0

We run a bunch of Ubuntu instances on Amazon EC2 and over the last two months we've observed a few of them becoming unresponsive and unreachable. No SSH, no ping responses, etc. Eventually rebooting those instances helped as a last resort.

All of the affected instances are EBS-based. There are no indications as to the reason of failure in the logs. We've been suspecting that maybe the instances could somehow drop the connection to their EBS volumes.

My questions are:

  1. Is this even possible (I mean: an instance losing the connection to its EBS volume)?
  2. How do I go about debugging this and ensuring this is the case?
  3. Most importantly: how do I prevent this?
5

As of now EBS is the most reliable option you have on AWS. It's not only better, its also easy to take snapshots and re-attach to another instance. We have very large Ec2 instances which are EBS based that are used by media site and have not had any issue like this. Apart from that Amazon guarantees 99.95% uptime for this service and will give a refund incase it goes down.

To answer your questions.

  1. Its quite unlikely, unless your instance health is bad for some reason. But similar issues, with bad IOPS has been reported in past. See
  2. Check the "read write IO" vs "pending IO in the queue". So if you have 0 IO when there is pending IO in queue you have a problem. See SLA.
  3. Check health of an instance + the health of EBS IOPs. You can use cloudwatch for this to some extend. Check this link.

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.