I'm curious if anyone else using us-east-2 has noticed this. To create new instances, we send a disk image to S3, run a aws ec2 import-snapshot <stuff>, and then aws ec2 register-imageto be able to stick that instance in our auto scaling group.
We've been using this process in us-east-2 for about 4 months now and the s3 upload took around 10 seconds and the AMI registration about the same. The disk image (around 3g, in a raw format) would take around 200 to 250 seconds for import-snapshot, which was acceptable.
On Monday (2019/1/28), we started having deployments fail because we hit a 15 minute timeout that we set for image imports. I added some extra logging and import-snapshots are around 1000 seconds and it nearly doubles our deployment turnaround time.
We haven't had any code changes, but since import-snapshot is a task that we fire off to amazon and then wait for it to complete by polling status, there's nothing really that I can change. Smaller image sizes don't appear to make a difference either.
AWS 官方已更新 1 年前