Schedule: No dest found

Brett E. brettspamacct at fastclick.com
Thu Apr 15 01:51:36 BST 2004


Martijn Grendelman wrote:
> Hi,
> 
> Every once in a while, my director denies connections on the VIP, even
> though both real servers are up. After a short while (seconds, up to a
> minute or so) everything turns back to normal. I think (but I am not 100%
> sure) that while one client gets a connection refused, another client can
> connect just fine. I have received complaints from users, while at that
> time, everything seemed fine from my end. At other times, I experienced
> the problem myself. Also, at times, the response from the web servers is
> slow, when accessed via the VIP, while the response on the real servers
> directly is fast, as it should be.
> 
> I use LVS/DR on Linux kernel 2.4.25, with the sh (source hashing)
> scheduler. Statistics show that both real servers are being accessed more
> or less equally, as could be expected with source hashing.Today, I turned
> on IPVS debugging in the kernel and set the debug level to 3. 'dmesg' now
> says "IPVS: Schedule: no dest found.", repeatedly. Maybe someone on the
> development team can explain this? In what situations should IPVS log that
> message? What's going wrong here? Please note that this is _not_ a problem
> with the real servers. At the times connections are denied at the VIP, the
> Apache servers respond on their RIPs just fine.
> 
> I hope someone can shed some light on this, because it's becoming more and
> more of a problem and I am considering turning off the LVS completely.
> 
> In response to a message about this problem I sent to the list on March
> 8th, someone said that he experienced the same problem, but it went away
> after he switched to RR scheduling. I wish I could do that too, but I
> don't have any secure way to share HTTP sessions between the real
> servers...
> 
> Best regards,
> Martijn.
> 
Maybe check that all realservers are up.  If all realservers go down and 
a packet comes in for the vip, you will see this message.  It's the 
result of the scheduler returning an empty realserver to the LVS code.

My patch to 2.6 which hashes source IP's to realservers would help you, 
you can find more information about it by searching the archives. It is 
beta but Wensong and other non-maintenance developers have not replied 
to me about it.  And after LVS went down again last night and took down 
the network, not to mention the increased CPU usage when LVS takes a 
realserver down, I've decided to ditch LVS and stick to the ServerIron's 
but I digress. Good luck.



More information about the lvs-users mailing list