classifyre-cli 0.4.60__tar.gz → 0.4.62__tar.gz

{classifyre_cli-0.4.60 → classifyre_cli-0.4.62}/.turbo/turbo-build.log

@@ -1,3 +1,3 @@
 $ uv sync
-Resolved 300 packages in 341ms
+Resolved 300 packages in 177ms
 Checked 51 packages in 1ms

{classifyre_cli-0.4.60 → classifyre_cli-0.4.62}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: classifyre-cli
-Version: 0.4.60
+Version: 0.4.62
 Summary: Classifyre CLI — scan and classify unstructured data sources
 License: MIT
 Keywords: data,ingestion,metadata,pii,secrets,unstructured

{classifyre_cli-0.4.60 → classifyre_cli-0.4.62}/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@classifyre/cli",
-  "version": "0.4.60",
+  "version": "0.4.62",
   "private": true,
   "scripts": {
     "build": "uv sync",

{classifyre_cli-0.4.60 → classifyre_cli-0.4.62}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "classifyre-cli"
-version = "0.4.60"
+version = "0.4.62"
 description = "Classifyre CLI — scan and classify unstructured data sources"
 readme = "README.md"
 requires-python = ">=3.12"

{classifyre_cli-0.4.60 → classifyre_cli-0.4.62}/src/detectors/custom/runners/_llm.py

@@ -6,6 +6,8 @@ import base64
 import json
 import logging
 import os
+import random
+import time
 from datetime import UTC, datetime
 from typing import Any
 
@@ -43,6 +45,23 @@ _MAX_VISION_IMAGES = 20
 # worker for litellm's ~10-minute default. Overridable per deployment.
 _COMPLETION_TIMEOUT_SECONDS = float(os.environ.get("CLASSIFYRE_LLM_TIMEOUT_SECONDS", "90"))
 
+# Retry budget for transient provider failures (429 rate limits, 5xx, timeouts).
+# Exponential backoff with jitter, honouring Retry-After when the provider
+# sends one, so a saturated endpoint is backed off from instead of hammered.
+_MAX_COMPLETION_ATTEMPTS = max(1, int(os.environ.get("CLASSIFYRE_LLM_MAX_ATTEMPTS", "4")))
+_RETRY_BACKOFF_BASE_SECONDS = float(os.environ.get("CLASSIFYRE_LLM_RETRY_BASE_SECONDS", "2"))
+_RETRY_BACKOFF_MAX_SECONDS = 60.0
+_RETRYABLE_STATUS_CODES = {408, 429, 500, 502, 503, 504}
+
+
+class LLMCompletionError(RuntimeError):
+    """Provider call failed after retries.
+
+    Carries only a string message so it survives pickling across the detector
+    worker-pool process boundary; the pipeline records the detector's outcome
+    as ERROR instead of a silent empty result.
+    """
+
 
 class LLMRunner(BaseRunner):
     """AI detector — sends content to a configured LLM provider for classification + extraction."""
@@ -130,18 +149,8 @@ class LLMRunner(BaseRunner):
         *,
         vision_pages: int | None = None,
     ) -> list[DetectionResult]:
-        schema = self._schema
         try:
-            response = self._litellm.completion(
-                model=self._model_string(),
-                api_key=self._runtime.api_key,
-                api_base=self._runtime.base_url or None,
-                temperature=schema.temperature if schema.temperature is not None else 0.0,
-                max_tokens=self._max_tokens(),
-                messages=messages,
-                response_format={"type": "json_object"},
-                timeout=_COMPLETION_TIMEOUT_SECONDS,
-            )
+            response = self._complete_with_backoff(messages)
             raw = response.choices[0].message.content or "{}"
             parsed = self._parse_json(raw)
         except Exception as exc:
@@ -152,10 +161,70 @@ class LLMRunner(BaseRunner):
                 exc,
                 exc_info=True,
             )
-            return []
+            # Propagate instead of returning [] — an empty result here is
+            # indistinguishable from "genuinely found nothing" and would let
+            # the pipeline record this detector OK on an asset it never saw.
+            raise LLMCompletionError(
+                f"LLM provider call failed for detector '{self._detector_key}' "
+                f"(model={self._runtime.model}): {exc}"
+            ) from None
 
         return self._results_from_payload(snippet, parsed, vision_pages=vision_pages)
 
+    def _complete_with_backoff(self, messages: list[dict[str, Any]]) -> Any:
+        schema = self._schema
+        for attempt in range(1, _MAX_COMPLETION_ATTEMPTS + 1):
+            try:
+                return self._litellm.completion(
+                    model=self._model_string(),
+                    api_key=self._runtime.api_key,
+                    api_base=self._runtime.base_url or None,
+                    temperature=schema.temperature if schema.temperature is not None else 0.0,
+                    max_tokens=self._max_tokens(),
+                    messages=messages,
+                    response_format={"type": "json_object"},
+                    timeout=_COMPLETION_TIMEOUT_SECONDS,
+                )
+            except Exception as exc:
+                if attempt >= _MAX_COMPLETION_ATTEMPTS or not self._is_retryable(exc):
+                    raise
+                delay = min(
+                    _RETRY_BACKOFF_MAX_SECONDS,
+                    _RETRY_BACKOFF_BASE_SECONDS * (2 ** (attempt - 1)),
+                )
+                retry_after = _retry_after_seconds(exc)
+                if retry_after is not None:
+                    delay = min(_RETRY_BACKOFF_MAX_SECONDS, max(delay, retry_after))
+                delay *= 0.75 + random.random() * 0.5  # ±25% jitter
+                logger.warning(
+                    "llm provider transient failure (detector=%s, model=%s, "
+                    "attempt %d/%d), retrying in %.1fs: %s",
+                    self._detector_key,
+                    self._runtime.model,
+                    attempt,
+                    _MAX_COMPLETION_ATTEMPTS,
+                    delay,
+                    exc,
+                )
+                time.sleep(delay)
+        raise LLMCompletionError("unreachable")  # pragma: no cover
+
+    def _is_retryable(self, exc: Exception) -> bool:
+        retryable_types = tuple(
+            t
+            for t in (
+                getattr(self._litellm, "RateLimitError", None),
+                getattr(self._litellm, "Timeout", None),
+                getattr(self._litellm, "ServiceUnavailableError", None),
+                getattr(self._litellm, "InternalServerError", None),
+                getattr(self._litellm, "APIConnectionError", None),
+            )
+            if isinstance(t, type)
+        )
+        if retryable_types and isinstance(exc, retryable_types):
+            return True
+        return getattr(exc, "status_code", None) in _RETRYABLE_STATUS_CODES
+
     def _vision_enabled(self) -> bool:
         return bool(getattr(self._runtime, "supports_vision", False))
 
@@ -298,3 +367,22 @@ class LLMRunner(BaseRunner):
     @staticmethod
     def _coerce_fields(raw: Any) -> dict[str, Any]:
         return {str(k): v for k, v in raw.items()} if isinstance(raw, dict) else {}
+
+
+def _retry_after_seconds(exc: Exception) -> float | None:
+    """Best-effort extraction of a Retry-After hint from a provider exception."""
+    response = getattr(exc, "response", None)
+    headers = getattr(response, "headers", None)
+    if headers is None:
+        return None
+    try:
+        raw = headers.get("retry-after")
+    except Exception:
+        return None
+    if raw is None:
+        return None
+    try:
+        value = float(raw)
+    except (TypeError, ValueError):
+        return None
+    return value if value > 0 else None

{classifyre_cli-0.4.60 → classifyre_cli-0.4.62}/src/models/generated_input.py

@@ -2122,6 +2122,10 @@ class CoreInput(BaseModel):
     )
     sampling: SamplingConfig
     resources: ResourceOverrides | None = None
+    cleanup_removed_detector_findings: bool | None = Field(
+        True,
+        description='When enabled (default), findings produced by detectors that are no longer configured on this source (removed or disabled) are automatically resolved at the start of the next run, keeping the findings list in step with the current detector set.',
+    )
 
 
 class SlackInput(CoreInput):
@@ -2142,6 +2146,10 @@ class SlackInput(CoreInput):
     )
     sampling: SamplingConfig
     resources: ResourceOverrides | None = None
+    cleanup_removed_detector_findings: bool | None = Field(
+        True,
+        description='When enabled (default), findings produced by detectors that are no longer configured on this source (removed or disabled) are automatically resolved at the start of the next run, keeping the findings list in step with the current detector set.',
+    )
 
 
 class EmailInput(CoreInput):
@@ -2160,6 +2168,10 @@ class EmailInput(CoreInput):
     )
     sampling: SamplingConfig
     resources: ResourceOverrides | None = None
+    cleanup_removed_detector_findings: bool | None = Field(
+        True,
+        description='When enabled (default), findings produced by detectors that are no longer configured on this source (removed or disabled) are automatically resolved at the start of the next run, keeping the findings list in step with the current detector set.',
+    )
 
 
 class S3CompatibleStorageInput(CoreInput):
@@ -2178,6 +2190,10 @@ class S3CompatibleStorageInput(CoreInput):
     )
     sampling: SamplingConfig
     resources: ResourceOverrides | None = None
+    cleanup_removed_detector_findings: bool | None = Field(
+        True,
+        description='When enabled (default), findings produced by detectors that are no longer configured on this source (removed or disabled) are automatically resolved at the start of the next run, keeping the findings list in step with the current detector set.',
+    )
 
 
 class LocalFolderInput(CoreInput):
@@ -2196,6 +2212,10 @@ class LocalFolderInput(CoreInput):
     )
     sampling: SamplingConfig
     resources: ResourceOverrides | None = None
+    cleanup_removed_detector_findings: bool | None = Field(
+        True,
+        description='When enabled (default), findings produced by detectors that are no longer configured on this source (removed or disabled) are automatically resolved at the start of the next run, keeping the findings list in step with the current detector set.',
+    )
 
 
 class SandboxInput(CoreInput):
@@ -2214,6 +2234,10 @@ class SandboxInput(CoreInput):
     )
     sampling: SamplingConfig
     resources: ResourceOverrides | None = None
+    cleanup_removed_detector_findings: bool | None = Field(
+        True,
+        description='When enabled (default), findings produced by detectors that are no longer configured on this source (removed or disabled) are automatically resolved at the start of the next run, keeping the findings list in step with the current detector set.',
+    )
 
 
 class AzureBlobStorageInput(CoreInput):
@@ -2232,6 +2256,10 @@ class AzureBlobStorageInput(CoreInput):
     )
     sampling: SamplingConfig
     resources: ResourceOverrides | None = None
+    cleanup_removed_detector_findings: bool | None = Field(
+        True,
+        description='When enabled (default), findings produced by detectors that are no longer configured on this source (removed or disabled) are automatically resolved at the start of the next run, keeping the findings list in step with the current detector set.',
+    )
 
 
 class GoogleCloudStorageInput(CoreInput):
@@ -2250,6 +2278,10 @@ class GoogleCloudStorageInput(CoreInput):
     )
     sampling: SamplingConfig
     resources: ResourceOverrides | None = None
+    cleanup_removed_detector_findings: bool | None = Field(
+        True,
+        description='When enabled (default), findings produced by detectors that are no longer configured on this source (removed or disabled) are automatically resolved at the start of the next run, keeping the findings list in step with the current detector set.',
+    )
 
 
 class WordPressInput(CoreInput):
@@ -2268,6 +2300,10 @@ class WordPressInput(CoreInput):
     )
     sampling: SamplingConfig
     resources: ResourceOverrides | None = None
+    cleanup_removed_detector_findings: bool | None = Field(
+        True,
+        description='When enabled (default), findings produced by detectors that are no longer configured on this source (removed or disabled) are automatically resolved at the start of the next run, keeping the findings list in step with the current detector set.',
+    )
 
 
 class PostgreSQLInput(CoreInput):
@@ -2286,6 +2322,10 @@ class PostgreSQLInput(CoreInput):
     )
     sampling: SamplingConfig
     resources: ResourceOverrides | None = None
+    cleanup_removed_detector_findings: bool | None = Field(
+        True,
+        description='When enabled (default), findings produced by detectors that are no longer configured on this source (removed or disabled) are automatically resolved at the start of the next run, keeping the findings list in step with the current detector set.',
+    )
 
 
 class MySQLInput(CoreInput):
@@ -2304,6 +2344,10 @@ class MySQLInput(CoreInput):
     )
     sampling: SamplingConfig
     resources: ResourceOverrides | None = None
+    cleanup_removed_detector_findings: bool | None = Field(
+        True,
+        description='When enabled (default), findings produced by detectors that are no longer configured on this source (removed or disabled) are automatically resolved at the start of the next run, keeping the findings list in step with the current detector set.',
+    )
 
 
 class MSSQLInput(CoreInput):
@@ -2322,6 +2366,10 @@ class MSSQLInput(CoreInput):
     )
     sampling: SamplingConfig
     resources: ResourceOverrides | None = None
+    cleanup_removed_detector_findings: bool | None = Field(
+        True,
+        description='When enabled (default), findings produced by detectors that are no longer configured on this source (removed or disabled) are automatically resolved at the start of the next run, keeping the findings list in step with the current detector set.',
+    )
 
 
 class OracleInput(CoreInput):
@@ -2340,6 +2388,10 @@ class OracleInput(CoreInput):
     )
     sampling: SamplingConfig
     resources: ResourceOverrides | None = None
+    cleanup_removed_detector_findings: bool | None = Field(
+        True,
+        description='When enabled (default), findings produced by detectors that are no longer configured on this source (removed or disabled) are automatically resolved at the start of the next run, keeping the findings list in step with the current detector set.',
+    )
 
 
 class HiveInput(CoreInput):
@@ -2358,6 +2410,10 @@ class HiveInput(CoreInput):
     )
     sampling: SamplingConfig
     resources: ResourceOverrides | None = None
+    cleanup_removed_detector_findings: bool | None = Field(
+        True,
+        description='When enabled (default), findings produced by detectors that are no longer configured on this source (removed or disabled) are automatically resolved at the start of the next run, keeping the findings list in step with the current detector set.',
+    )
 
 
 class DatabricksInput(CoreInput):
@@ -2380,6 +2436,10 @@ class DatabricksInput(CoreInput):
     )
     sampling: SamplingConfig
     resources: ResourceOverrides | None = None
+    cleanup_removed_detector_findings: bool | None = Field(
+        True,
+        description='When enabled (default), findings produced by detectors that are no longer configured on this source (removed or disabled) are automatically resolved at the start of the next run, keeping the findings list in step with the current detector set.',
+    )
 
 
 class SnowflakeInput(CoreInput):
@@ -2408,6 +2468,10 @@ class SnowflakeInput(CoreInput):
     )
     sampling: SamplingConfig
     resources: ResourceOverrides | None = None
+    cleanup_removed_detector_findings: bool | None = Field(
+        True,
+        description='When enabled (default), findings produced by detectors that are no longer configured on this source (removed or disabled) are automatically resolved at the start of the next run, keeping the findings list in step with the current detector set.',
+    )
 
 
 class MongoDBInput(CoreInput):
@@ -2430,6 +2494,10 @@ class MongoDBInput(CoreInput):
     )
     sampling: SamplingConfig
     resources: ResourceOverrides | None = None
+    cleanup_removed_detector_findings: bool | None = Field(
+        True,
+        description='When enabled (default), findings produced by detectors that are no longer configured on this source (removed or disabled) are automatically resolved at the start of the next run, keeping the findings list in step with the current detector set.',
+    )
 
 
 class Neo4jRequired(BaseModel):
@@ -2556,6 +2624,10 @@ class Neo4jInput(CoreInput):
     )
     sampling: SamplingConfig
     resources: ResourceOverrides | None = None
+    cleanup_removed_detector_findings: bool | None = Field(
+        True,
+        description='When enabled (default), findings produced by detectors that are no longer configured on this source (removed or disabled) are automatically resolved at the start of the next run, keeping the findings list in step with the current detector set.',
+    )
 
 
 class PowerBIInput(CoreInput):
@@ -2578,6 +2650,10 @@ class PowerBIInput(CoreInput):
     )
     sampling: SamplingConfig
     resources: ResourceOverrides | None = None
+    cleanup_removed_detector_findings: bool | None = Field(
+        True,
+        description='When enabled (default), findings produced by detectors that are no longer configured on this source (removed or disabled) are automatically resolved at the start of the next run, keeping the findings list in step with the current detector set.',
+    )
 
 
 class TableauInput(CoreInput):
@@ -2600,6 +2676,10 @@ class TableauInput(CoreInput):
     )
     sampling: SamplingConfig
     resources: ResourceOverrides | None = None
+    cleanup_removed_detector_findings: bool | None = Field(
+        True,
+        description='When enabled (default), findings produced by detectors that are no longer configured on this source (removed or disabled) are automatically resolved at the start of the next run, keeping the findings list in step with the current detector set.',
+    )
 
 
 class ConfluenceRequired(BaseModel):
@@ -3006,6 +3086,10 @@ class ConfluenceInput(CoreInput):
     )
     sampling: SamplingConfig
     resources: ResourceOverrides | None = None
+    cleanup_removed_detector_findings: bool | None = Field(
+        True,
+        description='When enabled (default), findings produced by detectors that are no longer configured on this source (removed or disabled) are automatically resolved at the start of the next run, keeping the findings list in step with the current detector set.',
+    )
 
 
 class JiraInput(CoreInput):
@@ -3024,6 +3108,10 @@ class JiraInput(CoreInput):
     )
     sampling: SamplingConfig
     resources: ResourceOverrides | None = None
+    cleanup_removed_detector_findings: bool | None = Field(
+        True,
+        description='When enabled (default), findings produced by detectors that are no longer configured on this source (removed or disabled) are automatically resolved at the start of the next run, keeping the findings list in step with the current detector set.',
+    )
 
 
 class ServiceDeskInput(CoreInput):
@@ -3042,6 +3130,10 @@ class ServiceDeskInput(CoreInput):
     )
     sampling: SamplingConfig
     resources: ResourceOverrides | None = None
+    cleanup_removed_detector_findings: bool | None = Field(
+        True,
+        description='When enabled (default), findings produced by detectors that are no longer configured on this source (removed or disabled) are automatically resolved at the start of the next run, keeping the findings list in step with the current detector set.',
+    )
 
 
 class SQLiteRequired(BaseModel):
@@ -3098,6 +3190,10 @@ class SQLiteInput(CoreInput):
     )
     sampling: SamplingConfig
     resources: ResourceOverrides | None = None
+    cleanup_removed_detector_findings: bool | None = Field(
+        True,
+        description='When enabled (default), findings produced by detectors that are no longer configured on this source (removed or disabled) are automatically resolved at the start of the next run, keeping the findings list in step with the current detector set.',
+    )
 
 
 class NotionRequired(BaseModel):
@@ -3229,6 +3325,10 @@ class NotionInput(CoreInput):
     )
     sampling: SamplingConfig
     resources: ResourceOverrides | None = None
+    cleanup_removed_detector_findings: bool | None = Field(
+        True,
+        description='When enabled (default), findings produced by detectors that are no longer configured on this source (removed or disabled) are automatically resolved at the start of the next run, keeping the findings list in step with the current detector set.',
+    )
 
 
 class KafkaSecurityProtocol(StrEnum):
@@ -3424,6 +3524,10 @@ class KafkaInput(CoreInput):
     )
     sampling: SamplingConfig
     resources: ResourceOverrides | None = None
+    cleanup_removed_detector_findings: bool | None = Field(
+        True,
+        description='When enabled (default), findings produced by detectors that are no longer configured on this source (removed or disabled) are automatically resolved at the start of the next run, keeping the findings list in step with the current detector set.',
+    )
 
 
 class NoAuthentication2(BaseModel):
@@ -3554,6 +3658,10 @@ class ElasticsearchInput(CoreInput):
     )
     sampling: SamplingConfig
     resources: ResourceOverrides | None = None
+    cleanup_removed_detector_findings: bool | None = Field(
+        True,
+        description='When enabled (default), findings produced by detectors that are no longer configured on this source (removed or disabled) are automatically resolved at the start of the next run, keeping the findings list in step with the current detector set.',
+    )
 
 
 class OpenSearchOptional(BaseModel):
@@ -3584,6 +3692,10 @@ class OpenSearchInput(CoreInput):
     )
     sampling: SamplingConfig
     resources: ResourceOverrides | None = None
+    cleanup_removed_detector_findings: bool | None = Field(
+        True,
+        description='When enabled (default), findings produced by detectors that are no longer configured on this source (removed or disabled) are automatically resolved at the start of the next run, keeping the findings list in step with the current detector set.',
+    )
 
 
 class NoAuthentication4(BaseModel):
@@ -3672,6 +3784,10 @@ class MeilisearchInput(CoreInput):
     )
     sampling: SamplingConfig
     resources: ResourceOverrides | None = None
+    cleanup_removed_detector_findings: bool | None = Field(
+        True,
+        description='When enabled (default), findings produced by detectors that are no longer configured on this source (removed or disabled) are automatically resolved at the start of the next run, keeping the findings list in step with the current detector set.',
+    )
 
 
 class LakehouseStorageConnection(BaseModel):
@@ -3908,6 +4024,10 @@ class Microsoft365Input(CoreInput):
     )
     sampling: SamplingConfig
     resources: ResourceOverrides | None = None
+    cleanup_removed_detector_findings: bool | None = Field(
+        True,
+        description='When enabled (default), findings produced by detectors that are no longer configured on this source (removed or disabled) are automatically resolved at the start of the next run, keeping the findings list in step with the current detector set.',
+    )
 
 
 class GoogleWorkspaceRequiredServiceAccount(BaseModel):
@@ -4048,6 +4168,10 @@ class GoogleWorkspaceInput(CoreInput):
     )
     sampling: SamplingConfig
     resources: ResourceOverrides | None = None
+    cleanup_removed_detector_findings: bool | None = Field(
+        True,
+        description='When enabled (default), findings produced by detectors that are no longer configured on this source (removed or disabled) are automatically resolved at the start of the next run, keeping the findings list in step with the current detector set.',
+    )
 
 
 class YouTubeInput(CoreInput):
@@ -4066,6 +4190,10 @@ class YouTubeInput(CoreInput):
     )
     sampling: SamplingConfig
     resources: ResourceOverrides | None = None
+    cleanup_removed_detector_findings: bool | None = Field(
+        True,
+        description='When enabled (default), findings produced by detectors that are no longer configured on this source (removed or disabled) are automatically resolved at the start of the next run, keeping the findings list in step with the current detector set.',
+    )
 
 
 class DeltaLakeOptional(BaseModel):
@@ -4092,6 +4220,10 @@ class DeltaLakeInput(CoreInput):
     )
     sampling: SamplingConfig
     resources: ResourceOverrides | None = None
+    cleanup_removed_detector_findings: bool | None = Field(
+        True,
+        description='When enabled (default), findings produced by detectors that are no longer configured on this source (removed or disabled) are automatically resolved at the start of the next run, keeping the findings list in step with the current detector set.',
+    )
 
 
 class IcebergOptional(BaseModel):
@@ -4118,6 +4250,10 @@ class IcebergInput(CoreInput):
     )
     sampling: SamplingConfig
     resources: ResourceOverrides | None = None
+    cleanup_removed_detector_findings: bool | None = Field(
+        True,
+        description='When enabled (default), findings produced by detectors that are no longer configured on this source (removed or disabled) are automatically resolved at the start of the next run, keeping the findings list in step with the current detector set.',
+    )
 
 
 class SourceInput(

{classifyre_cli-0.4.60 → classifyre_cli-0.4.62}/src/pipeline/detector_pipeline.py

@@ -150,6 +150,15 @@ class DetectorPipeline:
                 "application/x.asset-links",
             )
         ]
+        # A detector can qualify for both the text and the binary bucket — a
+        # vision-enabled LLM detector supports text plus image/pdf types — and
+        # would then run twice per asset (2x cost, 2x provider rate-limit
+        # pressure). Keep both passes only for IMAGE assets, where OCR text and
+        # pixels are complementary evidence; everywhere else (e.g. extensionless
+        # text files typed BINARY) the raw-bytes pass duplicates the text pass.
+        if text_detectors and binary_detectors and asset.asset_type != OutputAssetType.IMAGE:
+            text_ids = {id(d) for d in text_detectors}
+            binary_detectors = [d for d in binary_detectors if id(d) not in text_ids]
         # Any asset we resolved a text content type for is expected to yield
         # text, so warn whenever it yields none. This previously covered only
         # TXT/TABLE/URL — excluding exactly the types whose text is *derived*

{classifyre_cli-0.4.60 → classifyre_cli-0.4.62}/tests/detectors/custom/test_llm_runner.py

@@ -11,7 +11,7 @@ from unittest.mock import MagicMock
 
 import pytest
 
-from src.detectors.custom.runners._llm import LLMRunner
+from src.detectors.custom.runners._llm import LLMCompletionError, LLMRunner
 from src.models.generated_detectors import (
     LLMLabelDefinition,
     LLMOutputField,
@@ -161,6 +161,63 @@ def test_malformed_json_returns_no_findings() -> None:
     assert runner.detect(TEXT, "text/plain") == []
 
 
+# ── Provider-failure handling (BUG A) ─────────────────────────────────────────
+
+
+class _RateLimitError(Exception):
+    status_code = 429
+
+
+class _AuthError(Exception):
+    status_code = 401
+
+
+def test_provider_error_raises_instead_of_empty_result(monkeypatch) -> None:
+    # A dead provider must surface as an error the pipeline records, never as
+    # a silent zero-findings result indistinguishable from a clean scan.
+    monkeypatch.setattr("src.detectors.custom.runners._llm.time.sleep", lambda _s: None)
+    completion = MagicMock(side_effect=_RateLimitError("rate limited"))
+    runner = _runner(_schema(), completion)
+    with pytest.raises(LLMCompletionError, match="sentiment"):
+        runner.detect(TEXT, "text/plain")
+
+
+def test_transient_provider_error_retries_then_succeeds(monkeypatch) -> None:
+    sleeps: list[float] = []
+    monkeypatch.setattr("src.detectors.custom.runners._llm.time.sleep", sleeps.append)
+    good = _mock_completion({"labels": [{"name": "bad", "confidence": 0.9}]})
+    completion = MagicMock(
+        side_effect=[_RateLimitError("429"), _RateLimitError("503"), good.return_value]
+    )
+    runner = _runner(_schema(), completion)
+
+    results = runner.detect(TEXT, "text/plain")
+
+    assert [r.finding_type for r in results] == ["bad"]
+    assert completion.call_count == 3
+    assert len(sleeps) == 2
+    assert sleeps[1] > sleeps[0]  # exponential backoff
+
+
+def test_non_retryable_provider_error_fails_fast(monkeypatch) -> None:
+    monkeypatch.setattr("src.detectors.custom.runners._llm.time.sleep", lambda _s: None)
+    completion = MagicMock(side_effect=_AuthError("bad key"))
+    runner = _runner(_schema(), completion)
+    with pytest.raises(LLMCompletionError):
+        runner.detect(TEXT, "text/plain")
+    assert completion.call_count == 1
+
+
+def test_completion_error_is_picklable() -> None:
+    # The error crosses the detector worker-pool process boundary.
+    import pickle
+
+    err = LLMCompletionError("LLM provider call failed for detector 'x': 429")
+    restored = pickle.loads(pickle.dumps(err))
+    assert isinstance(restored, LLMCompletionError)
+    assert "429" in str(restored)
+
+
 # ── Vision / file input ───────────────────────────────────────────────────────
 
 

{classifyre_cli-0.4.60 → classifyre_cli-0.4.62}/tests/pipeline/test_detector_pipeline.py

@@ -554,6 +554,44 @@ async def test_pipeline_runs_mixed_detector_on_ocr_text_and_binary_payloads() ->
     assert mixed_detector.seen_content_types == ["text/plain", "image/png"]
 
 
+@pytest.mark.asyncio
+async def test_pipeline_runs_mixed_detector_once_on_binary_typed_text_assets() -> None:
+    # A detector supporting both text and binary types (e.g. a vision LLM
+    # detector) must not run twice on BINARY-typed assets whose text is the
+    # primary evidence — the raw-bytes pass would duplicate the text pass.
+    file_bytes = b"plain text stored without an extension"
+    source = OcrBinarySource(
+        {"type": "DUMMY"},
+        content="",
+        binary_data=file_bytes,
+        binary_mime="application/octet-stream",
+        ocr_pages=["extracted text"],
+    )
+    mixed_detector = RecordingDetector(["text/plain", "image/png"])
+
+    pipeline = DetectorPipeline(
+        detectors=[mixed_detector],
+        source=source,
+        runner_id="runner-binary-mixed",
+    )
+
+    now = datetime.now(UTC)
+    binary_asset = SingleAssetScanResults(
+        hash="bin-1",
+        checksum="checksum",
+        name="asset-bin-1",
+        external_url="urn:test/bin-1",
+        links=[],
+        asset_type=AssetType.BINARY,
+        created_at=now,
+        updated_at=now,
+    )
+    [_asset] = await pipeline.process([binary_asset])
+
+    assert mixed_detector.seen == ["extracted text"]
+    assert mixed_detector.seen_content_types == ["text/plain"]
+
+
 @pytest.mark.asyncio
 async def test_pipeline_resolves_effective_binary_mime_from_bytes_and_filename() -> None:
     image_bytes = b"\xff\xd8\xffjpeg-data"