[iddoavn]

Runs lakeFS Enterprise with Backblaze B2 as its blockstore, and uses lakeFS Datasets to publish a curated, versioned slice of the data stored there. Object data, lakeFS commit metadata and the dataset's own backing storage all live in the B2 bucket — nothing is stored locally.

Configuring lakeFS for B2

B2 speaks the S3 API, so the blockstore type is s3, but three settings are not the S3 defaults and it will not connect without them:

blockstore:
  type: "s3"
  s3:
    endpoint: "https://s3.<region>.backblazeb2.com"
    region: "<region>"
    force_path_style: true          # B2 does not serve virtual-host style buckets
    discover_bucket_region: false   # the bucket-region probe is an AWS-ism

Why Datasets, and not Iceberg or Multiple Storage Backends

Iceberg catalog and Metadata Search are not usable on B2 today. They depend on S3 conditional writes (If-None-Match), which B2 has in limited preview only, with broader rollout expected from Q1 2027. Worth knowing for anyone who tries: B2 does not return a clean 501 — it closes the connection mid-response, so clients report a transport error and it looks like a network problem. The README says this explicitly.

Multiple Storage Backends would need a second cloud to mean anything, and standing MinIO in for it would demo the cross-cloud claim against a container on your laptop. With Datasets the whole sample runs on B2 and nothing is simulated.

Handling the license

Mounted read-only from an absolute path given in a gitignored .env, so it never needs to sit inside the repository. Pairs with #294, which stops license files being committable at all.

Missing .env values fail docker compose immediately with a readable message rather than interpolating an empty string and failing later inside lakeFS:

required variable B2_REGION is missing a value: set B2_REGION in .env - e.g. us-east-005 (see .env.example)

Verified

Against a real Backblaze B2 account. From an empty bucket and a fresh stack, the notebook runs top to bottom with zero cell errors, leaving:

repo:    backblaze-b2-demo -> s3://lakefs-samples/backblaze-b2-demo
dataset: chihuahua-images v2

backblaze-b2-demo/      13 objects
datasets/                6 objects
TOTAL                   19 objects, 311 KB

Note this sample sits under 01_standalone_examples/, which CI's check-notebooks workflow ignores — it needs B2 credentials, so it cannot run there.

🤖 Generated with Claude Code

[kesarwam]

@iddoavn I made 2 small fixes in the README.