How to use multiple caches on GitLab-CI? - gitlab-ci

Question

I have three jobs:
job a: cache things in directory a/
job b: needs things from directory a/ and cache things in directory b/
job c: needs things from both directories
According to the GitLab CI caching documentation, we can have a maximum of 4 caches, so I tried this:
a:
cache:
key:
files:
- a.txt
paths:
- a/
policy: pull-push
script:
- mkdir a/
- echo "A" > a/a.txt
b:
needs: [a]
cache:
- key:
files:
- b.txt
paths:
- b/
policy: pull-push
- key:
files:
- a.txt
paths:
- a/
policy: pull
script:
- mkdir b/
- echo "B" > b/b.txt
- cat a/a.txt >> b/b.txt
c:
needs: [b]
cache:
- key:
files:
- b.txt
paths:
- b/
policy: pull
- key:
files:
- a.txt
paths:
- a/
policy: pull
script:
- cat a/*.txt
- cat b/*.txt
It doesn't work because when job b is finished, only the b/ directory is in the cache.
Indeed, on the server there is only one archive.zip file which contains only the directory b/. So, what's the aim of multiple caches if only the last pull is left? I thought the cache policy will get specified paths from the archive when pulling and add specified paths when pushing, not replace all the content.
The real directories are vendor/ from composer for job A and node_modules/ from yarn for job B, that's why I want to split them in order to only pull what I need without pushing to save time, but maybe I'm missing something.

Answer 1

I think I misunderstand the aim of key.files. I solved my problem by simply using different key names for my caches.
.a-cache: &a-cache
key: a-cache-$CI_COMMIT_REF_SLUG
paths:
- a/
policy: pull
.b-cache: &b-cache
key: b-cache-$CI_COMMIT_REF_SLUG
paths:
- b/
policy: pull
a:
cache:
- <<: *a-cache
policy: pull-push
script:
- mkdir a/
- echo "A" > a/a.txt
b:
needs: [a]
cache:
- <<: *a-cache
- <<: *b-cache
policy: pull-push
script:
- mkdir b/
- echo "B" > b/b.txt
- cat a/a.txt >> b/b.txt
c:
needs: [b]
cache:
- <<: *a-cache
- <<: *b-cache
script:
- cat a/*.txt
- cat b/*.txt