Cilium 1.15.5 on Azure reduces MTU on eth0, breaks connectivity

[csomm]

Cilium 1.15.5 on Azure reduces MTU on eth0, breaks connectivity

[csomm]

Can you please clarify from what version you upgraded? (meaning, what's the latest "known good" version)

The latest known good version is 1.15.4. The problems began about two days after upgrading from 1.15.4 to 1.15.5.

And after downgrading to 1.15.4 and rebooting all nodes, all eth0 interfaces have MTU 1500 again (checked on multiple clusters).

[borkmann]

Looking at the changes from 1.15.4 to 1.15.5, this one stands out around MTU: #32244

https://github.com/cilium/cilium/releases/tag/v1.15.5

Cc @learnitall could you take a look? Thanks a lot

[julianwiedmann]

We are using Cilium on our AKS clusters on Azure. After upgrading to Cilium 1.15.5 we sometimes

Can you please clarify from what version you upgraded? (meaning, what's the latest "known good" version)

[learnitall]

Taking a look 👁️

[learnitall]

Hey @csomm, this was a regression introduced by #32244. The changes in that PR surfaced a bug that we're investigating as part of #33303. We're reverting the blamed PR in #33304 which should fix the issue. Thank you for bug report, I really appreciate it.

Thanks @borkmann for calling out #32244. I hadn't considered it, but your bisect logic makes perfect sense.