Re: Persistent Services

To: "Thomas A. Morris" <morrist@xxxxxxxxxxxx>
Subject: Re: Persistent Services
Cc: lvs-users@xxxxxxxxxxxxxxxxxxxxxx
From: jbearce@xxxxxxxxxxxx
Date: Mon, 26 Jun 2000 13:15:54 -0400

That seems to work for the non-persistent services, but it doesn't seem to work
for the persistent ones.  After removing a realserver or marking the weight as
0, no new traffic will be sent to that server, but looking at 'ipchains -L -M
-n' still shows masq entries for existing connections and those connections
don't seem to get routed to a new server until the masq timer expires.  This is
with the Redhat 2.2.14-12 kernel plus the LVS 0.9.13 patch. (I've tried all the
earlier LVS patches down to 0.9.8 as well).


|        |          "Thomas A.   |
|        |          Morris"      |
|        |          <morrist@cont|
|        |>    |
|        |          Sent by:     |
|        |          morrist@cross|
|        |      |
|        |                       |
|        |                       |
|        |          06/26/2000   |
|        |          01:11 PM     |
|        |                       |
  |                                                                            |
  |       To:     jbearce@xxxxxxxxxxxx                                         |
  |       cc:     lvs-users@xxxxxxxxxxxxxxxxxxxxxx                             |
  |       Subject:     Re: Persistent Services                                 |

jbearce@xxxxxxxxxxxx wrote:

> Ok quick question....  I've noticed that when running LVS with persistent
> services, if one of the real servers goes down, anyone connected to that real
> server will not get routed to another real server until the timer for that
> persistent service runs out.  What I need to do is have the user immediately
> forwarded to another real server if the server they're connected to takes a
> (but still maintain the persistent feature).  The biggest problem is that due
> the application running behind the LVS box, the persistent timeout needs to be
> set at just over 10 minutes (which is a long time to have a user not able to
> access a box).  I've started looking at the LVS source, but I figured I'd
> and see if I was just missing something before I started trying to come up
> my own solution.
> Any help would be greatly appreciated.
> Thanks,
> Jim

In my experience, if the realserver is removed from the configuration with the
ipvsadm command, then
the clients are switched immedietly. This may help solve your problem.

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