HomeGitHub RepoBook Demo

Metadata Workflows

This feature requires the quilt3 API version 3.3 or higher.

Overview

A Quilt workflow is a quality gate that you set on a bucket to ensure the quality of your data and metadata before it becomes a Quilt package. You can create as many workflows as you like to accommodate all of your data creation patterns.

Why metadata quality matters

Under the hood, Quilt workflows use JSON Schema to check that package metadata have the right shape. Metadata shape determines which keys are defined, their values, and the types of the values.

Ensuring the quality of your data has long-lasting implications:

  1. Consistency — if labels and other metadata don't use a consistent, controlled vocabulary, reuse becomes difficult and trust in data declines

  2. Completeness — if your workflows do not require users to include files, documentation, labels, etc. then your data is on its way towards becoming mystery data and ultimately junk data that no one can use

  3. Context — data can only be reused if users know where it came from, what it means, who touched it, and what the related datasets are

From the standpoint of querying engines like Amazon Athena, metadata that lacks consistency and completeness is extremely difficult to query longitudinally and depreciates over time (as team members change, platforms change, and tribal knowledge is lost).

Use cases

  • Ensure that labels are correct and drawn from a controlled vocabulary (e.g. ensure that the only labels in a package of images are either "bird" or "not bird"; avoid data entry errors like "birb")

  • Ensure that users provide a README.md for every new package

  • Ensure that included files are non-empty

  • Ensure that every new package (or dataset) has enough labels so that it can be reused (e.g. Date, Creator, Type, etc.)

Get started

To get started, create a configuration file in your Quilt S3 bucket at s3://BUCKET-NAME/.quilt/workflows/config.yml.

Here's a complex example:

Setting a default workflow

Once a configuration file is present, you usually must specify a workflow when pushing a package from quilt3 to that bucket. You can however specify a default_workflow value at the top-level in your config.yml file:

This specifies which workflow will be used (experiment) if a workflow parameter in the Package.push() API call or CLI is not provided. Otherwise, all push calls must specify a workflow or the push will fail.

Bypassing workflow validation

If you wish for your users to be able to skip workflow validation altogether, you can make workflow validation optional with is_workflow_required: False at the top-level in your config.yml file:

Now your users can specify workflow=None in the Python API (or --workflow '' in the CLI) when they push packages (or drop the workflow parameter altogether).

Specifying a workflow

Otherwise, once you have a configuration you must specify a workflow when pushing a package:

Let's retry, adding the workflow= parameter:

The above QuiltException is caused by is_message_required: true. Here's how we can pass the workflow:

Now let's push with workflow='beta':

We encountered another exception because the beta workflow specifies metadata_schema: superheroes. Therefore, the test/package metadata must validate against the JSON Schema located at s3://quilt-dev-metadata/schemas/superheroes.schema.json:

Note that superhero is a required property of the package metadata, which is specified using set_meta.

For the gamma workflow, both is_message_required: true and metadata_schema are set, so both message and package metadata are validated:

Data quality controls

In addition to package-level metadata. Quilt workflows enable you to validate package names, and basic file metadata.

You must include the following schema version at the root of your config.yml in order for any catalog-specific features to function:

Package name defaults (Quilt catalog)

By default the Quilt catalog auto-fills the package handle prefix when creating or revising a package according to the following logic:

  • Packages tab: username (everything before the @ in your sign-in email). Equivalent to

  • Files tab: parent directory name. Equivalent to

You can customize the default prefix with package_handle key in one or both of the following places:

  • Set catalog.package_handle.(files|packages) at the root of config.yml to affect all workflows

  • Set workflows.WORKFLOW.catalog.package_handle.(files|packages) to affect the tabs and workflow in question

Example

Package name validation

You can validate package names with WORKFLOW.handle_pattern, which accepts JavaScript regular expression.

By default, patterns are not anchored. You can explicitly add start (^) and end ($) markers as needed.

Example

Package file validation

You can validate the names, sizes and metadata of files in the package with WORKFLOW.entries_schema. The provided schema runs against an array of objects known as package entries. Each package entry defines a logical key (its relative path and name in the parent package), size (in bytes) and metadata.

Example

s3://bucket/must-contain-readme.json

Requires a README

s3://bucket/must-contain-readme-summarize-at-least-1byte.json

s3://bucket/must-have-foo-bar-meta.json

Requires { "foo": "bar" } object as user specified metadata in README.md

Cross-bucket package push (Quilt catalog)

In Quilt, S3 buckets are like git branches but for data. With quilt3 you can browse any package and then push it to any bucket that you choose.

As a rule, cross-bucket pushes or "merges" reflect change in a package's lifecycle. For example, you might push a package from my-staging-bucket to my-production-bucket as it matures and becomes trusted.

The catalog's Push to bucket feature can be enabled by adding a successors property to the config. A successor is a destination bucket.

If copy_data is true (the default), all package entries will be copied to the destination bucket. If copy_data is false, all entries will remain in their current locations.

JSON Schema

  • Quilt workflows support the Draft 7 JSON Schema.

  • JSON schemas can be stored anywhere in any accessible Amazon S3 bucket. Provided the path to the file is accessible in config.yml, the schema will successfully validate your package metadata shape.

Default values

Quilt supports the default keyword.

Auto-fill dates

If you wish to pre-populate dates in the Quilt catalog, you can use the custom keyword dateformat in your schemas. For example:

The dateformat template follows Unicode Technical Standard #35.

Arrays, tuples and enums

Quilt supports the array data type. You can use array if you need to define a list of metadata values for a metadata key. These elements can be of any type.

If the order in the list is not significant, use "arrays" (using "items" and "anyOf"):

With this Schema you can create a list of metadata values such as: ["Any string A", 123, "Any string B"] or [123, "Any string", 456]

If the order in the list is important and the list is fixed in length, then use "tuples" (using "items", "minItems", and "maxItems"):

With this Schema you can create strictly ordered lists, such as ["Any string", 123].

An incorrect order will return an error [123, "Any string"] // invalid.

Remember that you should define "minItems" and "maxItems" or "minItems" and "additionalItems": false, because "tuples" must have a fixed size.

Instead of letting users set any metadata value, you can define list of available options with enum:

With this Schema you can create a list of any length with predefined values, such as ["Fixed 1", "Fixed 2", "Fixed 1"].

With this Schema users are allowed to create tuples like ["Fixed 1"] or ["Fixed 2"].

If you want to provide users with a list of predefined metadata values but additionally let them add any values outside of this list, you can use the anyOf keyword:

Metadata lists such as ["Fixed 1", "Fixed 2"], ["Fixed 1", "Any string"] or ["Any string 1", "Any string 2"] are all valid.

In certain use cases you may want to define metadata lists that have first-ordered items of predefined values, and the rest are any other outside of the predefined values. Then you create tuples with "additionalItems": true:

With this Schema lists such as ["Fixed 1", "Any string", 123] are valid but ["Any string", 123] are invalid.

Example properties

The following examples show how you can specify complex properties such as object, array, and compound enum types.

Objects

Compound enums: arrays

Compound enums: objects

Compound enums: arrays and objects

This allows for flexible and extensible schema definition, and hence validation, of complex metadata schemas to any depth.

Quilt currently uses the Draft 7 Json Schema where tuples are validated with items, and not prefixItems. The prefixItems keyword was added in Draft 2020-12, and is not currently supported.

config.yml JSON Schema

See workflows-config_catalog-1.0.0.json and workflows-config-1.1.0.json.

Known limitations

  • Only Draft 7 Json Schemas are supported

    • If a workflow schema includes a non-supported keyword, the user interface displays an unknown keyword: <non-supported keyword> error

  • Schemas with $ref are not supported

  • Schemas must be in an S3 bucket for which the Quilt user has read permissions

PreviousMetadata ManagementNextAbout the Catalog

Last updated

Was this helpful?