articat

articat

Minimal metadata catalog to store and retrieve metadata about data artifacts.

Getting started

At a high level, articat is simply a key-value store. Value being the Artifact metadata. Key a.k.a. "Artifact Spec" being:

• globally unique id
• optional timestamp: partition
• optional arbitrary string: version

To publish a file system Artifact (FSArtifact):

from articat import FSArtifact
from pathlib import Path
from datetime import date

# Apart from being a metadata containers, Artifact classes have optional
# convenience methods to help in data publishing flow:

with FSArtifact.partitioned("foo", partition=date(1643, 1, 4)) as fsa:
    # To create a new Artifact, always use `with` statement, and
    # either `partitioned` or `versioned` methods. Use:
    # * `partitioned(...)`, for Artifacts with explicit `datetime` partition
    # * `versioned(...)`, for Artifacts with explicit `str` version

    # Next we produce some local data, this could be a Spark job,
    # ML model etc.
    data_path = Path("/tmp/data")
    data_path.write_text("42")

    # Now let's stage that data, temporary and final data directories/buckets
    # are configurable (see below)
    fsa.stage(data_path)

    # Additionally let's provide some description, here we could also
    # save some extra arbitrary metadata like model metrics, hyperparameters etc.
    fsa.metadata.description = "Answer to the Ultimate Question of Life, the Universe, and Everything"

To retrieve the metadata about the Artifact above:

from articat.fs_artifact import FSArtifact
from datetime import date
from pathlib import Path

# To retrieve the metadata, use Artifact object, and `fetch` method:
fsa = FSArtifact.partitioned("foo", partition=date(1643, 1, 4)).fetch()

fsa.id # "foo"
fsa.created # <CREATION-TIMESTAMP>
fsa.partition # <CREATION-TIMESTAMP>
fsa.metadata.description # "Answer to the Ultimate Question of Life, the Universe, and Everything"
fsa.main_dir # Data directory, this is where the data was stored after staging
Path(fsa.joinpath("data")).read_text() # 42

Features

• store and retrieve metadata about your data artifacts
• no long running services (low maintenance)
• data publishing utils builtin
• IO/data format agnostic
• immutable metadata
• development mode

Artifact flavours

Currently available Artifact flavours:

• FSArtifact: metadata/utils for files or objects (supports: local FS, GCS, S3 and more)
• BQArtifact: metadata/utils for BigQuery tables
• NotebookArtifact: metadata/utils for Jupyter Notebooks

Development mode

To ease development of Artifacts, articat supports development/dev mode. Development Artifact can be indicated by dev parameter (preferred), or _dev prefix in the Artifact id. Dev mode supports:

• overwriting Artifact metadata
• configure separate locations (e.g. dev_prefix for FSArtifact), with potentially different retention periods etc

Backend

• local: mostly for testing/demo, metadata is stored locally (configurable, default: ~/.config/articat/local)
• gcp_datastore: metadata is stored in the Google Cloud Datastore

Configuration

articat configuration can be provided in the API, or configuration files. By default configuration is loaded from ~/.config/articat/articat.cfg and articat.cfg in current working directory. You can also point at the configuration file via environment variable ARTICAT_CONFIG.

You use local mode without configuration file. Available options:

[main]
# local or gcp_datastore, default: local
# mode =

# local DB directory, default: ~/.config/articat/local
# local_db_dir =

[fs]
# temporary directory/prefix
# tmp_prefix =
# development data directory/prefix
# dev_prefix =
# production data directory/prefix
# prod_prefix =

[gcp]
# GCP project
# project =

[bq]
# development data BigQuery dataset
# dev_dataset =
# production data BigQuery dataset
# prod_dataset =

Our/example setup

Below you can see a diagram of our setup, Articat is just one piece of our system, and solves a specific problem. This should give you an idea where it might fit into your environment: