feat(linbo-remote): append a JSON record per finished run

[HappyBasher]

Part 2 of #179. Stacked on #183 (base is that branch; retarget to main once #183 is merged).

Every -c run appends one JSON line to $LINBOLOGDIR/linbo-remote_runs.jsonl:

{"hostname":"r100-pc01","mode":"direct","commands":["sync:1"],"dry_run":false,"start":"2026-10-01T11:36:43+0200","log":"/var/log/linuxmuster/linbo/r100-pc01_linbo-remote_20261001113643.log","end":"2026-10-01T11:41:02+0200","rc":0}

• Written by the remote script via an EXIT trap, since only it knows end time and exit status. A trap also leaves a record for aborted runs (e.g. killed tmux session).
• Appends are serialized with flock (parallel hosts share the file).
• Tests run a rendered script for real with bash (linbo-ssh replaced by true/false) and check the record on success and failure.

Not covered: onboot (-p) runs. The server only queues them and never learns end time or status, so there is nothing to record on its side. Say if you want a "queued" record for those too.

Refs #179

🤖 Generated with Claude Code

[TomlDev]

Thanks for this. A review of the code, with the points a consumer (linuxmuster-api, webui7, edulution) runs into:

Line references are to the files on the PR branches (linbo_remote_lib.py = src/serverfs/usr/share/linuxmuster/linbo/linbo_remote_lib.py).

1. start/end use %z without a colon (linbo_remote_lib.py buildRunRecordPrefix, time.strftime('%Y-%m-%dT%H:%M:%S%z', ...), and date +%Y-%m-%dT%H:%M:%S%z in writeRunRecord). 2026-10-01T11:36:43+0200 is not RFC 3339/ISO 8601 extended; Date.parse in Safari/Firefox, Go's time.RFC3339, Java OffsetDateTime and Python <= 3.10 fromisoformat reject it.
Suggestion: date -Iseconds in the trap and datetime.fromtimestamp(time.mktime(started)).astimezone().isoformat(timespec='seconds') (or time.strftime(...) plus an inserted colon) in Python, giving +02:00.

2. A run that is still in progress, or ended by SIGKILL/OOM/server reboot, leaves no record at all (writeRunRecord is the only writer; the line carries start and end together). A client cannot list "running" or "lost" runs from the file alone and has to join with tmux ls.
Suggestion: document it in the README, or write the start line at launch (same flock append, "event":"start") and the end line from the trap, joined by a run id.

3. No field links the per-host lines of one invocation, and skipped hosts leave no trace. linbo-remote -g r101 -c ... writes N unrelated lines; an offline host (Not online, host skipped. in sendCmds) writes none. Consumers (linuxmuster-api #46) want "run on room R101 finished on 18 of 20 hosts".
Suggestion: add "run_id" (one per invocation, e.g. <pid>-<epoch> or uuid4, generated once before the host loop) and "user" (SUDO_USER/USER) to the prefix; optionally append a "rc":null,"skipped":"offline" line for skipped hosts.

4. rc is 0 for start*/reboot/halt regardless of outcome (renderRemoteScript: [ $RC = 0 ] && ssh ... & followed by sleep 10, the background job's status is never read). For all other commands it is 0/1 without saying which command failed.
Suggestion: state in the README that rc 0 after a backgrounded command only means "dispatched"; optionally add "failed_cmd":"<n>" from the loop.

5. Schema is undocumented beyond the field names. Add a "v":1 field to the prefix and document in the README: types, start/end format, log is an absolute path, rc meaning (0, 1, 129/130/143), dry_run records exist and carry the dry-run exit status, commands is the normalized list.