[corosync] information request

Vladislav Bogdanov bubble at hoster-ok.com
Fri Nov 29 06:06:26 UTC 2013

29.11.2013 08:36, Slava Bendersky wrote:
> Yes, this possible since everything running on vmware. Another point

I'd look closer at the eth driver and may be offloading parameters then
(on both sender and receiver sides). Something corrupts your packets (or
sends them initially corrupted).

> that might mtu problem involved, because vm located in two different geo

With MTU issues in place you wont see that packet on the receiving side
at all, it will be dropped at the MTU bottleneck (and ICMP message will
be sent back in order to make Path MTU Discovery work).

> locations.  We tried many think so far, include lower netmtu to 1000,
> but it is not really helped. Corosync still unstable and right now it is
> only vmware left on the way.
> Slava.
> ------------------------------------------------------------------------
> *From: *"Vladislav Bogdanov" <bubble at hoster-ok.com>
> *To: *discuss at corosync.org
> *Sent: *Friday, November 29, 2013 12:22:14 AM
> *Subject: *Re: [corosync] information request
> 26.11.2013 23:22, Slava Bendersky wrote:
>> Hello Honza,
>> Thank you for exact instruction how to find the trouble.
>> I followed step by step. Corosync-objectctl is not reporting errors I
>> was be able see data which you described earlier.
>> And I see interfaces both ends are bounded to proper ip address. Also I
>> see on tcpdump valid traffic which quite chatty.
>> > [bad udp cksum 6a0!] UDP, length 107
>> 14:52:09.333729 IP (tos 0x0, ttl 64, id 0, offset 0, flags [DF], proto
>> UDP (17), length 135)
> This ^^^ could be a source of the issue.
> Bad UDP checksumm usually means you have broken ethernet hardware or IP
> stack (and main reason for this could be a broken DIMM) somewhere on the
> path.
> Such packets are dropped by a kernel, so corosync wont see them.
> _______________________________________________
> discuss mailing list
> discuss at corosync.org
> http://lists.corosync.org/mailman/listinfo/discuss

More information about the discuss mailing list