Caching APT packages in GitHub Actions workflow

I use the following Github Actions workflow for my C project. The workflow finishes in ~40 seconds, but more than half of that time is spent by installing the valgrind package and its dependencies.

I believe caching could help me speed up the workflow. I do not mind waiting a couple of extra seconds, but this just seems like a pointless waste of GitHub's resources.

name: C Workflow

on: [push, pull_request]

jobs:
  build:
    runs-on: ubuntu-latest

    steps:
    - uses: actions/checkout@v1

    - name: make
      run: make

    - name: valgrind
      run: |
        sudo apt-get install -y valgrind
        valgrind -v --leak-check=full --show-leak-kinds=all ./bin

Running sudo apt-get install -y valgrind installs the following packages:

gdb
gdbserver
libbabeltrace1
libc6-dbg
libipt1
valgrind

I know Actions support caching of a specific directory (and there are already several answered SO questions and articles about this), but I am not sure where all the different packages installed by apt end up. I assume /bin/ or /usr/bin/ are not the only directories affected by installing packages.

Is there an elegant way to cache the installed system packages for future workflow runs?

Solution

The purpose of this answer is to show how caching can be done with github actions, not necessarily to show how to cache valgrind, (which it does). I also try to explain why not everything can/should be cached, because the cost (in terms of time) of caching and restoring a cache, vs reinstalling the dependency needs to be taken into account.

You will make use of the actions/cache action to do this.

Add it as a step (before you need to use valgrind):

- name: Cache valgrind
  uses: actions/cache@v2
  id: cache-valgrind
  with:
      path: "~/valgrind"
      key: ${{secrets.VALGRIND_VERSION}}

The next step should attempt to install the cached version if any or install from the repositories:

- name: Install valgrind
  env:
    CACHE_HIT: ${{steps.cache-valgrind.outputs.cache-hit}}
    VALGRIND_VERSION: ${{secrets.VALGRIND_VERSION}}
  run: |
      if [[ "$CACHE_HIT" == 'true' ]]; then
        sudo cp --verbose --force --recursive ~/valgrind/* /
      else
        sudo apt-get install --yes valgrind="$VALGRIND_VERSION"
        mkdir -p ~/valgrind
        sudo dpkg -L valgrind | while IFS= read -r f; do if test -f $f; then echo $f; fi; done | xargs cp --parents --target-directory ~/valgrind/
      fi

Explanation

Set VALGRIND_VERSION secret to be the output of:

apt-cache policy valgrind | grep -oP '(?<=Candidate:\s)(.+)'

this will allow you to invalidate the cache when a new version is released simply by changing the value of the secret.

dpkg -L valgrind is used to list all the files installed when using sudo apt-get install valgrind.

What we can now do with this command is to copy all the dependencies to our cache folder:

dpkg -L valgrind | while IFS= read -r f; do if test -f $f; then echo $f; fi; done | xargs cp --parents --target-directory ~/valgrind/

Furthermore

In addition to copying all the components of valgrind, it may also be necessary to copy the dependencies (such as libc in this case), but I don't recommend continuing along this path because the dependency chain just grows from there. To be precise, the dependencies needed to copy to finally have an environment suitable for valgrind to run in is as follows:

libc6
libgcc1
gcc-8-base

To copy all these dependencies, you can use the same syntax as above:

for dep in libc6 libgcc1 gcc-8-base; do
    dpkg -L $dep | while IFS= read -r f; do if test -f $f; then echo $f; fi; done | xargs cp --parents --target-directory ~/valgrind/
done

Is all this work really worth the trouble when all that is required to install valgrind in the first place is to simply run sudo apt-get install valgrind? If your goal is to speed up the build process, then you also have to take into consideration the amount of time it is taking to restore (downloading, and extracting) the cache vs simply running the command again to install valgrind.

And finally to restore the cache, assuming it is stored at /tmp/valgrind, you can use the command:

cp --force --recursive /tmp/valgrind/* /

Which will basically copy all the files from the cache unto the root partition.

In addition to the process above, I also have an example of "caching valgrind" by installing and compiling it from source. The cache is now about 63MB (compressed) in size and one still needs to separately install libc which kind of defeats the purpose.

Note: Another answer to this question proposes what I could consider to be a safer approach to caching dependencies, by using a container which comes with the dependencies pre-installed. The best part is that you can use actions to keep those containers up-to-date.

References: