amazon-eks-ami icon indicating copy to clipboard operation
amazon-eks-ami copied to clipboard

Published 20 hours ago • verified publisher icon awslabs

Add support for raid10

Open lassizci opened this issue 1 year ago • 4 comments

Description of changes: I would like to be able to migrate workloads away from a node gracefully in case of instance storage drive failure. Raid10 would provide redundancy and trade off disk space.

Adding support for creating raid10 in addition to raid0. This also removes the wait block for raid resync for two reasons:

  1. raid0 does not have redundancy and therefore no initial resync[1]
  2. with raid10 the resync time for 4x 1.9TB disks takes from tens of minutes to multiple hours, depending on sysctl params dev.raid.speed_limit_min and dev.raid.speed_limit_max and the speed of the disks. Initial resync for raid10 is not strictly needed[1]

filesystem creation: by default mkfs.xfs attempts to TRIM the drive. This is also something that can take tens of minutes or hours, depening on the size of drives. TRIM can be skipped, as instances are delivered with disks fully trimmed[2].

[1] https://raid.wiki.kernel.org/index.php/Initial_Array_Creation [2] https://docs.aws.amazon.com/AWSEC2/latest/UserGuide/ssd-instance-store.html#InstanceStoreTrimSupport

Testing Done

on m6id.metal with kernel defaults:

# uname -a
Linux ip-10-24-0-65.eu-west-1.compute.internal 5.10.199-190.747.amzn2.x86_64 #1 SMP Sat Nov 4 16:55:14 UTC 2023 x86_64 x86_64 x86_64 GNU/Linux

#  sysctl dev.raid.speed_limit_min
dev.raid.speed_limit_min = 1000

# sysctl dev.raid.speed_limit_max
dev.raid.speed_limit_max = 200000

# mdadm --create --force --verbose /dev/md/kubernetes --level=10 --name=kubernetes --raid-devices=4 /dev/nvme0n1 /dev/nvme1n1 /dev/nvme2n1 /dev/nvme3n1
mdadm: layout defaults to n2
mdadm: layout defaults to n2
mdadm: chunk size defaults to 512K
mdadm: size set to 1855337472K
mdadm: automatically enabling write-intent bitmap on large array
mdadm: Defaulting to version 1.2 metadata
mdadm: array /dev/md/kubernetes started.

# cat /proc/mdstat 
Personalities : [raid10] 
md127 : active raid10 nvme3n1[3] nvme2n1[2] nvme1n1[1] nvme0n1[0]
      3710674944 blocks super 1.2 512K chunks 2 near-copies [4/4] [UUUU]
      [>....................]  resync =  1.1% (41396352/3710674944) finish=304.3min speed=200910K/sec
      bitmap: 28/28 pages [112KB], 65536KB chunk

With increased resync limits:

# sysctl -w dev.raid.speed_limit_min=2146999999 ; sysctl -w dev.raid.speed_limit_max=2146999999
dev.raid.speed_limit_min = 2146999999
dev.raid.speed_limit_max = 2146999999

# cat /proc/mdstat 
Personalities : [raid10] 
md127 : active raid10 nvme3n1[3] nvme2n1[2] nvme1n1[1] nvme0n1[0]
      3710674944 blocks super 1.2 512K chunks 2 near-copies [4/4] [UUUU]
      [===>.................]  resync = 19.9% (740172096/3710674944) finish=20.4min speed=2418848K/sec
      bitmap: 23/28 pages [92KB], 65536KB chunk

Due to some... accidents, this replaces the old pull request: https://github.com/awslabs/amazon-eks-ami/pull/1549. There's further discussions about the details.

lassizci avatar Jul 26 '24 16:07 lassizci

The original PR has enough context, https://github.com/awslabs/amazon-eks-ami/pull/1549. The change LGTM. My question for this would be do we need to add this experimental feature for AL23(nodeadm) as well?

Issacwww avatar Jul 26 '24 17:07 Issacwww

The original PR has enough context, #1549. The change LGTM. My question for this would be do we need to add this experimental feature for AL23(nodeadm) as well?

We should be able to pass anything that setup-local-disks accepts via spec.instance.localStorage.strategy so i think it should work OOTB. A doc update there would be appreciated though!

ndbaker1 avatar Jul 27 '24 02:07 ndbaker1

The original PR has enough context, #1549. The change LGTM. My question for this would be do we need to add this experimental feature for AL23(nodeadm) as well?

We should be able to pass anything that setup-local-disks accepts via spec.instance.localStorage.strategy so i think it should work OOTB. A doc update there would be appreciated though!

Added a documentation update. A question regarding the CRD though: the validation changes, but the versioning isn't changed. Would that actually need v1alpha2 or what's the policy here?

lassizci avatar Oct 15 '24 12:10 lassizci

the versioning isn't changed. Would that actually need v1alpha2 or what's the policy here?

We only bump the version on breaking changes, this is purely additive so we're fine to stay on v1alpha1.

cartermckinnon avatar Oct 16 '24 17:10 cartermckinnon

Are there any further things needed to get the raid10 support?

lassizci avatar Jan 20 '25 19:01 lassizci

@lassizci thanks for your work on this! Giving the CI a final run before we merge.

/ci

cartermckinnon avatar Feb 21 '25 18:02 cartermckinnon

@cartermckinnon roger that! I've dispatched a workflow. 👍

github-actions[bot] avatar Feb 21 '25 18:02 github-actions[bot]

@cartermckinnon the workflow that you requested has completed. 🎉

AMI variantBuildTest
1.25 / al2success ✅success ✅
1.25 / al2023failure ❌skipped ⏭️
1.26 / al2success ✅success ✅
1.26 / al2023failure ❌skipped ⏭️
1.27 / al2success ✅success ✅
1.27 / al2023failure ❌skipped ⏭️
1.28 / al2success ✅success ✅
1.28 / al2023failure ❌skipped ⏭️
1.29 / al2success ✅success ✅
1.29 / al2023failure ❌skipped ⏭️
1.30 / al2success ✅success ✅
1.30 / al2023failure ❌skipped ⏭️
1.31 / al2success ✅success ✅
1.31 / al2023failure ❌skipped ⏭️
1.32 / al2success ✅success ✅
1.32 / al2023failure ❌skipped ⏭️

github-actions[bot] avatar Feb 21 '25 19:02 github-actions[bot]

@lassizci thanks for your work on this! Giving the CI a final run before we merge.

/ci

Thanks for coming back! Ok, it seems gotta still dive in to the build more, since they miserably fail for any AL2023. Sorry about that! Should be able to set up a test account to make sure.

lassizci avatar Feb 21 '25 20:02 lassizci

/ci

ndbaker1 avatar Feb 25 '25 22:02 ndbaker1

@ndbaker1 roger that! I've dispatched a workflow. 👍

github-actions[bot] avatar Feb 25 '25 22:02 github-actions[bot]

@ndbaker1 the workflow that you requested has completed. 🎉

AMI variantBuildTest
1.25 / al2success ✅success ✅
1.25 / al2023success ✅success ✅
1.26 / al2success ✅success ✅
1.26 / al2023success ✅success ✅
1.27 / al2success ✅success ✅
1.27 / al2023success ✅success ✅
1.28 / al2success ✅success ✅
1.28 / al2023success ✅success ✅
1.29 / al2success ✅success ✅
1.29 / al2023success ✅success ✅
1.30 / al2success ✅success ✅
1.30 / al2023success ✅success ✅
1.31 / al2success ✅success ✅
1.31 / al2023success ✅success ✅
1.32 / al2success ✅success ✅
1.32 / al2023success ✅success ✅

github-actions[bot] avatar Feb 25 '25 23:02 github-actions[bot]