- Introduced a new test file to validate the LLM bundle construction for streaming flows in chat.
- Implemented tests to ensure that both DB-backed and global models enable streaming correctly.
- Utilized mocking to isolate dependencies and verify the expected behavior of the LLM constructor.
Two integration tests pinning the connector index endpoint's authorization:
- cross-space index (attacker owns space B, connector lives in victim's
space A, request passes search_space_id=B) is rejected with 404 at the
search-space reconciliation, before the permission check (which would
otherwise pass for the attacker's own space).
- same-space index authorizes check_permission against the connector's
own search space, not the caller-supplied query param.
Mirrors the existing tests/integration harness (direct handler calls with
the savepoint-rolled-back db_session; check_permission patched so the test
needs no real RBAC wiring).
Real-DB tests assert unchanged chunk rows survive edits, only new text is
embedded, removed rows are deleted with positions compacted, and the kill
switch restores full-replace. An autouse fixture disables the ETL/embedding
caches so a developer's .env can't leak cache hits into unrelated tests.
Greedy multiset match on chunk text decides which rows keep their embeddings,
which texts need embedding, and which rows are deleted. No DB, no embeddings;
fully unit-tested (reuse, head insert, middle edit, deletion, duplicates,
reorder, full rewrite).
Covers the public cache surface against real Postgres and a real local file
backend (no mocks): recall miss, remember->recall vector/text/order round-trip,
the dimension-mismatch refusal, the repository SQL behind eviction and dedup
(size sum, coldest ordering, TTL cutoff, duplicate-key no-op, reuse counter),
and the blob store save/load round-trip and delete.
The cached payload is the indexing pipeline's embeddings (markdown is
chunked then embedded), so "embedding cache" names the expensive output
directly and removes the "index" ambiguity (DB index vs vector index vs
indexing phase). Renames the service, settings, eligibility, eviction
task, metrics, config flags (INDEX_CACHE_* -> EMBEDDING_CACHE_*), object
prefix, and the table (index_cache_embedding_sets -> embedding_cache_sets)
with its constraint and indexes. Migration 161 renamed accordingly.
Pin AUTH_TYPE=LOCAL (and REGISTRATION_ENABLED=TRUE) in the test bootstrap so
the email/password auth routers mount during integration tests regardless of a
developer's .env=GOOGLE; without this the upload tests 404 on registration.
Also update three tests to the current QuotaInsufficientError signature
(balance_micros) after used_micros/limit_micros were removed.
- Introduced LLMErrorCategory and adapt_llm_exception to normalize LLM exceptions.
- Updated llm_retryable_message and llm_permanent_message to utilize the new adaptation logic.
- Enhanced classify_stream_exception to classify provider errors and return user-friendly messages.
- Added tests for error classification and adaptation to ensure robustness.
- Updated frontend error handling to display appropriate messages based on new classifications.
