|
True, but is there any harm in reinstating it in the mean time?
You just removed its timeout table in 2.6.16. The real problem
is that timeouts are already per-protocol, i'm not sure it will compile
and work, one can try.
It did for me, however I will double check again in the evening.
On 5 Mar 2004 i sent you some ideas about
extending the timeout handling. The kernel has almost everything needed
to allow ipvsadm to set per-proto and per-app timeout, except if code
is already removed by someone as unused.
This dead code has been there for ages, without any notion of why it was
kept there. What are the use cases for per-proto and per-app?
per-proto: TCP, UDP, ICMP, IP?
per-app : ftp
Is this what you mean? And if so, is it simply a matter of splitting up
the reinstated code into those per-proto zones?
Time ago i didn't proceeded
because libipvs is not my area.
net/ipv4/ipvs/ip_vs_proto.c has funcs
to help per-state timeouts. Supporting state timeouts only for
secure_tcp mode (which is not useful) is bad idea. If you don't keep
this posting i can resend some ideas.
Please resend them again if you have time, since I can't find them. Also
note, that to me a pragmatic approach already helps, no over-engineering
needs to be done. The only real timeouts you ever need to set in 99.999%
of the cases are TCP related.
Best regards,
Roberto Nibali, ratz
--
echo '[q]sa[ln0=aln256%Pln256/snlbx]sb3135071790101768542287578439snlbxq'|dc
|
