https://aws.amazon.com/blogs/aws/elastic-network-adapter-high-performance-network-interface-for-amazon-ec2/
we support this in our kernel so we should probably enable it. We need to:
1 - enable it in fedimg and underlying tools if needed 2 - see if having it enabled on all AMIs affects being able to launch on certain instance types (we would like to be able to enable it on all AMIs and just have it be a no-op on instances that don't support it.
cc @sayanchowdhury
Metadata Update from @dustymabe: - Issue tagged with: infra
Metadata Update from @dustymabe: - Issue assigned to sayanchowdhury
another link: http://docs.aws.amazon.com/AWSEC2/latest/UserGuide/enhanced-networking-ena.html
Let's take into account several NVMe related requirements as well.
It is recommended that the nvme.io.timeout value be set to a high number. This requires the following patch, accepted in 4.14: http://lists.infradead.org/pipermail/linux-nvme/2017-September/012701.html
http://docs.aws.amazon.com/AWSEC2/latest/UserGuide/nvme-ebs-volumes.html shows setting a high timeout for the nvme devices as well. This is set to the uint_max in Amazon Linux because it is expected that when the instance becomes available, the volume will be available as well.
With Amazon Linux AMI 2017.09.01 or later, you can run the ebsnvme-id command to map the NVMe device name to a volume ID and device name. This is a part of the ec2-utils applications included in the Amazon Linux source and will be a helpful thing to incorporate for support of the C5 and future instance types.
with the f28 AMIs released today this should be resolved. Please let us know if that doesn't reflect reality.
Metadata Update from @dustymabe: - Issue close_status updated to: Fixed - Issue status updated to: Closed (was: Open)
Metadata Update from @sayanchowdhury: - Issue status updated to: Open (was: Closed)
The AMIs seems not to reflect the ENA changes so opening the ticket.
➜ ~ aws ec2 describe-images --image-id ami-eb63fa94 --query 'Images[].EnaSupport' ➜ ~
hmm. I thought @davdunc confirmed for me before I closed the ticket.
I did verify that the instance types were not blocked from launch with the images at one time. That requires the ena flag to be set . The kernel support for ena is confirmed. I will need to investigate the requirements for libcloud.
I did verify that the instance types were not blocked from launch with the images at one time. That requires the ena flag to be set .
Can you check the f28 images from getfedora.org ? I think i linked you to them and you gave me a :thumbsup: , which is why I closed this ticket.
https://libcloud.readthedocs.io/en/latest/compute/drivers/ec2.html#libcloud.compute.drivers.ec2.BaseEC2NodeDriver.ex_register_image shows the boolean "ena_support=None" as a component. Is it possible to deregister and then register the snapshot with the ena_support parameter set to true?
images from getfedora.org verified. I launched an i3.metal instance and that requires the ena_support=True attribute to launch at all. Confirming that the ena_support attribute is set on current images.
Confirmed.
I did a provision of the stg boxes last night and re-ran the command seems like the images do have the ena support. I'll go ahead and close the issue.
[root@fedimg01 ~][STG]# aws ec2 describe-images --image-id ami-eb63fa94 --query 'Images[].EnaSupport' [ true ] [root@fedimg01 ~][STG]# aws ec2 describe-images --image-id ami-f54c298a --query 'Images[].EnaSupport' [ true ]
Thanks @davdunc for looking into it.
Metadata Update from @sayanchowdhury: - Issue close_status updated to: Fixed - Issue status updated to: Closed (was: Open)
Log in to comment on this ticket.