In the development version, the netdev collector fails on kernel 5.19 (and later)

[siebenmann]

Host operating system: output of uname -a

Linux hawkwind.cs 5.19.4-200.fc36.x86_64 #1 SMP PREEMPT_DYNAMIC Thu Aug 25 17:42:04 UTC 2022 x86_64 x86_64 x86_64 GNU/Linux

node_exporter version: output of node_exporter --version

node_exporter, version 1.4.0-rc.0 (branch: master, revision: 46748fd229e249c0d7a9198a98e6f9450d437478)
  build user:       cks@hawkwind.cs
  build date:       20220901-15:41:49
  go version:       go1.19
  platform:         linux/amd64

node_exporter command line flags

Only --web.listen-address :8100 so that it can run at the same time as my normal install of node_exporter.

Are you running node_exporter in Docker?

No.

What did you do that produced an error?

Requested metrics from the /metrics endpoint (http://localhost:8100/metrics with my command line flag). On kernel 5.19.4 (now the default kernel in Fedora), this produced the error: ts=2022-09-01T15:42:27.342Z caller=collector.go:169 level=error msg="collector failed" name=netdev duration_seconds=0.006418612 err="couldn't get netstats: incorrect size, want: 184 or 192"

This error comes from github.com/jsimonetti/rtnetlink and is already fixed in v1.2.2 of rtnetlink, through jsimonetti/rtnetlink#146 so I think all that's needed to fix this is to update the rtnetlink version in node_exporter's go.mod. A version of the development version built with this change appears to work for me.

This issue doesn't happen with the released node_exporter 1.3.1 as far as I can tell, although I don't know if its netdev metrics are correct on kernel 5.19. (I'm not actually running 1.3.1 as my live node_exporter due to issue #2068.)

[SuperQ]

Yes, we made changes to get more data over netlink. Updating the netlink library should be easy enough.

[SuperQ]

Ok, I bumped the library, it'll be in the next release (soon).