Hi all
Merry <hic> holidays to you all :)
Joseph Mack NA3T wrote:
On Fri, 23 Dec 2005, Jan Abraham wrote:
From time to time, my director sends a tcp reset packet into an
established
connection, thus terminating the connection on the initiators side. The
balanced realserver doesn't know anything about the reset packet and
tries to
resend its last unacknowledged tcp packet until it times out.
good detective work here. I expect you're going to have to wait for
Horms and Julian to float to the surface after the holidays before
anyone will have a look here
Just a thought - is this a relatively old kernel, are you using
persistence, have you configured FTP on your LVS? This could conceivably
be related to the situation which got fixed earlier this year where the
FTP helper module broke persistence and caused sessions to flap between
realservers; I saw all sorts of weird packets floating around at the
time but ignored them because they didn't seem that relevent (and in
fact appeared to be related to outbound, rather than inbound, connections).
I'm wondering if the unusual RST packets are occurring because the
director is responding to some part of an established connection which
falls out of the LVS tables and therefore is getting processed by the
director itself. I'd expect to see some ICMP error messages in that
case, but if the director happens to be listening on the port you're
talking about then you could see some very unusual and apparently
unmatched traffic.
Graeme
|