feat(document): configurable OCR timeout and fail-fast PDF size guard

Two ingest-robustness fixes from card 309 (OHR-Bench smoke-test triage).

The OCR backend timeout was a hardcoded 180s module constant, so a tenant
whose gateway has its own shorter ceiling couldn't tune it. Promote it to
DOCUMENT_OCR_TIMEOUT_SECONDS (default 180), resolved per call via get_settings
so an override applies without a restart.

Large, awkward PDFs (e.g. a 42 MB scanned DUDE) were handed straight to the
fast/OCR tiers, where they burned the full OCR timeout for zero recovered
text. Add a pre-parse size guard in the tiered PDF pipeline: a PDF over
DOCUMENT_MAX_PDF_SIZE_MB (default 50, 0 disables) fails fast with
parse_failed_reason="oversize" before any tier runs, so the existing
permanent-failure path marks the placeholder failed and records
astrolabe_document_parse_failed_total{reason="oversize"} instead of retrying.

Both knobs go through Settings + dynaconf validators (env-var keys verified by
regression tests) and are documented under Background Indexing Configuration.

Refs: Deck board 12 card 309 (AC #3 OCR timeout + size guard).

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
This commit is contained in:
Chris Coutinho
2026-06-11 05:09:58 +02:00
co-authored by Claude Opus 4.8
parent 457c115ef4
commit 523e4cb7b5
7 changed files with 163 additions and 2 deletions
+18
View File
@@ -213,6 +213,24 @@ class TestChunkConfigValidation:
_reload_config()
assert get_settings().document_chunk_page_aware is False
def test_ocr_timeout_default_and_env_override(self):
"""document_ocr_timeout_seconds defaults to 180 and reads its env var.
Guards the _DEFAULTS-key-must-match-env-var footgun: a mismatch would
leave the override silently ignored.
"""
assert Settings().document_ocr_timeout_seconds == 180.0
with patch.dict(os.environ, {"DOCUMENT_OCR_TIMEOUT_SECONDS": "45"}, clear=True):
_reload_config()
assert get_settings().document_ocr_timeout_seconds == 45.0
def test_max_pdf_size_default_and_env_override(self):
"""document_max_pdf_size_mb defaults to 50 and reads its env var."""
assert Settings().document_max_pdf_size_mb == 50.0
with patch.dict(os.environ, {"DOCUMENT_MAX_PDF_SIZE_MB": "12.5"}, clear=True):
_reload_config()
assert get_settings().document_max_pdf_size_mb == 12.5
def test_valid_chunk_settings(self):
"""Test valid chunk size and overlap configuration."""
settings = Settings(
+39
View File
@@ -14,6 +14,7 @@ def _settings(**kw) -> Any: # a Settings stand-in (only the read fields matter)
base = dict(
document_ocr_provider="auto",
document_ocr_model="mistral/mistral-ocr-latest",
document_ocr_timeout_seconds=180.0,
embedding_gateway_url=None,
embedding_gateway_client_id=None,
embedding_gateway_client_secret=None,
@@ -128,3 +129,41 @@ async def test_processor_backend_error_returns_success_false(monkeypatch):
r = await ocr.OcrProcessor().process(b"%PDF-1.7", "application/pdf")
assert r.success is False
assert r.metadata["parse_failed_reason"] == "error"
async def test_gateway_backend_uses_configured_timeout(monkeypatch):
"""The gateway OCR call must use DOCUMENT_OCR_TIMEOUT_SECONDS (resolved per
call), not the old hardcoded 180s constant."""
captured: dict[str, Any] = {}
class _FakeResponse:
def raise_for_status(self):
pass
def json(self):
return {"pages": [{"index": 0, "markdown": "ok"}]}
class _FakeClient:
def __init__(self, *, timeout=None, **kw):
captured["timeout"] = timeout
async def __aenter__(self):
return self
async def __aexit__(self, *exc):
return False
async def post(self, url, json=None, headers=None):
return _FakeResponse()
monkeypatch.setattr(ocr.httpx, "AsyncClient", _FakeClient)
monkeypatch.setattr(
ocr, "get_settings", lambda: _settings(document_ocr_timeout_seconds=42.0)
)
backend = ocr._GatewayOcrBackend("http://gw", "mistral/mistral-ocr-latest")
await backend.ocr(b"%PDF-1.7", "application/pdf")
# httpx.Timeout(42.0, connect=10.0): the read/overall budget is the setting.
assert captured["timeout"].read == 42.0
assert captured["timeout"].connect == 10.0
+39
View File
@@ -74,6 +74,7 @@ class _Settings:
page_fraction=0.5,
min_page_chars=16,
detect_scanned=False,
max_pdf_size_mb=0.0,
):
self.document_tier1_engine = engine
self.document_classify_enabled = classify
@@ -82,6 +83,7 @@ class _Settings:
self.document_ocr_page_fraction = page_fraction
self.document_ocr_min_page_chars = min_page_chars
self.document_ocr_detect_scanned = detect_scanned
self.document_max_pdf_size_mb = max_pdf_size_mb
def _registry(*procs: tuple[DocumentProcessor, int]) -> ProcessorRegistry:
@@ -98,6 +100,43 @@ async def test_pdf_routes_to_fast_tier(monkeypatch):
assert res.processor == "fast"
async def test_oversize_pdf_fails_fast_without_parsing(monkeypatch):
"""A PDF over the size cap must fail fast as 'oversize' before any tier runs."""
monkeypatch.setattr(
reg_mod, "get_settings", lambda: _Settings(max_pdf_size_mb=0.001)
)
fast = _Fake("fast", "fast")
ran = False
orig = fast.process
async def _tracking(*a, **k):
nonlocal ran
ran = True
return await orig(*a, **k)
fast.process = _tracking # type: ignore[method-assign]
r = _registry((fast, 20))
# ~2 KB > 0.001 MB (~1 KB) cap.
res = await r.process(b"%PDF-1.7" + b"0" * 2048, "application/pdf", "big.pdf")
assert res.success is False
assert res.metadata["parse_failed_reason"] == "oversize"
assert res.processor == "size_guard"
assert ran is False, "size guard must short-circuit before the fast tier runs"
async def test_under_cap_pdf_still_parses(monkeypatch):
"""A PDF under the cap is unaffected by the guard."""
monkeypatch.setattr(
reg_mod, "get_settings", lambda: _Settings(max_pdf_size_mb=10.0)
)
r = _registry((_Fake("fast", "fast"), 20))
res = await r.process(b"%PDF-1.7", "application/pdf")
assert res.success is True
assert res.processor == "fast"
async def test_engine_rollback_uses_structured(monkeypatch):
monkeypatch.setattr(reg_mod, "get_settings", lambda: _Settings(engine="pymupdf"))
r = _registry((_Fake("fast", "fast"), 20), (_Fake("structured", "structured"), 10))