This wires up cleanup code shared memory segments.
It uses slightly modified copies of pthread_shared_mem.cc/h, containing a change
proposal to support our ordering of events during a configuration reload.
It can be tested and reviewed through this branch.
I did some extensive tests, it seems segment cleanup is not perfect yet
(segments for named locks, per vhost statistics segments). I think we need to
fix that in the trunk-tracking branch or in a separate pull though. It seems the
current trunk of psol has improvements that may fix this, and fixing it in this
pull will make it larger and more complex then I would like it to be.
Squash-merge of Otto's #179.
This allows viewing the last few messages written to the SharedCircularBuffer in
NgxMessageHandler through url /ngx_pagespeed_message
Related issue: https://github.com/pagespeed/ngx_pagespeed/issues/153
Squash-merge of Otto's #180.
This adds /ngx_pagespeed_statistics. It also uses SharedMemoryStatistics instead
of the SimpleStatistics currently used, and sets them as the default. It also
adds support for tracking per server{} statistics.
Cleaning up of shared segments is disabled, because the current psol version we
link in has a problem with the ordering of nginx's events during a configuration
reload.
The referrer statistics will be ported in a separate pull.
Related to issue #133
Squash-merge of Otto's work in #176.
This adds an NgxMessageHandler, that uses ngx_log_error to output messages. It
is heavily based on apache_message_handler from mod_pagespeed.
It falls back to an internal GoogleMessageHandler if the required initialisation
has not been performed yet.
It can also install a crash handler, that will try to log a backtrace upon
receiving certain signals. In preparation of us porting /ngx_pagespeed_messages,
it also keeps the last few messages in a SharedCircularBuffer
(Squash-merge of Otto's #163)
We need to have the NgxRewriteDriverFactory available at configuration parse
time to set options on it. This change moves instantiation of the factory in
front of configuration parsing. Afterwards, a check is done to see if we where
enabled anywhere. If we where not enabled / do not have any active server
contexts, the NgxRewriteDriverFactory is deleted. It's not very elegant, but we
can't think of anything better currently.
Squash merge of Otto's #165.
A normal .pagespeed. url looks like:
256x192vPuzzle.jpg.pagespeed.ic.alA0AdlR-2.jpg
If someone turns on RunExperiment, however, we generate urls like:
256x192vPuzzle.jpg.pagespeed.a.ic.alA0AdlR-2.jpg
256x192vPuzzle.jpg.pagespeed.b.ic.alA0AdlR-2.jpg
The new section is always a single letter between a and z. We need to catch
these and pass them through to PageSpeed.
Before the experiment code was just setting the need_to_store_experiment_data
option but nothing was looking at that option. Rewrote the code, basing it on
InstawebContext::SetFuriousStateAndCookie, in order to actually set the cookie
to support the experiment. There are tests for this functionality, but we're
waiting on issue #40 to be able to run them.
Manual testing (with line wrapping):
$ curl -s -D- http://localhost:8050/mod_pagespeed_example/ | grep GFURIOUS
Set-Cookie: _GFURIOUS=1; Expires=Fri, 01 Feb 2013 21:38:31 GMT;
Domain=.localhost; Path=/
$ curl -s -D- http://localhost:8050/mod_pagespeed_example/ | grep GFURIOUS
Set-Cookie: _GFURIOUS=2; Expires=Fri, 01 Feb 2013 21:42:56 GMT;
Domain=.localhost; Path=/
And it responds properly on receiving the cookie, although it did that before
this change too.
guard against server only options being configured at location scope this
introduces a duplicated list of server-level-only options from mod_instaweb.cc,
which is a short term solution and should be removed later on
Fixes#99.
Shut down the driver factory before the proxy fetch factory.
shutting down the driver factory will make sure that all worker
threads are stopped, This prevents in-flight callbacks to
destroyed proxy fetch factories, so we should be able to safely
delete them
This fixes#109.
If we're proxying the origin server might advertise `Accept-Ranges: bytes` but
we don't support range requests. So we need to remove the header. We also need
to not add `Vary: Accept-Encoding` twice, and because the gzip encoding will add
it later we need to remove it here.
The file cache needs a slow worker or it will never run the cleanup code. In
configurations with multiple server blocks, like jefftk.com, the rewrite
driver factory could cause a a segfault by deleting the slow worker when setting
up caches. So this fixes#126.
Otto van der Schaaf