[lvs-users] Sync issue after master reboot
dan.markhasin at intel.com
Tue Feb 25 13:03:04 GMT 2014
Thanks for the reply.
Let me clarify, it doesn't stop synchronizing information after the reboot. The synchronization continues, but after the old master is rebooted its connections table is obviously wiped clean.
At this point, the old slave (now acting as master) has a full connections table, and new connects are being synchronized with the old master, but only the *new* connections are synchronized. Connections that existed on the slave before the master failed will not be synchronized, and if I initiate failback, they will be disconnected.
Is this clearer now? :)
From: Julian Anastasov [mailto:ja at ssi.bg]
Sent: Tuesday, February 25, 2014 10:26 AM
To: Markhasin, Dan
Cc: lvs-users at linuxvirtualserver.org
Subject: Re: [lvs-users] Sync issue after master reboot
On Wed, 12 Feb 2014, Markhasin, Dan wrote:
> We are currently building an LVS cluster with two ldirector servers to ensure HA. Each of these servers has two ipvsadm 1.26 sync daemons running, which we start by:
> On master:
> --start-daemon master --syncid 50
> --start-daemon backup --syncid 51
> On backup:
> --start-daemon master --syncid 51
> --start-daemon backup --syncid 50
> During normal operation there are no issues, connections are made to the master and are synced to the backup. In case the master goes down, the connections seamlessly transfer over to the backup.
> However, in the case of a reboot of the master, after the master is brought back online and the ipvsadm daemons are started, the existing connections from the backup server are not synced to the master - causing them all to disconnect whenever a failback is performed. New connections made to the slave are synced, but the old ones are not.
> I have seen references from the past decade to this problem, but have found no solution for syncing the master and the backup server after the master's reboot. Any help would be greatly appreciated!
Yes, there is no trigger implemented to sync all
connections at once. But looking at the code I don't see
why sync should stop transferring information. I don't
know what kernel you are using but even connections with
IP_VS_CONN_F_SYNC bit (created while we were backup)
should be synced when we are master. Of course, it can depend
on forwarding methods, fwmark usage, kernel (IPVS sync
protocol) version, netfilter (stateful firewalling dropping
the synced conns), etc. May be you can give more information
and to share more observations.
Julian Anastasov <ja at ssi.bg>
Intel Israel (74) Limited
This e-mail and any attachments may contain confidential material for
the sole use of the intended recipient(s). Any review or distribution
by others is strictly prohibited. If you are not the intended
recipient, please contact the sender and delete all copies.
More information about the lvs-users