All notable changes to this project should be documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

• http: Rewrite specific URL patterns to DBpedia resource URIs:
  • http://<locale>.dbpedia.org/page/<slug>
  • https?://<locale>.wikipedia.org/wiki/<slug>
• http: Allow lookup of multiple URI clusters at once with the uris parameter.
• http: log in the Uvicorn logger that app startup is ready.

• Install the latest RocksDB release from GitHub archive instead of git (see #6).
• Update python to v3.7 and install the latest version of python packages.
• Added mypy to development dependencies.

• db: lock contention (RocksIOError, see #8), fixed by reusing the admin DB connection in restore.create_backup().

• http: purge old data DBs on Gunicorn/Uvicorn initialization.

• http: let Gunicorn preload the app (instead of per worker).

• loader: ability to restore from backup by registering them in the admin DB.
• loader: CLI to interactively select a backup to restore.

• loader: hot-swap DB when the current DB never completed loading.

• db.get_connection(): RocksDB API change.

• loader: find and fetch latest snapshot via Databus SPARQL query (see #5).

• http: Gunicorn performance tuning.
• loader: RocksDB performance tuning.

• http: move meta info to its own top-level key.
• Docs: updated readme.

• loader: don't overlook decompressed chunks that are smaller than one line.
• loader: only split on the secondary seperator once.

• Docs: deployment and maintenance notes.
• parser: parse RDF literals from ntriple lines.
• loader: progress bars.
• http: include singleton IDs and meta info in response (see #1).

• Adjust DBp Global namespace to global.dbpedia.org/id/.
• loader: only load and backup if a new snapshot is available.
• loader: load compressed TSV snapshot instead of ntriples part files.
• loader: separate concerns with producer/consumer pattern.
• http: replace ApiStar with Starlette.
• Restructured Docker configuration.

• pinned python version at 3.6.
• loader: close the event loop after load and backup are done.

• Docs: installation and usage notes.

• loader: only load and backup if new parts are available.

• Docker image path.
• loader: close the event loop after load and backup are done.

• http: run with Gunicorn server in production.
• loader: don't attempt manual compaction (prone to stalling).

• loader: admin part key encoding.

• URI cluster loader from Spark output.
• http interface based on ApiStar.