: tests/perl/test_event_close_recovered.pl (17 cases; 11 fail on master's Event.pm).

Behaviour changes

Change	Who sees it	Intended / collateral	Alternatives looked into	Chosen, why	Tested by
A dead zmc's open events are closed seconds after the next zmc is up, named <prefix><Id>(r), note Recovered.	event lists, montage review, filters, API clients; zmeventnotificationNg's isEventEnded() now sees them end	intended	zmaudit's pass (off by default, one-day age); zmc in C++ (another copy of the close logic); zmdc before fork (misses a hand-started zmc)	zmwatch: already watches each zmc's shared memory and has Event::Close()	probe, events
zmwatch Info line Closed event N on monitor M …	log readers	intended	Debug level	visible once per recovery	probe
zmeventtool.pl close handles events with no Frames rows	admins	intended	—	—	Perl test
One indexed SELECT per monitor per zmc start	DB	collateral, negligible	—	—	events

After

Branch at 993b0d80d (base 821030e17). kill -9 of monitor 2's zmc at 00:01:38; event 60 closed 13 s later, the new zmc's event 61 untouched:

60 | 00:01:15 | 00:01:35 | 20.00 | 300 | incomplete.h264.mp4 | Event-60(r) | Recovered.

Check	Result
holodeck checks for the diff	unit 249 cases; script tests PASS=30; smoke PASS=11 XFAIL=1 (stock zmonvif-probe.pl, known); events PASS=10; 0 open events after the run

Not run: clients (no API change), multi-server, release-1.38 (backport prepared, not run).

Possible follow-ups

Offered for discussion, not part of this change:

• Length from the video. The row's Length can trail the video by the last unflushed batch (20.00 s in the row vs 22.0 s by ffprobe here). Closing with the video's duration would be more accurate. Benefit: recovered events show their real length. Risk: an ffprobe run per recovered event.
• Finish the HLS playlist. A recovered index.m3u8 has no #EXT-X-ENDLIST, so some players treat it as live. Rewriting it as VOD would fix that. Risk: file rewrites in event directories.
• Fill DiskSpace and keep a user-set name in zmeventtool close. Risk: a naming change for admins who rely on the (r) suffix.
• Gaps left: monitors with Capturing=None keep any open event (zmwatch skips them), and zmaudit's unclosed-event pass keeps its own copy of the close logic.

A branch with the first three exists and was lab-tested; happy to open it separately if wanted.

refs #5200

🤖 Generated with Claude Code

[connortechnology]

Close() always renames the event to <prefix><Id>(r), so it overwrites a name the user gave the event while it was recording. That was fine while only zmeventtool.pl close and zmaudit called it, by hand or a day later, but now zmwatch runs it on its own every time a zmc restarts.

zmc's own close only renames an event whose name is still the placeholder (~Event() in src/zm_event.cpp):

Name = IF(Name='New Event', '<prefix><Id>', Name)

Could Close() do the same here: add the (r) name only when Name is still New Event, and otherwise keep the user's name? The Recovered. note still marks the event either way. I'd fold this into this PR rather than leave it as a follow-up, since this is the change that makes the rename automatic.