Chef-solo should be able to persist state attribute changes like chef-client does in node.save

Details

  • Type: Improvement Improvement
  • Status: Open Open
  • Priority: Major Major
  • Resolution: Unresolved
  • Affects Version/s: 0.9.8
  • Fix Version/s: 0.9.12
  • Component/s: Chef Solo
  • Labels:
    None
  • Environment:

    Ubuntu 10.4 LTS

Description

Including the #chef-hacking chat that lead up to this:

<`jatan> hmm
<`jatan> running the mysql recipe twice on chef-solo breaks
<`jatan> cause node attributes are not persisted properly and secure_password gets regenerated
<`jatan> is this known behavior or should considered to be a bug? ie - not idempotent
<`jatan> aah, http://tickets.opscode.com/browse/COOK-314
<bawt> COOK-314 (1/3): 'mysql::server bombs when no passwords set' updated at 2010-07-11T08:21:16+00:00 https://tickets.opscode.com/browse/COOK-314
<`jatan> hmm
<`jatan> I'm thinking we need to hack node save function to write to Chef::Config[:json_attribs]
<`jatan> why would that be bad?
<`jatan> chef client already does the same
<`jatan> as an safety feature, it should be handled via the usual checksum/backup routine
<Mithrandir> it should probably have a state file somewhere, but I don't think :json_attribs is a good choice
<`jatan> Mithrandir, so another config option Chef::Config[:solo_state_attributes] ?
<`jatan> but where should it be located then?
<`jatan> in the same dir as json_attribs?
<`jatan> the argument for json_attribs is that all the changes goto one place
<`jatan> easier to review
<`jatan> con is that when you have your config in git and then update it.. merges become messy
<`jatan> on the whole, separate file is better.. you can still have it in git and you can diff it for changes
<Mithrandir> isn't there a state dir on the client?
<`jatan> Mithrandir, what's the config option for it then?
<`jatan> looked at lib/chef/config.rb ..didn't see it
<Mithrandir> maybe we need to make one, "localstatedir" or something and then have some of the others hang off it, like the checksum dir.
<`jatan> you mean node_path perhaps?
<Mithrandir> nope, doesn't look like there's anything, really.
<`jatan> node_path seems to fit
<`jatan> since node.rb uses it to # Find a recipe for this Chef::Node by fqdn.  Will search first for 
<`jatan>     # Chef::Config["node_path"]/fqdn.rb, then hostname.rb, then default.rb.
<`jatan> so it looks like some overrides already go there
<`jatan> it's weird
<`jatan> but that's the only place using it
<Mithrandir> ok, sounds like a sensible place, then

Activity

Ascending order - Click to sort in descending order
Hide
Permalink
Toomas Pelberg added a comment -

Available at http://github.com/toomasp/chef/tree/CHEF-1608

Not sure I liked what I ended up with - but it does work.

Show
Toomas Pelberg added a comment - Available at http://github.com/toomasp/chef/tree/CHEF-1608 Not sure I liked what I ended up with - but it does work.
Hide
Permalink
Toomas Pelberg added a comment -

Added some more tests and refactored a bit - looks better now.

<devils advocate>
According to tests and re-running http://tickets.opscode.com/browse/COOK-314 it works.

Still, it's a pretty radical change of behavior - and should be tested with a larger number of cookbooks to ensure it doesn't break anything too obvious. It works for me(tm).
</devils advocate>

Show
Toomas Pelberg added a comment - Added some more tests and refactored a bit - looks better now. <devils advocate> According to tests and re-running http://tickets.opscode.com/browse/COOK-314 it works. Still, it's a pretty radical change of behavior - and should be tested with a larger number of cookbooks to ensure it doesn't break anything too obvious. It works for me(tm). </devils advocate>
Hide
Permalink
Daniel DeLeo added a comment -

bulk updating fix versions to next release

Show
Daniel DeLeo added a comment - bulk updating fix versions to next release
Hide
Permalink
Bryan McLellan [Opscode] added a comment -

Opscode TSE tells me that the mysql cookbook has been fixed so it doesn't require the save to work.

Is node.save a feature that we still want in chef-solo?

Is there a remaining use case in which one would want to be able to persist chef-solo data? Is this changing the purpose of chef-solo to be too much like chef-client?

Show
Bryan McLellan [Opscode] added a comment - Opscode TSE tells me that the mysql cookbook has been fixed so it doesn't require the save to work. Is node.save a feature that we still want in chef-solo? Is there a remaining use case in which one would want to be able to persist chef-solo data? Is this changing the purpose of chef-solo to be too much like chef-client?
Hide
Permalink
Alex Soto added a comment -

I used to use chef-solo a lot for cookbook development so I wouldn't need a server. However, this caused me some frustration because cookbooks and lwrp's that would call node.save would error out with exceptions when run under chef-solo.

I've since changed my behavior and no longer use chef-solo for cookbook development so I can't say I 'need' this feature anymore. However I do think you should be able to code your recipes under chef-solo using node attributes (just persist to a local disk cache) so you can pass along state from one chef-run to the next using the same coding patterns that recipes using a server would.

Show
Alex Soto added a comment - I used to use chef-solo a lot for cookbook development so I wouldn't need a server. However, this caused me some frustration because cookbooks and lwrp's that would call node.save would error out with exceptions when run under chef-solo. I've since changed my behavior and no longer use chef-solo for cookbook development so I can't say I 'need' this feature anymore. However I do think you should be able to code your recipes under chef-solo using node attributes (just persist to a local disk cache) so you can pass along state from one chef-run to the next using the same coding patterns that recipes using a server would.
Hide
Permalink
Jordan added a comment -

This is a feature I would really like to see in chef-solo - I've been working with it for a few weeks now building up configurations for our development and staging environments to see if this is something we could use to manage our production infrastructure (600+ machines). Since they are all in different security zones we are unable to run a chef server, so chef-solo is the clear winner for us. Unfortunately without this feature chef-solo is only useful for doing the initial configuration of a server, and not keeping settings up to date.

Show
Jordan added a comment - This is a feature I would really like to see in chef-solo - I've been working with it for a few weeks now building up configurations for our development and staging environments to see if this is something we could use to manage our production infrastructure (600+ machines). Since they are all in different security zones we are unable to run a chef server, so chef-solo is the clear winner for us. Unfortunately without this feature chef-solo is only useful for doing the initial configuration of a server, and not keeping settings up to date.
Hide
Permalink
James Sanders added a comment -

Our situation is similar to Jordan's. We have some hacks in place for the mysql recipe in particular and for other issues along these lines. A generic solution for this would be swell for us, regardless of the exact implementation.

It looks like there hasn't been any movement on this in quite awhile - was the decision that this is not desired in chef-solo or has it just fallen by the wayside?

Show
James Sanders added a comment - Our situation is similar to Jordan's. We have some hacks in place for the mysql recipe in particular and for other issues along these lines. A generic solution for this would be swell for us, regardless of the exact implementation. It looks like there hasn't been any movement on this in quite awhile - was the decision that this is not desired in chef-solo or has it just fallen by the wayside?
Hide
Permalink
Andrey Stepachev added a comment -

Hi.
I need this feature to implement different way of node state distribution (without central server).
Why this feature can't be optional (if someone need to save state, appropriate parameter can be specified on chef-solo command line, f.e. --local-state-dir where node state will be dumped). Other logic (like checksumming, moving state around) can be implement by external user of this feature.

Show
Andrey Stepachev added a comment - Hi. I need this feature to implement different way of node state distribution (without central server). Why this feature can't be optional (if someone need to save state, appropriate parameter can be specified on chef-solo command line, f.e. --local-state-dir where node state will be dumped). Other logic (like checksumming, moving state around) can be implement by external user of this feature.

People

  • Assignee:
    Unassigned
    Reporter:
    Toomas Pelberg
Vote (3)
Watch (5)

Dates

  • Created:
    Updated: