Incorrect handling of nonexistent instances

[ampedandwired]

Incorrect handling of nonexistent instances

[ampedandwired]

BTW, in the current implementation if the instance can't be found in the AWS API, it does not get deleted in Rancher. I figured it best to err on the side of caution in cases where it is not definitively clear that the instance has terminated.

If you feel it is useful though, I could add a flag to force deletion in cases like this?

[ampedandwired]

Thanks for the bug report. The code currently handles an Aws::EC2::Errors::InvalidInstanceIDNotFound exception, which is what I expected would be thrown if the instance has been purged from AWS. Looks like there might be another scenario though. I'll have a look into it.

[Troyhy]

I have same situation. I have staging environment that is only active from 8-16. In the morning when staging environment goes up, amazon has already purged all info about machine that was serving last afternoon and I get this error. Environment consist only one machine at the moment so reaping cannot happen 16:00 when environment goes down.

It would be helpful to add flag to reap host in this situation.

[ampedandwired]

The original issue is actually due to a bug in the AWS SDK: aws/aws-sdk-ruby#1449 - this should be fixed in 0.4.0.

Also, after giving this some thought, I have updated the way rancher-reaper determines if a host is terminated. It now assumes that if the host was not returned by the AWS API, then it is terminated. This should solve the use case mentioned by @Troyhy.