Use one connection for all communication with extension.

[purple4reina]

What does this PR do?

Changes all communication with the extension to using a persistent http connection.

Motivation

This change reduces runtime duration by somewhere between 5-10%.

Testing Guidelines

⚠️ This is a high risk change. Sufficient testing must be done before merging! ⚠️

As I run tests on this branch, I will post the results below as comments.

Additional Notes

Types of Changes

• Bug fix
• New feature
• Breaking change
• Misc (docs, refactoring, dependency upgrade, etc.)

Check all that apply

• This PR's description is comprehensive
• This PR contains breaking changes that are documented in the description
• This PR introduces new APIs or parameters that are documented and unlikely to change in the foreseeable future
• This PR impacts documentation, and it has been updated (or a ticket has been logged)
• This PR's changes are covered by the automated tests
• This PR collects user input/sensitive content into Datadog
• This PR passes the integration tests (ask a Datadog member to run the tests)

[purple4reina]

I created two apps, one using v85 of the prod layer and another using a layer built from this branch. I then curled the two apps on a loop:

while true ; do curl [URL1] ; curl [URL2] ; done

I let this run for several hours. Here are stats from a 4 hour window of this run:

[purple4reina]

I created two apps, one using v85 of the prod layer and another using a layer built from this branch. I then curled the two apps on a loop with a sleep in between:

while true ; do curl [URL1] ; curl [URL2] ; sleep 120 ; done

I let this run for about 30 minutes. Here are stats:

As you can see, the difference is not as large as it was for the busy function. Since there are no errors, I will say that this PR neither hurts nor helps apps like these.

[purple4reina]

I created two apps, one using v85 of the prod layer and another using a layer built from this branch. The functions will timeout during init on the first try, but succeed on the second.

import os
import time

filename = '/tmp/testfile'
if not os.path.exists(filename):
    # cause timeout on first invocation only
    with open(filename, 'w') as f:
        f.write('hello')
    time.sleep(600)

def handler(event, context):
    return 'ok'

I then curled the two apps on a loop, changing the max memory in between in order to force a cold start.

In neither the prod nor testing layer versions was I able to receive the aws.lambda.enhanced.init_duration metric. Therefore, I am unable to state the overhead impact on this scenario.

On the other hand, the function still performs as expected when using the new layer.

[purple4reina]

Here is a test with functions that are forced to cold start on each invocation. We can see that there is no significant change in overhead on cold start.

[purple4reina]

Closing in favor of #406. The flush endpoint on the extension does nothing, therefore we shouldn't be calling it in the first place.