conda-forge / conda-forge/sqlite-feedstock
There's some kind of bug where an old version of sqlite is getting installed all of a sudden and it's breaking JupyterHub
- Dominant language
- Batchfile
- Stars
- 1
- Forks
- 44
- PR merge metrics
- No merged PRs in 30d
Description
I'm sorry this isn't more helpful information to reproduce, but we've seen some weird behavior with sqlite recently, and I just wanted to share what happened and what we did to fix it, in case it helps uncover something deeper here.
_Note: this is not for ALL jupyterhubs, just some - I'm sharing two examples below in case that helps pin down what's the root cause_
- We're running JupyterHub, building user environments with `repo2docker`
- Recently two different hubs suddenly started failing their launches
- We tracked this down to a dependency shift where an _old version of sqlite_ seems to get installed now.
- Min-version pinning the sqlite version fixed the issue, here are two PRs where this works:
- https://github.com/ProjectPythia/gdex-cookbook/pull/10
- https://github.com/HigherEdData/Du-Bois-STEM/pull/4
I'm not sure if the change would need to be in this repo, or in some other repo dependent on sqlite, but I just wanted to record it _somewhere_ 😅
I think this has something to do with the "split" between [libsqlite](https://anaconda.org/channels/conda-forge/packages/libsqlite/overview) and sqlite - maybe they're both being installed alongside each other an the older one started winning out for some reason? Either way, if we could pin `sqlite` to `>3.39` here then that might fix the problem.
@agoose77 does any of this sound wrong to you?
Contributor guide
Assessment
This issue has not been assessed yet.