Re: Bleed-over between VIPs? Connection routing issue?
Horms <[email protected]>
| Newsgroups | gmane.linux.highavailability.ultramonkey |
|---|---|
| Message-ID | <[email protected]> |
On Mon, Sep 12, 2005 at 12:53:52PM -0500, Kelly Corbin wrote: > I'm having a strange issue that I've been unable to figure out the last > few days. > My setup consists of a load balancer and 6 real servers in the High > Capacity High Availability and Load Balancing configuration. I've been > running in this configuration for several years now with mostly flawless > service. > > All of the real servers are configured exactly the same with one small > exception. To facilitate testing, in the load balancer I've set 2 of > the web servers to be served by a different VIP than the 4 "production" > web servers. All of the web servers are setup with both of the VIPs > however. Normally, I set the apache conf to be identical on all the > servers regardless of whether they will be hosting the test domains or > not but this time I happened to not enable them on the production > servers which allowed me to catch this problem easier. > > Every now and then (maybe 1 in 10-50 clicks) I will get the default page > for one of the random production web servers. > > At first I thought this was an ARPing issue but after checking > everything thoroughly, the real servers are not ARPing for the VIPs. > > After much log-watching, I've found that it usually (but not always) > coincides with removing or quiescing real servers or when adding the > fall back (127.0.0.1 on the load balancer). It seems that connections > are being briefly routed incorrectly in the instant between when a > server is removed and at other various seemingly random times. > > I've turned quiescent on and off and persistence off and on as an > attempted work-around but the problem still continues. Oddly enough, I > don't see this happening with my other VIPs (i.e. the production domains > are not inadvertently being routed to the test servers). > > Has anyone seen this before? I'm trying to figure out how this could > possibly happen. That does seem quite strange. I suspect that you are seeing a bug in LVS, which kernel are you running? -- Horms -- Ultra Monkey - http://www.ultramonkey.org/ To UNSUBSCRIBE, email to [email protected], with a body: unsubscribe ultramonkey-users [email protected] where "[email protected]" is YOUR email address.