Ease GitLab CI Pipelines set-up by running your jobs locally in Docker containers.
Why ? Because I did not wanted to commit, push, and wait for my jobs to run on the GitLab UI to figure I forgot to install make
before running make build
.
📣 Disclaimer: this is a helper tool aiming to facilite the process of setting up GitLab CI Pipelines. glci does NOT aim to replace any other tool.
yarn global add glci
At the root of your project (where your .gitlab-ci.yml
is):
glci
Limiting the jobs to run to the comma-separated list of jobs name given. Handy when setting up that stage-three job depending on that first-stage job artifacts.
Example:
glci --only-jobs=install,test:e2e
# "build" and "test:unit" won't be ran here
#
# ----------- --------- -------------
# | install | --- | build | --- | test:unit |
# ----------- --------- | -------------
# |
# | ------------
# --- | test:e2e |
# ------------
- If a
.env
file exists next to your.gitlab-ci.yml
, variables inside it get automatically parsed and added to the containers - Most of the pre-defined environment variables normally set by GitLab CI are also set here: see pre-defined.js
It's pretty straightforward:
- it parses your
.gitlab-ci.yml
file (and its "includes") - it runs each job of each stage (serially) in a Docker container created on the fly using the right image
- it logs the results in the console
- it shares the
cache
and theartifacts
between jobs using Docker volumes - it automatically stops and removes the containers and the volumes created
- Find a way to bind single files (not dirs) with volumes (need to figure out how)
- Handle glob in
cache:paths
andartifacts:paths
(need to figure out how) - Handle
artifacts:untracked
andcache:untracked
(need to figure out how) - Add
--env
to allow defining / overriding env variables - Add
--in-vagrant
to run docker in Vagrant (not faster even on Mac for what I've tried) - Prevent sharing artifacts between same-stage jobs