Catalog Workflows¶
Catalog workflows govern which published datasets the serving layer can discover.
The catalog is the discoverable registry of published datasets. A serving store without a valid catalog is not a complete serving surface.
Catalog Lifecycle¶
flowchart LR
Publish[dataset publish] --> Promote[catalog promote]
Promote --> Discover[Dataset discoverable]
Discover --> Rollback[catalog rollback if needed]
Promote --> Alias[latest alias update if policy allows]
This lifecycle diagram shows the catalog’s real job: making published dataset state discoverable to the serving layer. Publication alone is not enough if the catalog never points the runtime at the dataset.
Main Catalog Operations¶
catalog validate: validate a catalog documentcatalog publish: write a catalog into a store rootcatalog promote: add a published dataset to the catalogcatalog rollback: remove a dataset from the catalogcatalog latest-alias-update: update the latest alias after promotion
Recommended Normal Flow¶
flowchart TD
Build[Build and validate dataset] --> Publish[dataset publish]
Publish --> Promote[catalog promote]
Promote --> Serve[Serving store is ready]
This normal flow matters because it separates dataset publication from dataset discovery. Many user issues that look like runtime bugs are really catalog-state omissions.
For most users, catalog promote is the important day-to-day action after a dataset is successfully published.
Example Commands¶
Promote a published dataset into the catalog:
cargo run -p bijux-atlas --bin bijux-atlas -- catalog promote \
--store-root artifacts/getting-started/tiny-store \
--release 110 \
--species homo_sapiens \
--assembly GRCh38
Remove it again if needed:
cargo run -p bijux-atlas --bin bijux-atlas -- catalog rollback \
--store-root artifacts/getting-started/tiny-store \
--release 110 \
--species homo_sapiens \
--assembly GRCh38
What Can Go Wrong¶
- the dataset was never published into the store
- the catalog is missing or malformed
- the latest alias is updated before promotion
- the serving store is mistaken for the ingest build root
What This Page Helps You Confirm¶
- whether a dataset is actually discoverable by the server
- whether the catalog reflects the dataset state you think is published
- whether alias changes happened in the right order
Rule of Thumb¶
If the question is “can the server discover this dataset,” the answer usually lives in the catalog state, not only in the existence of artifact files.
Purpose¶
This page explains the Atlas material for catalog workflows and points readers to the canonical checked-in workflow or boundary for this topic.
Stability¶
This page is part of the canonical Atlas docs spine. Keep it aligned with the current repository behavior and adjacent contract pages.