v1.0.0-beta.2

Added

• HTTP requests made by the feed-fetcher worker can be limited with the new
  request_timeout setting.
• Optimistic locking mechanism for GORM models.
  See: models.OptimisticLockModel, models.OptimisticSave and
  models.ErrStaleObject.

Changed

• The base GORM model (models.Model) now includes a Version field and
  satisfies the models.OptimisticLockModel interface, allowing
  optimistic locking.
• Each worker has been modified avoiding long-lasting transactions, extracting
  from them heavy operations, and opting for optimistic locking when
  record updates are involved. Reducing the transactions' duration and removing
  the explicit row-level locks can produce tremendous improvements
  on the performance of the whole system, when under heavy loads (i.e.
  at least thousands of sources).
• Give ordering priority to sources never retrieved before (i.e.
  last_retrieved_at is null) when looping through sources to schedule from
  feed-scheduler and twitter-scheduler tasks.
• Improve hnswcloent.Client.SearchKNN performance, making the requests to
  each candidate daily HNSW index concurrently.
• Minor refactoring and improvements to some log messages and their severity
  level.
• Upgrade dependencies.

v1.0.0-beta.1

Added

• hnswclient.Client.FlushAllIndices function.
• HNSW-Purger task now flushes all remaining indices after deleting the old
  ones.
• Some tests.

Changed

• Use weaker database row-level locks wherever possible (FOR SHARE instead
  of FOR UPDATE) to prevent possible slowdowns.
• Upgrade dependencies.

Removed

• hnswclient.Client.Index does not flush an HNSW index anymore at each
  vector insertion. This was possibly causing slowdowns in case of
  large indices and many concurrent jobs inserting new vectors.

Fixed

• A failing configuration test.

v1.0.0-beta

Added

• HNSW-Purger task (command purge-hnsw).

Changed

• Avoid long-living gRPC connections moving the dialing from commands
  initialization to workers (see commit a61fcbc for details).
• Upgrade dependencies.

v1.0.0-alpha.3

Added

• Add documentation to the README.
• AUTHORS.md

Changed

• Move cmd/whatsnew.go to the project's root path, so that the tool can be
  installed more easily with go install command.
• Use golang:1.17.1-alpine3.14 as base Builder image in the Dockerfile.
• Provide a complete docker-compose file and related configurations, now under
  docker-compose folder.
• Upgrade dependencies.

v1.0.0-alpha.2

Changed

• Enable client-side round-robin DNS load balancing for all gRPC connections.

v1.0.0-alpha.1

Added

• Allow setting reservation timeout and number of retries for each Faktory job
  from configuration.

Changed

• Use the WebArticle translated title, when available, as preferred text data
  source in text-classifier, vectorizer, and zero-shot-classifier workers.
• Upgrade dependencies.

v1.0.0-alpha

Changed

• The whole project has been completely rewritten. Most notably, the simplistic
  way of handling workers' jobs with RabbitMQ has been replaced with more
  reliable jobs scheduling using Faktory.

v0.5.0

Handler CORS

v0.4.0

Upgrade spago to v0.5.2 and adapt the code

v0.3.3

Add max tweets number to configuration