Metro/Automation

From Funtoo
Jump to: navigation, search

Metro also includes a number of advanced features that can be used to automate builds and set up distributed build servers. These features require you to emerge sqlalchemy, as SQLite is used as a dependency and also emerge dev-python/lxml as this is needed for index file generation.

Repository Management

Metro includes a script in the scripts directory called buildrepo. Buildrepo serves as the heart of Metro's advanced repository management features.

Initial Setup

To use buildrepo, you will first need to create a .buildbot configuration file. Here is the file I use on my AMD Jaguar build server:

   /root/.buildbot (python source code)
builds = (
	"funtoo-current",
	"funtoo-current-hardened",
)

arches = (
	"x86-64bit",
	"pure64"
)

subarches = (
	"amd64-jaguar",
	"amd64-jaguar-pure64",
)

def map_build(build, subarch, full, full_date):
	# arguments refer to last build...
	if full == True:
		buildtype =  ( "freshen", )
	else:
		buildtype =  ("full", )
        # return value can be a string like "full+openvz" or a sequence type like [ "freshen", "openvz" ]
	return buildtype

This file is actually a python source file that defines the tuples builds, arches and subarches. These variables tell buildrepo which builds, arches and subarches it should manage. A map_build() function is also defined which buildbot uses to determine what kind of build to perform. The arguments passed to the function are based on the last successful build. The function can read these arguments and return a string to define the type of the next build. In the above example, the map_build() function will cause the next build after a freshen build to be a full build, and the next build after a full build to be a freshen build, so that the build will alternate between full and freshen.

Automated Builds

Once the .buildbot file has been created, the buildrepo and buildbot.sh tools are ready to use. Here's how they work. These tools are designed to keep your repository (path/mirror in /root/.metro up-to-date by inspecting your repository and looking for stages that are out-of-date.

To list the next build that will be performed, do this -- this is from my ARM build server:

root # ./buildrepo nextbuild
build=funtoo-current
arch_desc=arm-32bit
subarch=armv7a_hardfp
fulldate=2015-02-08
nextdate=2015-02-20
failcount=0
target=full
extras=''

If no output is displayed, then all your builds are up-to-date.

To actually run the next build, run buildbot.sh:

root # ./buildbot.sh

If you're thinking that buildbot.sh would be a good candidate for a cron job, you've got the right idea!

List Builds

To get a quick look at our repository, let's run the buildrepo fails command:

root # ./buildrepo fails
   0   2015-02-18 /home/mirror/funtoo/funtoo-current/x86-64bit/amd64-jaguar
   0   2015-02-18 /home/mirror/funtoo/funtoo-current/pure64/amd64-jaguar-pure64
   0   2015-02-18 /home/mirror/funtoo/funtoo-current-hardened/x86-64bit/amd64-jaguar
   0   2015-02-18 /home/mirror/funtoo/funtoo-current-hardened/pure64/amd64-jaguar-pure64

On my AMD Jaguar build server, on Feb 20, 2015, this lists all the builds that buildrepo has been configured to manage. The first number on each line is a failcount, which is the number of consecutive times that the build has failed. A zero value indicates that everything's okay. The failcount is an important feature of the advanced repository management features. Here are a number of behaviors that are implemented based on failcount:

  • If buildbot.sh tries to build a stage and the build fails, the failcount is incremented.
  • If the build succeeds for a particular build, the failcount is reset to zero.
  • Builds with the lowest failcount are prioritized by buildrepo to build next, to steer towards builds that are more likely to complete successfully.
  • Once the failcount reaches 3 for a particular build, it is removed from the build rotation.

Resetting Failcount

If a build has issues, the failcount for a build will reach 3, at which point it will be pulled out of build rotation. To clear failcount, so that these builds are attempted again -- possibly fixed by new updates to the Portage tree -- use buildrepo zap:

root # /root/metro/scripts/buildrepo zap
Removing /mnt/data/funtoo/funtoo-current/arm-32bit/armv7a_hardfp/.control/.failcount...
Removing /mnt/data/funtoo/funtoo-current/arm-32bit/armv6j_hardfp/.control/.failcount...
Removing /mnt/data/funtoo/funtoo-current/arm-32bit/armv5te/.control/.failcount...

Repository Maintenance

A couple of repository maintenance tools are provided:

  • buildrepo digestgen will generate hash files for the archives in your repository, and clean up stale hashes.
  • buildrepo index.xml will create an index.xml file at the root of your repository, listing all builds available.
  • buildrepo clean will output a shell script that will remove old stages. No more than the three most recent stage builds for each build/arch/subarch are kept.

Distributed Repositories

In many situation, you will have a number of build servers, and each will build a subset of your master repository, and then upload builds to the master repository. This is an area of Metro that is being actively developed. For now, automated upload functionality is not enabled, but is expected to be implemented in the relatively near future. However, it is possible to have your master repository differentiate between subarches that are built locally, and thus should be part of that system's buildbot build rotation, and those that are stored locally and built remotely. These builds should be cleaned when buildrepo clean is run, but should not enter the local build rotation. To set this up, modify /root/.buildbot and use the subarches and all_subarches variables:

   /root/.buildbot - Excerpt of .buildbot config for master repository
# subarches we are building locally:

subarches = ( 
        "pentium4",
        "athlon-xp",
        "corei7",
        "corei7-pure64",
        "generic_32", 
        "i686", 
        "amd64-k8",
        "amd64-k8-pure64",
        "core2_64",
        "core2_64-pure64",
        "generic_64",
        "generic_64-pure64",
) 
  
# Things we need to clean, even if we may not be building:
  
all_subarches = subarches + (
        "atom_32",
        "atom_64",
        "atom_64-pure64",
        "amd64-k10",
        "amd64-k10-pure64",
        "amd64-bulldozer",
        "amd64-bulldozer-pure64",
        "amd64-steamroller",
        "amd64-steamroller-pure64",
        "amd64-piledriver",
        "amd64-piledriver-pure64",
        "amd64-jaguar",
        "amd64-jaguar-pure64",
        "intel64-haswell",
        "intel64-haswell-pure64",
        "intel64-ivybridge-pure64",
        "intel64-ivybridge",
        "armv7a_hardfp",
        "armv6j_hardfp",
        "armv5te"
)

Using binary cache

Metro has built-in feature which allows to use binary packages cache rather then building same list of packages from sources. For example, core packages, such as @system are updated at slower pace and it makes sense to enable binary cache to make stage building blazing fast. However, the real disadvantage with using binary cache could be a core package update that due to internal ABI changes require rebuilding of numerous packages from sources. Good example is sys-libs/ncurses-5 to sys-libs/ncurses-6 major update. This is the case when you would need to disable binary cache and use regular ebuild installation from sources. To enable binary cache, in your metro git repository copy, edit the common.conf

   /etc/builds/common.conf - Excerpt of default common.conf
[section metro]

options:
options/stage:
target: gentoo

and set cache/package

   /etc/builds/common.conf - Excerpt of common.conf with binary cache enabled
[section metro]

options:
options/stage: cache/package
target: gentoo

During stage build metro will save package cache in /var/tmp/metro/cache/package-cache. With any next builds this binary package cache will be used.