Products pipeline: multi-product multi-configuration builds¶
In the previous section we computed a conan graph build-order
with several simplifications, we didn’t take the mapviewer
product into account, and we processed only 1 configuration.
In real scenarios, it will be necessary to manage more than one product and the most common case is that there is more than one configuration for every product. If we build these different cases sequentially it will be much slower and inefficient, and if we try to build them in parallel there will easily be many duplicated and unnecessary builds of the same packages, wasting resources and even producing issues as race conditions or traceability problems.
To avoid this issue, it is possible to compute a single unified “build-order” that aggregates all the different build-orders that are computed for the different products and configurations.
Let’s start as usual cleaning the local cache and defining the correct repos:
# First clean the local "build" folder
$ pwd # should be <path>/examples2/ci/game
$ rm -rf build # clean the temporary build folder
$ mkdir build && cd build # To put temporary files
$ conan remove "*" -c # Make sure no packages from last run
# NOTE: The products repo is first, it will have higher priority.
$ conan remote enable products
Now, we will start computing the build-order for game/1.0
for the 2 different configurations that we are building in this tutorial, debug and release:
$ conan graph build-order --requires=game/1.0 --build=missing --order-by=recipe --format=json > game_release.json
$ conan graph build-order --requires=game/1.0 --build=missing --order-by=recipe -s build_type=Debug --format=json > game_debug.json
These commands are basically the same as in the previous section, each one with a different configuration and creating a different output file game_release.json
and game_debug.json
. These files will be similar to the previous ones, but as we haven’t used the --reduce
argument (this is important!) they will actually contain a “build-order” of all elements in the graph, even if only some contain the binary: Build
definition, and others will contain other binary: Download|Cache|etc
.
Now, let’s compute the build-order for mapviewer/1.0
:
$ conan graph build-order --requires=mapviewer/1.0 --build=missing --order-by=recipe --format=json > mapviewer_release.json
$ conan graph build-order --requires=mapviewer/1.0 --build=missing --order-by=recipe -s build_type=Debug --format=json > mapviewer_debug.json
Note that in the generated mapviewer_xxx.json
build-order files, there will be only 1 element for mapviewer/1.0
that contains a binary: Download
, because there is really no other package to be built, and as mapviewer
is an application linked statically, Conan knows that it can “skip” its dependencies binaries. If we had used the --reduce
argument we would have obtained an empty order
. But this is not an issue, as the next final step will really compute what needs to be built.
Let’s take all the 4 different “build-order” files (2 products x 2 configurations each), and merge them together:
$ conan graph build-order-merge --file=game_release.json --file=game_debug.json --file=mapviewer_release.json --file=mapviewer_debug.json --reduce --format=json > build_order.json
Now we have applied the --reduce
argument to produce a final build_order.json
that is ready for distribution to the build agents and it only contains those specific packages that need to be built:
{
"order_by": "recipe",
"reduced": true,
"order": [
[
{
"ref": "engine/1.0#fba6659c9dd04a4bbdc7a375f22143cb",
"packages": [
[
{
"package_id": "de738ff5d09f0359b81da17c58256c619814a765",
"filenames": ["game_release"],
"build_args": "--requires=engine/1.0 --build=engine/1.0",
},
{
"package_id": "cbeb3ac76e3d890c630dae5c068bc178e538b090",
"filenames": ["game_debug"],
"build_args": "--requires=engine/1.0 --build=engine/1.0",
}
]
]
}
],
[
{
"ref": "game/1.0#1715574045610faa2705017c71d0000e",
"packages": [
[
{
"package_id": "bac7cd2fe1592075ddc715563984bbe000059d4c",
"filenames": ["game_release"],
"build_args": "--requires=game/1.0 --build=game/1.0",
},
{
"package_id": "01fbc27d2c156886244dafd0804eef1fff13440b",
"filenames": ["game_debug"],
"build_args": "--requires=game/1.0 --build=game/1.0",
}
]
]
}
]
],
"profiles": {
"game_release": {"args": ""},
"game_debug": {"args": "-s:h=\"build_type=Debug\""},
"mapviewer_release": {"args": ""},
"mapviewer_debug": {"args": "-s:h=\"build_type=Debug\""}
}
}
This build order summarizes the necessary builds. First it is necessary to build all different binaries for engine/1.0
. This recipe contains 2 different binaries, one for Release and the other for Debug. These binaries belong to the same element in the packages
list, which means they do not depend on each other and can be built in parallel. Each binary tracks its own original build-order file with "filenames": ["game_release"],
so it is possible to deduce the necessary profiles to apply to it. The build_order.json
file contains a profiles
section that helps recovering the profile and settings command line arguments that were used to create the respective original build-order files.
Then, after all binaries of engine/1.0
have been built, it is possible to proceed to build the different binaries for game/1.0
. It also contains 2 different binaries for its debug and release configurations, which can be built in parallel.
In practice, this would mean something like:
# This 2 could be executed in parallel
# (in different machines, or different Conan caches)
$ conan install --requires=engine/1.0 --build=engine/1.0
$ conan install --requires=engine/1.0 --build=engine/1.0 -s build_type=Debug
# Once engine/1.0 builds finish, it is possible
# to build these 2 binaries in parallel (in different machines or caches)
$ conan install --requires=game/1.0 --build=game/1.0
$ conan install --requires=game/1.0 --build=game/1.0 -s build_type=Debug
In this section we have still omitted some important implementation details that will follow in next sections. The goal was to focus on the conan graph build-order-merge
command and how different products and configurations can be merged in a single “build-order”. The next section will show with more details how this build-order can be really distributed in CI, using lockfiles to guarantee constant dependencies.