LVS
lvs-users
Google
 
Web LinuxVirtualServer.org

Re: Ldirectord checktime issue

To: "LinuxVirtualServer.org users mailing list." <lvs-users@xxxxxxxxxxxxxxxxxxxxxx>
Subject: Re: Ldirectord checktime issue
From: Malcolm Turnbull <malcolm.turnbull@xxxxxxxxxxxx>
Date: Thu, 03 Apr 2003 22:48:13 +0100
Horms wrote:
On Wed, Apr 02, 2003 at 10:13:44PM +0100, Malcolm Turnbull wrote:

Howdy,

I've got a client with 4 iis servers behind LVS & Ldirectord.
checktime = 10
checktimeout = 5
And his servers keep on getting taken on & offline... :-(
I've asked him to increase settings to 20/10 to see what happens
but has anyone else needed to increase the settings this high ?



[Wed Apr 2 21:56:12 2003|ldirectord] Quiescent real server:
209.189.52.228:80 (4 x 209.189.52.240:80) (Weight set to 0)
[Wed Apr 2 21:56:22 2003|ldirectord] Restored real server: 209.189.52.228:80
(3 x 209.189.52.240:80) (Weight set to 100)
[Wed Apr 2 21:56:42 2003|ldirectord] Quiescent real server:
209.189.52.228:80 (4 x 209.189.52.240:80) (Weight set to 0)
[Wed Apr 2 21:57:02 2003|ldirectord] Restored real server: 209.189.52.228:80
(3 x 209.189.52.240:80) (Weight set to 100)
[Wed Apr 2 21:57:12 2003|ldirectord] Quiescent real server:
209.189.52.228:80 (4 x 209.189.52.240:80) (Weight set to 0)
[Wed Apr 2 21:57:22 2003|ldirectord] Restored real server: 209.189.52.228:80
(3 x 209.189.52.240:80) (Weight set to 100)

Current load is approx. 1600 active connections per real server.  The real
servers themselves are all around 33% average CPU use.  Uptime on the load
balancer reports:  0.00, 0.00, 0.00 (no real load at all).


Hi Malcolm,

that is very strange and obviously is something that shouldn't be
happening. The only time that I have seen behaviour like that before
is in a situation where the Real Servers where under extreme load.

I think that your idea To extend the checktime and checktimeout
is sounds. However I am curious to know if you think that
the Real Servers are genuinely taking more than 5 seconds to
respond or if ldirectord is spinning out of control for some reason.

If it is indeed a resopnse time problem with the Real Servers
then you may want to consider setting checktype to connect
rather than negotiate. I would imagine this effected less by
a Real Server that is responding slowly for some reason.


Horms,

It was on connect... :-(
But then I got this response from the client with the issue :

>I *think* it was just my own ineptitude. I had a gateway defined on >the loopback adapter on the real servers (I went back and re-read the >setup doc and noticed you didn't have one in there). I removed it, > >rebooted all the servers, and it seems to be running fine now. Thanks >for all your help.

Which is weird 'cause that would imply an ARP issue (he's using DR ) but then again I've seen NT do some weird things when you hide an interface on the loopback adapater ;-).

BTW.. Have you every thought about adding a slow start option to ldirectord ? or should I switch from WLC to WRR to stop servers getting hammered when they are brought back online ?







--
Regards,

Malcolm Turnbull.
Crocus.co.uk Ltd
01344 629661
07715 770523

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