Problem is that while this is always the most correct way to do things, that it may not be the correct way for the business that is using chef to get things done right now. Its quite possible that you've got a huge mess on your hands and the easiest way to alleviate 90% of the pain is to simply enforce one line of state in a config file that has a thousand unique snowflakes that have been deployed by hand over the last decade. The 'right' thing to do is clean up the whole mess, but that often takes 10x the effort, since you often need to audit the whole thing and figure out what deltas in all of the config options are actual necessary business requirements and what is cargo cult and deltas that are unnecessary. Leaving that unmanaged state out there is often messy, but I've made that decision a lot in order to move forward in increments. We also leave most of the state of the system unmanaged because most of the state doesn't matter.
There's also other use cases, such as using bundled chef-solo to manage state for an application that needs to ship and be integrated with uncontrolled external config management systems. Such code may need to enforce state in a file, but the rest of the state of that file is complete out of your control and fundamentally cannot be brought under your own management.
Within the same company you can wind up seeing this problem where departments may have written tooling that edits files already and config management systems may need to integrate with those and chef may need to make partial edits to the portion of the file that it 'controls'. The right solution there is always to fix that separation of concerns and make chef authoritative and to call a tool in order so that chef could render the whole template, but stupid thing like that can make for political snarls in large organizations with legacy infrastructure and legacy teams, and I don't think chef should be in the business of inserting itself into politics.
And this goes to the whole meme of being "opinionated", and I'm already highly skeptical and tired of that meme and think its a bad one for this kind of product. Its also why I don't use CFEngine anymore because of some of the opinions of that crowd that made it almost deliberately difficult to integrate that product with a central database. I really don't like using tools that have opinions that force me into making certain cost/benefit decisions in my job, I need to be able to weigh those decisions myself.
So, yes, I've run into this decision over and over through the years, and have quite often determined that it is *far* from trivial to take the whole file and templatize it (its often far from trivial to understand the actual requirements from the cargo culting), and have chosen to either do the expedient thing for the business and clean up a fragment of it first, or simple to ignore the rest of the config. In some cases I've actually written code that does "if the file exists, edit it to enforce this one line, if the file does not exist, then use my nice standardized template" to gain eventual convergence and reduce config deltas over time while at the same time punting the existing mess, and allowing me to test my standardized template every time a new server goes into prod.
And the whole use of "our model" and such is language that is kind of annoying. Chef didn't invent that, its just common practice that its the better way of doing things in the unix sysadmin world, but its not the /only/ configuration management design problem out there -- although for a small startup with 40 nodes in the cloud that has been using chef since day #1 it may be the only one that you need. When talking about walking into thousands of unmanaged servers grown organically over a decade and on different major flavors of operating systems.... You will need every trick in the book.
And being able to use every trick in the book and be able to easily get the job done -- no matter how broken the existing systems and no matter how ingrained the departmental politics -- is what is going to make chef awesome to work with for practicing/suffering devops/admins out there...
On 6/15/12 7:29 AM, Christopher Brown wrote:
This exchange (Aaron and Lamont) is almost exactly the conversation
Adam and I had when I first had Nuo write file_edit and that's why we
decided to bury it. It's useful, but cuts against the grain.
Pragmatically though, there are files over which you may not have full
control. We constantly tell people that you can ease into as much
Chef as you can take, without having to rigidly accept our model.
Doesn't saying "but you can only manage whole files, rendered by Chef"
blunt that argument?
-C
On Fri, Jun 15, 2012 at 7:21 AM, Aaron Peterson< > wrote:Lamont "I promise I'll only use it correctly, tactfully, and rarely"
Granquist, your argument does not persuade. :). But even if you held
true to that...
-1, I could never recommend this, and it fails most methodological
tests that Chef works under.
It should be trivial to take the whole file and templatize it. If you
cannot just easily take the whole file under management, that means
there is interestingly different content "out there" that represents
unmanaged yet important operational state. Now you are making
successive error-prone edits to a collection of interestingly
different files, and there is no guarantee that you will not have
drift unless you have already analyzed the differences across the
collection, at which point you might as well manage the whole thing.
Even if it is consistent, or the file differences are reproduced
normally by an underlying OS or software process, I had better only
add but never delete file edits to recipes, because if ever a machine
misses a past edit because it is off or whatever, or we launch a new
machine, we won't be creating the file correctly.
Generally, if you cannot always guarantee the recreation of the file
in its current state, you have entombed state and risk and have failed
to create an infrastructure you can reason about or rebuild.
Making it a full resource is the Chef imprimatur to use it, and this
isn't just rope, it's already tied into a noose and permanently
attached to a scaffold. IMO.
-a
On Jun 14, 2012, at 14:02, Adam
Jacob< >
wrote:
I wonder if it's time to finally turn that into a resource, even
though I think there is near universal agreement it's not a great
idea?
Adam
On Thu, Jun 14, 2012 at 4:54 AM, AJ Christensen< > wrote:https://github.com/opscode/chef/blob/master/chef/lib/chef/util/file_edit.rb
On 14 June 2012 23:25, Bryan
Berry< >
wrote:
minimin, u can use the template resource for this purpose
questions like this are probably better suited to the main chef mailing list
On Thu, Jun 14, 2012 at 1:11
PM,< >
wrote:
Hi,
I saw a file resource which can be used to create and delete a file,but i
want
to edit a file with values provided by the user.
Does chef provides such facility??
--
Opscode, Inc.
Adam Jacob, Chief Customer Officer
T: (206) 619-7151 E:
Archive powered by MHonArc 2.6.16.