[pve-devel] corosync, multicast problem because of vmbr multicast_snooping enabled

Alexandre DERUMIER aderumier at odiso.com
Sun Mar 3 18:02:15 CET 2013


Hi,

I had worked on a problem this week,

Since some month, I have corosync random failures.
Thanks to omping, I had check the multicast layer, and it seem that sometime it's hanging.

Also, I have had a strange thing, shutdown down a node had killed multicast between all my others nodes ....

I had found article here:
https://lists.linux-foundation.org/pipermail/openais/2011-July/016491.html

Telling that multicast_snooping on bridge can lead to problems.
(Multicast snooping is used to filter multicast, listen igmp packet to create groups, and avoid to send multicast to all ports).

I don't known how exactly it's works with linux bridge, But as my proxmox host ip is on vmbr0 (the default proxmox setup),

 echo 0 > /sys/class/net/vmbr0/bridge/multicast_snooping

has resolved all my problems.


I'm not sure, but I think we should disable multicast_snooping by default. (Don't known if it can cause regression ?)





More information about the pve-devel mailing list