Router - Lambda Starter

[huntharo]

Motivations

• Unpredictable issues can cause hundreds to thousands of Lambda invokes, which can be very costly
• Router - Lambda failure to connect back to pods (or other error) results in tight invoke loop #154
  • Demonstrates a case of hundreds of thousands of invokes
• Extension - Error handling improvements and tests #182
  • Added information on the Extension response to indicate whether the Router needs to back off

Acceptance Criteria

• Need to put the requests to start instances into a queue
• Need configurable limits to the number of lambdas to run concurrently
• This will still get into a tight Lambda Invoke loop if the lambda is invoked correctly but cannot connect back to the Router
  • Need to backoff retry on that so we don't cause huge AWS bills
  • Could consider throwing an exception out of the Lambda, which would then cause AWS's SDK to perform the backoff

[huntharo]

This was partly handled as part of #35