Support for Metrics

[astorm]

We should decide on a strategy for reporting metrics (such as memory usage, total available memory, billing duration/init duration, etc.). This may mean making the agent's metrics work inside a lambda function, it may mean using the Lambda Logs API to look for and report metrics, it may means leveraging metric beat or function beat, or may mean some other thing.

• Decide on Metrics Strategy
• Decide which Metrics to Collect
• Expose Lambda specific metrics via UI

[AlexanderWert]

Community & customer feedback:

• a solution that does not involve interaction with CloudWatch is preferred for costs and rate limiting reasons
• possibility to write metrics to logstash (instead of directly to the APM server)
• metrics indicating efficiency would be very valuable (throttles, CPU time/usage, Memory consumption)

Proposal: let's start as simple as collecting the metrics provided by the Lambda platform logs:

• Duration
• Billed Duration
• Memory Size
• Memory Used
• Init Duration

[jlvoiseux]

Was closed by mistake upon merging #202 - We still need to update the APM model to properly include faas information alongside the metrics (ie. replace labels.faas_xxxx by faas.xxxx).

[axw]

With elastic/apm-server#8400 merged, I think we can close this. Server support will be in 8.4.