hj/pleroma
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity

fix(search): set gin_fuzzy_search_limit only for search queries

Browse source 
The limit was a connection parameter, so it applied to every GIN index
scan, not only full-text search: recipient lookups and Oban job queries
could return incomplete, random results. Set it with SET LOCAL around
the search query only, configurable under Pleroma.Search.DatabaseSearch.

The default stays 500 for search: measured on a 74M-object database,
higher limits or none make searches for common words exceed the query
timeout. Remote fetches for URL queries run outside the transaction.

Based on Akkoma 811f0a468.

Co-authored-by: Oneric <oneric@oneric.stub>
This commit is contained in:
Lain Soykaf 2026-10-29 11:26:02 +04:00
commit fa8541309c
No known key found for this signature in database
6 changed files with 98 additions and 17 deletions
Download patch file Download diff file Expand all files Collapse all files

8
docs/configuration/search.md
View file

@ -10,6 +10,14 @@ To use built-in search that has no external dependencies, set the search module
While it has no external dependencies, it has problems with performance and relevancy.
With a GIN index (the default), search queries limit how many matches PostgreSQL considers (`gin_fuzzy_search_limit`, see [PostgreSQL's docs](https://www.postgresql.org/docs/current/gin.html#GIN-TIPS)). This keeps searches for common words fast on large databases, but the results are incomplete and can change between requests. On smaller instances you can turn it off for exact results:
> config :pleroma, Pleroma.Search.DatabaseSearch, gin_fuzzy_search_limit: nil
Without a limit, searches for common words can take longer than the database timeout on large databases, and then return nothing.
Earlier versions of Pleroma set this limit for every database connection, which also made other GIN index scans (for example recipient lookups) incomplete. It is now only set for search queries.
## QdrantSearch
This uses the vector search engine [Qdrant](https://qdrant.tech) to search the posts in a vector space. This needs a way to generate embeddings and uses the [OpenAI API](https://platform.openai.com/docs/guides/embeddings/what-are-embeddings). This is implemented by several project besides OpenAI itself, including the python-based fastembed-server found in `supplemental/search/fastembed-api`.