[ClusterLabs] Redudant Ring Network failure

[Jan Friesse]

Michael,

> Jan,
> 
> actually we using this.
> 
> [root at lvm-nfscpdata-05ct::~ 100 ]# apt show corosync
> Package: corosync
> Version: 3.0.1-2+deb10u1
> 
> [root at lvm-nfscpdata-05ct::~]# apt show libknet1
> Package: libknet1
> Version: 1.8-2
> 
> This are the newest version provided on Mirror.

yup, but these are pretty old anyway and there is quite a few bugs (and 
many of them may explain behavior you see).

I would suggest you to either try a upstream code compilation or (if you 
need to stick with packages) you may give a try sid packages or Proxmox 
repositories (where knet is at version 1.15 and corosync 3.0.3).

Regards,
   Honza

> 
> 
> 
> 
> Sitz der Gesellschaft / Corporate Headquarters: Deutsche Lufthansa Aktiengesellschaft, Koeln, Registereintragung / Registration: Amtsgericht Koeln HR B 2168
> Vorsitzender des Aufsichtsrats / Chairman of the Supervisory Board: Dr. Karl-Ludwig Kley
> Vorstand / Executive Board: Carsten Spohr (Vorsitzender / Chairman), Thorsten Dirks, Christina Foerster, Harry Hohmeister, Dr. Detlef Kayser, Dr. Michael Niggemann
> 
> 
> -----Ursprüngliche Nachricht-----
> Von: Jan Friesse <jfriesse at redhat.com>
> Gesendet: Mittwoch, 10. Juni 2020 09:24
> An: Cluster Labs - All topics related to open-source clustering welcomed <users at clusterlabs.org>; ROHWEDER-NEUBECK, MICHAEL (EXTERN) <michael.rohweder-neubeck.sp at dlh.de>; users at lists.clusterlabs.org
> Betreff: Re: [ClusterLabs] Redudant Ring Network failure
> 
> Michael,
> what version of knet you are using? We had quite a few problems with older versions of knet, so current stable is recommended (1.16). Same applies for corosync because 3.0.4 has vastly improved display of links status.
> 
>> Hello,
>> We have massive problems with the redundant ring operation of our Corosync / pacemaker 3 Node NFS clusters.
>>
>> Most of the nodes either have an entire ring offline or only 1 node in a ring.
>> Example: (Node1 Ring0 333 Ring1 n33 | Node2 Ring0 033 Ring1 3n3 |
>> Node3 Ring0 333 Ring 1 33n)
> 
> Doesn't seem completely wrong. You can ignore 'n' for ring 1, because that is localhost which is connected only on Ring 0 (3.0.4 has this output more consistent) so all nodes are connected at least via Ring 1.
> Ring 0 on node 2 seems to have some trouble with connection to node 1 but node 1 (and 3) seems to be connected to node 2 just fine, so I think it is ether some bug in knet (probably already fixed) or some kind of firewall blocking just connection from node 2 to node 1 on ring 0.
> 
> 
>>
>> corosync-cfgtool -R don't help
>> All nodes are VMs that build the ring together using 2 VLANs.
>> Which logs do you need to hopefully help me?
> 
> syslog/journal should contain everything needed especially when debug is enabled (corosync.conf - logging.debug: on)
> 
> Regards,
>     Honza
> 
>>
>> Corosync Cluster Engine, version '3.0.1'
>> Copyright (c) 2006-2018 Red Hat, Inc.
>> Debian Buster
>>
>>
>> --
>> Mit freundlichen Grüßen
>>     Michael Rohweder-Neubeck
>>
>> NSB GmbH – Nguyen Softwareentwicklung & Beratung GmbH Röntgenstraße 27
>> D-64291 Darmstadt
>> E-Mail:
>> mrn at nsb-software.de<mailto:mrn at nsb-software.de<mailto:mrn at nsb-software
>> .de%3cmailto:mrn at nsb-software.de>>
>> Manager: Van-Hien Nguyen, Jörg Jaspert
>> USt-ID: DE 195 703 354; HRB 7131 Amtsgericht Darmstadt
>>
>>
>>
>>
>> Sitz der Gesellschaft / Corporate Headquarters: Deutsche Lufthansa
>> Aktiengesellschaft, Koeln, Registereintragung / Registration:
>> Amtsgericht Koeln HR B 2168 Vorsitzender des Aufsichtsrats / Chairman
>> of the Supervisory Board: Dr. Karl-Ludwig Kley Vorstand / Executive
>> Board: Carsten Spohr (Vorsitzender / Chairman), Thorsten Dirks,
>> Christina Foerster, Harry Hohmeister, Dr. Detlef Kayser, Dr. Michael
>> Niggemann
>>
>>
>>
>>
>> _______________________________________________
>> Manage your subscription:
>> https://lists.clusterlabs.org/mailman/listinfo/users
>>
>> ClusterLabs home: https://www.clusterlabs.org/
>>
>