syslog poller: lock timeout, signal handler, and earlier partition rotation

[somethingwithproof]

Description

Per review feedback on PR #290 (functions.php:285), three improvements are needed for the syslog poller process management:

1. Lock timeout

The GET_LOCK() call in syslog_preprocess_incoming_records() has no timeout. If a process dies while holding the lock, subsequent pollers block indefinitely. Add a reasonable timeout (e.g., 60 seconds) and handle the timeout case.

2. Signal handler

Add pcntl_signal() handlers for SIGTERM/SIGINT that release the lock and call unregister_process() before exiting. Prevents stale locks and orphaned process registrations after unclean shutdowns.

3. Earlier partition rotation

Move partition create/rotate to run ahead of the rotation boundary (e.g., an hour before midnight) instead of at the boundary itself. Avoids contention between ALTER TABLE and concurrent INSERT traffic from incoming syslog messages.

References

• PR Consolidated hardening, bug fixes, and CI improvements #290 review comment from TheWitness (2026-03-21)
• functions.php syslog_preprocess_incoming_records()
• functions.php syslog_partition_create()

[somethingwithproof]

Suggested approach for the lock/signal handler items:

Cacti core's register_process() / unregister_process() in lib/poller.php already handles process interlocking with heartbeat-based stale detection. syslog_process.php already uses it. Replacing GET_LOCK() with this mechanism would:

1. Replace GET_LOCK() with register_process() (already exists in syslog_process.php)
2. Set $timeout on the registration so stale entries expire automatically
3. Drop the MySQL lock entirely

This avoids the pcntl_signal() complexity and the connection-scoped lock issues (if the DB connection drops, GET_LOCK() silently releases and another process can grab it).