undo parallel CI jobs (tecken)
Categories
(Tecken :: General, task, P2)
Tracking
(Not tracked)
People
(Reporter: willkg, Assigned: willkg)
Details
Attachments
(1 file)
Earlier this year, we switched from one set of images to the cimg images. When I did that, I also redid the CI configuration and broke up our monolithic job into a series of jobs. The thinking was that we could take advantage of parallelization and reduce the overall time it takes to do a CI run.
We're using a couple of CircleCI cachey things:
- docker layer caching
- layers are kept around for some period of time
- only one executor has access to the layers so when an executor doesn't have access, it builds everything
- workspace
- things can get saved to the workspace
- right now we're saving a docker image to the workspace and it's big so it takes a while to save and load
The end result of this is that on average, builds were faster before than they are now.
We could yak-shave the jobs. It's likely we could set up Python virtual environments and reduce the complexities of parallelization. This is a non-trivial rethink of our local dev, CI, and server environments.
Or, we could ditch parallelization and go back to a single monolithic job.
I want to do that. This bug covers that work for Tecken.
Assignee | ||
Comment 1•2 years ago
|
||
Assignee | ||
Comment 2•2 years ago
|
||
Assignee | ||
Comment 3•2 years ago
|
||
I pushed this out in bug #1780235. Marking as FIXED.
Description
•