Puppet

Possible fix in svn rev 2217

After further testing, this apparently doesn’t fix it completely. It just slows the accumulation of open sockets/file handles.

I guess we’ll need to find the spot (or spots) where all the storeconfigs and export/collection operations finish, and close each connection explicitly. Alternately, we may also be able to wrap all the rails code in something that ensures that all threads using rails use the same connection handle, this is probably the best approach, but it’s also the one that requires the most intimate knowledge of puppet internals.

This seems better, but still leaking on the sqlite (debian package 0.24.1-2, rails 1.1.6-3)

  1 
  3 /dev/null
  1 /dev/urandom
  1 /etc/group
  1 /etc/passwd
  2 pipe:r29756375
  1 socket:r29756390
  1 socket:r29756393

1008 /var/lib/puppet/state/clientconfigs.sqlite3

  3 /var/log/puppet/masterhttp.log
  1 /var/log/puppet/rails.log

This looks related: http://blog.moertel.com/articles/2006/08/24/database-connection-leak-in-typo-4-0-3-problem-solved

It leaks by 1 FD each node run, with the patch below it stops leaking. I’ll leave this running for a little while. I don’t know what unexpected side effects this may have.

—– /usr/lib/ruby/1.8/puppet/rails.rb.orig 2008-03-27 10:08:54.000000000 +0000 +++ /usr/lib/ruby/1.8/puppet/rails.rb 2008-03-27 10:07:26.000000000 +0000 @@ -21,7 +21,7 @@

         [[ActiveRecord]]::Base.logger.level = Logger::DEBUG
     end

•  [[ActiveRecord]]::Base.allow_concurrency = true
  

•  [[ActiveRecord]]::Base.allow_concurrency = false
   [[ActiveRecord]]::Base.verify_active_connections!
  
   begin
  

let’s try that again, remembering the formatting this time (can’t see an edit unfortunately)

This seems better, but still leaking on the sqlite (debian package 0.24.1-2, rails 1.1.6-3)

  3 /dev/null
  1 /dev/urandom
  1 /etc/group
  1 /etc/passwd
  2 pipe:r29756375
  1 socket:r29756390
  1 socket:r29756393
  1008 /var/lib/puppet/state/clientconfigs.sqlite3
  3 /var/log/puppet/masterhttp.log
  1 /var/log/puppet/rails.log

This looks related: http://blog.moertel.com/articles/2006/08/24/database-connection-leak-in-typo-4-0-3-problem-solved

It leaks by 1 FD each node run, with the patch below it stops leaking. I’ll leave this running for a little while. I don’t know what unexpected side effects this may have.

--- /usr/lib/ruby/1.8/puppet/rails.rb.orig  2008-03-27 10:08:54.000000000 +0000
+++ /usr/lib/ruby/1.8/puppet/rails.rb   2008-03-27 10:07:26.000000000 +0000
@@ -21,7 +21,7 @@
             [[ActiveRecord]]::Base.logger.level = Logger::DEBUG
         end

-        [[ActiveRecord]]::Base.allow_concurrency = true
+        [[ActiveRecord]]::Base.allow_concurrency = false
         [[ActiveRecord]]::Base.verify_active_connections!

         begin