|
Hello,
On Wed, 30 Aug 2000, Joachim Wolff wrote:
> Hi all,
>
> i try to test LVS / ipvsadm since a few days.
>
> I send requests with http_load to the virtual IP-adress, and in works.
>
> When i change the weight of one real-server to 0 with
> ipvsadm -e -t 192.168.64.11:80 -r 192.168.64.15 -g -w 0
> the server is under load as well.
but the counters are not increased, right?
-w 0 stops the schedulers to select the real server but
the existing connections continue to work. In your test the TCP
connections are reused and the scheduler is not used. All next
connections hit existing entries, i.e. no new connections are
created and the scheduler is not called. This is because your test
is limited to one client host and to the number of its local ports
used in this test, i.e. 1024..4999
The possible solutions to this problem could be:
1. select new real server
But it is not possible to have two connections with same ends
2. stop the traffic for this connection
This is solved by deleting the real server, not by -w 0
3. continue to serve existing connections but stop the new connections
to this real server
This is the current behaviour.
Try this test:
1. Start sending requests from one client host
2. Set weight to 0
3. Stop the first client and start sending requests from second
host immediately
Are the counters for the stopped real server increased?
>
> IP Virtual Server version 0.9.15 (size=65536)
> Prot LocalAddress:Port Scheduler Flags
> -> RemoteAddress:Port Forward Weight ActiveConn InActConn
> TCP www42.mobile.test:http rr
> -> www42-3.mobile.test:http Route 1 36 1308
> -> www42-2.mobile.test:http Route 0 21 1280
> -> www42-1.mobile.test:http Route 1 27 1297
>
> After the load is stopped _AND_ the InActConn go to 0 after some
> seconds, the result for the next test will be fine, like
Switch to an another client host just after the weight change.
>
> IP Virtual Server version 0.9.15 (size=65536)
> Prot LocalAddress:Port Scheduler Flags
> -> RemoteAddress:Port Forward Weight ActiveConn InActConn
> TCP www42.mobile.test:http rr
> -> www42-3.mobile.test:http Route 1 44 1940
> -> www42-2.mobile.test:http Route 0 0 0
> -> www42-1.mobile.test:http Route 1 54 1930
>
> The other way to set the weight to 1 has the same behavior.
> Where is my mistake?
>
> At the other hand i searching for an developer newsgroup or an place for
> the newest patches, or is there really nothing changed since June 28 ?
> That would be fine ;-)
There is no other place
> And an description for the /proc files would be useful, sometimes ?
There is one description here:
http://www.linuxvirtualserver.org/defense.html
>
>
> By,
>
> Joachim
>
> --
> Joachim Wolff
>
> mailto:jwolff@xxxxxxxxx
> http://www.mobile.de Ihr Automarkt im Internet
Regards
--
Julian Anastasov <ja@xxxxxx>
|
