gistfile1.txt

------------------------------------------------------------------------
r5217 | phk | 2010-09-15 21:01:42 +0200 (on., 15 sep. 2010) | 7 lines

Fix a cornercase of chunked encoding and malloc stevedore:

Don't try to trim a storage segment we filled completely.

Fixes: #776

------------------------------------------------------------------------
r5216 | phk | 2010-09-15 21:00:29 +0200 (on., 15 sep. 2010) | 3 lines

Make it possible to supress the automatic content-length header.


------------------------------------------------------------------------
r5211 | martin | 2010-09-15 09:54:50 +0200 (on., 15 sep. 2010) | 4 lines

Set the int status when setting the txt header status in VRT_l_[obj|resp|beresp]_status.

Fixes: #769

------------------------------------------------------------------------
r5210 | phk | 2010-09-15 09:22:14 +0200 (on., 15 sep. 2010) | 5 lines

Lock the sms counters properly

Fixes: #768

------------------------------------------------------------------------
r5195 | phk | 2010-09-13 12:00:41 +0200 (ma., 13 sep. 2010) | 11 lines

Fix an embarrasing typo in the WM-Aware binary heap layout.

The effect of this is that the root object in the binheap could be
wrong and thus hold the expiry thread hostage, even though other objects
were ripe for expiry. 

This would show up as inflated obj/objcore/objhdr counts, but have no
other deleritous effects.

Detected, conclusively, by:     sky

------------------------------------------------------------------------
r5192 | tfheen | 2010-09-12 20:18:47 +0200 (sø., 12 sep. 2010) | 4 lines

Change vcl_pipe comment to talk about bereq, not req

Thanks to DocWilco for spotting this.

------------------------------------------------------------------------
r5186 | phk | 2010-09-08 12:04:17 +0200 (on., 08 sep. 2010) | 3 lines

The expire thread don't actuall need a VCL reference any more.

------------------------------------------------------------------------
r5178 | phk | 2010-09-08 09:32:53 +0200 (on., 08 sep. 2010) | 4 lines

Always include <sys/param.h> on FreeBSD so we are sure we have a
definition for __FreeBSD_version.

------------------------------------------------------------------------
r5176 | martin | 2010-09-07 15:12:12 +0200 (ti., 07 sep. 2010) | 4 lines

Apply optional unbuffered output on stdout before call to do_order.

Fixes: #760

------------------------------------------------------------------------
r5175 | kristian | 2010-09-06 16:57:27 +0200 (ma., 06 sep. 2010) | 2 lines

More VCL tests for the DNS director
------------------------------------------------------------------------
r5174 | kristian | 2010-09-06 16:38:27 +0200 (ma., 06 sep. 2010) | 2 lines

Don't output old .ident string for DNS director .list

------------------------------------------------------------------------
r5173 | kristian | 2010-09-06 16:24:19 +0200 (ma., 06 sep. 2010) | 2 lines

Correct typo in DNS director example

------------------------------------------------------------------------
r5172 | kristian | 2010-09-06 16:23:30 +0200 (ma., 06 sep. 2010) | 6 lines

DNS director test

It correctly identifies that the .list option is currently out-of-sync with
the rest of VCC and needs to be adjusted. It's on the todo.


------------------------------------------------------------------------
r5170 | martin | 2010-09-06 15:21:26 +0200 (ma., 06 sep. 2010) | 3 lines

Introduce req.hash_always_miss (force cache miss) and req.hash_ignore_busy (ignore busy objects) variables available in vcl_fetch.


------------------------------------------------------------------------
r5169 | perbu | 2010-09-06 14:40:48 +0200 (ma., 06 sep. 2010) | 1 line

more on beresp.cacheable
------------------------------------------------------------------------
r5168 | perbu | 2010-09-06 14:15:08 +0200 (ma., 06 sep. 2010) | 1 line

cleanup + add return()
------------------------------------------------------------------------
r5167 | perbu | 2010-09-06 14:03:43 +0200 (ma., 06 sep. 2010) | 1 line

rewrite beresp/obj.cacheable (thanks DocWilco) and fixed some formatting errors

------------------------------------------------------------------------
r5162 | martin | 2010-08-31 12:35:16 +0200 (ti., 31 aug. 2010) | 8 lines

Make varnishncsa only consider log information between the open and close
log tags for a connection, and clear the collected log data on seeing the open
tag. This resolves an issue with log data where varnishncsa would consider a
set of data as bogus because of extra data in the log belonging to the previous
connection.

Support case 164
.
------------------------------------------------------------------------
r5160 | martin | 2010-08-31 09:45:29 +0200 (ti., 31 aug. 2010) | 4 lines

Add explicit log flushing when backend connections close to keep log chronology. This is to avoid log lines for the same (reused) FD coming before the closing connection when load is high.

Fixes #739

------------------------------------------------------------------------
r5158 | phk | 2010-08-30 14:52:09 +0200 (ma., 30 aug. 2010) | 4 lines

Don't compile the pthread based memory-barrier fall back unless we 
actually need it.

------------------------------------------------------------------------
r5135 | ingvar | 2010-08-26 15:49:35 +0200 (to., 26 aug. 2010) | 1 line

fixed a typo and changed to a more recent version number example in README.redhat
------------------------------------------------------------------------
r5129 | phk | 2010-08-25 11:47:48 +0200 (on., 25 aug. 2010) | 3 lines

Add comments with BNF-ish syntax since this is starting to get hairy.


------------------------------------------------------------------------
r5128 | ingvar | 2010-08-25 11:30:04 +0200 (on., 25 aug. 2010) | 1 line

makes trunk build again, including new docs
------------------------------------------------------------------------
r5127 | phk | 2010-08-25 11:19:43 +0200 (on., 25 aug. 2010) | 3 lines

white-space anal-retentive fixup


------------------------------------------------------------------------
r5126 | phk | 2010-08-25 11:01:05 +0200 (on., 25 aug. 2010) | 3 lines

More svn:keywords fixup


------------------------------------------------------------------------
r5125 | phk | 2010-08-25 10:57:04 +0200 (on., 25 aug. 2010) | 3 lines

Add missing svn:keywords


------------------------------------------------------------------------
r5124 | phk | 2010-08-25 10:52:44 +0200 (on., 25 aug. 2010) | 3 lines

An XXX comment


------------------------------------------------------------------------
r5123 | phk | 2010-08-25 10:49:27 +0200 (on., 25 aug. 2010) | 10 lines

The -T argument can be specified as :0 which will just find one or
two (IPv4/6) available sockets.  In this case, writing the :0 into
the VSM will not tell varnishadm where to look.

Write the actual socket addresses, one per line instead, and have
varnishadm try them all.

Make -n mutually exclusive with -S and -T on varnishadm.

------------------------------------------------------------------------
r5122 | phk | 2010-08-25 10:42:45 +0200 (on., 25 aug. 2010) | 4 lines

Look for separating space before colon, so that we can grok ":: 50239"
as a socketaddr.


------------------------------------------------------------------------
r5121 | ingvar | 2010-08-25 10:08:22 +0200 (on., 25 aug. 2010) | 1 line

Upping this test to 1s. It times out on some builds on ppc64.
------------------------------------------------------------------------
r5120 | ingvar | 2010-08-25 10:07:04 +0200 (on., 25 aug. 2010) | 1 line

A stack size of 128k is too small for rhel6/ppc64. Upping to 256k.
------------------------------------------------------------------------
r5119 | phk | 2010-08-24 17:56:49 +0200 (ti., 24 aug. 2010) | 6 lines

Always report sockaddr's as
        IP# <space> PORT#

Explain why.


------------------------------------------------------------------------
r5118 | ingvar | 2010-08-24 17:49:52 +0200 (ti., 24 aug. 2010) | 2 lines

Just a typofix

------------------------------------------------------------------------
r5117 | phk | 2010-08-23 11:34:33 +0200 (ma., 23 aug. 2010) | 9 lines

Have VSS_resolve() call VSS_parse() and only use the provided port argument
as a default, if the host argument does not specify a port.

Elminate most separate calls to VSS_parse()

Make it possible to specify address and port of backends in VCL using
only the .host field.


------------------------------------------------------------------------
r5116 | phk | 2010-08-21 16:57:13 +0200 (lø., 21 aug. 2010) | 8 lines

Whenever the CLI comms with the child process runs of the rails, we
have to kill the child:  We cannot get the CLI pipes back in sync,
even if the child process was just preoccupied with something for 
much longer than cli_timeout.

Fixes: #744


------------------------------------------------------------------------
r5115 | phk | 2010-08-21 14:34:15 +0200 (lø., 21 aug. 2010) | 3 lines

Only provide help on commands authorrized for the current CLI session


------------------------------------------------------------------------
r5114 | phk | 2010-08-21 14:21:48 +0200 (lø., 21 aug. 2010) | 3 lines

Add a couple of documenting asserts.


------------------------------------------------------------------------
r5113 | phk | 2010-08-21 14:03:19 +0200 (lø., 21 aug. 2010) | 5 lines

Don't mess with X-forwarded-for on restarts.

Fixes: #758


------------------------------------------------------------------------
r5112 | phk | 2010-08-21 14:02:17 +0200 (lø., 21 aug. 2010) | 3 lines

Log the length of cli responses to VSL


------------------------------------------------------------------------
r5111 | phk | 2010-08-21 14:00:59 +0200 (lø., 21 aug. 2010) | 3 lines

Emit varnishd ident as part of banner.


------------------------------------------------------------------------
r5110 | phk | 2010-08-16 12:59:18 +0200 (ma., 16 aug. 2010) | 3 lines

Polish while looking for clues to #754/#744


------------------------------------------------------------------------
r5109 | kristian | 2010-08-13 20:53:42 +0200 (fr., 13 aug. 2010) | 3 lines

tools/cookie-stripper: Script to generate simple VCL that strips specific
cookies.

------------------------------------------------------------------------
r5108 | perbu | 2010-08-13 14:07:24 +0200 (fr., 13 aug. 2010) | 1 line

described critbit in the man page
------------------------------------------------------------------------
r5107 | kristian | 2010-08-13 13:52:04 +0200 (fr., 13 aug. 2010) | 5 lines

Add a varnishtests/Makefile.kristian for parallel tests

Usage: cd bin/varnishtest/; make -f Makefile.kristian -j20


------------------------------------------------------------------------
r5106 | tfheen | 2010-08-13 13:15:20 +0200 (fr., 13 aug. 2010) | 7 lines

Correct embedded dot about obj.* in vcl_fetch

The embedded dot claimed obj.* was available in vcl_fetch, something
it's not.  Fix this.

Thanks to Yves Hwang for spotting this.

------------------------------------------------------------------------
r5105 | phk | 2010-08-13 12:40:12 +0200 (fr., 13 aug. 2010) | 3 lines

Remove this old test file


------------------------------------------------------------------------
r5104 | phk | 2010-08-13 12:39:40 +0200 (fr., 13 aug. 2010) | 3 lines

Add a testcase for backend retries.


------------------------------------------------------------------------
r5103 | phk | 2010-08-13 12:19:40 +0200 (fr., 13 aug. 2010) | 7 lines

Add a missing assert.

Spotted by: pprocacci  (Thanks!)

Fixes: #752


------------------------------------------------------------------------
r5102 | phk | 2010-08-13 12:14:05 +0200 (fr., 13 aug. 2010) | 6 lines

Retry backend fetches one time if we got a recycled conncetion and
the transfer failed before we received anything.

Fixes: #749


------------------------------------------------------------------------
r5101 | phk | 2010-08-13 11:40:18 +0200 (fr., 13 aug. 2010) | 10 lines

Rework the FetchHdr() and surroundings a litle bit, to get us a return
value that indicates if a failure should be retried.

Condition for retry is that the connection is recycled and fails before
we receive any data from the other end.

(The actual retry is not implemented yet)
------------------------------------------------------------------------
r5099 | phk | 2010-08-12 10:58:21 +0200 (to., 12 aug. 2010) | 5 lines

Use bprintf() into static buffers to get overflow assert.

Whitespace nits

------------------------------------------------------------------------
r5096 | phk | 2010-08-12 10:13:15 +0200 (to., 12 aug. 2010) | 3 lines

Set backend health good if we stop polling the health.

------------------------------------------------------------------------
r5091 | phk | 2010-08-11 15:22:31 +0200 (on., 11 aug. 2010) | 3 lines

Widen ident field to 64 byte.

------------------------------------------------------------------------
r5086 | phk | 2010-08-11 10:59:40 +0200 (on., 11 aug. 2010) | 3 lines

Add string representation of backend ipv4/ipv6 addresses to compiled VCL.


------------------------------------------------------------------------
r5085 | perbu | 2010-08-10 17:39:36 +0200 (ti., 10 aug. 2010) | 1 line

removed a rst warning
------------------------------------------------------------------------
r5084 | perbu | 2010-08-10 17:38:20 +0200 (ti., 10 aug. 2010) | 1 line

cleaned up the index, must restructure the tutorial some day to balance out the chapters
------------------------------------------------------------------------
r5083 | perbu | 2010-08-10 17:21:42 +0200 (ti., 10 aug. 2010) | 1 line

missing set
------------------------------------------------------------------------
r5082 | phk | 2010-08-10 11:40:18 +0200 (ti., 10 aug. 2010) | 8 lines

Fix a corner-case of pipe mode close-down processing, by simplifying
the logic and letting vca_close_session() and VBE_CloseFd() carry out
the last rites.

Closes: #746



------------------------------------------------------------------------
r5081 | phk | 2010-08-10 11:23:39 +0200 (ti., 10 aug. 2010) | 5 lines

VBE_ClosedFd() was badly named, loose the 'd' to indicate that it
will close the Fd, rather than being an indication this already
happened.


------------------------------------------------------------------------
r5080 | phk | 2010-08-10 11:08:21 +0200 (ti., 10 aug. 2010) | 10 lines

Add a new VCL variable "client.identity" which defaults to client.ip
(without port number).

This variable is used by the "client" director to distribute sessions
across multiple backends.

Having it be a separate variable from client.ip makes it possible to
fill it from X-Forwarded-For: or Cookie headers.

------------------------------------------------------------------------
r5077 | phk | 2010-08-09 17:05:52 +0200 (ma., 09 aug. 2010) | 3 lines

upd


------------------------------------------------------------------------
r5076 | perbu | 2010-08-09 14:11:19 +0200 (ma., 09 aug. 2010) | 1 line

add initial and duplcation of probes
------------------------------------------------------------------------
r5075 | phk | 2010-08-09 10:14:15 +0200 (ma., 09 aug. 2010) | 9 lines

If we get a HTTP/1.1 response with no indicatation of length, assume EOF
encoding, rather than zero length.

This has implications for a number of testcases which were written
using the previous assumption.

Fixes: #733


------------------------------------------------------------------------
r5074 | phk | 2010-08-09 08:55:52 +0200 (ma., 09 aug. 2010) | 6 lines

Do not suppress the Content-Length:, Age:, Range: and Proxy-Auth headers
from the backend on a pass operation.

Fixes: #730


------------------------------------------------------------------------
r5073 | phk | 2010-08-09 08:22:27 +0200 (ma., 09 aug. 2010) | 5 lines

Don't explode in vcl.show if VCL code contains a '%' sign.

Fixes: 742


------------------------------------------------------------------------
r5072 | phk | 2010-08-09 08:10:17 +0200 (ma., 09 aug. 2010) | 3 lines

Fix the error emitted if quotes on a line are not properly balanced.

------------------------------------------------------------------------
r5071 | perbu | 2010-08-04 15:13:13 +0200 (on., 04 aug. 2010) | 1 line

Remove the command line. Incorrect and not really needed.
------------------------------------------------------------------------
r5070 | tfheen | 2010-08-04 14:48:54 +0200 (on., 04 aug. 2010) | 4 lines

Mention pkg-config in build requirements too

Fixes #740

------------------------------------------------------------------------
r5069 | phk | 2010-08-03 22:41:39 +0200 (ti., 03 aug. 2010) | 3 lines

Ramblings about why we do not support Zilog S-8000 Zeus as a platform.


------------------------------------------------------------------------
r5068 | kristian | 2010-08-02 21:46:49 +0200 (ma., 02 aug. 2010) | 2 lines

Clean up obj/beresp left overs in vcl(7).

------------------------------------------------------------------------
r5067 | kristian | 2010-08-02 21:36:28 +0200 (ma., 02 aug. 2010) | 2 lines

Update vcl(7) for 2.1: return(foo) is mandatory.

------------------------------------------------------------------------
r5064 | kristian | 2010-08-02 20:30:38 +0200 (ma., 02 aug. 2010) | 2 lines

DNS Director: Include the necessary headers to build on FreeBSD too.

------------------------------------------------------------------------
r5063 | kristian | 2010-08-02 12:40:35 +0200 (ma., 02 aug. 2010) | 6 lines

Clean out some silly imprecisions, clearing the build

Some minor nitpicking and some slightly ugly stuff. It currently fails a
regression test (v00017.vtc). I'm on it.


------------------------------------------------------------------------
r5062 | kristian | 2010-08-01 16:45:22 +0200 (sø., 01 aug. 2010) | 10 lines

Add a DNS director

The DNS director allows Varnish to pick backend based on the Host header
provided by the client, and how it resolves in DNS. A suffix can be added
to make it "internal" (see vcl(7)).

There's still some quirks that I want to work out, but this seems fairly
commit-ready and non-intrusive.


------------------------------------------------------------------------
r5061 | ingvar | 2010-07-31 01:17:36 +0200 (lø., 31 juli 2010) | 6 lines

* Thu Jul 29 2010 Ingvar Hagelund <[email protected]> - 2.1.4-0.svn20100730
- Replaced specific include files with a wildcard glob
- Needs python-sphinx to build sphinx documentation
- Builds html and latex documentation. Put that in a subpackage varnish-docs


------------------------------------------------------------------------
r5060 | ingvar | 2010-07-30 15:26:48 +0200 (fr., 30 juli 2010) | 3 lines

- Synced bits from Fedora
- Made to work with trunk again, see specfile changelog

------------------------------------------------------------------------
r5059 | ingvar | 2010-07-30 15:24:24 +0200 (fr., 30 juli 2010) | 2 lines

Simpler script to build rpm files from trunk

------------------------------------------------------------------------
r5058 | ingvar | 2010-07-30 14:04:26 +0200 (fr., 30 juli 2010) | 3 lines

missing ifdefs for pagesize/pointersize s390 (32bit)