I deployed a new version of my Go-based lambda which uses public.ecr.aws/lambda/provided:al2023 as a base image and noticed higher latencies. I did a bisect between the good version and the bad version and deployed various images between them to isolate the issue. I isolated the regression to be caused by the base image change. The same Go binary using public.ecr.aws/lambda/provided:al2023.2025.10.02.14 as the base image has higher latency compared to using public.ecr.aws/lambda/provided:al2023.2025.09.30.00 as the base image.
When I do a diff of the docker inspect output of the two base images, I notice:
"Labels": {
"com.amazonaws.lambda.platform.kernel": "k510ga",
- "com.amazonaws.lambda.platform.version": "sbxv2brave"
+ "com.amazonaws.lambda.platform.version": "sbxv2ga"
},
I can't find more information about what sbxv2brave or sbxv2ga mean. Or whether it's something else inside the base image that's causing the latency regression. But there's a noticeable difference in latencies (both P50 and P90) when using the newer base image.
Here's a graph that shows the latencies (and my various attempts deploying different versions to isolate which change caused the increase):
