Metadata Warehouse

Docker Compose Setup

This repo contains a docker-compose.yml file which configures the containers and their interaction. To run the containers:

• users and passwords (adjust env variables as needed and set new passwords):

  cp env.template .env

  Optionally add the following env variables for postgres and/or OpenSearch (not needed for local dev):
  • POSTGRES_ADDRESS (default "postgres") and POSTGRES_PORT (default 5432)
  • OPENSEARCH_ADDRESS (default "opensearch") and OPENSEARCH_PORT (default 9200)
  • FASTAPI_ADDRESS (default "127.0.0.1") and FASTAPI_PORT (default 8080)
• API keys for search API server:

  cp keys.env.template keys.env

• Dev config for docker containers:

  cp docker-compose.override.yml.template docker-compose.override.yml

• docker compose up -d

• create postgreSQL table structure, see below.
• create OpenSearch index, see below.
• run transformation process, see below.

pgAdmin

• when using pgAdmin, register a new server with Host name "postgres" (container name in docker network) with port "5432".
• provide credentials as defined in .env.

Basic Setup

• cd scripts

• Install uv and run

  uv sync --frozen

Prepare Data For Local Import

In production, the DB is populated by running the crawler. In development, it may be more convenient to load pre-harvested static data:

• create a folder per repository, e.g., scripts/postgres_data/data/dans_arch

• create an XML file containing records such as

  <Records xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance">
     <record xmlns="http://www.openarchives.org/OAI/2.0/">
     ...
     </record>
     ...
  </Records>

• fetch additional metadata using scripts/postgres_data/dataverse.py (Dataverse) or scripts/postgres_data/get_meta.py (HAL, Zenodo) Combine additional metadata files in one virtual structure using

  Dataverse:

   import json, glob
   
   lookup = {}
   for f in glob.glob('*.json'):
       with open(f) as fh:
           obj = json.load(fh)
       key = obj["data"]["datasetPersistentId"]
       lookup[key] = obj  # or just the fields you need
   
   # Optionally save it
   with open('lookup.json', 'w') as out:
       json.dump(lookup, out)

  HAL:

  import json, glob
    
    lookup = {}
    for f in glob.glob('*.json'):
        with open(f) as fh:
            obj = json.load(fh)
        if len(obj["response"]["docs"]) == 1:
            key = obj["response"]["docs"][0]["halId_s"]
            lookup[key] = obj  # or just the fields you need
        else:
            print(f)
            print(obj)
    
    # Optionally save it
    with open('lookup.json', 'w') as out:
        json.dump(lookup, out)      

• check the settings in scripts/postgres_data/import_data.py:

  HARVEST_ENDPOINTS = [
    ('DANS', 'https://archaeology.datastations.nl/oai', Path('data/dans_arch/dans_arch.xml'), Path('doi_dataverse/lookup.json'), None)

  where data/dans_arch/dans_arch.xml contains the OAI-PMH records and doi_dataverse/lookup.json the additional metadata.

Create Postgres DB and Load and Transform Data

• cd scripts/postgres_data

• create table structure and repo config as defined in scripts/postgres_data/create_sql/$dbname

  uv run create_db.py --db $dbname [--reset]

  This will create and init the specified DB if it does not exist yet. If it already exists and should be dropped and reinitialized, additionally provide the flag --reset.

• load XML data from scripts/postgres_data/data (populates table harvest_events):

   uv run import_data.py

  See Prepare Data For Local Import for further details about local data preparation.

• transform data from scripts/postgres_data/data to a local dir (to test transformation, alternative to using the Celery process):

  uv run transform.py -i harvests_{repo_suffix} -o {repo_suffix}_json -s JSON_schema_file [-n] [-v]

  If the -n flag is provided, the JSON data will be normalized (the raw JSON may look differently based on the input XML, see these specs). If the -v flag is set, the JSON will be validated against the JSON schema file utils/schema.json

Create OpenSearch Index

• cd scripts/opensearch_data

• create test_datacite index (deletes existing test_datacite index):

  uv run create_index.py

• for sample OpenSearch queries, see open_search_queries

• to test queries requiring vector embeddings, run

  uv run query_index.py

Run Transformation Process

The transformer container provides an API to start the transformation and indexing process.

A transformation requires a harvest_run_id. When running the script import_data.py (scripts/postgres_data/data), for each endpoint a harves run is created, the single OAI-PMH records are registered as harvest events, and the harvest run is then closed. Note that a transformation can only be performed for a closed harvest run.

• check if transformer container is up and running:

  http://127.0.0.1:8080/health

• To obtain a harvest run id and status for a given endpoint (https://dabar.srce.hr/oai):

  http://127.0.0.1:8080/harvest_run?harvest_url=https%3A%2F%2Fdabar.srce.hr%2Foai

• start transformation process:

  http://127.0.0.1:8080/index?harvest_run_id=xyz

• see transformation task results in flower:

  http://127.0.0.1:5555/tasks

After starting the stack with docker compose up, you can run the harvester for a given repository URL, e.g.:

docker compose run harvester https://lifesciences.datastations.nl/oai

Scheduler

The scheduler automates the full ingestion workflow: harvesting → transformation → indexing. It is designed to be executed periodically via CRON.

Dependent endpoints

Some repositories are configured as dependent endpoints in the endpoints table by setting depends_on_endpoint_id. A dependent endpoint uses identifiers collected from its master endpoint, so the scheduler always triggers independent endpoints first and moves dependent endpoints to the end of the same harvesting batch. This ensures HAL is harvested before Zenodo when both endpoints are due for harvesting.

The rest of the scheduler pipeline is unchanged: every selected endpoint is still crawled sequentially, the scheduler waits until harvest runs are closed, and then closed runs are sent to transformation/indexing.

Run scheduler

uv run python -m scheduler.run

Environment variables

Optionally add the following env variables (not needed for local dev):

• WAREHOUSE_API_URL (default "http://transform:80")
• INDEX_NAME (default "test_datacite")

Linting

To format all files properly, run:

• uv run ruff format
• uv run ruff check --select I --fix

Run E2E Tests

Before running the e2e tests locally, set the env vars POSTGRES_DB and FILE_DB to testdatasetdb and testfiledb, respectively, since the e2e tests and the API must use the same DBs.

Note that the e2e tests reset testdatasetdb and testfiledb on each run. Because the test DB names are hardcoded in the e2e tests, your production DBs will not be overwritten.

To run the e2e tests:

uv run pytest -s e2e

Commit Message Conventions

Keep to this commit message style. For semantic versioning, see these release-please. Set up pre-commit hooks to check your messages before commiting them to the repo:

• uv sync --frozen --all-extras --dev
• uv run pre-commit install --hook-type commit-msg

See .pre-commit-config.yaml for further details.