1

I have an application that is very fsync intensive. When I run it on a traditional VM the application is quite slow. When I run it in a Docker container on CoreOS it gets much faster, by at least a factor of 10.

At least, until recently. Sometime around the end of November 2018 the Docker performance suddenly dropped becoming about the same as the VM-hosted version.

CoreOS runs in AWS on a fairly standard EC2 instance. The filesystem that Docker writes to is mounted from a directory in an ext4 partition on CoreOS. The underlying file system is an EBS volume.

Was there a change in Docker (or CoreOS ) around that time that might explain the sudden change in my container's performance? The CoreOS version running now is 4.14.81-coreos. CoreOS updates by itself, so I don't know the version(s) of CoreOS when I was having better performance.

user35042
  • 2,681
  • 12
  • 34
  • 60
  • Version numbers would help. It sounds like you're describing a scenario in which your fsync() calls weren't making it down the stack (being acknowledged early), and now they operate as "expected". Some information on what your storage stack looks like (both at the FS and block layer) would help to identify if that's the case as well. – Spooler Dec 20 '18 at 15:47
  • Added more information on the version and filesystem. – user35042 Dec 20 '18 at 16:03

0 Answers0