What's a Totem "Retransmit List" all about in Corosync?

Posted on Thu 15 March 2012 in hints-and-kinks • 2 min read

Occasionally, you may see errors similar to this in your system logs:

corosync [TOTEM ] Retransmit List: e4 e5 e7 e8 ea eb ed ee

Here’s what causes them, and what you can do to fix the issue.

Corosync, more specifically its Totem protocol implementation, defines a maximum number of cluster messages that can be sent during one token rotation. By default, that number is 50, but you may modify this value by setting the window_size parameter in your corosync.conf configuration file.

When among several fast cluster nodes (“processors” in Totem speak) there are one or few slow ones, the kernel receive buffers can’t cope, messages get lost, and they then need to be retransmitted. This is what causes the Retransmit List notifications in the syslogs. This doesn’t mean you’re losing any messages or data. But it does mean that your cluster performance degrades when this happens, and thus you should really fix that problem.

There are a few considerations that apply to tuning Corosync’s window_size:

  • If you have a small cluster (say, 8 nodes or less), and they all can be expected to perform equally well because they have identical or nearly-identical hardware, then setting a large window_size of up to 300 should be fine.

  • If your cluster is rather heterogeneous, then you should probably stick with the default of 50. Definitely don’t go higher than 256000/MTU, where MTU is that of the network interface(s) Corosync communicates over. For a standard Ethernet interface the default MTU is 1500, which would make for a maximum window_size of 170.

  • If you’re running on the generally safe default of 50, and you’re still getting Retransmit List notifications, then one of your nodes is most likely significantly slower than the others, and you had better find the cause of that and fix it. The node could be under constant excessive load, or have a problem with its network driver, or may be plugged into an incorrectly-configured switch port.

This article originally appeared on the hastexo.com website (now defunct).