Hi all,
I've just released 0.24.4. This is a major bug-fix release, and
everyone who's running any 0.24.x release is highly recommended to
upgrade; you otherwise risk file corruption on your clients.
You can get links to the packages from here:
http://reductivelabs.com/trac/puppet/wiki/DownloadingPuppet
Although we've fixed multiple sources of memory leaks, there are still
indications that the server leaks memory, and especially so when the
http connection gets killed. If you suffer these memory problems and
can help debug this, we'd really appreciate the help.
And here's the changelog:
Pass source to pkg_add via the PKG_PATH environment variable if
it ends in a '/' indicating it is a directory. Allows pkg_add
to resolve dependancies, and make it possible to specify packages
without version numbers.
Fixing #571 -- provider suitability is now checked at resource
evaluation time, rather than resource instantiation time. This
means that you don't catch your "errors" as early, but it also
means you should be able to realistically configure a whole host
in one run.
Moved the configuration of the Node cache to the puppetmasterd
executable, since it otherwise causes caches to be used in all
cases, which we don't want (e.g., bin/puppet was using them).
Ported #198 man page creation functionality to 0.24.x branch and
added man pages and man page creation logic to install.rb. The
man pages are stored in man/man8 and will install to config::CONFIG
mandir/man8.
Fixing #1138 -- the yamldir is automatically created by the
server now that it's in the :puppetmasterd section rather than
a separate :yaml section.
Disabling http keep-alive as a means of preventing #1010.
There is now a constant in Puppet::Network::HttpPool that will
disable or enable this feature, but note that we determined
that it can cause corruption, especially in file serving (but
it's client-side corruption).
Applying patch by Ryan McBride to fix OpenBSD package
matching. The actual problem was caused by the fix to #1001.
Found all instances of methods where split() is used without
any local variables and added a local variable -- see
http://snurl.com/21zf8. My own testing showed that this
caused memory growth to level off at a reasonable level.
Note that the link above says the problem is only with class
methods, but my own testing showed that it's any method that
meets these criteria. This is not a functional change, but
should hopefully be the last nail in the coffin of #1131.
Found an array that leaked pretty quickly between reparsing
files, thanks to work by Adam Jacob and Arjuna Christenson
(the finding, not the leak). I'm going to act like this
fixes #1131, at least for now, but I doubt it does,
since that shows general memory growth over time, whereas
the leak here should go away as soon as files are reparsed
(because the parser is holding the reference to the leaking
array).
Fixed #1147: Cached nodes are correctly considered out of
date if the node facts have been updated (thus causing
node facts to again be available in manifests, for those
cases where they were not).
Fixed #1137: The certificate name is correctly being added
to the facts hash.
Fixed #1136: Verbose and Debug no longer clobber each other.
Hopefully *finally* fixed the "already being managed" problem
(#1036). The problem only cropped up if there was a failure
when trying to manage the system -- this would cause the
setting-based resources not to get cleaned up.
--
Experience is that marvelous thing that enables you recognize a
mistake when you make it again. -- F. P. Jones
---------------------------------------------------------------------
Luke Kanies | http://reductivelabs.com | http://madstop.com
|
