Support custom loggers

[onlyplo]

This is a Feature Proposal

Description

• Need a way to log out timeout error messages as using our own repo's/companies logging convention. Currently the middleware lambda-powertools-middleware-log-timeout locks us down to the following error message 'invocation timed out'
• Need a way to also track the occurrence of timeouts as a metric. Ideally a way to plug in datadog or newrelic clients to track custom metrics

Proposed solution:

• Add override parameter to enable consumers to use their own logging function when a timeout occurs.
• example of how it could be implemented: d4f873b

[theburningmonk]

Is the problem the invocation timed out message, if so, there's a simpler solution to let you customize the log message itself.

But sounds like the problem you actually have, and the solution you're looking for, is a way to use your own logger instead of the one from the powertools, right? The solution for this problem would need to work for more than just this middleware. Maybe a possible solution here is to let the powertools logger take in a proxy for console.log that it calls instead of console.log? That way, you can inject your company's logging convention between the powertool logger and stdout.

As for custom metrics, you have a few solutions here:

• event hooks in the middleware: downside - you gotta consider this in code EVERYWHERE, upside - no additional infra)
• CloudWatch metric filter: downside - you gotta configure the metric filter for EVERY lambda log group (you can automate this though), upside - no code change
• post-process the log messages and turn them into custom metrics <like this guy>: downside - extra infra complexity, upside - deploy once per region and can be automated with a regional template stack (or maybe something you can do through AWS Control Tower?)

[onlyplo]

Actually the core problem is the inability to do a simple metrics.increment('failed.timeout', 1); with the datadog logging package whenever there is a timeout. Not looking to overcomplicate the solution with metric filters and log trawling.

Yes the message invocation timed out also doesn't match our repo's log naming convention hence I expanded the issue and the sample to allow the consumers to do add a custom function to do what we wish when a timeout occurs. (we can live the log message for now).

I agree powertools in general should allow consumers to inject their own logger as the current conventions are quite prescriptive, but perhaps we should splinter that off into a separate issue.

[theburningmonk]

Ok, let's keep this ticket for the custom logger thread since that's where we started off on. I'm gonna start a new issue for metrics.

For that, the simplest thing I can think of is to let you pass in a callback function, but let's carry on the discussed on the other thread.