On 1/27/2018 2:21 AM, Igor Cicimov wrote: Thank you! That was it.
On 27 Jan 2018 4:44 pm, "TomK" <[email protected] <mailto:[email protected]>> wrote:On 1/26/2018 7:49 PM, Igor Cicimov wrote: On Fri, Jan 26, 2018 at 2:28 PM, TomK <[email protected] <mailto:[email protected]> <mailto:[email protected] <mailto:[email protected]>>> wrote: Hey All, We have UCARP and HAproxy configured and setup between two servers. HAproxy is bound to the UCARP VIP between the nodes. There are four services per hoer: four on SRV1 (primary) and same four apps on SRV2 (secondary) We need active / passive behavior, since apps don't support an active / active config. We have one UCARP VIP for each application. SRV1 primary SRV2 secondary We need all four VIP's and HAproxy processes to failover to the standby if: 1) One of the four processes on the primary fails. 2) Primary host fails ( This piece is easier. ) When all fail over to the standby, we need them capable of failover back if the secondary (standby) fails in the future. We can't seem to have all of them stick on the standby (now primary) when the primary comes back up or even when the one failed service comes back on SRV1 (former primary). They end up flipping back and we end up in a situation where some of the traffic goes to SRV1 and some to SRV2. We tried the: stick-table type ip size 1 nopurge peers LB stick on dst As well as rise 9999999 but those eventually fail over. Wanted to know what is the best practice in this case? I think you should solve this in your design instead in haproxy. Keepalived for example supports iptables rules, you can use them to block the traffic completely on the secondary so the primary will still think the secondary instances are out of service even when the server is up and services running. I don't think this is an option with CARP which doesn't support even executing scripts on state change. You can also use Heartbeat or Pacemaker which will have control over the services too so they will be only running if the server owns the VIP. Agreed, CARP is limited. We see that already. We're increasingly looking at keepalived instead of UCARP however we already have alarge portion of our infrastructure now using UCARP and HAPROXY. (However, that won't take long to change.)We just tested using sticky tables today and like the results so far however we notice we can't start haproxy on the second node of a cluster as long as haproxy was already started on the first node and already bound to the UCARP VIP. It will start if you set net.ipv4.ip_nonlocal_bind = 1 kernel prameter I do recall this is possible where UCARP runs on both nodes and HAproxy starts up on both nodes and successfully binds to the same VIP on both nodes. But I can't access that environment anymore to see what the config difference might be. Is this more of a UCARP question or an HAproxy config to allow both HAproxy to bind to the same VIP and same port off of two different hosts? Let me know if I'm not being clear. Second question we have is how to split up HAproxy processes into separate start and stop scripts? Currently we stop and start using only the main restart script /etc/init.d/haproxy but that stops all or starts all. Has anyone split these up into separate start stop scripts to control individual HAproxy instances? In other environments we find that they start multiple copies of the same HAproxy definition. We need better fine grained control of each one. Not sure I understand this question and what do you refer to when saying haproxy processes? If you mean separate haproxy per service I don't think thats a good idea and sounds like waste of resources. Anyway, changing your design should solve you problems and the need of needlessly messing around with haoroxy.-- Cheers,Tom K. ------------------------------------------------------------------------------------- Living on earth is expensive, but it includes a free trip around the sun.
-- Cheers, Tom K. ------------------------------------------------------------------------------------- Living on earth is expensive, but it includes a free trip around the sun.
