I'm investigating latency between API GW and Lambda. Although the latency may seem low, there is a source of latency that I cannot discern.
It appears to be integration latency from when API GW makes a request to Lambda, to when Lambda begins execution.
We have ruled out cold start by enabling provisioned concurrency and have no spill over into un-provisioned containers.
Example Access log & Corresponding Lambda
{
"requestTime": "09/Jul/2021:12:18:48 +0000",
"requestId": "redacted",
"httpMethod": "POST",
"path": "/redacted",
"resourcePath": "/redacted",
"status": 200,
"responseLatency": 56,
"wafError": "-",
"wafStatus": "-",
"wafLatency": "-",
"authnError": "-",
"authnStatus": "200",
"authnLatency": "2",
"authzError": "-",
"authzStatus": "200",
"authzLatency": "0",
"integrationRequestId": "redacted",
"integrationResponseStatus": "200",
"integrationLatency": "52",
"integrationServiceStatus": "200",
"identitySourceIp": "redacted",
"identityUserAgent": "redacted",
"identityUser": "redacted"
}
REPORT RequestId: redacted Duration: 37.83 ms Billed Duration: 38 ms Memory Size: 2048 MB Max Memory Used: 77 MB
Has anyone encountered this gap before? The overhead from API GW seems minimal.