LVS
lvs-users
Google
 
Web LinuxVirtualServer.org

Re: [lvs-users] Cannot figure out failover balancing

To: David Coulson <david@xxxxxxxxxxxxxxxx>
Subject: Re: [lvs-users] Cannot figure out failover balancing
Cc: "LinuxVirtualServer.org users mailing list." <lvs-users@xxxxxxxxxxxxxxxxxxxxxx>
From: Peter Scott <Peter@xxxxxxxx>
Date: Mon, 23 Jan 2012 13:40:23 -0800
Thanks!  We're going to look at Pacemaker (which does seem to be 
supported on RHEL6 after all).  Looks like what we want.

On 1/20/2012 5:43 PM, David Coulson wrote:
> Why not just setup a resource in rgmanager (I'd say use pacemaker, but 
> RedHat doesn't support that yet), and move the MySQL IP between the 
> hosts, rather than messing around with IPVS?
>
> David
>
> On 1/20/12 8:13 PM, Peter Scott wrote:
>> Hello.  We are using the Redhat load balancing add-on that comes on RHEL
>> 6.  We want to have highly available MySQL; we have two co-master MySQL
>> servers that replicate to each other.  MySQL says that that updates
>> should only come in to one of these at a time because of the possibility
>> that conflicting updates arrive at each server simultaneously.  So we
>> want to use LVS to direct all traffic to one of the co-masters (call it
>> mysql01) except when it is down, in which case to direct all traffic to
>> the other one (call it mysql02).
>>
>> We thought this requirement would be common enough to be addressable via
>> stock LVS capability.  Our initial thought was to set the weight of
>> mysql01 to 1 and mysql02 to 0.  However, new connections still went to
>> mysql02.  We discovered
>> http://archive.linuxvirtualserver.org/html/lvs-users/2008-10/msg00101.html 
>>
>> and set the /etc/sysctl.conf parameters, also tried setting them via the
>> equivalent files in /proc/sys/net/ipv4/vs/, still connections went to
>> both servers even when we waited beyond the persistence timeout of 600s.
>>
>> Because of this behavior, we are not able to test whether our approach
>> is sound to begin with; i.e., if mysql01 goes down will traffic be
>> redirected to mysql02 even though its weight is 0?  We thought that if
>> it was the only machine up in its group the weight would not matter.
>> But would like to know if that assumption was accurate also.
>>
>> Here's that section of our lvs.cf:
>>
>> virtual MySQL {
>>        active = 1
>>        address = 192.168.185.115 eth0:6
>>        vip_nmask = 255.255.255.0
>>        fwmark = 3306
>>        port = 3306
>>        persistent = 600
>>        use_regex = 0
>>        load_monitor = none
>>        scheduler = wlc
>>        protocol = tcp
>>        timeout = 6
>>        reentry = 15
>>        quiesce_server = 0
>>        server MySQL01 {
>>            address = 192.168.185.51
>>            active = 1
>>            port = 3306
>>            weight = 1
>>        }
>>        server MySQL02 {
>>            address = 192.168.185.52
>>            active = 1
>>            port = 3306
>>            weight = 0
>>        }
>> }
>>
>> Would appreciate any advice.  Thanks.
>>
>> _______________________________________________
>> Please read the documentation before posting - it's available at:
>> http://www.linuxvirtualserver.org/
>>
>> LinuxVirtualServer.org mailing list - lvs-users@xxxxxxxxxxxxxxxxxxxxxx
>> Send requests to lvs-users-request@xxxxxxxxxxxxxxxxxxxxxx
>> or go to http://lists.graemef.net/mailman/listinfo/lvs-users


_______________________________________________
Please read the documentation before posting - it's available at:
http://www.linuxvirtualserver.org/

LinuxVirtualServer.org mailing list - lvs-users@xxxxxxxxxxxxxxxxxxxxxx
Send requests to lvs-users-request@xxxxxxxxxxxxxxxxxxxxxx
or go to http://lists.graemef.net/mailman/listinfo/lvs-users

<Prev in Thread] Current Thread [Next in Thread>