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.

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

update flannel v0.24.2 #5071

Merged
merged 5 commits into from
Mar 12, 2024
Merged

update flannel v0.24.2 #5071

merged 5 commits into from
Mar 12, 2024

Conversation

laverya
Copy link
Member

@laverya laverya commented Feb 9, 2024

What this PR does / why we need it:

Adds a systemd drop-in to disable TCP checksum offloading for VMware VMXNET3 NIC drivers - this is defaulted to ON in modern Linux kernels, but in certain VMWare deployments, the checksum offloading is handled incorrectly leading to dropped packets on the interface. Detecting it is difficult because it only manifests in multi-node configurations once additional nodes have been added, but host networking works just fine so kURL does not detect a problem at installation time. Disabling this for all VMXNET3 driver users seems like a safe assumption and disabling TCP checksum offloading incurs very little performance penalty.

The ethtool command can be used to fix the problem in-situ but changes made with that tool are not persistent, so this systemd unit hooks into the flannel device unit file and disables the offloading option whenever the flannel interface is configured.

Researching the issue

See also

Which issue(s) this PR fixes:

Fixes https://app.shortcut.com/replicated/story/91846/hostpreflight-detect-if-host-nic-uses-vmxnet3-driver-from-vmware-and-if-so-warn-about-vxlan-checksum-offloading

Special notes for your reviewer:

Steps to reproduce

  1. Install a VMWare ESXi cluster with NSX networking enabled for software-defined-networking at the VMWare cluster level
  2. configure >2 VMs with the VMXNet3 NIC driver
  3. install kurl on primary
  4. add secondary
  5. observe that pod-pod traffic between nodes fails silently

Does this PR introduce a user-facing change?

NONE

Does this PR require documentation?

NONE

@laverya laverya added the type::feature An enhancement to an existing add on or feature label Feb 9, 2024
@laverya laverya requested a review from a team as a code owner February 9, 2024 19:19
Copy link
Contributor

github-actions bot commented Feb 9, 2024

@ricardomaraschini
Copy link
Contributor

I don't want this to become bigger and bigger but shouldn't we be also shipping the ethtool host package if we are using it ?

@laverya
Copy link
Member Author

laverya commented Feb 12, 2024

Possibly, but also any server that needs this change (running in vmware with virtualization packages) probably already has it

@adamancini
Copy link
Member

adamancini commented Feb 13, 2024

I created #5076 because I had to add some fixes to my code, and I don't have push access to replicatedhq/kURL repo

Copy link
Contributor

github-actions bot commented Mar 8, 2024

@adamancini
Copy link
Member

@laverya ready for review again, thanks!

Copy link
Contributor

github-actions bot commented Mar 8, 2024

Copy link
Member Author

@laverya laverya left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Looks good - I can't approve because I'm the PR author

Copy link
Member

@adamancini adamancini left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@laverya approves

@adamancini adamancini merged commit 82513b3 into main Mar 12, 2024
9 of 10 checks passed
@adamancini adamancini deleted the adamancini/flannel-ethtool branch March 12, 2024 15:03
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
flannel type::feature An enhancement to an existing add on or feature
Projects
None yet
Development

Successfully merging this pull request may close these issues.

3 participants