[GSK-1589] Analytics infinite loop

[mattbit]

No description provided.

[linear]

GSK-1589 Infinite loop in analytics (mixpanel)

When api.mixpanel.com is not reachable:

• analytics makes millions of connection attempts in the background
• analytics threads block the python process exit
• analytics eats lots of CPU after just importing giskard, and forever in the background as it desperately tries to connect to mixpanel

How to reproduce: block connections to api.mixpanel.com and run import giskard.

Issues

analytics_method uses analytics.track but analytics.track is wrapped by analytics_method. This generates an infinite loop.

One solution can be moving the current track method to a private _track, then define the public track as:

    analytics_method
    def track(self, event_name, properties=None, meta=None, force=False):
        return self._track(event_name, properties=properties, meta=meta, force=force)

So that the _track can still be used without the analytics_method wrapping. Thus in the decorator we can call the unwrapped _track

try:
    analytics._track("tracking error", {"error": str(e)})
except BaseException:  # NOSONAR
    pass

This avoids falling in the infinite loop.

But there is a second problem: we launch analytics._track in a thread, so the try-catch does not really work (the call will return immediately, the thread will run in the background and has no access to the context anymore). The thread exceptions will be handled by threading.excepthook that we currently overwrite to track the errors, causing again an infinite loop…

The whole threading code should be revised. Maybe the higher level concurrent.futures.ThreadPoolExecutor can provide a better way to handle this (with futures collection at shutdown inside the GiskardAnalyticsCollector instance).

[sonarqubecloud]

Kudos, SonarCloud Quality Gate passed!   

0 Bugs
0 Vulnerabilities
0 Security Hotspots
0 Code Smells

77.8% Coverage
0.0% Duplication

[sonarqubecloud]

Kudos, SonarCloud Quality Gate passed!   

0 Bugs
0 Vulnerabilities
0 Security Hotspots
0 Code Smells

No Coverage information
No Duplication information

[Hartorn]

Using report error filter out the exception only when it's Giskard one.
If connection to mix panel fails, it will not raise a giskard exception so it will no be caught again

[mattbit]

@Hartorn ok for me, just ensure you have the pre-commit hook installed to run black & ruff.

(Can’t approve because I contributed to the PR)

[mattbit]

code style