* fix: roll back document rows when FAISS insert fails
Validate embedding shape/dimension before local writes. If DocumentStorage
commits but FAISS write fails, best-effort delete the partial FAISS ids and
document rows so the two stores do not diverge.
* fix: clean up partial KB upload state before metadata commit
Track metadata_committed right after commit succeeds. On failure before
metadata commit, roll back chunks/vectors first, then residual KB rows and
media files. After metadata is committed, keep the document even if stats
or refresh fails.
* test: cover knowledge base upload atomicity and rollback
Add unit tests for insert_batch FAISS failure rollback, dimension validation
before document writes, upload cleanup order, no-rollback after metadata
commit, and real DocumentStorage+FAISS no-orphan cases.
* fix: log media directory cleanup failures after upload rollback
Best-effort rmdir of the per-document media directory should still emit a
warning when removal fails so incomplete cleanup is diagnosable.
* fix(kb): clean up KBMedia records on document delete, fix update_kb_stats
- delete_document_by_id now deletes associated KBMedia records before
removing the KBDocument row, preventing orphaned media data.
- update_kb_stats now passes kb_id filter to vec_db.count_documents,
fixing chunk count that previously counted across all KBs.
- add tests for document deletion cleanup and update_kb_stats behavior.
* fix(kb): merge media and document deletion into single transaction
* feat: add EPUB parsing support for knowledge base and file reader
* feat: update supported file formats for document upload in knowledge base
* feat: enhance EPUB parser to support spine order and generic containers
* makeitdown parse epub
* update parser
* fix
* feat: implement FTS5 support in DocumentStorage and SparseRetriever with tokenizer enhancements
* feat: optimize FTS row handling in DocumentStorage and update query tokenization in SparseRetriever
* fix: improve error handling for knowledge base upload
- Log details field in KnowledgeBaseUploadError for better debugging
- Distinguish between empty pre-chunked text and empty chunking result
with appropriate error messages
* style: format code
* fix: 改进 KnowledgeBaseManager 和 KBHelper 中的初始化错误处理
* fix: 改进知识库初始化和重排序错误处理，增强日志记录
* fix: 改进知识库模块初始化和检索错误处理
* fix(ui): handle kb init errors in list cards
display a dedicated error state for knowledge base cards that fail
initialization, including a visible badge and error details
prevent navigation and edit actions for failed cards while keeping
delete available, and hide normal stats/description for error items
add list.initError locale strings for en-US, ru-RU, and zh-CN
* fix(kb): avoid replacing helper on init failure
Initialize a new KB helper before swapping instances so a failed re-init
does not break the active knowledge base service.
If initialization fails, restore in-memory KB settings and keep the
existing helper and previous init error state.
Also clear stale init_error after successful vector DB initialization to
prevent outdated error reporting.
* test(kb): add kb manager resilience tests
cover initialization failure and recovery scenarios to guard
against regressions in kb error handling
include reference assets under refs for test validation
---------
Co-authored-by: RC-CHN <1051989940@qq.com>
* perf: batch metadata query in KB retrieval to fix N+1 problem
Replace N sequential get_document_with_metadata() calls with a single
get_documents_with_metadata_batch() call using SQL IN clause.
Benchmark results (local SQLite):
- 10 docs: 10.67ms → 1.47ms (7.3x faster)
- 20 docs: 26.00ms → 2.68ms (9.7x faster)
- 50 docs: 63.87ms → 2.79ms (22.9x faster)
* refactor: use set[str] param type and chunk IN clause for SQLite safety
Address review feedback:
- Change doc_ids param from list[str] to set[str] to avoid unnecessary conversion
- Chunk IN clause into batches of 900 to stay under SQLite's 999 parameter limit
- Remove list() wrapping at call site, pass set directly
* ci: remove Electron desktop build from release pipeline
* chore: remove electron desktop and switch to tauri release trigger
* ci: remove desktop workflow dispatch trigger
* refactor: migrate data paths to astrbot_path helpers
* fix: point desktop update prompt to AstrBot-desktop releases
* fix: ensure atomic creation of knowledge base with proper cleanup on failure
- Added pre-validation for embedding_provider_id parameter
- Added check for existing knowledge base with same name
- Implemented proper rollback mechanism when KBHelper initialization fails
- Uses same session for cleanup to ensure data consistency
- Fixes#4403
* fix: ensure atomic KB creation with session.flush() to remove race condition risks
* fix: ensure change the annotation back
