Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Support attaching multiple network interfaces to the same network card #2855

Merged
merged 1 commit into from
Dec 19, 2024
Merged

Support attaching multiple network interfaces to the same network card #2855

merged 1 commit into from
Dec 19, 2024

Conversation

hanwen-cluster
Copy link
Contributor

@hanwen-cluster hanwen-cluster commented Dec 18, 2024 •
edited
Loading

Prior to this commit, the code had two assumptions:

  1. Single network card instances can have only one network interface
    1.1. therefore, when there are more than one network interface, the code made a IMDS call to retrieve device index (network interface index), which is only available on instances with multiple network cards. Having a secondary network interface on single network card instance failed the code and caused instance launch failures.
  2. Each network card can have only one network interface
    2.1. therefore, the route table is unique to each network card. Having multiple network interfaces on a network card confused the code and generated wrong route tables.

To fix (1), this commit uses a fallback value 0 when retrieval of device index fails.
To fix (2), this commit names the route tables in a way that is unique to network interface and network card.

FYI:

  1. Network card is the physical card. Network interface is the virtual concept. Each network card can have multiple network interfaces (which in AWS is 1 or 2)
  2. "Network interface" is synonym to "device"

Tests

  • Manually modified launch template to add secondary network interface to compute nodes. Jobs ran successfully. SSH works on both interfaces.

Checklist

  • Make sure you are pointing to the right branch.
  • If you're creating a patch for a branch other than develop add the branch name as prefix in the PR title (e.g. [release-3.6]).
  • Check all commits' messages are clear, describing what and why vs how.
  • Make sure to have added unit tests or integration tests to cover the new/modified code.
  • Check if documentation is impacted by this change.

Please review the guidelines for contributing and Pull Request Instructions.

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

@hanwen-cluster
Support attaching multiple network interfaces to the same network card
d1c7232 
Prior to this commit, the code had two assumptions:
1. Single network card instances can have only one network interface
 a. therefore, when there are more than one network interface, the code made a IMDS call to retrieve device index (network interface index), which is only available on instances with multiple network cards. Having a secondary network interface on single network card instance failed the code and caused instance launch failures.
2. Each network card can have only one network interface
 a. therefore, the route table is unique to each network card. Having multiple network interfaces on a network card confused the code and generated wrong route tables.

To fix (1), this commit uses a fallback value 0 when retrieval of device index fails.
To fix (2), this commit names the route tables in a way that is unique to network interface and network card.

FYI:
1. Network card is the physical card. Network interface is the virtual concept. Each network card can have multiple network interfaces (which is AWS is 1 or 2)
2. "Network interface" is synonym to "device"

Signed-off-by: Hanwen <hanwenli@amazon.com>
@hanwen-cluster hanwen-cluster requested review from a team as code owners December 18, 2024 20:43
@hanwen-cluster hanwen-cluster merged commit cb83f4c into aws:develop Dec 19, 2024
33 of 44 checks passed
hanwen-cluster added a commit to hanwen-cluster/aws-parallelcluster-cookbook that referenced this pull request Dec 20, 2024
@hanwen-cluster
Raise priority of ParallelCluster configured route tables
2606279 
This commit fixes a bug from aws#2855, which made the route table/metric number larger (meaning lower priority). Thereafter, some unwanted default rules on AL2023 took priority and failed test_multiple_nics integration test on AL2023.

This commit makes the number smaller (meaning higher priority) to fix the issue.
e.g.
Prior to this commit, the number for table for 1,1 is 1001001. After this commit, the number is 101+10=111. The "+10" is to properly handle table for 0,0, which has number 10. Without "+10", the table would conflict with table 0 from OS.

FYI: the number of unwanted default AL2023 rule starts with 10101

Signed-off-by: Hanwen <hanwenli@amazon.com>
@hanwen-cluster hanwen-cluster mentioned this pull request Dec 20, 2024
Merged
hgreebe pushed a commit that referenced this pull request Dec 24, 2024
@hanwen-cluster
Raise priority of ParallelCluster configured route tables (#2857)
e479a86 
This commit fixes a bug from #2855, which made the route table/metric number larger (meaning lower priority). Thereafter, some unwanted default rules on AL2023 took priority and failed test_multiple_nics integration test on AL2023.

This commit makes the number smaller (meaning higher priority) to fix the issue.
e.g.
Prior to this commit, the number for table for 1,1 is 1001001. After this commit, the number is 101+10=111. The "+10" is to properly handle table for 0,0, which has number 10. Without "+10", the table would conflict with table 0 from OS.

FYI: the number of unwanted default AL2023 rule starts with 10101

Signed-off-by: Hanwen <hanwenli@amazon.com>
hanwen-cluster added a commit to hanwen-cluster/aws-parallelcluster-cookbook that referenced this pull request Dec 31, 2024
@hanwen-cluster
Configure proper priority for route tables
ac3e225 
aws#2855 made the pcluster route table/metric number larger (meaning lower priority). Thereafter, some unwanted default rules on AL2023 ec2-net-utils took priority and failed test_multiple_nics integration test on AL2023.
Then, aws#2857 made the number too small, interfering route table configurations from IMDS on AL2.

Therefore, this commit tries to imitate the priority prior to these two PRs. This is not the cleanest fix, because it is staying in the lucky priority rand instead of fully resolving the issue (i.e. prevent IMDS and ec2-net-utils from configuring the route tables). However, this commit is the least breaking change. So I propose to go with this commit.

Metric number range before the two PRs
Network card (0,0): 1000
Network card (0,1): 1000 (which was causing conflicts and the reason for all these PRs)
Network card (n,1): 100n (for p5, which has 32 network card, it will be 1000-10031)

Metric number range after the first PR:
Network card (0,0): 1000000
Network card (0,1): 1000001 (conflict fixed :) )
Network card (n,1): 1000001+n*1000 (for p5, it will be 1000000-1031001)

Metric number range after the second PR:
Network card (0,0): 10
Network card (0,1): 75
Network card (n,1): 0x(hexadecimal number)n01+10 (for p5, it will be 10-12555. The hexadecimal number was accidentally introduced because bash automatically interpret numbers start with "00" as hexadecimal number)

Metric number range after this commit:
Network card (0,0): 1000
Network card (0,1): 1001
Network card (n,1): n01+1000 (for p5, it will be 1000-4101)

Signed-off-by: Hanwen <hanwenli@amazon.com>
@hanwen-cluster hanwen-cluster mentioned this pull request Dec 31, 2024
Merged
hanwen-cluster added a commit that referenced this pull request Dec 31, 2024
@hanwen-cluster
Configure proper priority for route tables
e3949f3 
#2855 made the pcluster route table/metric number larger (meaning lower priority). Thereafter, some unwanted default rules on AL2023 ec2-net-utils took priority and failed test_multiple_nics integration test on AL2023.
Then, #2857 made the number too small, interfering route table configurations from IMDS on AL2.

Therefore, this commit tries to imitate the priority prior to these two PRs. This is not the cleanest fix, because it is staying in the lucky priority rand instead of fully resolving the issue (i.e. prevent IMDS and ec2-net-utils from configuring the route tables). However, this commit is the least breaking change. So I propose to go with this commit.

Metric number range before the two PRs
Network card (0,0): 1000
Network card (0,1): 1000 (which was causing conflicts and the reason for all these PRs)
Network card (n,1): 100n (for p5, which has 32 network card, it will be 1000-10031)

Metric number range after the first PR:
Network card (0,0): 1000000
Network card (0,1): 1000001 (conflict fixed :) )
Network card (n,1): 1000001+n*1000 (for p5, it will be 1000000-1031001)

Metric number range after the second PR:
Network card (0,0): 10
Network card (0,1): 75
Network card (n,1): 0x(hexadecimal number)n01+10 (for p5, it will be 10-12555. The hexadecimal number was accidentally introduced because bash automatically interpret numbers start with "00" as hexadecimal number)

Metric number range after this commit:
Network card (0,0): 1000
Network card (0,1): 1001
Network card (n,1): n01+1000 (for p5, it will be 1000-4101)

Signed-off-by: Hanwen <hanwenli@amazon.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers

@hgreebe hgreebe hgreebe approved these changes

Assignees
No one assigned
Labels
skip-changelog-update
Projects
None yet
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.
2 participants
@hanwen-cluster @hgreebe