|
On Thu, Feb 05, 2004 at 10:14:58AM -0500, John Reuning wrote:
> On Thu, 2004-02-05 at 06:44, Horms wrote:
>
> > By definition the InActConn count for a given connection
> > won't be decremented until the conection expires from
> > the connection table.
>
> This is consistent with the behavior from my recent tests. The
> InActConn count seemed to be decremented when the TIME_WAIT connection
> countdown hits zero and disappears from /proc/net/ip_vs_conn.
Yes, that is the way it should work.
> The problem of pounding a newly added real server seems to have more to
> do with the ActiveConn count, though. Is that the figure used by the
> director to determine which real server has the least number of
> connections for lc load balancing?
Yes. There is a ratio that takes into account the number of
ActiveConn and InActConn (the former is weighted very heavily).
You might want to try chaning this in the source. Look in ip_vs_lc.c
> Here's another question: Where can I see the persistence timeout
> countdown when connection persistence is enabled?
Yes, it should also show up as an entry with a client port of 0
when you run ipvsadm -L -c -n.
--
Horms
|
