difference between CUBEJS_DB_EXPORT_BUCKET, CUBESTORE_GCS_BUCKET and Local storage

[jxperf]

Hi I'm trying to understand a couple of things

My understanding is that with each cube-router/worker there are 2 levels of storage:

• remote storage
• local storage Theres also another storage configuration:
• export buckets
  • currently I have set CUBEJS_DB_EXPORT_BUCKET to a gcs bucket and I can see .csv.gz files within it.

The documentation points to the fact that remote storage should use a remote blob storage e.g. gcs to guarantee consistency.

My questions are

• what is the relationship between export buckets (storing .csv.gz files) and the remote storage (storing .parquet files)?
  • why does cube store things in an export bucket then import it into a pod just to export it into another bucket again (if I understand the process correctly)?
• is remote storage accessed by all cubestore-worker/router nodes?
• what is local storage?
  • do we need persisted volumes or can we just mount an empty volume each time a cubestore pod is created?
  • does each worker/router pod need to point to the same local storage volume?

[ovr]

Hello @jxperf,

1. Cube uses the export bucket to transfer data from DWH to Cube Store. Technically Cube asks DWH to export data (UNLOAD), and then it asks Cube Store to import data directly from cloud storage. Cube Store uses remote storage as persistent storage.
2. Yes.
3. It's temporary storage for each pod.

Thanks

[jxperf]

Hi @ovr thanks for this! For the Local storage do you recommend setting persistent volume in the case of a pod restart or does it not really matter? if so should it be one volume per worker or should it be a shared persistent volume i.e. one which is accessed by all workers?

Thanks!!

[paveltiunov]

@jxperf Please see https://cube.dev/docs/caching/running-in-production#storage