Large slowdowns in Python cold start times about newrelic-lambda-layers HOT 4 OPEN

I've been communicating with New Relic support and they suggested setting
NEW_RELIC_PACKAGE_REPORTING_ENABLED: false
https://docs.newrelic.com/docs/apm/agents/python-agent/configuration/python-agent-configuration/#package-reporting-enabled

To bypass this (new) code:
https://github.com/newrelic/newrelic-python-agent/blame/2b14392a19517a20012d281fbaaedfc2497f4fc3/newrelic/core/environment.py#L207-L245

This makes a big difference to us. The start time is now more 0.1s, rather than 1.5s (for the reduced lambda) or 5s (for our real ones).

from newrelic-lambda-layers.

As a workaround, we might have to downgrade to arn:aws:lambda:ap-southeast-2:451483290750:layer:NewRelicPython310:1, but that's seems like it would be quite old. Are there change logs for each layer version so we can work out the risks?

from newrelic-lambda-layers.

For a really simple lambda:

print("imported")
def handler(*args):
  print("started")

The numbers are a fair bit smaller, and somewhat different (regression from v1 -> v2, and v15 -> 16), but still bad:

There's also ~0.6s of init time overhead on a cold start, from INIT_START to print("imported"), which seems to be constant across all versions.

That is, a cold start of a lambda with New Relic instrumentation seems to spend a fair chunk of time (>2s) on the New Relic overhead.

from newrelic-lambda-layers.

Observing the same slowdown (from few hundreds of milliseconds to 8-15 seconds) for arn:aws:lambda:eu-central-1:451483290750:layer:NewRelicPython311:11

from newrelic-lambda-layers.