I have a Lambda function (initiated by API Gateway) that accesses an Aurora Cluster in private subnets (no errors in CloudWatch, just that the function timed out).
If I invoke the function several times, at around 5 concurrent executions it is timing out and returning 502 errors from API Gateway.
I know there can be issues with cold start times when accessing a VPC, due to ENI's being created, and I need to make sure to have enough IP's for the ENI's to be created. But using the formula from AWS (my function has 512MB, and the private subnets has a /24 range, meaning 254 usable IP's):
IP's for ENI's = Projected peak concurrent executions * (Memory in GB / 3GB)
254 = Projected peak concurrent executions * (512/3000)
Projected peak concurrent executions = 1400+
Way higher than ~5. Have I missed something? Do I somehow need to manually create the ENI's, or is making sure I have enough available IP's enough?
Any guidance would be appreciated.