Think I got it - ARP problem.
On 10-09-10 04:53 AM, Mike wrote: > Thank you for the response Simon. Helpful as always. > > I did some testing on this issue tonight and I've discovered an > interesting issue. > > First of all I am using LVS-Tun and I have created a tunnel interface on > each of my back end servers. This tunnel interface is of course the VIP > you see below - 172.28.185.54. > > When we started our tests this evening all requests were going to one > back end server, 172.28.185.57. As a test I decided to completely > shutdown the LVS servers by shutting down LinuxHA which shuts down > ldirectord. > > To my surprise, I was still able to send connections to the VIP. When I > ssh'ed into the VIP I ended up on the back end server that was receiving > the connections. Of course the VIP should be down since LVS was down. I > rebooted both back end servers, started up LVS and like magic everything > worked. Load balancing was working perfectly. > > So the problem was the backend server somehow took ownership of the VIP > and as a result was grabbing all requests. A reboot resolved it. Can you > tell me why this may have happened? What could be wrong on my backend > servers that they would grab the VIP like this? > > Any help would be appreciated greatly. > > Mike > On Thu, 2010-09-09 at 18:15 +0900, Simon Horman wrote: > >> On Tue, Sep 07, 2010 at 03:57:52PM -0300, Mike wrote: >> >>> Hello all, >>> >>> I've implemented a LVS cluster using ldirectord and LinuxHA. Here is a >>> snippet from my ldirectord.cf file: >>> >>> virtual=172.28.185.54:8080 >>> protocol=tcp >>> scheduler=wrr >>> checktype=connect >>> checkport=8080 >>> #service=ldap >>> real=172.28.185.57:8080 ipip >>> real=172.28.185.58:8080 ipip >>> #service=http >>> checktimeout=10 >>> checkinterval=10 >>> >>> While running tests that hit the index.html page on the VIP above I >>> noticed that we seem to be hitting the same backend server repeatedly. >>> When I run the test from another server, we start hitting the other back >>> end server repeatedly. I'm curious if anyone can tell me if there is a >>> built in affinity here that I'm missing. I'm thinking that ldirecotor >>> seems to be smart enough (or dumb enough depending on your point of >>> view) to notice that a connection attempt from ip address guest will be >>> sent to the same backend server for productivity reasons. >>> >>> Anyone know if thee is a natural built in affinity here? >>> >> If you have persistence enabled, then yes that is the expected behaviour. >> Otherwise its a bit odd. >> > > _______________________________________________ > Linux-HA mailing list > [email protected] > http://lists.linux-ha.org/mailman/listinfo/linux-ha > See also: http://linux-ha.org/ReportingProblems > > _______________________________________________ Linux-HA mailing list [email protected] http://lists.linux-ha.org/mailman/listinfo/linux-ha See also: http://linux-ha.org/ReportingProblems
