+1, very insightful.
We also plan giving a try to the literate plugin in the upcoming weeks. So
we're also interested to have up-to-date informations about where/how to
report things, etc.

Cheers


2014-03-10 13:33 GMT+01:00 Mark Waite <[email protected]>:

> That is a great vision.  I'd like to help the vision with some testing and
> can provide you some feedback.  How would you prefer the feedback?  I can
> submit bug reports through JIRA, or send mail to the list, or some other
> technique.
>
> Testing time is limited, and must be squeezed around my real job and my
> family, but I'm sure I can provide some testing and some feedback if there
> are indications they will be helpful to your efforts.
>
> Mark Waite
>
>
> On Mon, Mar 10, 2014 at 6:28 AM, Stephen Connolly <
> [email protected]> wrote:
>
>> I will give you my vision.
>>
>> In my vision there are two types of things:
>>
>> 1. Things that depend on the stuff in a build job itself
>> 2. Things that depend on the inter-relationship of jobs within a CI
>> server.
>>
>> Traditionally, Jenkins takes the view that there is just one type of
>> thing. So you end up configuring everything in the Jenkins job (or you use 
>> the
>> evil 
>> one<http://javaadventure.blogspot.ie/2013/11/jenkins-maven-job-type-considered-evil.html>
>>  and
>> have slightly less to configure)
>>
>> This leads people to want to lock down access to some portions of the
>> configuration... in a sense you need to let some people tweak the first
>> type of things (because they are changing the code being built and thus may
>> break the instructions as to how to build it) but you don't want to let
>> them tweak the second type of things (e.g. how the built application gets
>> put into the staging environment and which job(s) to trigger so that the
>> pre-production tests run against the staging environment)
>>
>> The folly people then run down is trying to get a division of permission
>> scopes.
>>
>> There is, however, a split... if you let people commit to source control,
>> then in reality you let them change the build script that lives in source
>> control (unless you are using SVN and lock the build script down with
>> permissions... but at some point a refactoring needs a tweak to the build
>> script and giving developers a world of pain to make those changes only
>> pisses them off)... or they write a unit test that does the changes on the
>> fly as a hack around the fact you haven't given them the ability to tweak
>> the build script... so now your build script is part build script part unit
>> test hacks that "fix" the build script.
>>
>> So the simple fact is that however you set up your build script, you have
>> to accept that developers will be able to change the build script. This
>> probably means that you have code review for changes to the build script...
>> not that the devs are making malicious changes... more that they are making
>> appropriate changes.
>>
>> In my view a CI build job should not need configuration changes to
>> reflect a change in the build script... the CI job should be able to pick
>> up the configuration relevant to the build script from the source tree
>> itself... hence the literate job type.
>>
>> This unlocks a lot of great power... tracking multiple branches is now
>> trivial, as each branch stores the detail of how that branches version of
>> the build script should be interpreted...
>>
>> All the Type 1 things are configuration that naturally should sit within
>> source control. People who can change the build script can then change the
>> corresponding jenkins configuration in the same atomic commit.
>>
>> The Type 2 things are about the greater context. That context relies on
>> other projects. It relies on knowledge that is outside the scope of SCM
>> about which branch is the current mainline... The SCM may know on Day X
>> this branch was tagged as being the mainline... but it has no way to link
>> against the other SCM repos that hold the side projects with their
>> independent release schedules.
>>
>> The Type 2 things are invariably the bits that you don't want the
>> developers messing with. With the literate project type those things remain
>> in the Jenkins UI.
>>
>> Literate (and I think it needs a better name BTW) is about moving the
>> Type 1 things out of the Jenkins UI and into the SCM where they belong.
>>
>> What is literate missing before I call it 1.0 (without the alpha or beta):
>>
>> * Support for GitHub pull requests
>> * Support for Maven multi-module reporting (without invoking the curse of
>> the evil one)
>> * Support for untrusted builds (partially there... just need something
>> that people can more easily use)
>> * It would be super nice if Vincent can get his Yaml parser stuff
>> committed before 1.0 also so that people who don't like the markdown build
>> description can use the yaml based alternative (literate has always had a
>> "yaml" format... just one that only supported a very very small subset of
>> the "yaml" syntax that people expect)
>>
>> The real joy of literate will be when you can have pull requests get
>> their own branch job on demand that gets built with a commit note being
>> added back to the pull request and the branch job being removed once the
>> pull request is resolved.
>>
>> The functionality comes with a risk... namely the drive-by pull request
>> that f*cks your build server... oh let's just add a `rm -rf /` to the build
>> script via a pull request using a throwaway github account I created just
>> to screw you over.
>>
>> So before I give people the GitHub pull request UI for literate... I need
>> to give people an easy route to protect themselves, e.g. by letting them
>> say that pull request builds will run in a chroot environment, or an LXC
>> container, or whatever set of protections they want.
>>
>> On top of that, weaving in the Maven support that I want to add may make
>> changes to the literate job type that could be problematic to migrate, so I
>> would like to get those features bedded down before calling something 1.0.
>>
>> So that is my vision... there is still some work left in it... but a vast
>> chunk has been completed...
>>
>>
>> On 10 March 2014 11:56, Stephen Connolly <[email protected]
>> > wrote:
>>
>>> Like that's going to stop them...
>>>
>>>       <plugin>
>>>         <groupId>org.codehaus.mojo</groupId>
>>>         <artifactId>exec-maven-plugin</artifactId>
>>>         <version>1.2.1</version>
>>>         <executions>
>>>           <execution>
>>>             <goals>
>>>               <goal>exec</goal>
>>>              </goals>
>>>             <configuration>
>>>               <executable>/bin/rm</executable>
>>>               <workingDirectory>/</workingDirectory>
>>>                <arguments>
>>>                 <argument>-rf</argument>
>>>                 <argument>/</argument>
>>>               </arguments>
>>>             </configuration>
>>>           </execution>
>>>         </executions>
>>>       </plugin>
>>>
>>>
>>>
>>> On 10 March 2014 00:39, Christian Willman <[email protected]> wrote:
>>>
>>>> Unfortunately it is not possible. A common suggestion is to template
>>>> out your specific job type, but even then, there is no authorization
>>>> strategy to prevent users from simply creating a new FreestyleBuild with a
>>>> console build step.
>>>>
>>>> Our organization has the same requirements -- we cannot allow untrusted
>>>> developers to run arbitrary shell commands. We've solved this by switching
>>>> to an external templating system (which I wrote in-house). Now developers
>>>> cannot create or modify *any* jobs directly through the Jenkins UI.
>>>> This works for us because 99% of our Java projects can be built with a
>>>> simple "mvn clean deploy". The oddball jobs are isolated to a
>>>> heavily-audited Jenkins instance.
>>>>
>>>> Another potential solution is to implement your own job types and then
>>>> write a custom authorization strategy which revokes access to the built-in
>>>> job types. But this is not maintainable and will probably require a
>>>> dedicated admin to babysit the implementation for the foreseeable future.
>>>>
>>>> I've searched through the source and couldn't figure out any other
>>>> implementations which don't require modifications to the Jenkins core.
>>>> Unfortunately nobody (including Cloudbees) seems interested in this use
>>>> case right now.
>>>>
>>>> Cheers,
>>>> Christian
>>>>
>>>> --
>>>> You received this message because you are subscribed to the Google
>>>> Groups "Jenkins Users" group.
>>>> To unsubscribe from this group and stop receiving emails from it, send
>>>> an email to [email protected].
>>>> For more options, visit https://groups.google.com/d/optout.
>>>>
>>>
>>>
>>  --
>> You received this message because you are subscribed to the Google Groups
>> "Jenkins Users" group.
>> To unsubscribe from this group and stop receiving emails from it, send an
>> email to [email protected].
>> For more options, visit https://groups.google.com/d/optout.
>>
>
>
>
> --
> Thanks!
> Mark Waite
>
> --
> You received this message because you are subscribed to the Google Groups
> "Jenkins Users" group.
> To unsubscribe from this group and stop receiving emails from it, send an
> email to [email protected].
> For more options, visit https://groups.google.com/d/optout.
>



-- 
Baptiste <Batmat> MATHUS - http://batmat.net
Sauvez un arbre,
Mangez un castor !

-- 
You received this message because you are subscribed to the Google Groups 
"Jenkins Users" group.
To unsubscribe from this group and stop receiving emails from it, send an email 
to [email protected].
For more options, visit https://groups.google.com/d/optout.

Reply via email to