]>
err.no Git - varnish/log
phk [Tue, 12 Sep 2006 20:17:35 +0000 (20:17 +0000)]
Fix timestamps in shm tag StatSess for sessions with no requests.
git-svn-id: svn+ssh://projects.linpro.no/svn/varnish/trunk@974
d4fa192b -c00b-0410-8231-
f00ffab90ce4
phk [Tue, 12 Sep 2006 20:06:55 +0000 (20:06 +0000)]
chunksize is in kilobytes
git-svn-id: svn+ssh://projects.linpro.no/svn/varnish/trunk@973
d4fa192b -c00b-0410-8231-
f00ffab90ce4
phk [Tue, 12 Sep 2006 19:41:52 +0000 (19:41 +0000)]
There may be error returns from kevent, don't assert there are none.
git-svn-id: svn+ssh://projects.linpro.no/svn/varnish/trunk@972
d4fa192b -c00b-0410-8231-
f00ffab90ce4
des [Tue, 12 Sep 2006 10:12:34 +0000 (10:12 +0000)]
Implement sendfile() support for Linux.
git-svn-id: svn+ssh://projects.linpro.no/svn/varnish/trunk@971
d4fa192b -c00b-0410-8231-
f00ffab90ce4
phk [Tue, 12 Sep 2006 09:38:28 +0000 (09:38 +0000)]
fix typo
git-svn-id: svn+ssh://projects.linpro.no/svn/varnish/trunk@970
d4fa192b -c00b-0410-8231-
f00ffab90ce4
andersb [Mon, 11 Sep 2006 23:30:07 +0000 (23:30 +0000)]
Getting somewhere. I now have a "full" logline. Apart from writing usernames caught by the auth header, the logline is complete.
I have now cleared up my obvious memoryleaks also.
The code looks like it has been through a shredder, and needs massive cleaning before alpha.
git-svn-id: svn+ssh://projects.linpro.no/svn/varnish/trunk@969
d4fa192b -c00b-0410-8231-
f00ffab90ce4
phk [Mon, 11 Sep 2006 17:18:31 +0000 (17:18 +0000)]
If we have more threads than the max, kill one if we can.
git-svn-id: svn+ssh://projects.linpro.no/svn/varnish/trunk@968
d4fa192b -c00b-0410-8231-
f00ffab90ce4
phk [Mon, 11 Sep 2006 15:12:05 +0000 (15:12 +0000)]
remember to close pipes
git-svn-id: svn+ssh://projects.linpro.no/svn/varnish/trunk@967
d4fa192b -c00b-0410-8231-
f00ffab90ce4
phk [Mon, 11 Sep 2006 14:50:56 +0000 (14:50 +0000)]
Use a pipe for synchronization instead.
git-svn-id: svn+ssh://projects.linpro.no/svn/varnish/trunk@966
d4fa192b -c00b-0410-8231-
f00ffab90ce4
phk [Mon, 11 Sep 2006 14:34:48 +0000 (14:34 +0000)]
Redo the worker thread pool locking a bit, all the while cursing
pthreads deficient API design for it's shortcomings.
git-svn-id: svn+ssh://projects.linpro.no/svn/varnish/trunk@965
d4fa192b -c00b-0410-8231-
f00ffab90ce4
phk [Mon, 11 Sep 2006 12:00:50 +0000 (12:00 +0000)]
Embellish the mutex debugging a bit
git-svn-id: svn+ssh://projects.linpro.no/svn/varnish/trunk@964
d4fa192b -c00b-0410-8231-
f00ffab90ce4
phk [Mon, 11 Sep 2006 11:55:00 +0000 (11:55 +0000)]
Move LOCK/UNLOCK macros to cache.h where they belong.
git-svn-id: svn+ssh://projects.linpro.no/svn/varnish/trunk@963
d4fa192b -c00b-0410-8231-
f00ffab90ce4
phk [Mon, 11 Sep 2006 10:35:26 +0000 (10:35 +0000)]
Ignore kevent error return entries
git-svn-id: svn+ssh://projects.linpro.no/svn/varnish/trunk@962
d4fa192b -c00b-0410-8231-
f00ffab90ce4
phk [Mon, 11 Sep 2006 10:16:02 +0000 (10:16 +0000)]
Only expire when the timer fires
git-svn-id: svn+ssh://projects.linpro.no/svn/varnish/trunk@961
d4fa192b -c00b-0410-8231-
f00ffab90ce4
phk [Mon, 11 Sep 2006 10:14:45 +0000 (10:14 +0000)]
Piggy-back kevent changes on kevent wait call
git-svn-id: svn+ssh://projects.linpro.no/svn/varnish/trunk@960
d4fa192b -c00b-0410-8231-
f00ffab90ce4
phk [Mon, 11 Sep 2006 10:14:29 +0000 (10:14 +0000)]
Avoid a startup-race
git-svn-id: svn+ssh://projects.linpro.no/svn/varnish/trunk@959
d4fa192b -c00b-0410-8231-
f00ffab90ce4
andersb [Mon, 11 Sep 2006 09:35:43 +0000 (09:35 +0000)]
Cleaned up User-Agent and Referer. Both should now work correct. As far as I can see, I only have the timestamp left for a sane loggline.
git-svn-id: svn+ssh://projects.linpro.no/svn/varnish/trunk@958
d4fa192b -c00b-0410-8231-
f00ffab90ce4
des [Mon, 11 Sep 2006 09:16:59 +0000 (09:16 +0000)]
Check for sendfile().
git-svn-id: svn+ssh://projects.linpro.no/svn/varnish/trunk@957
d4fa192b -c00b-0410-8231-
f00ffab90ce4
phk [Mon, 11 Sep 2006 09:14:29 +0000 (09:14 +0000)]
Reduce syscall footprint of expiry and hangman
git-svn-id: svn+ssh://projects.linpro.no/svn/varnish/trunk@956
d4fa192b -c00b-0410-8231-
f00ffab90ce4
phk [Mon, 11 Sep 2006 09:05:39 +0000 (09:05 +0000)]
Avoid a clock_gettime() call
git-svn-id: svn+ssh://projects.linpro.no/svn/varnish/trunk@955
d4fa192b -c00b-0410-8231-
f00ffab90ce4
phk [Mon, 11 Sep 2006 09:00:44 +0000 (09:00 +0000)]
braino in last commit.
git-svn-id: svn+ssh://projects.linpro.no/svn/varnish/trunk@954
d4fa192b -c00b-0410-8231-
f00ffab90ce4
phk [Mon, 11 Sep 2006 08:58:42 +0000 (08:58 +0000)]
Save more time() calls
git-svn-id: svn+ssh://projects.linpro.no/svn/varnish/trunk@953
d4fa192b -c00b-0410-8231-
f00ffab90ce4
phk [Mon, 11 Sep 2006 08:53:53 +0000 (08:53 +0000)]
A avoid a time(NULL) call
git-svn-id: svn+ssh://projects.linpro.no/svn/varnish/trunk@952
d4fa192b -c00b-0410-8231-
f00ffab90ce4
phk [Mon, 11 Sep 2006 08:42:35 +0000 (08:42 +0000)]
Do stats on sendfile/write split.
Add param for minimum size of object before we will sendfile it.
git-svn-id: svn+ssh://projects.linpro.no/svn/varnish/trunk@951
d4fa192b -c00b-0410-8231-
f00ffab90ce4
andersb [Sun, 10 Sep 2006 22:23:59 +0000 (22:23 +0000)]
Started to fix output of User-Agent. It was all over the place. Found what I was doing wrong and fixed it.
Should work with correct User-Agent now.
git-svn-id: svn+ssh://projects.linpro.no/svn/varnish/trunk@950
d4fa192b -c00b-0410-8231-
f00ffab90ce4
phk [Fri, 8 Sep 2006 06:47:55 +0000 (06:47 +0000)]
Make the fetchers chunksize a parameter, but keep the default at 128k
git-svn-id: svn+ssh://projects.linpro.no/svn/varnish/trunk@949
d4fa192b -c00b-0410-8231-
f00ffab90ce4
phk [Fri, 8 Sep 2006 05:58:33 +0000 (05:58 +0000)]
Add an extra bucket and make it clear how the number was chosen.
Remove extra sanitychecking.
git-svn-id: svn+ssh://projects.linpro.no/svn/varnish/trunk@948
d4fa192b -c00b-0410-8231-
f00ffab90ce4
phk [Thu, 7 Sep 2006 10:14:43 +0000 (10:14 +0000)]
Missing commit message from previous commit:
Fix a logic bug which allowed storage objects to be trampled on:
the last bucket may contain object smaller than we want.
git-svn-id: svn+ssh://projects.linpro.no/svn/varnish/trunk@947
d4fa192b -c00b-0410-8231-
f00ffab90ce4
phk [Thu, 7 Sep 2006 10:12:54 +0000 (10:12 +0000)]
Handle zero length storage objects correctly: Don't let them exist.
git-svn-id: svn+ssh://projects.linpro.no/svn/varnish/trunk@946
d4fa192b -c00b-0410-8231-
f00ffab90ce4
phk [Thu, 7 Sep 2006 08:25:05 +0000 (08:25 +0000)]
Insert a bunch of asserts to try to catch anything going wrong
git-svn-id: svn+ssh://projects.linpro.no/svn/varnish/trunk@945
d4fa192b -c00b-0410-8231-
f00ffab90ce4
phk [Thu, 7 Sep 2006 08:01:14 +0000 (08:01 +0000)]
Pick up uint32_t from <stdint.h> instead of <sys/types.h>
git-svn-id: svn+ssh://projects.linpro.no/svn/varnish/trunk@944
d4fa192b -c00b-0410-8231-
f00ffab90ce4
phk [Thu, 7 Sep 2006 07:50:48 +0000 (07:50 +0000)]
Logging for status & response was swapped for generated errors.
git-svn-id: svn+ssh://projects.linpro.no/svn/varnish/trunk@943
d4fa192b -c00b-0410-8231-
f00ffab90ce4
phk [Thu, 7 Sep 2006 07:33:23 +0000 (07:33 +0000)]
Remove debugging messages.
git-svn-id: svn+ssh://projects.linpro.no/svn/varnish/trunk@942
d4fa192b -c00b-0410-8231-
f00ffab90ce4
phk [Thu, 7 Sep 2006 07:27:21 +0000 (07:27 +0000)]
Add stats counters for large and small free fragments.
Remove debugging
git-svn-id: svn+ssh://projects.linpro.no/svn/varnish/trunk@941
d4fa192b -c00b-0410-8231-
f00ffab90ce4
phk [Thu, 7 Sep 2006 07:17:31 +0000 (07:17 +0000)]
remove "age", it was effectively unused. Revisit the idea later.
git-svn-id: svn+ssh://projects.linpro.no/svn/varnish/trunk@940
d4fa192b -c00b-0410-8231-
f00ffab90ce4
phk [Thu, 7 Sep 2006 07:12:18 +0000 (07:12 +0000)]
Don't clear alloc, it will be done in due time.
git-svn-id: svn+ssh://projects.linpro.no/svn/varnish/trunk@939
d4fa192b -c00b-0410-8231-
f00ffab90ce4
phk [Thu, 7 Sep 2006 07:09:37 +0000 (07:09 +0000)]
Split the freelist by size.
Use 32 buckets for now, with a 4k pagesize that takes us to 128k
which matches the fetchers default chunksize.
git-svn-id: svn+ssh://projects.linpro.no/svn/varnish/trunk@938
d4fa192b -c00b-0410-8231-
f00ffab90ce4
phk [Thu, 7 Sep 2006 05:41:13 +0000 (05:41 +0000)]
Remove debug entry
git-svn-id: svn+ssh://projects.linpro.no/svn/varnish/trunk@937
d4fa192b -c00b-0410-8231-
f00ffab90ce4
phk [Thu, 7 Sep 2006 05:40:09 +0000 (05:40 +0000)]
Read more than one session pointer at a time from the pipe and
handle them all with one kevent call.
git-svn-id: svn+ssh://projects.linpro.no/svn/varnish/trunk@936
d4fa192b -c00b-0410-8231-
f00ffab90ce4
phk [Wed, 6 Sep 2006 22:09:32 +0000 (22:09 +0000)]
Don't use binary address anyway, we only want the IP number part
and splitting that out of the binary part is not worth the
trouble.
git-svn-id: svn+ssh://projects.linpro.no/svn/varnish/trunk@935
d4fa192b -c00b-0410-8231-
f00ffab90ce4
phk [Wed, 6 Sep 2006 22:02:07 +0000 (22:02 +0000)]
Account for header bytes in error and 304 responses.
Remove unhelpful debug entry.
git-svn-id: svn+ssh://projects.linpro.no/svn/varnish/trunk@934
d4fa192b -c00b-0410-8231-
f00ffab90ce4
phk [Wed, 6 Sep 2006 21:52:50 +0000 (21:52 +0000)]
Move some setup code to first state now that we have it.
git-svn-id: svn+ssh://projects.linpro.no/svn/varnish/trunk@933
d4fa192b -c00b-0410-8231-
f00ffab90ce4
phk [Wed, 6 Sep 2006 21:41:13 +0000 (21:41 +0000)]
Move a bunch of work from the acceptor thread to the worker thread
for better scaling.
git-svn-id: svn+ssh://projects.linpro.no/svn/varnish/trunk@932
d4fa192b -c00b-0410-8231-
f00ffab90ce4
phk [Wed, 6 Sep 2006 21:15:26 +0000 (21:15 +0000)]
Move CRC32 into libvarnish
Use it for hashing srcaddr as well. Hash the sockaddr instead
of the ascii representation.
Add a primitive mutex tracking facility which writes debug
records to the shmlog if enabled.
git-svn-id: svn+ssh://projects.linpro.no/svn/varnish/trunk@931
d4fa192b -c00b-0410-8231-
f00ffab90ce4
phk [Wed, 6 Sep 2006 20:45:49 +0000 (20:45 +0000)]
Give the stats summation its own mutex
git-svn-id: svn+ssh://projects.linpro.no/svn/varnish/trunk@930
d4fa192b -c00b-0410-8231-
f00ffab90ce4
phk [Wed, 6 Sep 2006 19:35:33 +0000 (19:35 +0000)]
Reduce traffic on the wrk_mtx, saving some syscalls along the way:
Make a thread-reaper-thread which examines the tail end of the queue
and kicks threads which are too old so the wake up and die.
This allows us to assume that a thread on the free queue is always
waiting on its condvar, so we don't need to hold the mutex when we
signal the condvar.
As a result, the woken thread stands a chance to grab the mutex
and a little song and dance between the two threads is avoided.
git-svn-id: svn+ssh://projects.linpro.no/svn/varnish/trunk@929
d4fa192b -c00b-0410-8231-
f00ffab90ce4
phk [Wed, 6 Sep 2006 19:08:06 +0000 (19:08 +0000)]
Grab the r_recv timestamp at the right time.
git-svn-id: svn+ssh://projects.linpro.no/svn/varnish/trunk@928
d4fa192b -c00b-0410-8231-
f00ffab90ce4
phk [Wed, 6 Sep 2006 18:47:06 +0000 (18:47 +0000)]
Don't bother with SO_NOSIGPIPE, it doesn't always work and we
have ignored the signal anyway.
git-svn-id: svn+ssh://projects.linpro.no/svn/varnish/trunk@927
d4fa192b -c00b-0410-8231-
f00ffab90ce4
phk [Wed, 6 Sep 2006 18:46:26 +0000 (18:46 +0000)]
Remove unused timestamp "t0" and save a syscall per request.
git-svn-id: svn+ssh://projects.linpro.no/svn/varnish/trunk@926
d4fa192b -c00b-0410-8231-
f00ffab90ce4
phk [Wed, 6 Sep 2006 18:37:17 +0000 (18:37 +0000)]
Save a systemcall by using the same timestamp twice.
git-svn-id: svn+ssh://projects.linpro.no/svn/varnish/trunk@925
d4fa192b -c00b-0410-8231-
f00ffab90ce4
phk [Wed, 6 Sep 2006 18:32:05 +0000 (18:32 +0000)]
Give pool threads 120 second timeout, 10 was close to silly.
git-svn-id: svn+ssh://projects.linpro.no/svn/varnish/trunk@924
d4fa192b -c00b-0410-8231-
f00ffab90ce4
phk [Wed, 6 Sep 2006 18:23:24 +0000 (18:23 +0000)]
Fix the poll acceptor, we need to cache the fd as sp->fd might
change to -1 under us.
git-svn-id: svn+ssh://projects.linpro.no/svn/varnish/trunk@923
d4fa192b -c00b-0410-8231-
f00ffab90ce4
phk [Wed, 6 Sep 2006 11:21:40 +0000 (11:21 +0000)]
Fix buglet in -i/-x name matching
git-svn-id: svn+ssh://projects.linpro.no/svn/varnish/trunk@922
d4fa192b -c00b-0410-8231-
f00ffab90ce4
phk [Wed, 6 Sep 2006 10:45:55 +0000 (10:45 +0000)]
Allow '-' in identifiers.
For further study: should we accept RFC2616's definition of "token" ?
git-svn-id: svn+ssh://projects.linpro.no/svn/varnish/trunk@921
d4fa192b -c00b-0410-8231-
f00ffab90ce4
phk [Wed, 6 Sep 2006 09:58:24 +0000 (09:58 +0000)]
Mark responses with Set-Cookie as "pass"
git-svn-id: svn+ssh://projects.linpro.no/svn/varnish/trunk@920
d4fa192b -c00b-0410-8231-
f00ffab90ce4
phk [Wed, 6 Sep 2006 09:54:49 +0000 (09:54 +0000)]
Add support for inspecting response headers.
git-svn-id: svn+ssh://projects.linpro.no/svn/varnish/trunk@919
d4fa192b -c00b-0410-8231-
f00ffab90ce4
phk [Wed, 6 Sep 2006 09:18:06 +0000 (09:18 +0000)]
Fix the same issue as in the kqueue acceptor: By the time we get
here the filedescriptor may already be closed, so accept EBADF.
Reported by: Xing Li <xing@litespeedtech.com>
git-svn-id: svn+ssh://projects.linpro.no/svn/varnish/trunk@918
d4fa192b -c00b-0410-8231-
f00ffab90ce4
phk [Wed, 6 Sep 2006 07:37:34 +0000 (07:37 +0000)]
Fix a bug in chunked fetching:
With very short chunks, in this case 3 characters, our buffer may
contain not only the chunk length and the chunk data, but also the
next chunk length.
If the short chunk is the last chunk before the zero length chunk
at the end, unconditionally trying to fill the buffer before parsing
the length may hang because we already have everything there is to
have in the buffer.
The fix is to always try to parse the buffer before adding to it.
While here, tighten up and improve error checks of the code.
Reported by: Xing Li <xing@litespeedtech.com>
git-svn-id: svn+ssh://projects.linpro.no/svn/varnish/trunk@917
d4fa192b -c00b-0410-8231-
f00ffab90ce4
phk [Wed, 6 Sep 2006 06:39:52 +0000 (06:39 +0000)]
escape control characters in SLT_Debug as %XX
git-svn-id: svn+ssh://projects.linpro.no/svn/varnish/trunk@916
d4fa192b -c00b-0410-8231-
f00ffab90ce4
andersb [Tue, 5 Sep 2006 22:11:05 +0000 (22:11 +0000)]
Added more outputdata, fixed memory leaks (I think).
Code still not Alpha ready, but I think most concepts work now.
git-svn-id: svn+ssh://projects.linpro.no/svn/varnish/trunk@915
d4fa192b -c00b-0410-8231-
f00ffab90ce4
phk [Thu, 24 Aug 2006 07:57:07 +0000 (07:57 +0000)]
Go through all asserts and mark those which indicate missing code with
xxx or XXX.
git-svn-id: svn+ssh://projects.linpro.no/svn/varnish/trunk@914
d4fa192b -c00b-0410-8231-
f00ffab90ce4
phk [Thu, 24 Aug 2006 07:17:35 +0000 (07:17 +0000)]
Introduce LOCK() and UNLOCK() macros which does the right thing
with pthread_mutex_{lock,unlock}()
git-svn-id: svn+ssh://projects.linpro.no/svn/varnish/trunk@913
d4fa192b -c00b-0410-8231-
f00ffab90ce4
phk [Thu, 24 Aug 2006 07:10:35 +0000 (07:10 +0000)]
Give xxxasserts their own backend with a different message.
git-svn-id: svn+ssh://projects.linpro.no/svn/varnish/trunk@912
d4fa192b -c00b-0410-8231-
f00ffab90ce4
phk [Thu, 24 Aug 2006 06:58:37 +0000 (06:58 +0000)]
Split assert into "static check" and "missing code" variants.
The "missing code" variants have xxx prefix
Introduce AN() (assert non-null) variant as well.
git-svn-id: svn+ssh://projects.linpro.no/svn/varnish/trunk@911
d4fa192b -c00b-0410-8231-
f00ffab90ce4
phk [Thu, 24 Aug 2006 06:15:13 +0000 (06:15 +0000)]
This is a workaround for what is probably a race in FreeBSD RELENG_6
socket dismantling.
There is no way that close(2) should ever be able to return EINVAL,
but we've seen it.
Specifically assert on EBADF which is the check we're really after.
git-svn-id: svn+ssh://projects.linpro.no/svn/varnish/trunk@910
d4fa192b -c00b-0410-8231-
f00ffab90ce4
phk [Wed, 23 Aug 2006 14:30:06 +0000 (14:30 +0000)]
Handle backend failure in pass mode
git-svn-id: svn+ssh://projects.linpro.no/svn/varnish/trunk@909
d4fa192b -c00b-0410-8231-
f00ffab90ce4
phk [Wed, 23 Aug 2006 13:11:14 +0000 (13:11 +0000)]
These are not used.
git-svn-id: svn+ssh://projects.linpro.no/svn/varnish/trunk@908
d4fa192b -c00b-0410-8231-
f00ffab90ce4
phk [Wed, 23 Aug 2006 12:10:11 +0000 (12:10 +0000)]
don't service cached objects the last second of their lifetime.
git-svn-id: svn+ssh://projects.linpro.no/svn/varnish/trunk@907
d4fa192b -c00b-0410-8231-
f00ffab90ce4
phk [Wed, 23 Aug 2006 11:42:16 +0000 (11:42 +0000)]
Set response timestamp when we emit an error.
git-svn-id: svn+ssh://projects.linpro.no/svn/varnish/trunk@906
d4fa192b -c00b-0410-8231-
f00ffab90ce4
phk [Wed, 23 Aug 2006 11:27:54 +0000 (11:27 +0000)]
Implement error in miss also
git-svn-id: svn+ssh://projects.linpro.no/svn/varnish/trunk@905
d4fa192b -c00b-0410-8231-
f00ffab90ce4
phk [Wed, 23 Aug 2006 07:32:19 +0000 (07:32 +0000)]
unused variable.
git-svn-id: svn+ssh://projects.linpro.no/svn/varnish/trunk@904
d4fa192b -c00b-0410-8231-
f00ffab90ce4
phk [Wed, 23 Aug 2006 07:30:42 +0000 (07:30 +0000)]
Handle backend connection error in pass mode with a 503
git-svn-id: svn+ssh://projects.linpro.no/svn/varnish/trunk@903
d4fa192b -c00b-0410-8231-
f00ffab90ce4
phk [Wed, 23 Aug 2006 07:29:28 +0000 (07:29 +0000)]
Fill shmem log with the reply
git-svn-id: svn+ssh://projects.linpro.no/svn/varnish/trunk@902
d4fa192b -c00b-0410-8231-
f00ffab90ce4
phk [Wed, 23 Aug 2006 07:22:28 +0000 (07:22 +0000)]
grammar police.
git-svn-id: svn+ssh://projects.linpro.no/svn/varnish/trunk@901
d4fa192b -c00b-0410-8231-
f00ffab90ce4
phk [Wed, 23 Aug 2006 07:16:03 +0000 (07:16 +0000)]
Teach RES_Error() about the canonical response code texts from
RFC2616.
Add the XID as "guru meditation" in the error HTML.
git-svn-id: svn+ssh://projects.linpro.no/svn/varnish/trunk@900
d4fa192b -c00b-0410-8231-
f00ffab90ce4
phk [Wed, 23 Aug 2006 06:55:18 +0000 (06:55 +0000)]
Move defaults from varnishd.c to mgt_param.c and use regular functions
for setting them.
Collapse all the 'timeout' functions.
Add pipe_timeout parameter.
git-svn-id: svn+ssh://projects.linpro.no/svn/varnish/trunk@899
d4fa192b -c00b-0410-8231-
f00ffab90ce4
phk [Wed, 23 Aug 2006 06:53:28 +0000 (06:53 +0000)]
If CLI is NULL, use stdout.
git-svn-id: svn+ssh://projects.linpro.no/svn/varnish/trunk@898
d4fa192b -c00b-0410-8231-
f00ffab90ce4
phk [Tue, 22 Aug 2006 19:14:01 +0000 (19:14 +0000)]
Implement "error" in vcl_hit()
git-svn-id: svn+ssh://projects.linpro.no/svn/varnish/trunk@897
d4fa192b -c00b-0410-8231-
f00ffab90ce4
phk [Tue, 22 Aug 2006 10:56:06 +0000 (10:56 +0000)]
Redraw must be checked when we update, we cannot trust
that we will actually have a timeout.
git-svn-id: svn+ssh://projects.linpro.no/svn/varnish/trunk@896
d4fa192b -c00b-0410-8231-
f00ffab90ce4
phk [Tue, 22 Aug 2006 10:46:16 +0000 (10:46 +0000)]
Optimize shmlog writing:
If we know the record length, only hold the mutex while we reserve
the space. Until we change the first byte, nothing bad can happen.
XXX: a memory barrier is strictly speaking necessary before we assign
the first byte.
If there are no '%' in the format string, treat as fixed length for
speed.
git-svn-id: svn+ssh://projects.linpro.no/svn/varnish/trunk@895
d4fa192b -c00b-0410-8231-
f00ffab90ce4
phk [Tue, 22 Aug 2006 10:40:55 +0000 (10:40 +0000)]
Consistent naming of mutex
git-svn-id: svn+ssh://projects.linpro.no/svn/varnish/trunk@894
d4fa192b -c00b-0410-8231-
f00ffab90ce4
des [Tue, 22 Aug 2006 10:14:39 +0000 (10:14 +0000)]
Allow the delay between screen updates to be specified with -w.
git-svn-id: svn+ssh://projects.linpro.no/svn/varnish/trunk@893
d4fa192b -c00b-0410-8231-
f00ffab90ce4
des [Tue, 22 Aug 2006 09:37:58 +0000 (09:37 +0000)]
Correct the error check for write().
git-svn-id: svn+ssh://projects.linpro.no/svn/varnish/trunk@892
d4fa192b -c00b-0410-8231-
f00ffab90ce4
des [Tue, 22 Aug 2006 09:31:34 +0000 (09:31 +0000)]
Slight optimization: use strlcpy() to avoid calloc().
git-svn-id: svn+ssh://projects.linpro.no/svn/varnish/trunk@891
d4fa192b -c00b-0410-8231-
f00ffab90ce4
phk [Tue, 22 Aug 2006 09:30:23 +0000 (09:30 +0000)]
Swap the '|' and '#' symbols so '#' means "miss", that's much more
graphically useful.
git-svn-id: svn+ssh://projects.linpro.no/svn/varnish/trunk@890
d4fa192b -c00b-0410-8231-
f00ffab90ce4
phk [Tue, 22 Aug 2006 09:16:33 +0000 (09:16 +0000)]
Use different marks for hit & miss
Autoscale vertical axis.
git-svn-id: svn+ssh://projects.linpro.no/svn/varnish/trunk@889
d4fa192b -c00b-0410-8231-
f00ffab90ce4
phk [Tue, 22 Aug 2006 08:55:15 +0000 (08:55 +0000)]
Eliminate the MD5 optional code. There is no sufficiently strong
statistical basis for using MD5 that will outweigh the performance
penalty or "IT USES THE BR0K3N MD5 ALGORITM" cries on slashdot.
The only known artifact in CRC32 is that hashing it with a power
of two is slightly inefficient (a few percent in bucket length
standard deviation) if you have URLs with fixed width fields of
a limited charset (such as numeric) fields in them.
Avoid this by checking the hash width specified for power of two,
and reduce it by one telling the user why.
Using a hash width that is a prime number does provable advantage
over just not using power-of-two width. I have heard a mathematician
say that this is inherent in the design of the polynomia chosen
for CRC algorithms error (optimized for bit error detection) but
the actual math is way beyond me.
Increase default hash width to 16383, which is probably still
smaller than it should be.
git-svn-id: svn+ssh://projects.linpro.no/svn/varnish/trunk@888
d4fa192b -c00b-0410-8231-
f00ffab90ce4
des [Tue, 22 Aug 2006 08:18:15 +0000 (08:18 +0000)]
Expand keywords.
git-svn-id: svn+ssh://projects.linpro.no/svn/varnish/trunk@887
d4fa192b -c00b-0410-8231-
f00ffab90ce4
des [Tue, 22 Aug 2006 08:17:54 +0000 (08:17 +0000)]
Allow an empty address if a port is specified; thus ":80" is a valid
listening address ("port 80 on all interfaces")
git-svn-id: svn+ssh://projects.linpro.no/svn/varnish/trunk@886
d4fa192b -c00b-0410-8231-
f00ffab90ce4
phk [Tue, 22 Aug 2006 08:06:46 +0000 (08:06 +0000)]
Additional marginal improvement:
Sort on length of key, then on digest.
git-svn-id: svn+ssh://projects.linpro.no/svn/varnish/trunk@885
d4fa192b -c00b-0410-8231-
f00ffab90ce4
phk [Tue, 22 Aug 2006 07:52:33 +0000 (07:52 +0000)]
Rework the classic hasher to reduce lock contention and CPU usage.
Make nbuckets=nhash mandatory.
Order hash lists by the length of key instead of the key contents,
comparing the length is much faster.
Also compare disgest before we take the expensive content compare.
Use memcmp() for content compare instead of strcmp().
Use a two-pass algorithm for inserts to lower mutex contention.
git-svn-id: svn+ssh://projects.linpro.no/svn/varnish/trunk@884
d4fa192b -c00b-0410-8231-
f00ffab90ce4
des [Tue, 22 Aug 2006 07:24:07 +0000 (07:24 +0000)]
Add cache_acceptor.h to noinst_HEADERS.
git-svn-id: svn+ssh://projects.linpro.no/svn/varnish/trunk@883
d4fa192b -c00b-0410-8231-
f00ffab90ce4
des [Tue, 22 Aug 2006 07:11:59 +0000 (07:11 +0000)]
Add strndup() to libcompat.
git-svn-id: svn+ssh://projects.linpro.no/svn/varnish/trunk@882
d4fa192b -c00b-0410-8231-
f00ffab90ce4
phk [Mon, 21 Aug 2006 20:42:47 +0000 (20:42 +0000)]
Flexelint inpired cleanups
git-svn-id: svn+ssh://projects.linpro.no/svn/varnish/trunk@881
d4fa192b -c00b-0410-8231-
f00ffab90ce4
phk [Mon, 21 Aug 2006 20:30:29 +0000 (20:30 +0000)]
Give VBE_ClosedFd() an argument to tell if the fd has already
been closed.
Pipe does this and would panic otherwise.
git-svn-id: svn+ssh://projects.linpro.no/svn/varnish/trunk@880
d4fa192b -c00b-0410-8231-
f00ffab90ce4
phk [Mon, 21 Aug 2006 20:25:28 +0000 (20:25 +0000)]
Fix these two up to current standard.
Poll is tested, epoll isn't.
While the three implementations share a lot of identical code
right now, I will wait a bit before unifying more of them, at
least until performance proves that this is the right way for
kqueue.
XXX: they're really not acceptors any more, they're herders.
git-svn-id: svn+ssh://projects.linpro.no/svn/varnish/trunk@879
d4fa192b -c00b-0410-8231-
f00ffab90ce4
phk [Mon, 21 Aug 2006 20:23:47 +0000 (20:23 +0000)]
Make it easier to experiement with acceptors by having a single
pointer to the one in use.
git-svn-id: svn+ssh://projects.linpro.no/svn/varnish/trunk@878
d4fa192b -c00b-0410-8231-
f00ffab90ce4
phk [Mon, 21 Aug 2006 20:23:14 +0000 (20:23 +0000)]
Remove unused #includes
git-svn-id: svn+ssh://projects.linpro.no/svn/varnish/trunk@877
d4fa192b -c00b-0410-8231-
f00ffab90ce4
phk [Mon, 21 Aug 2006 19:05:57 +0000 (19:05 +0000)]
Add asserts.
Return if we pass on directly.
git-svn-id: svn+ssh://projects.linpro.no/svn/varnish/trunk@876
d4fa192b -c00b-0410-8231-
f00ffab90ce4
phk [Mon, 21 Aug 2006 18:55:24 +0000 (18:55 +0000)]
Overhaul kqueue acceptor in light of todays learnings.
Use the pipe trick to inject sessions into the system, as far as I
can tell it is cheaper because of the low rate it happens and the
high rate of mutex operations avoided.
Ignore the timer event, but purge the list every time we wake up
to reduce lumpyness of timeout'ing.
Centralize the polling of a session so we don't have the same two
messages spread out all over the place.
Centralize the acceptor thread and send things directly to the worker
thread, leaving only the session-herder in the split out files.
poll & epoll not yet updated accordingly.
git-svn-id: svn+ssh://projects.linpro.no/svn/varnish/trunk@875
d4fa192b -c00b-0410-8231-
f00ffab90ce4