Unique violations on packages_name_key happen on a statement that
cannot logically touch name - log the exact query and a longer
backtrace to identify what really fails.
Drop the threaded worker pool (queue, mutexes, connection pooling).
The task now processes packages one by one via find_each - plenty
fast since the external embedding service dominates latency, and
much easier to reason about.
A single failing package used to kill its whole worker thread silently
because only Vectorizer::Error was rescued per package. Now every
StandardError is caught and logged with class and backtrace, and the
embedding write is narrowed to update_column(:embedding) so the task
only ever issues a minimal UPDATE - it cannot touch any other column
or fire model callbacks. Progress stats no longer serialize database
writes through the mutex.
Package text search now combines several strategies: exact name match,
name prefix promotion, compound word splitting ("sqlite browser" finds
"sqlitebrowser") and semantic nearest neighbor search over description
embeddings computed by an external embedding service (all-MiniLM-L6-v2,
384 dims) stored with the pgvector extension. Classic PostgreSQL
full-text search remains as fallback when the vector service is
unreachable. Gibberish queries without lexical overlap with the package
data return empty results instead of random matches.
Embeddings can be backfilled with bin/rails debshots:compute_vectors.
Also drops the unused lograge gem from the Gemfile.
To migrate the application as of April 2020:
• leave models/screenshot.rb with “has_attached_file”
• bundle exec rake db:migrate
• bin/move_paperclip_to_activestorage
• set models/screenshot.rb to “has_one_attached”
