[GHC] #13716: Move CI to Jenkins

[GHC]

#13716: Move CI to Jenkins
-------------------------------------+-------------------------------------
        Reporter:  bgamari           |                Owner:  (none)
            Type:  task              |               Status:  new
        Priority:  normal            |            Milestone:
       Component:  None              |              Version:  8.0.1
      Resolution:                    |             Keywords:
Operating System:  Unknown/Multiple  |         Architecture:
                                     |  Unknown/Multiple
 Type of failure:  None/Unknown      |            Test Case:
      Blocked By:  13897             |             Blocking:
 Related Tickets:  #11958            |  Differential Rev(s):
       Wiki Page:                    |
-------------------------------------+-------------------------------------
Description changed by simonpj:

Old description:

> Currently we use Harbormaster to build Differentials and commits. While
> this works, it leaves much to be desired,
>
>  * job scheduling is (literally) random: this means that some patch
> authors end up waiting literally ways for their patches to be built
>  * there is little support for automatic provisioning of builders: this
> means we can't scale to meet demand and make poor use of our
> computational resources
>  * periodic builds (e.g. nightlies) are not supported.
>  * it lacks a sensible interface for integration with external tools:
> this means that efforts like CI-before-merge have been pushed off
>  * status reporting is poor: even answering the question of what a given
> builder is currently working on is surprisingly difficult
>  * common maintenance tasks are byzantine: Adding a new builder requires
> adding at least six different objects (a Harbormaster Build Plan, a
> Drydock Working Copy, a Drydock Blueprint, an Almanac Service, an Almanac
> Device, and an Almanac Binding) in various Phabricator applications. None
> of this configuration can be tracked under version control nor can most
> of it be cloned from an existing builder's configuration.
>  * Design assumptions don't match GHC's constraints: Harbormaster was
> designed under the assumption that builds are cheap and computation
> plentiful. The maintainers have stated that they have little interest in
> supporting environments where this doesn't hold.
>
> Jenkins, while far from perfect, seems a bit better suited to our needs,
> more mature, and far more flexible. This ticket will serve as the
> checklist for our move to Jenkins.
>
> In the end we want,
>
>  * Builders (static or dynamically-provisioned, as appropriate) for
>    * x86-64, i386 Linux
>    * x86-64, i386 Windows
>    * x86-64 Darwin
>    * x86-64 OpenBSD
>    * Cross compile from x86-64 to ARM
>    * Native ARM
>
>  * Differential builds with sensible scheduling (e.g. first build on
> 64-bit Linux where machines are cheap, then build on the others)
>  * Per-commit builds on all
>  * Nightly builds on all, including
>    * Collection of binary distribution for user download
>    * Update of `master` documentation mirror on downloads.haskell.org
>    * Slow validation (including tests requiring Hackage packages)
>    * nofib run on some platforms?
>  * Test-before-merge-to-master

New description:

 Currently we use Harbormaster to build Differentials and commits. While
 this works, it leaves much to be desired:

  * job scheduling is (literally) random: this means that some patch
 authors end up waiting literally ways for their patches to be built
  * there is little support for automatic provisioning of builders: this
 means we can't scale to meet demand and make poor use of our computational
 resources
  * periodic builds (e.g. nightlies) are not supported.
  * it lacks a sensible interface for integration with external tools: this
 means that efforts like CI-before-merge have been pushed off
  * status reporting is poor: even answering the question of what a given
 builder is currently working on is surprisingly difficult
  * common maintenance tasks are byzantine: Adding a new builder requires
 adding at least six different objects (a Harbormaster Build Plan, a
 Drydock Working Copy, a Drydock Blueprint, an Almanac Service, an Almanac
 Device, and an Almanac Binding) in various Phabricator applications. None
 of this configuration can be tracked under version control nor can most of
 it be cloned from an existing builder's configuration.
  * Design assumptions don't match GHC's constraints: Harbormaster was
 designed under the assumption that builds are cheap and computation
 plentiful. The maintainers have stated that they have little interest in
 supporting environments where this doesn't hold.

 Jenkins, while far from perfect, seems a bit better suited to our needs,
 more mature, and far more flexible. This ticket will serve as the
 checklist for our move to Jenkins.

 See [https://ghc.haskell.org/trac/ghc/blog/jenkins-ci Ben's blog post (Aug
 17)].

 In the end we want,

  * Builders (static or dynamically-provisioned, as appropriate) for
    * x86-64, i386 Linux
    * x86-64, i386 Windows
    * x86-64 Darwin
    * x86-64 OpenBSD
    * Cross compile from x86-64 to ARM
    * Native ARM

  * Differential builds with sensible scheduling (e.g. first build on
 64-bit Linux where machines are cheap, then build on the others)
  * Per-commit builds on all
  * Nightly builds on all, including
    * Collection of binary distribution for user download
    * Update of `master` documentation mirror on downloads.haskell.org
    * Slow validation (including tests requiring Hackage packages)
    * nofib run on some platforms?
  * Test-before-merge-to-master

--

-- 
Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/13716#comment:5>
GHC <http://www.haskell.org/ghc/>
The Glasgow Haskell Compiler