NethVoice: Tancredi stays stopped after a dependency startup failure #8245

Steps to reproduce

1. Use a configured NethVoice instance with Tancredi enabled.
2. During startup, make a required dependency fail temporarily. For example, delay MariaDB startup until its service startup timeout expires.
3. Allow MariaDB and FreePBX to recover without manually starting Tancredi.
4. Check Tancredi's service state.
5. Update the instance and check whether the update recovers Tancredi.

Expected behavior

Once FreePBX is ready, enabled Tancredi should start automatically.

An update should recover required, enabled dependent services without starting disabled optional services or unnecessarily restarting healthy telephony services.

Actual behavior

Tancredi's start job fails with result dependency and the service remains inactive after its prerequisite recovers.

The update includes Tancredi in its restart list, but the restart helper uses systemctl --user try-restart, which skips inactive services. Consequently, the old stopped container can remain after the module image settings have been updated.

Sanitized evidence from an affected instance:

• 14:17:13 CEST: MariaDB startup began.
• 14:18:43 CEST: MariaDB reached its 90-second startup timeout.
• 14:19:32 CEST: Tancredi's start job failed with result dependency.
• 14:24:18 CEST: MariaDB started successfully.
• 18:46:28 CEST: the update passed Tancredi to the conditional restart helper.
• Tancredi remained inactive, referencing the previous container image.

The original reason for MariaDB's slow startup remains unconfirmed. Disk I/O pressure was observed later, but has not been proven to have caused the earlier timeout.

Components

• Observed with NethVoice 1.7.10, followed by an update to 1.7.11.
• The relevant recovery behavior is also present on fpbx17.

See also

Implementation PR: nethesis/ns8-nethvoice#1058