|
Hi All,
In an LVS-DR cluster we remove one of the real servers (using ldirectord)
and clients are still sent to the real server even though it is not in the
IPVS table--even after 12+ hours.
(A tcpdump shows that the connection is going through the Director and not
just going directly to the LVS-DR real server due to an ARP problem.)
If I remove the loopback alias that has the cluster IP address from the
LVS-DR Real Server (with it still removed from the IPVS table) the next
attempt by the same client to connect to the cluster hangs for awhile then
fails. If you then attempt to connect again from the client the LVS Director
sends you to a real server properly listed in the IPVS table.
I've upgraded to the latest 2.4.19 kernel and IPVS code (1.0.6) and we still
have this problem.
Why would the Director send clients to a real server not in the table even
after several hours of inactivity from the client?
My simple IPVS table looks like this:
IP Virtual Server version 1.0.6 (size=4096)
Prot LocalAddress:Port Scheduler Flags
-> RemoteAddress:Port Forward Weight ActiveConn InActConn
TCP 172.24.150.90:23 wrr
-> 172.24.150.92:23 Route 1 0 0
-> 127.0.0.1:23 Local 1 1 2
The IP address of the real server removed from the cluster is 172.24.150.94.
--L
_________________________________________________________________
Join the world?s largest e-mail service with MSN Hotmail.
http://www.hotmail.com
|
