yahoo-eng-team team mailing list archive

[Nils Meyer <1828611@xxxxxxxxxxxxxxxxxx>]

Public bug reported:

Cloud Provider: Amazon Web Services

As is documented in [1], instances on the nitro type hypervisor don't
attach the NVME disks in deterministic order, yielding a different order
of disks, example [2]. This makes it somewhat difficult to format and
partition volumes since you don't know the volume ids beforehand when
creating an instance (in an Autoscaling group for example).

My current thinking is that maybe a sort of special device name (much
like swap / ephemeralX) coul be used to locate a device, for example
ebs:root for the root drive - which is easy to detect, ebs:size=12G[0]
for the first volume found with 12GiB size. With an appropriate instance
profile and boto3 more elaborate selectors are conceivable (for example
based on tags).

Further complicating things is that the metadata endpoint doesn't expose
the correct device names, opting instead for fantasy names (sda1 for the
root volume, sdX for other volumes).

My Workaround for a 2 volume instance: Try and format both devices, then mount the disk by label:
#cloud-config
fs_setup:
  - label: mylabel
    device: /dev/nvme0n1
    filesystem: xfs
    partition: none
    overwrite: false
  - label: mylabel
    device: /dev/nvme1n1
    filesystem: xfs
    partition: none
    overwrite: false

mounts:
- ["/dev/disk/by-label/mylabel", "/mnt/label", "xfs", "defaults"]

[1] https://docs.aws.amazon.com/AWSEC2/latest/UserGuide/nvme-ebs-volumes.html
[2] https://gist.github.com/nilsmeyer/eddcfa4b7fc5b04ebc0be9eaa3c7b7dd

** Affects: cloud-init
     Importance: Undecided
         Status: New

-- 
You received this bug notification because you are a member of Yahoo!
Engineering Team, which is subscribed to cloud-init.
https://bugs.launchpad.net/bugs/1828611

Title:
  disk_setup could deal with non-deterministic device naming on EC2
  nitro instances

Status in cloud-init:
  New

Bug description:
  Cloud Provider: Amazon Web Services

  As is documented in [1], instances on the nitro type hypervisor don't
  attach the NVME disks in deterministic order, yielding a different
  order of disks, example [2]. This makes it somewhat difficult to
  format and partition volumes since you don't know the volume ids
  beforehand when creating an instance (in an Autoscaling group for
  example).

  My current thinking is that maybe a sort of special device name (much
  like swap / ephemeralX) coul be used to locate a device, for example
  ebs:root for the root drive - which is easy to detect, ebs:size=12G[0]
  for the first volume found with 12GiB size. With an appropriate
  instance profile and boto3 more elaborate selectors are conceivable
  (for example based on tags).

  Further complicating things is that the metadata endpoint doesn't
  expose the correct device names, opting instead for fantasy names
  (sda1 for the root volume, sdX for other volumes).

  My Workaround for a 2 volume instance: Try and format both devices, then mount the disk by label:
  #cloud-config
  fs_setup:
    - label: mylabel
      device: /dev/nvme0n1
      filesystem: xfs
      partition: none
      overwrite: false
    - label: mylabel
      device: /dev/nvme1n1
      filesystem: xfs
      partition: none
      overwrite: false

  mounts:
  - ["/dev/disk/by-label/mylabel", "/mnt/label", "xfs", "defaults"]

  [1] https://docs.aws.amazon.com/AWSEC2/latest/UserGuide/nvme-ebs-volumes.html
  [2] https://gist.github.com/nilsmeyer/eddcfa4b7fc5b04ebc0be9eaa3c7b7dd

To manage notifications about this bug go to:
https://bugs.launchpad.net/cloud-init/+bug/1828611/+subscriptions