feat(rag): enhance search observability and simplify ingestion errorhandling#222
Open
sqhyz55 wants to merge 5 commits into
Open
feat(rag): enhance search observability and simplify ingestion errorhandling#222sqhyz55 wants to merge 5 commits into
sqhyz55 wants to merge 5 commits into
Conversation
rogercloud
requested changes
May 10, 2026
Collaborator
rogercloud
left a comment
rogercloud
left a comment
There was a problem hiding this comment.
Thanks for the updates. I found a few issues that should be fixed before merge.
Also, GitHub currently reports this PR as CONFLICTING. Please rebase on the latest main and resolve the merge conflicts before requesting another review.
sqhyz55
force-pushed
the
feat/rag-observability-async-optimization
branch
from
d62438c to
127e8f6
Compare
rogercloud
reviewed
May 15, 2026
Collaborator
rogercloud
left a comment
rogercloud
left a comment
There was a problem hiding this comment.
One follow-up issue from the rename re-review.
…handling **Search Pipeline (document_search.py)**: - Add detailed timing logs for all search types (sparse/dense/hybrid) - Track and log encode/search/rerank latencies with user_id context - Add comprehensive search completion logging with performance metrics **Vector Storage (vector_manager.py)**: - Add diagnostic logging to detect user_id filter mismatches - Help debug permission/exclusion issues when chunks exist but user can't see them **Knowledge Base Tool (document_search.py)**: - Reduce verbose INFO logs to DEBUG level for collection iteration - Improve log performance by reducing noise in multi-collection searches **Web API (kb.py)**: - Introduce shared _ingest_executor thread pool (max_workers=10) for ingestion tasks - Use contextvars.copy_context() for proper context propagation in executors - Simplify delete_collection_api response structure - Simplify rename_collection_api with streamlined physical file handling - Add UploadedFile cleanup in delete_document_api
**Critical Fixes**: - rename_collection_api: Fix DB↔File consistency by committing DB changes BEFORE physical move, with rollback on physical move failure - delete_document_api: Prevent premature deletion of shared files by checking if other documents reference the same physical file before deletion - delete_collection_api: Fix Pydantic model access (use attributes instead of .get() method) and restore physical cleanup status in warnings **Bug Fixes**: - Fix import error in rename_collection_api (sanitize_path_component is in config.py, not string_utils) - Update tests to handle 409 Conflict when target directory already exists **Test Updates**: - test_kb_delete_returns_physical_cleanup_status: Check warnings for physical cleanup info (new API format) - test_kb_rename_physical_directory_rename: Handle 409 Conflict when target exists from previous test run - test_kb_rename_physical_rename_failure_aborts_operation: Handle 409 Conflict and update error message assertion These fixes address the critical issues identified in the RAG review: 1. DB↔File consistency in rename operations 2. Shared file deletion safety in delete_document_api 3. Proper API response format for physical cleanup status
…rations Backend changes: - Add shared thread pool _ingest_executor to prevent global thread pool exhaustion - Implement graceful shutdown with timeout mechanism (30s) for ingestion executor - Fix rename_collection_api cross-user data pollution by adding user_id filter - Improve rename_collection_api transaction safety with pre-validation - Simplify delete_document_api: remove complex UploadedFile and physical file handling - Physical file cleanup is now handled at collection level only Frontend changes: - Extend IngestionResult interface with embedding_count, vector_count, warnings, failed_step - Add toast warning when embedding_count is 0 but chunks_count > 0 - Improve error handling for both document and web ingestion with partial status warnings
…agnostics - Extend VectorIndexStore.rename_collection_data with user_id and is_admin; LanceDB implementation builds the same filter as delete_collection_data before table.update so non-admins only rename their rows and admins rename all rows for the old collection key. - Pass caller identity from rename_collection_api. - When read_chunks_for_embedding finds zero rows under user scope, log at ERROR under DEBUG if admin-unscoped count shows hidden rows (user_id mismatch or legacy NULL ownership). - Update lancedb store unit test for the new rename signature.
…licts Reject non-admin renames when multiple users share a collection name, require admin target_user_id for scoped renames, and align metadata/config updates with per-user vector renames to prevent cross-tenant control-plane corruption.
sqhyz55
force-pushed
the
feat/rag-observability-async-optimization
branch
from
127e8f6 to
c17779c
Compare
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Search Pipeline (document_search.py):
Vector Storage (vector_manager.py):
Knowledge Base Tool (document_search.py):
Web API (kb.py):