Extract core httpfs retrieval optimizations into clean implementation…#58
Extract core httpfs retrieval optimizations into clean implementation…#58cr458 wants to merge 1 commit into
Conversation
There was a problem hiding this comment.
💡 Codex Review
geovibes/geovibes/ui/data_manager.py
Lines 421 to 423 in 01d24cc
This early return prevents the remote Tier-4 scan from running when manifest.csv is absent, even when include_remote=True. In remote-only environments (or packaged installs where the manifest is not present), initialization will end with no discovered databases and raise FileNotFoundError, despite remote datasets being available via discover_remote_databases.
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| self.data.current_database_path = db_url | ||
| self.data.current_database_info = db_info | ||
| self.data.current_faiss_path = faiss_url | ||
| self.data.current_geometry_cache_path = geometry_url | ||
| self.data.tile_spec = db_info.get("tile_spec") |
There was a problem hiding this comment.
Clear old connection pools before progressive remote loading
This progressive-loading path mutates self.data directly instead of going through DataManager.switch_database/_connect_to_database_internal, so prior auxiliary state is never torn down. If a user switches remote databases after prefetch has already created a background pool, DataManager.configure_background_connection_pool can reuse existing pooled connections (same size) that still point at the previous DB, and background embedding prefetch may read vectors for overlapping IDs from the wrong dataset, corrupting the query vector and search ranking.
Useful? React with 👍 / 👎.
… branch