[Cirrus CI] Using cache and fingerprint

Introduction

Imagine: You want to create two tasks in your Cirrus pipeline:

build_task
test_task (depends on build_task)

You should not expect that the output of build_task is available in test_task. Some other CI does that some not such as Cirrus.

To address that use case, Cirrus recommends to use cache:

cache: instruction to persist files between task runs.

cf: https://cirrus-ci.org/guide/writing-tasks/#supported-instructions

So basically the build_task will declare a cache:

#.cirrus.yml
build_task:
  container:
    image: node:latest
  node_modules_cache: # <--- declare a cache
    folder: node_modules # <--- this cache will contains this particular folder
    fingerprint_script: cat package-lock.json # <--- This is the most important notion that I will explain later
  script:
    - npm install

And then this cache can be reused in the test_task.

Importance of fingerprint_script

There are only two hard things in computer science: cache invalidation and naming things. – Phil Karlton

This single line is responsible for the cache invalidation.

For cirrus, as long as the evaluation of fingerprint_script stays the same -> it reuse the cache.

The only way to make it repopulate the cache is to have a different value here.

Situations

(Real life issue) As a beginner I used the following: fingerprint_script: echo $CIRRUS_OS

This was a terrible mistake. Here is why: my cache was based on the OS used by the cirrus agent running the pipeline. As our cluster was based on linux, that value stayed constantly the same. So whatever I could have build or changed the cached folder was still to reuse from cirrus perspective.

It can be useful to use the CIRRUS_OS as part of the finger print but much more rarely as a single value.

Here are some good candidates for a cirrus cache fingerprint:

cat yarn.lock // It will change as soon as you update the version of a dependency/add/remove dependencies pretty nice to reuse the same node_modules dir
cat package-lock.json