[trytond] Tracing and transaction info

[n1ngu]

Add tracing to the Trytond integration and extract transaction information from RPC request objects.

[n1ngu]

Any insight on how to test this is very welcome

[n1ngu]

@pokoli sorry to bother you. Just pushed the tracing feature commit with a fair amount of metaprogramming black magic. I am still figuring out some typing annotation details but I'd enjoy your review on that part.

It is working just fine in some production deployments but I may have missed some thoughts for multi-tenancy deployments.

[pokoli]

@n1ngu starting from 6.8 series there is a new log feature, which is used by the tryton server to log events from the user.
Maybe you will like to trace such events also in sentry. Do you think this is doable?

[n1ngu]

@n1ngu starting from 6.8 series there is a new log feature, which is used by the tryton server to log events from the user. Maybe you will like to trace such events also in sentry. Do you think this is doable?

Because that logging is implemented with a ModelStorage, any "ir.model.log":create call will get its own span with the present PR! Model.log could be instrumented just as create, copy, delete, etc, but there is very few code there. I think it would be more interesting to create spans for Transaction.commit or Transaction._store_log_records but that can be done out of the box with

sentry_sdk.init(
    traces_sample_rate=1/128,
    functions_to_trace=[
        {"qualified_name": "trytond.transaction.Transaction.commit"},
        {"qualified_name": "trytond.transaction.Transaction._store_log_records"},
    ],
)

because, unlike pool models, Transaction can be statically instrumented. Transaction.start is explicitly patched in this PR because I find the whole context manager it creates deserves a span but couldn't figure out how to do that with functions_to_trace (I feel I am missing a more obvious solution here)

I don't know if you meant to capture log events in sentry with capture_message to send events irrespective of unhandled exceptions and tracing sample rate? That'd be certainly interesting and easily doable! I'll give the idea a try in my staging deployments while this PR settles. I feel the major challenge will be configuring this not to wipe your sentry event ingestion quota 🥲