AL2022 Neuron Support

[Realmonia]

Summary

Merge feature/al2022neuron branch to main

Implementation details

• Add new target to Makefile
• Create al2022neu.pkr.hcl and modify al2022.pkr.hcl build sources to include al2022neu
• Modify scripts/enable-ecs-agent-inferentia-support.sh to also run when AMI_TYPE is al2022neu. Skip neuron tools installation when it is on AL2022.
• Run scripts/enable-ecs-agent-inferentia-support.sh to install neuron packages on AL2022 base instance and enable INF on instance.
• Add a reboot before enable-ecs-agent-inferentia-support.sh script to install possible kernel upgrade (when release distribution is not latest on base AMI).
• Some other updates

Testing

New tests cover the changes: No

Manual Tests:
Ran REGION=us-west-2 make al2022neu and successfully created an private AMI with name unofficial-amzn2022-ami-ecs-neu-hvm-2022.0.20220831-x86_64-ebs, id ami-0cb05523683ed2eb6

Launched inf1/trn1 instances with this AMI and ran a busybox task on this instance.
trn1.2xlarge instance gets 1 neuron device
trn1.32xlarge instance gets 16 neuron devices

Functional test:

./bin/run-test -domain desktop -realm us-west-2 -platform al2022neu --imageId ami-0cb05523683ed2eb6

• --- FAIL: TestContainerAccessIntrospection (21.73s). This is a known AL2022 Base AMI issue with route table and not related the specific changes made in this PR.

    run.go:147: Expected callIntrospection to exit with 42; actually exited with 11. logGroup=macis-application-logs logStream=application-logs/callIntrospection/73871a3a91624c1c804ab7a175a8850c
  

• --- FAIL: TestEFS/test-efs-rw-awsvpc (166.33s). Error:

  StoppedReason: "Error response from daemon: create ecs-ecsftest-test-efs-rw-awsvpc-2799b080208bbcb584a2f18b29920839-1-task-efs-write-c0dec88888cedca66e00: VolumeDriver.Create: mounting volume failed: b'mount.nfs4: Failed to resolve server fs-4eb22fe4.efs.us-west-2.amazonaws.com: Name or service not known'",
  

  However, on the affected host, efs url is able to be resolved.

  sh-5.1$ ping fs-4eb22fe4.efs.us-west-2.amazonaws.com
  PING fs-4eb22fe4.efs.us-west-2.amazonaws.com (10.0.65.39) 56(84) bytes of data.
  ^C
  --- fs-4eb22fe4.efs.us-west-2.amazonaws.com ping statistics ---
  1 packets transmitted, 0 received, 100% packet loss, time 0ms
  

  This is likely due to virtual interface issue (ec2-net-utils for AL2022) udev rules configuration incorrectly handles virtual interfaces amazonlinux/amazon-ec2-net-utils#67

• --- FAIL: TestEFS/test-efs-iam-awsvpc (166.30s). This is the sad case test for TestEFS/test-efs-rw-awsvpc and we expect a failure when TestEFS/test-efs-rw-awsvpc failed.

  StoppedReason: "Error response from daemon: create ecs-ecsftest-test-efs-iam-awsvpc-ad6ab14d62cc80293a953381263afeac-4-task-efs-iam-fc9adcaba8c7e1e88e01: Post \"http://%2Frun%2Fdocker%2Fplugins%2Famazon-ecs-volume-plugin.sock/VolumeDriver.Create\": context deadline exceeded"
  

Functional test passed for 20220610 AMI I built before, but test AMI was unfortunately deregistered and the base AL2022 AMI and it's distribution release is also deregistered.

AMI test:

./bin/ami-tests -domain desktop -realm us-west-2 -platform al2022neu --imageId ami-0cb05523683ed2eb6

passed.

Description for the changelog

AL2022 Neuron Support

Licensing

By submitting this pull request, I confirm that my contribution is made under the terms of the Apache 2.0 license.