Until wiki page updates are working, updates to 2.0 changes file will
not show, but you can see them through the repository browser at:
http://code.google.com/p/modwsgi/source/browse/wiki/ChangesInVersion0200.wiki
Because of wiki page problems, also keeping link to tar ball off main
page, but you can see it by going to:
http://code.google.com/p/modwsgi/downloads/list
The main changes since version 2.0c4 which are contained in this
release candidate are:
1. Improvements to network performance for daemon mode processes. For
a simple hello world test application the improvement was about 40%.
In relation to the performance estimates previously quoted on mod_wsgi
site, this takes WSGIDaemonProcess case from 500 to about 700.
2. Support wsgi.file_wrapper with mmap/sendfile optimisations where
possible on UNIX systems.
3. Ability to change displayed name of daemon processes in 'ps' command output.
Various other changes have also been made which others may view as
equally important, but those above are the ones which would be most
attractive to most.
It is really hoped that this is now the final release candidate for
2.0 and in a few weeks the final official 2.0 can be released. It may
take me a little while to expand the wiki documentation to cover new
changes in this release candidate.
Anyway, please let me know of any major issues ASAP.
Enjoy.
Graham
Graham
Try it with a simple hello world application as described in
configuration guidelines. Make sure it is a simple GET request. Ensure
that LogLevel directive in Apache is set to 'debug' and see what the
Apache error log files are saying. Time exactly how long it takes to
do a request.
Graham
When loading a new version of module, it is sometimes necessary to do
an Apache 'stop' and 'start'. A 'restart' or 'graceful' is not enough.
Were you doing a full 'stop'?
Graham
carl....@gmail.com
-------------------------------------------------------------------------------
It perhaps depends on what platform you are using and how that
distribution has modified the standard Apache apachectl script. The
'httpd' executable based on source code that comes from ASF doesn't
support the 'reload' option that you talk about.
In general, for an unmodified build of Apache from ASF source,
'restart' does not result in Apache parent process being killed.
Instead, it kills off all Apache child processes, unloads the
dynamically loaded Apache modules from parent Apache process and then
starts over again, applying configuration and reloading modules
(potentially new versions) as appropriate.
Even if a module has changed versions, this restart process should
work okay, but in some odd cases, for one reason or another it may not
be able to unload a module or some shared library it depends upon.
This can cause problems when starting over again.
Thus, why a full 'stop' is required so that Apache parent process is killed off.
Anyway, one thing I may have to do some checking on is a 'graceful'
restart. When mod_wsgi is communicating between Apache child process
and daemon process it uses a secret key to validate that it is
receiving data from where it expects to. This key includes the Apache
generation, which is incremented on a 'restart' or 'graceful'.
What I need to check is what lifetime the existing Apache child
processes have and check whether they could still handle requests. If
they can, problem is that daemon has been restarted and is expecting a
newer generation number than old Apache child processes. As a result
the daemon would reject request and so requests going through older
Apache child process would eventually fail with service unavailable.
It was intended to work this way, just not sure how long Apache child
processes can hang around and keep serving requests on 'graceful'
restart or whether guaranteed they will not handle new requests.
Graham
the package was just uploaded to Debian/experimental. Sorry for the
delay - I'm still not completely settled in my new home.
> It is really hoped that this is now the final release candidate for
> 2.0 and in a few weeks the final official 2.0 can be released. It may
> take me a little while to expand the wiki documentation to cover new
> changes in this release candidate.
That would be great, so we can get it into Debian's Lenny before the freeze.
Cheers,
Bernd
--
Bernd Zeimetz
<be...@bzed.de> <http://bzed.de/>
When is the cut off? Would they still except 2.0 even though it is
still in experimental stream? In other words, is getting it into
testing stream sufficient or does it also need to pass through that
stream as well?
Graham