Catch, log, and handle exception in pipeline

[A-Artemis]

Summary

Enhance resilience of event dispatching and stream loading by handling errors gracefully

Checklist

• The PR title and description are clear and descriptive
• I have manually tested the changes
• All changes are covered by automated tests
• All related issues are linked to this PR (if applicable)
• Documentation has been updated (if applicable)

[github-actions]

⏱️ Backend import time — app.main

Accelerator	Cumulative (s)	Self (s)
cpu	4.546	0.004
xpu	4.550	0.003
cuda	4.627	0.004

[Copilot]

Pull request overview

This PR improves backend pipeline resilience by catching and handling exceptions in key worker loops (event dispatching, stream loading, inference callback processing, and inference-server TTL monitoring) so that transient or isolated failures don’t crash long-running workers.

Changes:

• Add retry/backoff + consecutive-error thresholding for frame acquisition and reopen attempts in StreamLoader.
• Make event emission and dispatching more fault-tolerant by isolating failing handlers/destinations and keeping loops alive.
• Extend unit tests to cover “survives failure” behavior across stream loading, dispatching, event bus, and inference-server monitoring.

Reviewed changes

Copilot reviewed 9 out of 9 changed files in this pull request and generated 3 comments.

Show a summary per file

File	Description
application/backend/app/workers/stream_loading.py	Adds consecutive-error handling, exponential backoff, and stream lifecycle guards for stream acquisition.
application/backend/app/workers/dispatching.py	Makes sink initialization/connection and dispatch loop more resilient to exceptions; isolates per-destination failures.
application/backend/app/services/event/base.py	Prevents one failing subscriber from breaking event emission to other subscribers.
application/backend/app/workers/inference.py	Wraps inference completion callback to prevent exceptions from wedging the async inference queue.
application/backend/app/workers/inference_server_monitor.py	Prevents TTL monitor thread from dying when model unload fails.
application/backend/tests/unit/workers/test_stream_loading.py	Updates/extends tests for consecutive frame error thresholding and transient failure recovery.
application/backend/tests/unit/workers/test_dispatching.py	Adds resilience tests for sink connect/build failures and data collector failures.
application/backend/tests/unit/services/event/test_event_bus.py	Adds test ensuring emit_event continues after a failing handler.
application/backend/tests/unit/workers/test_inference_server_monitor.py	Adds test ensuring unload failure doesn’t kill the monitor loop.

[github-actions]

📊 Test coverage report

Metric	Coverage
Lines	60.3%
Functions	57.0%
Branches	52.1%
Statements	59.9%

[github-actions]

🐳 Docker image sizes

Device	Size
xpu	10801.8 MB (10.55 GB)
cpu	4013.2 MB (3.92 GB)
cuda	10208.4 MB (9.97 GB)

[leoll2]

Looks good to me. This very defensive exception handling may seem excessive at first glance, but I think it's necessary in these delicate yet critical pipeline components. I appreciate the very descriptive inline comments, I find them really helpful to understand what every protection mechanism is intended for.