Vadim_Malanov/LegacyHUB
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

fix(indexing): unblock chunk indexing and gate reranker model load

Browse Source 
Three defects surfaced during the first live ingest:

1. OpenSearch index mapping is dynamic:"strict", so the first chunk
   carrying metadata.section_heading was rejected with
   strict_dynamic_mapping_exception and zero chunks were indexed. The
   metadata and quality_flags sub-objects now opt into
   dynamic:true so chunker-specific keys insert without a fixed schema,
   while the top-level mapping stays strict.

2. hybrid_search.run_search called get_reranker() unconditionally, so
   every search - even search_mode=lexical with RERANKER_ENABLED=false
   - triggered a multi-GB FlagReranker download and hung the request.
   The loader is now lazy behind the reranker_enabled flag.

3. docker-compose x-common-env never forwarded RERANKER_ENABLED, so the
   container always defaulted to true regardless of .env. Added
   RERANKER_ENABLED: ${RERANKER_ENABLED:-true} to common-env.

After these fixes POST /search (lexical) returns HTTP 200 in ~50ms with
a ranked hit and citation; OpenSearch _count reports the indexed chunk.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
This commit is contained in:
Vadim Malanov
2026-06-14 13:42:10 +03:00
parent 2806aaf441
commit 01ba710c08
3 changed files with 19 additions and 10 deletions
Download Patch File Download Diff File Expand all files Collapse all files

18
app/indexing/hybrid_search.py
View File

@@ -72,14 +72,18 @@ def run_search(req: SearchRequest) -> SearchResponse:
merged = _merge(lexical, semantic, mode)
candidates = merged[: settings.rerank_candidates]
reranker = get_reranker()
reranked_flag = False
if settings.reranker_enabled and reranker.available and candidates:
scores = reranker.score(req.query, [c.text for c in candidates])
for c, s in zip(candidates, scores, strict=True):
c.dense_score = s
candidates.sort(key=lambda c: (c.dense_score or 0.0), reverse=True)
reranked_flag = True
if settings.reranker_enabled and candidates:
# Lazy-load only when the flag is on; FlagReranker construction
# downloads the model and we do not want that on the cold path when
# the operator has disabled reranking.
reranker = get_reranker()
if reranker.available:
scores = reranker.score(req.query, [c.text for c in candidates])
for c, s in zip(candidates, scores, strict=True):
c.dense_score = s
candidates.sort(key=lambda c: (c.dense_score or 0.0), reverse=True)
reranked_flag = True
final = candidates[: req.limit]

8
app/indexing/opensearch_client.py
View File

@@ -72,8 +72,12 @@ INDEX_SETTINGS: dict[str, Any] = {
},
"ocr_confidence": {"type": "float"},
"language_hint": {"type": "keyword"},
"metadata": {"type": "object", "enabled": True},
"quality_flags": {"type": "object", "enabled": True},
# metadata + quality_flags carry chunker-specific keys
# (section_heading, table_index, low_ocr_confidence, ...) that we do
# not want to enumerate up-front. Top-level mapping is dynamic:
# "strict", but these two sub-objects opt-in to dynamic insertion.
"metadata": {"type": "object", "enabled": True, "dynamic": True},
"quality_flags": {"type": "object", "enabled": True, "dynamic": True},
"created_at": {"type": "date"},
},
},