[LU-17729] LNET goes down often in AKS deployments

• Type: Bug
• Resolution: Unresolved
• Priority: Minor
• Fix Version/s: Lustre 2.16.0
• Labels: msft
• Severity: 3
• Assignee: Ellis Wilson
• Reporter: Ellis Wilson

Description

LNET goes down every time network interfaces change.  This happens with IPv6 every time a pod is added or removed from the system when using the kubenet networking plugin.

The dmesg output looks like this:
[68345.628473] IPv6: ADDRCONF(NETDEV_CHANGE): eth0: link becomes ready
[68345.628503] IPv6: ADDRCONF(NETDEV_CHANGE): vethde7a610d: link becomes ready
[68345.628535] cbr0: port 1(vethde7a610d) entered blocking state
[68345.628536] cbr0: port 1(vethde7a610d) entered forwarding state
[68433.354593] cbr0: port 1(vethde7a610d) entered disabled state
[68433.357102] device vethde7a610d left promiscuous mode
[68433.357112] cbr0: port 1(vethde7a610d) entered disabled state
[68518.266748] LNet: Added LNI 10.224.0.5@tcp [8/256/0/180]    ***LNET ACTIVE HERE**
[68518.266823] LNet: Accept secure, port 988
[68536.743566] cbr0: port 1(veth3c5b73b5) entered blocking state  **POD CHANGE**
[68536.743568] cbr0: port 1(veth3c5b73b5) entered disabled state
[68536.743800] device veth3c5b73b5 entered promiscuous mode
[68536.749545] IPv6: ADDRCONF(NETDEV_CHANGE): eth0: link becomes ready  **POD CHANGE**
[68536.749570] IPv6: ADDRCONF(NETDEV_CHANGE): veth3c5b73b5: link becomes ready
[68536.749593] cbr0: port 1(veth3c5b73b5) entered blocking state
[68536.749595] cbr0: port 1(veth3c5b73b5) entered forwarding state
[68767.344431] cbr0: port 1(veth3c5b73b5) entered disabled state
[68767.349742] device veth3c5b73b5 left promiscuous mode
[68767.349751] cbr0: port 1(veth3c5b73b5) entered disabled state

This is the result of ksocknal_handle_link_state_change and ksocknal_handle_inetaddr_change not being namespace aware.

Patch to be sent shortly.