One option is definitely having a base image that has the things needed. Anything else? Thanks!
This is a bit complicated, to get the cache to kick in you have to mount an NFS file into the pod as the cache (to create a persistent cache)
Basically, spin NFS pod to store the cache, change the glue job template yaml to mount it into the pod (see default cache folders:
/root/.cache/pip and /root/.clearml/pip-download-cache)
Make sense ?
Would adding support for some sort of post task script help? Is something already there?
Gitlab has support for S3 based cache btw.
This might still be considered "slow" compared to local-dist/cluster mount
Would adding support for some sort of post task script help? Is something already there?
Interesting, can you expand on the use case? (currently there is only pre-task script, for setup)
IF there’s a post-task script, I can add a way to zip and upload pip cache etc to s3 - as in do any caching that I want without having first class support in clearml
pip cache & git cache & venvs cache
Are all supported, you just need to map the folders.
If you do not want to spin a PVC with NFS mount, you can just mount an S3 bucket with s3fs as part of the container extra bash script,
https://github.com/allegroai/clearml-agent/blob/b39b54bbafab39e6731cb742fdf317bc6dcae54a/docs/clearml.conf#L140
s3 FUSE fuse filesystems:
https://github.com/kahing/goofys
https://github.com/s3fs-fuse/s3fs-fuse
WDYT?
Interesting. How do you do PVC? by using the yaml template optionn?