corosync - KNET - MTU Warnings (MTU: 1500 vs 1397)

[mohnewald]

Hello,

i get a lot of logs like this. Is it something to worry about?:

Dec 20 08:23:48 node01 corosync[2647]:   [KNET  ] udp: detected kernel MTU: 1500

Dec 20 08:23:48 node01 corosync[2647]:   [KNET  ] pmtud: PMTUD completed for host: 3 link: 0 current link mtu: 1397

Dec 20 08:23:48 node01 corosync[2647]:   [KNET  ] pmtud: Starting PMTUD for host: 3 link: 1

Dec 20 08:23:48 node01 corosync[2647]:   [KNET  ] udp: detected kernel MTU: 1500

Dec 20 08:23:48 node01 corosync[2647]:   [KNET  ] pmtud: PMTUD completed for host: 3 link: 1 current link mtu: 1397

Dec 20 08:23:48 node01 corosync[2647]:   [KNET  ] pmtud: Starting PMTUD for host: 2 link: 0

Dec 20 08:23:48 node01 corosync[2647]:   [KNET  ] udp: detected kernel MTU: 1500

Dec 20 08:23:48 node01 corosync[2647]:   [KNET  ] pmtud: PMTUD completed for host: 2 link: 0 current link mtu: 1397

Dec 20 08:23:48 node01 corosync[2647]:   [KNET  ] pmtud: Starting PMTUD for host: 2 link: 1

Dec 20 08:23:48 node01 corosync[2647]:   [KNET  ] udp: detected kernel MTU: 1500

Dec 20 08:23:48 node01 corosync[2647]:   [KNET  ] pmtud: PMTUD completed for host: 2 link: 1 current link mtu: 1397

I turned debugging on, changed token and token_retransmits_before_loss_const since i had some link up/down events.
Since my physical MTU ist greater than the corosync mtu 1397 i dont see a problem?

~# cat /etc/pve/corosync.conf

logging {
  debug: on
  to_syslog: yes
}

nodelist {
  node {
    name: node01
    nodeid: 1
    quorum_votes: 1
    ring0_addr: 172.16.30.230
    ring1_addr: 10.10.70.1
  }
  node {
    name: node02
    nodeid: 2
    quorum_votes: 1
    ring0_addr: 172.16.30.231
    ring1_addr: 10.10.70.2
  }
  node {
    name: node03
    nodeid: 3
    quorum_votes: 1
    ring0_addr: 172.16.30.232
    ring1_addr: 10.10.70.3
  }
}

quorum {
  provider: corosync_votequorum
}

totem {
  token: 5000
  token_retransmits_before_loss_const: 10

  cluster_name: cluster
  config_version: 4
  interface {
    linknumber: 0
  }
  interface {
    linknumber: 1
  }
  ip_version: ipv4-6
  link_mode: passive
  secauth: on
  version: 2
}

Cheers,
Michael

 

[spirit]

This log is the autodetection of the best mtu by corosync itself, no need to worry about it.

link/down events could be problematic (if you really don't have real down).

your cluster is really small, until you have your nodes on long distance remote locations, you shouldn't have too tune the token && token_retransmits_before_loss_const. (the token value is auto increasing with the number the nodes of the cluster, so be carefull if you fix it manually)

Maybe check if you don't have any network saturation when you see "down" events. (when backup are running for example)

 

[mohnewald]

Hello spirit.

Thanks for the feedback.