|
On 09/25/2012 02:42 AM, Graeme Fowler wrote:
> On Sun, 2012-09-23 at 02:51 -0600, Pete Ashdown wrote:
>> Thanks for the localhost addressing tip. I've got my lvs dns in production
>> now and it is cranking along. I had to lower the udp timeout to 1 second
>> from my predicted 15 because the latter was crashing the server with OOM.
> Really? How many queries/sec is your system handling?
1000-2000 qps.
>> Even with 1 second timeout, the InActConn are piling up (it hovers around
>> 300):
>>
>> Prot LocalAddress:Port Scheduler Flags
>> -> RemoteAddress:Port Forward Weight ActiveConn InActConn
>> UDP cns1.domain.com:domain wlc
>> -> localhost:domain Route 12 0 282
>> -> rec3.domain.com:domain Route 4 0 84
>>
>> I don't understand why lvs is tracking these connections, because I presume
>> with UDP they don't flow back through. Is there a trigger to have it just
>> "fire and forget" UDP after it is released to the real-IP?
> Those numbers are small enough to be practically insignificant. You're
> right though, in DR mode there's no record of an active connection as
> the three-way-handshake (for TCP connections) doesn't traverse the
> director; for UDP in any mode there's no concept of state so the only
> "connection" the director ever sees is "inactive" (that is, a packet was
> routed).
>
> I'm troubled by the OOM though. That would, for most modern systems,
> imply something is in error in the ipvs code - unless you're running on
> a system with a relatively small amount of RAM.
>
8GB of RAM.
_______________________________________________
Please read the documentation before posting - it's available at:
http://www.linuxvirtualserver.org/
LinuxVirtualServer.org mailing list - lvs-users@xxxxxxxxxxxxxxxxxxxxxx
Send requests to lvs-users-request@xxxxxxxxxxxxxxxxxxxxxx
or go to http://lists.graemef.net/mailman/listinfo/lvs-users
|
