|
Hi,
the entries themselv can't be the problem (directly) - I see something
between 500 and the theoretic maximum is about 2000. I now established a
cron-job to take a closer look at the numbers.
The connect itselv is very short lvs just does the connection, the answer
goes from the real-server directly back to the client - but both are very
short-lived.
Could the problem lie somewhere deeper inside the IP-Stack?
About 1000 Clients permanently reconnecting - are there some IP-timeouts I
should consider setting lower?
Kind regards,
Nils
> -----Original Message-----
> From: Joseph Mack NA3T [mailto:jmack@xxxxxxxx]
> Sent: Thursday, August 31, 2006 3:42 PM
> To: LinuxVirtualServer.org users mailing list.
> Subject: Re: Connection table clears and ipvsadm stops
> distributing the load
>
> On Thu, 31 Aug 2006, Hildebrand, Nils, 122 wrote:
>
> > We are doing load-balancing for http with DR/WLC and had a
> persistency
> > of one hour. After we raised the persistency to 8 hours and later
> > reduced it to
> > 3 hours we ran into a strange problem:
> >
> > After some time "ipvsadm -Lc" does not show any
> connections, while all
> > connections go to only one real server (the other real
> server is not
> > contacted any longer).
>
> people running long persistence times report weird stuff like
> this. Since it's hard to do experiments with a setup that
> takes 8hrs (or whatever) to do a test, we've never found out
> whether the problems are unique to that user or a bug.
>
> You don't tell us how many connections are involved
> (millions, half a dozen?). There are explanations for what
> you see. Your clients could be coming in from an AOL like
> proxy and all be directed to the same realserver. You'd need
> to connect from clients with different IPs to check this out.
>
> Otherwise I have no idea what the problem is
>
> Joe
>
> [...]
|
