On Sat, Sep 03, 2011 at 02:25:37AM +0200, Alexander Bluhm wrote:
During socket splicing the relayd session timeouts could not be
measured exactly in user land. Use the new idle timeout for socket
splicing in the kernel to make it correct.
I think, I got the flag handling wrong. Make sure
Hi,
During socket splicing the relayd session timeouts could not be
measured exactly in user land. Use the new idle timeout for socket
splicing in the kernel to make it correct.
ok?
bluhm
Index: usr.sbin/relayd//parse.y
===
RCS
On Tue, Mar 08, 2011 at 02:11:31AM +0100, Alexander Bluhm wrote:
Hmm, perhaps too small. There is a session timeout and relayctl
show sessions idle time. Relayd does not realize when the kernel
is transferring data automatically. It should check the splice
data length with getsockopt()
On Sun, Mar 06, 2011 at 09:12:41AM +0100, Reyk Floeter wrote:
this diff will break chunked encoding and keep-alive connections where
we need to enable splicing for a specified amount of data only and
return for the next HTTP header.
I don't think so. I only set F_SPLICE for RELAY_PROTO_TCP
I do not think splice should be an option that is exposed to
users.
Either the kernel support works, or doesn't.
The option nosplice is useful for performance comparison and
debugging.
+ { nosplice, NOSPLICE },
That is crazy notion. Either it is faster and works, and
On Mon, Mar 07, 2011 at 04:23:08PM -0700, Theo de Raadt wrote:
I do not think splice should be an option that is exposed to
users.
That makes the diff much smaller.
ok?
Index: usr.sbin/relayd/relay.c
===
RCS file:
On Tue, Mar 08, 2011 at 01:00:48AM +0100, Alexander Bluhm wrote:
On Mon, Mar 07, 2011 at 04:23:08PM -0700, Theo de Raadt wrote:
I do not think splice should be an option that is exposed to
users.
That makes the diff much smaller.
Hmm, perhaps too small. There is a session timeout and
hi,
this diff will break chunked encoding and keep-alive connections where
we need to enable splicing for a specified amount of data only and
return for the next HTTP header. the env variable should be replaced
with a permanent config option in parse.y.
reyk
On Wed, Mar 02, 2011 at 09:34:14PM
This diff implements socket splicing for relayd. Instead of copying
data in userland from one TCP socket into another, the kernel is
told to move the data himself.
The environment variable RELAY_NOSPLICE works like EVENT_NOKQUEUE
from libevent. It can be used to easily turn it on and off for
I just relized that I did send out an outdated diff. This one has
an additional check for F_SSL and F_SSLCLIENT to avoid splicing ssl
connections.
The RELAY_NOSPLICE environment variable is only for testing and
will be removed in the final version.
bluhm
Index: usr.sbin/relayd/parse.y
10 matches
Mail list logo