don't push out facter-1.6.11 without testing ; causes puppetd hang

[Jo Rhett]

At least on CentOS 5 and CentOS 6, after upgrading to facter 1.6.11 our hosts stopped checking in. Stale puppetdlock problem.

-- 

Jo Rhett

Net Consonance : net philanthropy to improve open source and internet projects.

[Stuart Cracraft]

Can you kick them somehow?

--
You received this message because you are subscribed to the Google Groups "Puppet Users" group.
To post to this group, send email to puppet...@googlegroups.com.
To unsubscribe from this group, send email to puppet-users...@googlegroups.com.
For more options, visit this group at http://groups.google.com/group/puppet-users?hl=en.

[Jo Rhett]

Nope, they think they are running. We had to reset the policy to downgrade facter, then login to each host and "service puppet stop ; puppet agent --test --ignoreschedules" (our systems are set to only upgrade packages in certain hours) to get the hosts back online.

This looks similar to the old problem with a kernel that changed proc semantics, but it's not the kernel this time. Reverting facter to 1.6.10 resolves the issue.

[Justin Stoller]

Hey Jo,

Do you have any debugging information about this issue? Stack traces,
systems, versions, ruby, custom facts, etc would all be helpful. It
certainly seems from what you said that the version of Facter had
something to do with this error, but I'm not exactly sure how Facter
would affect Puppet's lock file....

- Justin

[Jo Rhett]

Lots of people have logged information in ticket #10418 http://projects.puppetlabs.com/issues/10418

I don't have much time to track this today, but our environment is fairly bone-stock CentOS 5.6, Ruby 1.8.7, Puppet 1.7.18. The only custom facts come from puppetlabs-stdlib. 

[Eric Sorenson]

Ugh, the only code changes between 1.6.10 and 1.6.11 are /proc related changes for selinux detection.

https://github.com/puppetlabs/facter/commit/a836eeee764790be14acc276ab039080316f65b2

So this is somehow tickling the buggy select() behaviour. Will spend more time on tracking this down today; could anyone with affected systems please ping me (eric0) on irc?

-=Eric