nv-ingest 2025.10.22.dev20251022__py3-none-any.whl → 2025.11.19.dev20251119__py3-none-any.whl

nv_ingest/framework/orchestration/process/termination.py

@@ -19,20 +19,45 @@ logger = logging.getLogger(__name__)
 
 
 def _safe_log(level: int, msg: str) -> None:
-    """Best-effort logging that won't crash during interpreter shutdown."""
+    """Best-effort logging that won't emit handler tracebacks on closed streams.
+
+    Temporarily disables logging.raiseExceptions to prevent the logging module
+    from printing "--- Logging error ---" to stderr if a handler's stream is
+    already closed (common during process teardown). Falls back to writing to
+    sys.__stderr__ if available.
+    """
     try:
-        logger.log(level, msg)
+        import logging as _logging
+
+        prev = getattr(_logging, "raiseExceptions", True)
+        # Suppress handler errors being printed to stderr
+        _logging.raiseExceptions = False
+
+        # If there are no handlers, skip and use stderr fallback
+        if logger.handlers:
+            logger.log(level, msg)
+            return
     except Exception:
+        # Intentionally ignore and try stderr fallback
+        pass
+    finally:
         try:
-            # Fallback to stderr if available
-            import sys
+            import logging as _logging  # re-import safe even if earlier failed
 
-            if hasattr(sys, "__stderr__") and sys.__stderr__:
-                sys.__stderr__.write(msg + "\n")
-                sys.__stderr__.flush()
+            _logging.raiseExceptions = prev  # type: ignore[name-defined]
         except Exception:
             pass
 
+    # Fallback to stderr if available
+    try:
+        import sys
+
+        if hasattr(sys, "__stderr__") and sys.__stderr__:
+            sys.__stderr__.write(msg + "\n")
+            sys.__stderr__.flush()
+    except Exception:
+        pass
+
 
 def kill_pipeline_process_group(process) -> None:
     """
@@ -74,7 +99,17 @@ def kill_pipeline_process_group(process) -> None:
 
     try:
         # Send graceful termination to the entire process group
-        os.killpg(os.getpgid(pid), signal.SIGTERM)
+        try:
+            pgid = os.getpgid(pid)
+        except Exception:
+            # Process already gone
+            _safe_log(logging.DEBUG, f"Process group for PID {pid} not found during SIGTERM phase")
+            return
+        try:
+            os.killpg(pgid, signal.SIGTERM)
+        except ProcessLookupError:
+            _safe_log(logging.DEBUG, f"Process group for PID {pid} no longer exists (SIGTERM)")
+            return
 
         # If we have a Process handle, give it a chance to exit cleanly
         if proc is not None and hasattr(proc, "join"):
@@ -95,7 +130,12 @@ def kill_pipeline_process_group(process) -> None:
         if still_alive:
             _safe_log(logging.WARNING, "Process group did not terminate gracefully, using SIGKILL")
             try:
-                os.killpg(os.getpgid(pid), signal.SIGKILL)
+                try:
+                    pgid2 = os.getpgid(pid)
+                except Exception:
+                    _safe_log(logging.DEBUG, f"Process group for PID {pid} vanished before SIGKILL")
+                    return
+                os.killpg(pgid2, signal.SIGKILL)
             finally:
                 if proc is not None and hasattr(proc, "join"):
                     try:

nv_ingest/framework/orchestration/ray/examples/pipeline_test_harness.py

@@ -152,11 +152,11 @@ if __name__ == "__main__":
     os.environ["OCR_MODEL_NAME"] = "paddle"
     os.environ["NEMORETRIEVER_PARSE_HTTP_ENDPOINT"] = "https://integrate.api.nvidia.com/v1/chat/completions"
     os.environ["VLM_CAPTION_ENDPOINT"] = "https://integrate.api.nvidia.com/v1/chat/completions"
-    os.environ["VLM_CAPTION_MODEL_NAME"] = "nvidia/llama-3.1-nemotron-nano-vl-8b-v1"
+    os.environ["VLM_CAPTION_MODEL_NAME"] = "nvidia/nemotron-nano-12b-v2-vl"
     logger.info("Environment variables set.")
 
     image_caption_endpoint_url = "https://integrate.api.nvidia.com/v1/chat/completions"
-    model_name = "nvidia/llama-3.1-nemotron-nano-vl-8b-v1"
+    model_name = "nvidia/nemotron-nano-12b-v2-vl"
     yolox_grpc, yolox_http, yolox_auth, yolox_protocol = get_nim_service("yolox")
     (
         yolox_table_structure_grpc,

nv_ingest/framework/orchestration/ray/stages/extractors/audio_extractor.py

@@ -5,7 +5,6 @@
 
 import logging
 from typing import Optional
-
 import ray
 
 from nv_ingest.framework.orchestration.ray.stages.meta.ray_actor_stage_base import RayActorStage
@@ -67,7 +66,6 @@ class AudioExtractorStage(RayActorStage):
         # Extract the DataFrame payload.
         df_ledger = control_message.payload()
         self._logger.debug("Extracted payload with %d rows.", len(df_ledger))
-
         # Remove the "audio_data_extract" task from the message to obtain task-specific configuration.
         task_config = remove_task_by_type(control_message, "extract")
         self._logger.debug("Extracted task config: %s", sanitize_for_logging(task_config))

nv_ingest/framework/orchestration/ray/stages/extractors/ocr_extractor.py

@@ -0,0 +1,71 @@
+# SPDX-FileCopyrightText: Copyright (c) 2024-25, NVIDIA CORPORATION & AFFILIATES.
+# All rights reserved.
+# SPDX-License-Identifier: Apache-2.0
+
+import logging
+import ray
+
+from nv_ingest.framework.orchestration.ray.stages.meta.ray_actor_stage_base import RayActorStage
+from nv_ingest.framework.util.flow_control import filter_by_task
+from nv_ingest_api.internal.extract.image.ocr_extractor import extract_text_data_from_image_internal
+from nv_ingest_api.internal.primitives.ingest_control_message import IngestControlMessage, remove_task_by_type
+from nv_ingest_api.internal.primitives.tracing.tagging import traceable, set_trace_timestamps_with_parent_context
+from nv_ingest_api.internal.schemas.extract.extract_ocr_schema import OCRExtractorSchema
+from nv_ingest_api.util.exception_handlers.decorators import nv_ingest_node_failure_try_except
+from typing import Optional
+
+from nv_ingest.framework.util.flow_control.udf_intercept import udf_intercept_hook
+
+logger = logging.getLogger(__name__)
+
+
+@ray.remote
+class OCRExtractorStage(RayActorStage):
+    """
+    A Ray actor stage that extracts text data from image content.
+
+    It expects an IngestControlMessage containing a DataFrame with image data. It then:
+      1. Removes the "text_data_extract" task from the message.
+      2. Calls the text extraction logic using a validated configuration.
+      3. Updates the message payload with the extracted text DataFrame.
+    """
+
+    def __init__(self, config: OCRExtractorSchema, stage_name: Optional[str] = None) -> None:
+        super().__init__(config, log_to_stdout=False, stage_name=stage_name)
+        try:
+            self.validated_config = config
+            self._logger.info("OCRExtractorStage configuration validated successfully.")
+        except Exception as e:
+            self._logger.exception(f"Error validating Text extractor config: {e}")
+            raise
+
+    @nv_ingest_node_failure_try_except()
+    @traceable()
+    @udf_intercept_hook()
+    @filter_by_task(required_tasks=["ocr_data_extract"])
+    def on_data(self, control_message: IngestControlMessage) -> IngestControlMessage:
+        # Extract DataFrame payload
+        df_ledger = control_message.payload()
+        if df_ledger.empty:
+            return control_message
+
+        # Remove the "text_data_extract" task from the message
+        task_config = remove_task_by_type(control_message, "ocr_data_extract")
+
+        execution_trace_log = {}
+        new_df, extraction_info = extract_text_data_from_image_internal(
+            df_extraction_ledger=df_ledger,
+            task_config=task_config,
+            extraction_config=self.validated_config,
+            execution_trace_log=execution_trace_log,
+        )
+
+        control_message.payload(new_df)
+        control_message.set_metadata("ocr_extraction_info", extraction_info)
+
+        do_trace_tagging = control_message.get_metadata("config::add_trace_tagging") is True
+        if do_trace_tagging and execution_trace_log:
+            parent_name = self.stage_name if self.stage_name else "ocr_extractor"
+            set_trace_timestamps_with_parent_context(control_message, execution_trace_log, parent_name, logger)
+
+        return control_message

nv_ingest/framework/orchestration/ray/stages/sources/message_broker_task_source.py

@@ -30,6 +30,7 @@ from nv_ingest_api.internal.schemas.meta.ingest_job_schema import validate_inges
 from nv_ingest_api.util.message_brokers.simple_message_broker.simple_client import SimpleClient
 from nv_ingest_api.util.service_clients.redis.redis_client import RedisClient
 from nv_ingest_api.util.logging.sanitize import sanitize_for_logging
+from nv_ingest_api.util.message_brokers.qos_scheduler import QosScheduler
 
 logger = logging.getLogger(__name__)
 
@@ -89,8 +90,10 @@ class MessageBrokerTaskSourceConfig(BaseModel):
 
     # Use the discriminated union for broker_client
     broker_client: Union[RedisClientConfig, SimpleClientConfig] = Field(..., discriminator="client_type")
-    task_queue: str = Field(..., description="The name of the queue to fetch tasks from.")
-    poll_interval: float = Field(default=0.1, gt=0, description="Polling interval in seconds.")
+    task_queue: str = Field(
+        ..., description="The base name of the queue to fetch tasks from. Derives sub-queues for fair scheduling."
+    )
+    poll_interval: float = Field(default=0.0, gt=0, description="Polling interval in seconds.")
 
 
 @ray.remote
@@ -134,7 +137,29 @@ class MessageBrokerTaskSourceStage(RayActorSourceStage):
         self._current_backoff_sleep: float = 0.0
         self._last_backoff_log_time: float = 0.0
 
-        self._logger.debug("MessageBrokerTaskSourceStage initialized. Task queue: %s", self.task_queue)
+        # Initialize QoS scheduler. Use a simple base-queue strategy for SimpleClient.
+        strategy = "simple" if isinstance(self.client, SimpleClient) else "lottery"
+        self.scheduler = QosScheduler(
+            self.task_queue,
+            num_prefetch_threads=6,  # one per category (no-op for simple strategy)
+            total_buffer_capacity=96,  # e.g., ~16 per thread
+            prefetch_poll_interval=0.002,  # faster polling for responsiveness
+            prefetch_non_immediate=True,  # enable prefetch for non-immediate categories
+            strategy=strategy,
+        )
+
+        self._logger.info(
+            "MessageBrokerTaskSourceStage initialized. Base task queue: %s | Derived queues: %s",
+            self.task_queue,
+            {
+                "immediate": f"{self.task_queue}_immediate",
+                "micro": f"{self.task_queue}_micro",
+                "small": f"{self.task_queue}_small",
+                "medium": f"{self.task_queue}_medium",
+                "large": f"{self.task_queue}_large",
+                "default": f"{self.task_queue}",
+            },
+        )
 
     # --- Private helper methods ---
     def _create_client(self):
@@ -265,14 +290,21 @@ class MessageBrokerTaskSourceStage(RayActorSourceStage):
 
         return control_message
 
-    def _fetch_message(self, timeout=100):
+    def _fetch_message(self, timeout=0):
         """
-        Fetch a message from the message broker.
+        Fetch a message from the message broker using fair scheduling across derived queues.
+        This is a non-blocking sweep across all queues for the current scheduling cycle. If no
+        message is found across any queue, return None so the caller can sleep briefly.
         """
         try:
-            job = self.client.fetch_message(self.task_queue, timeout)
+            # Use scheduler to fetch next. In simple strategy this will block up to poll_interval on base queue.
+            job = self.scheduler.fetch_next(self.client, timeout=self.config.poll_interval)
             if job is None:
-                self._logger.debug("No message received from '%s'", self.task_queue)
+                self._logger.debug(
+                    "No message received from derived queues for base "
+                    "'%s' (immediate, micro, small, medium, large, default)",
+                    self.task_queue,
+                )
                 # Do not treat normal empty polls as failures
                 self._fetch_failure_count = 0
                 self._current_backoff_sleep = 0.0
@@ -336,7 +368,8 @@ class MessageBrokerTaskSourceStage(RayActorSourceStage):
         Instead of reading from an input edge, fetch a message from the broker.
         """
         self._logger.debug("read_input: calling _fetch_message()")
-        job = self._fetch_message(timeout=100)
+        # Perform a non-blocking sweep across all queues for this cycle
+        job = self._fetch_message(timeout=0)
         if job is None:
             # Sleep for either the configured poll interval or the current backoff, whichever is larger
             sleep_time = max(self.config.poll_interval, getattr(self, "_current_backoff_sleep", 0.0))

nv_ingest/framework/util/service/impl/ingest/redis_ingest_service.py

@@ -218,12 +218,33 @@ class RedisIngestService(IngestServiceMeta):
             ttl_for_result: Optional[int] = (
                 self._result_data_ttl_seconds if self._fetch_mode == FetchMode.NON_DESTRUCTIVE else None
             )
+            # Determine target queue based on optional QoS hint
+            queue_hint = None
+            try:
+                routing_opts = job_spec.get("routing_options") or {}
+                tracing_opts = job_spec.get("tracing_options") or {}
+                queue_hint = routing_opts.get("queue_hint") or tracing_opts.get("queue_hint")
+            except Exception:
+                queue_hint = None
+            allowed = {"default", "immediate", "micro", "small", "medium", "large"}
+            if isinstance(queue_hint, str) and queue_hint in allowed:
+                if queue_hint == "default":
+                    channel_name = self._redis_task_queue
+                else:
+                    channel_name = f"{self._redis_task_queue}_{queue_hint}"
+            else:
+                channel_name = self._redis_task_queue
+            logger.debug(
+                f"Submitting job {trace_id} to queue '{channel_name}' (hint={queue_hint}) "
+                f"with result TTL: {ttl_for_result}"
+            )
+
             logger.debug(
                 f"Submitting job {trace_id} to queue '{self._redis_task_queue}' with result TTL: {ttl_for_result}"
             )
             await self._run_bounded_to_thread(
                 self._ingest_client.submit_message,
-                channel_name=self._redis_task_queue,
+                channel_name=channel_name,
                 message=job_spec_json,
                 ttl_seconds=ttl_for_result,
             )
@@ -436,12 +457,13 @@ class RedisIngestService(IngestServiceMeta):
         metadata_key = f"parent:{parent_job_id}:metadata"
 
         try:
-            # Store subjob IDs as a set
-            await self._run_bounded_to_thread(
-                self._ingest_client.get_client().sadd,
-                parent_key,
-                *subjob_ids,
-            )
+            # Store subjob IDs as a set (only if there are subjobs)
+            if subjob_ids:
+                await self._run_bounded_to_thread(
+                    self._ingest_client.get_client().sadd,
+                    parent_key,
+                    *subjob_ids,
+                )
 
             # Store metadata as hash (including original subjob ordering for deterministic fetches)
             metadata_to_store = dict(metadata)
@@ -500,21 +522,21 @@ class RedisIngestService(IngestServiceMeta):
         metadata_key = f"parent:{parent_job_id}:metadata"
 
         try:
-            # Check if this is a parent job
+            # Check if this is a parent job (check metadata_key since non-split PDFs may not have parent_key)
             exists = await self._run_bounded_to_thread(
                 self._ingest_client.get_client().exists,
-                parent_key,
+                metadata_key,  # Check metadata instead of parent_key for non-split PDF support
             )
 
             if not exists:
                 return None
 
-            # Get subjob IDs
+            # Get subjob IDs (may be empty for non-split PDFs)
             subjob_ids_bytes = await self._run_bounded_to_thread(
                 self._ingest_client.get_client().smembers,
                 parent_key,
             )
-            subjob_id_set = {id.decode("utf-8") for id in subjob_ids_bytes}
+            subjob_id_set = {id.decode("utf-8") for id in subjob_ids_bytes} if subjob_ids_bytes else set()
 
             # Get metadata
             metadata_dict = await self._run_bounded_to_thread(

nv_ingest/pipeline/default_libmode_pipeline_impl.py

@@ -318,8 +318,8 @@ stages:
     actor: "nv_ingest.framework.orchestration.ray.stages.transforms.image_caption:ImageCaptionTransformStage"
     config:
       api_key: $NGC_API_KEY|$NVIDIA_API_KEY
-      endpoint_url: $VLM_CAPTION_ENDPOINT|"https://integrate.api.nvidia.com/v1/chat/completions"
-      model_name: $VLM_CAPTION_MODEL_NAME|"nvidia/llama-3.1-nemotron-nano-vl-8b-v1"
+      endpoint_url: $VLM_CAPTION_ENDPOINT|"http://vlm:8000/v1/chat/completions"
+      model_name: $VLM_CAPTION_MODEL_NAME|"nvidia/nemotron-nano-12b-v2-vl"
       prompt: "Caption the content of this image:"
     replicas:
       min_replicas: 0

nv_ingest/pipeline/default_pipeline_impl.py

@@ -192,6 +192,27 @@ stages:
         strategy: "static"
         value: 1
 
+  - name: "ocr_extractor"
+    type: "stage"
+    phase: 1  # EXTRACTION
+    actor: "nv_ingest.framework.orchestration.ray.stages.extractors.ocr_extractor:OCRExtractorStage"
+    config:
+      endpoint_config:
+        ocr_endpoints: [
+          $OCR_GRPC_ENDPOINT|"ocr:8001",
+          $OCR_HTTP_ENDPOINT|"http://ocr:8000/v1/infer",
+        ]
+        ocr_infer_protocol: $OCR_INFER_PROTOCOL|grpc
+        auth_token: $NGC_API_KEY|$NVIDIA_API_KEY
+    replicas:
+      min_replicas: 0
+      max_replicas:
+        strategy: "static"
+        value: 4
+      static_replicas:
+        strategy: "static"
+        value: 3
+
   - name: "infographic_extractor"
     type: "stage"
     phase: 1  # EXTRACTION
@@ -317,7 +338,8 @@ stages:
     actor: "nv_ingest.framework.orchestration.ray.stages.transforms.image_caption:ImageCaptionTransformStage"
     config:
       api_key: $NGC_API_KEY|$NVIDIA_API_KEY
-      model_name: $VLM_CAPTION_MODEL_NAME|"nvidia/llama-3.1-nemotron-nano-vl-8b-v1"
+      model_name: $VLM_CAPTION_MODEL_NAME|"nvidia/nemotron-nano-12b-v2-vl"
+      endpoint_url: $VLM_CAPTION_ENDPOINT|"http://vlm:8000/v1/chat/completions"
       prompt: "Caption the content of this image:"
     replicas:
       min_replicas: 0
@@ -427,76 +449,79 @@ edges:
   # Intake
   - from: "source_stage"
     to: "metadata_injector"
-    queue_size: 32
+    queue_size: 4
 
   # Document Extractors
   - from: "metadata_injector"
     to: "pdf_extractor"
-    queue_size: 32
+    queue_size: 8
   - from: "pdf_extractor"
     to: "audio_extractor"
-    queue_size: 32
+    queue_size: 4
   - from: "audio_extractor"
     to: "docx_extractor"
-    queue_size: 32
+    queue_size: 4
   - from: "docx_extractor"
     to: "pptx_extractor"
-    queue_size: 32
+    queue_size: 4
   - from: "pptx_extractor"
     to: "image_extractor"
-    queue_size: 32
+    queue_size: 4
   - from: "image_extractor"
     to: "html_extractor"
-    queue_size: 32
+    queue_size: 4
   - from: "html_extractor"
     to: "infographic_extractor"
-    queue_size: 32
+    queue_size: 4
 
   # Primitive Extractors
   - from: "infographic_extractor"
     to: "table_extractor"
-    queue_size: 32
+    queue_size: 4
   - from: "table_extractor"
     to: "chart_extractor"
-    queue_size: 32
+    queue_size: 4
   - from: "chart_extractor"
+    to: "ocr_extractor"
+    queue_size: 8
+  - from: "ocr_extractor"
     to: "image_filter"
-    queue_size: 32
+    queue_size: 4
 
   # Primitive Mutators
   - from: "image_filter"
     to: "image_dedup"
-    queue_size: 32
+    queue_size: 4
   - from: "image_dedup"
     to: "text_splitter"
-    queue_size: 32
+    queue_size: 4
 
   # Primitive Transforms
   - from: "text_splitter"
     to: "image_caption"
-    queue_size: 32
+    queue_size: 4
   - from: "image_caption"
     to: "text_embedder"
-    queue_size: 32
+    queue_size: 4
   - from: "text_embedder"
     to: "image_storage"
-    queue_size: 32
+    queue_size: 4
 
   # Primitive Storage
   - from: "image_storage"
     to: "embedding_storage"
-    queue_size: 32
+    queue_size: 4
   - from: "embedding_storage"
     to: "broker_response"
-    queue_size: 32
+    queue_size: 4
 
   # Response and Telemetry
   - from: "broker_response"
     to: "otel_tracer"
-    queue_size: 32
+    queue_size: 4
   - from: "otel_tracer"
     to: "default_drain"
-    queue_size: 32
+    queue_size: 4
 
 # Pipeline Runtime Configuration
 pipeline:

{nv_ingest-2025.10.22.dev20251022.dist-info → nv_ingest-2025.11.19.dev20251119.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: nv-ingest
-Version: 2025.10.22.dev20251022
+Version: 2025.11.19.dev20251119
 Summary: Python module for multimodal document ingestion
 Author-email: Jeremy Dyer <jdyer@nvidia.com>
 License:                                  Apache License
@@ -226,7 +226,6 @@ Requires-Dist: isodate>=0.7.2
 Requires-Dist: langdetect>=1.0.9
 Requires-Dist: minio>=7.2.12
 Requires-Dist: librosa>=0.10.2
-Requires-Dist: openai>=1.82.0
 Requires-Dist: opentelemetry-api>=1.27.0
 Requires-Dist: opentelemetry-exporter-otlp>=1.27.0
 Requires-Dist: opentelemetry-sdk>=1.27.0

{nv_ingest-2025.10.22.dev20251022.dist-info → nv_ingest-2025.11.19.dev20251119.dist-info}/RECORD

@@ -7,27 +7,27 @@ nv_ingest/api/v1/__init__.py,sha256=wQSlVx3T14ZgQAt-EPzEczQusXVW0W8yynnUaFFGE3s,
 nv_ingest/api/v1/health.py,sha256=pV-RoVq5y0iBPp0qZoLzd1xKpd0JiHAi0UMyMj99LqU,4740
 nv_ingest/api/v1/ingest.py,sha256=LWk3LN4lBd3uO8h30EN42g3LHCVcO00avVd5ohVK7NI,19392
 nv_ingest/api/v1/metrics.py,sha256=ZGVRApYLnzc2f2C7wRgGd7deqiXan-jxfA-33a16clY,981
-nv_ingest/api/v2/README.md,sha256=tbQOcD_67YWedboAcDRlZJgjvVZZTW1-ZodcqP0iynk,7133
+nv_ingest/api/v2/README.md,sha256=VhpdjEmCyr3qIOhwqISFx9C5WezJFcxYc-NB9S98HMg,7562
 nv_ingest/api/v2/__init__.py,sha256=wQSlVx3T14ZgQAt-EPzEczQusXVW0W8yynnUaFFGE3s,143
-nv_ingest/api/v2/ingest.py,sha256=v5l1c1BdmgyPqMzRj8CezI3dR6HpKOuevfomT1v4RGc,37313
+nv_ingest/api/v2/ingest.py,sha256=ikbZE2eAjSnFmt5CcpTduY1t9DsUQBhnBQlsd3HaBww,53103
 nv_ingest/framework/__init__.py,sha256=wQSlVx3T14ZgQAt-EPzEczQusXVW0W8yynnUaFFGE3s,143
 nv_ingest/framework/orchestration/__init__.py,sha256=wQSlVx3T14ZgQAt-EPzEczQusXVW0W8yynnUaFFGE3s,143
 nv_ingest/framework/orchestration/execution/__init__.py,sha256=wQSlVx3T14ZgQAt-EPzEczQusXVW0W8yynnUaFFGE3s,143
 nv_ingest/framework/orchestration/execution/helpers.py,sha256=-F8SZh7ISWtzJz6X1O2LQ133t-17Jxi8lL-NHz4rwj0,2818
 nv_ingest/framework/orchestration/execution/options.py,sha256=Ms1t4591EIv4ZrMRdhsCYPgLnMVXJosG3MURCbPXUoA,3983
 nv_ingest/framework/orchestration/process/__init__.py,sha256=wQSlVx3T14ZgQAt-EPzEczQusXVW0W8yynnUaFFGE3s,143
-nv_ingest/framework/orchestration/process/dependent_services.py,sha256=ERf2M4O6pvbLDFrvayBHHL7M-FIwECeDEDTY3bi7MBg,2940
+nv_ingest/framework/orchestration/process/dependent_services.py,sha256=s0j_rsFtCKHFIuvOkBe9NEAkPNPhSYse_ApeHka8gyg,3032
 nv_ingest/framework/orchestration/process/execution.py,sha256=P1kzpYV23e4QYrKw9Td1TCZK3CK1ENVqqnI_axRCqBk,19814
 nv_ingest/framework/orchestration/process/lifecycle.py,sha256=L5NDwnzSMQPGjqJDC8jC75L1YqWey-dtK8N_HgBzb0E,8001
-nv_ingest/framework/orchestration/process/strategies.py,sha256=D7fdTPA7uuteoj6McA6hm1J5ArqoDdSZ7W6_ONDX7N0,7845
-nv_ingest/framework/orchestration/process/termination.py,sha256=_aI2ZzCasGfqwu0fcvufOlr1BGAay_Noxq5pAu67gv4,3593
+nv_ingest/framework/orchestration/process/strategies.py,sha256=Q1Q04PPseF775omeS0FoXfK187NiS_bbqTaaJRwzKn8,7972
+nv_ingest/framework/orchestration/process/termination.py,sha256=PAogFeW0FATFS6Mcp_UkZgq_SbWV18RtdZN-0NbComw,5042
 nv_ingest/framework/orchestration/ray/__init__.py,sha256=wQSlVx3T14ZgQAt-EPzEczQusXVW0W8yynnUaFFGE3s,143
 nv_ingest/framework/orchestration/ray/edges/__init__.py,sha256=wQSlVx3T14ZgQAt-EPzEczQusXVW0W8yynnUaFFGE3s,143
 nv_ingest/framework/orchestration/ray/edges/async_queue_edge.py,sha256=PQliU_kyGbO9o42njpb8FrDMLrbLqwZzmBNXifxyG5Y,2312
 nv_ingest/framework/orchestration/ray/edges/ray_queue_edge.py,sha256=VFii2yxJuikimOxie3edKq5JN06g78AF8bdHSHVX8p8,2677
 nv_ingest/framework/orchestration/ray/edges/threaded_queue_edge.py,sha256=N6NH4KgZJ60e_JkGRcSmfQtX37qtX4TMcavOR-n3heE,2549
 nv_ingest/framework/orchestration/ray/examples/__init__.py,sha256=wQSlVx3T14ZgQAt-EPzEczQusXVW0W8yynnUaFFGE3s,143
-nv_ingest/framework/orchestration/ray/examples/pipeline_test_harness.py,sha256=hnRLybIpVTj3mXkLW0ErWVn4vRsInjNZmA80JqDiQuw,16473
+nv_ingest/framework/orchestration/ray/examples/pipeline_test_harness.py,sha256=Bn4rjkO14BwvvUNG_HBCSVXetYk7DKqRRsYHJADWqjc,16455
 nv_ingest/framework/orchestration/ray/examples/task_source_harness.py,sha256=Yt7uxThg7s8WuMiaHLKC8r1XAG7QixegfkT-juE5oNw,1953
 nv_ingest/framework/orchestration/ray/examples/task_source_sink_harness.py,sha256=XkvsoIzH5ftXvAZ4ox7mxbx7ESVx6D8Xupcwbqgd52w,3277
 nv_ingest/framework/orchestration/ray/primitives/__init__.py,sha256=wQSlVx3T14ZgQAt-EPzEczQusXVW0W8yynnUaFFGE3s,143
@@ -38,12 +38,13 @@ nv_ingest/framework/orchestration/ray/primitives/ray_pipeline.py,sha256=t9lf6zTj
 nv_ingest/framework/orchestration/ray/primitives/ray_stat_collector.py,sha256=GGY6_i6_g5xTFzdo9Qmsu9i4knMTq6pJfgm-aaPEt_o,17226
 nv_ingest/framework/orchestration/ray/stages/__init__.py,sha256=wQSlVx3T14ZgQAt-EPzEczQusXVW0W8yynnUaFFGE3s,143
 nv_ingest/framework/orchestration/ray/stages/extractors/__init__.py,sha256=wQSlVx3T14ZgQAt-EPzEczQusXVW0W8yynnUaFFGE3s,143
-nv_ingest/framework/orchestration/ray/stages/extractors/audio_extractor.py,sha256=4SdgvzI9oJ_OK5oWGir9wXVIPV4Pont2EKv9mwcWMC0,3631
+nv_ingest/framework/orchestration/ray/stages/extractors/audio_extractor.py,sha256=UVp_kDmkaBlfO0Mbl_IxKq6imzLvs4-DKHgUHJIh3mo,3629
 nv_ingest/framework/orchestration/ray/stages/extractors/chart_extractor.py,sha256=rfaDx6PqRCguhSYkJI6iVmMMtAlJNxzKfUrLmw_fKqs,4381
 nv_ingest/framework/orchestration/ray/stages/extractors/docx_extractor.py,sha256=R4vshPcAUN2U6BIv8BCZQ862wLx8RJhCGXfpQ3K09Bs,3627
 nv_ingest/framework/orchestration/ray/stages/extractors/html_extractor.py,sha256=7JrZSVIrK4_wr2s7TOTss7pgTY2F9GPQ7Ze3F_WFlKU,3642
 nv_ingest/framework/orchestration/ray/stages/extractors/image_extractor.py,sha256=iY9fEfucfgCmO2ixX6qwn418J97nJz_FQGh7B6yziVo,3980
 nv_ingest/framework/orchestration/ray/stages/extractors/infographic_extractor.py,sha256=v5J7dnJBEaDfjoTz_N_yC3RAt6lwMLgLT28V-ahquLE,3261
+nv_ingest/framework/orchestration/ray/stages/extractors/ocr_extractor.py,sha256=pwVoA5-CF9GVWusoFZOMGBvSyW5udD9bdxVJXA_SghE,3188
 nv_ingest/framework/orchestration/ray/stages/extractors/pdf_extractor.py,sha256=QagIA99AsHLihjRbXm-2BphdoQGHwzOHlqLyz7oDOSk,4992
 nv_ingest/framework/orchestration/ray/stages/extractors/pptx_extractor.py,sha256=RMbbl7Cuj4BT-TcgUx_0k8R-DLdw-o3fHxcIBIgrWt4,3776
 nv_ingest/framework/orchestration/ray/stages/extractors/table_extractor.py,sha256=p71ktv6v5T-9npYpCbgbwW6-fS-65UWS7rCm8OWr2Bc,4170
@@ -61,7 +62,7 @@ nv_ingest/framework/orchestration/ray/stages/sinks/__init__.py,sha256=wQSlVx3T14
 nv_ingest/framework/orchestration/ray/stages/sinks/default_drain.py,sha256=_USW1Vq8G2Wn-QFdPfFQCrtKG46hHeJvkEGbBxdpbVM,1488
 nv_ingest/framework/orchestration/ray/stages/sinks/message_broker_task_sink.py,sha256=QcvMQXIJ7EWIxty76Mo5Xv38Oj6X2KuS8qXQlf7E1uA,11676
 nv_ingest/framework/orchestration/ray/stages/sources/__init__.py,sha256=wQSlVx3T14ZgQAt-EPzEczQusXVW0W8yynnUaFFGE3s,143
-nv_ingest/framework/orchestration/ray/stages/sources/message_broker_task_source.py,sha256=Qm9XtTNX2CcUAlZRw33BS3Ql0djcsMGp52FPA2zHu3Q,22340
+nv_ingest/framework/orchestration/ray/stages/sources/message_broker_task_source.py,sha256=LrqaWpWyuiAHlpXWKYSyHZJBFegGXfNlpCXrucbK5NM,24067
 nv_ingest/framework/orchestration/ray/stages/storage/__init__.py,sha256=wQSlVx3T14ZgQAt-EPzEczQusXVW0W8yynnUaFFGE3s,143
 nv_ingest/framework/orchestration/ray/stages/storage/image_storage.py,sha256=WZN_-3Li-izDaPtk8IMrtn2os1ckT3U8Rb2PsfOWrcI,4009
 nv_ingest/framework/orchestration/ray/stages/storage/store_embeddings.py,sha256=EUtwhSDf-qGLVEhWEInr1VaLsvpcHUSyzCmHQVai-Ps,3547
@@ -103,22 +104,22 @@ nv_ingest/framework/util/flow_control/udf_intercept.py,sha256=zQ9uuCcHLEd0P52Eiw
 nv_ingest/framework/util/service/__init__.py,sha256=wQSlVx3T14ZgQAt-EPzEczQusXVW0W8yynnUaFFGE3s,143
 nv_ingest/framework/util/service/impl/__init__.py,sha256=wQSlVx3T14ZgQAt-EPzEczQusXVW0W8yynnUaFFGE3s,143
 nv_ingest/framework/util/service/impl/ingest/__init__.py,sha256=wQSlVx3T14ZgQAt-EPzEczQusXVW0W8yynnUaFFGE3s,143
-nv_ingest/framework/util/service/impl/ingest/redis_ingest_service.py,sha256=OuGC3FFhkLQLR3x4s-tyxGguYYn8ORKr2xkzMy2br0g,22552
+nv_ingest/framework/util/service/impl/ingest/redis_ingest_service.py,sha256=59P-BMWnFY37GJm5w23-TMxgLhiZGZpJogC0gjDBaTA,23835
 nv_ingest/framework/util/service/meta/__init__.py,sha256=wQSlVx3T14ZgQAt-EPzEczQusXVW0W8yynnUaFFGE3s,143
 nv_ingest/framework/util/service/meta/ingest/__init__.py,sha256=wQSlVx3T14ZgQAt-EPzEczQusXVW0W8yynnUaFFGE3s,143
 nv_ingest/framework/util/service/meta/ingest/ingest_service_meta.py,sha256=QS3uNxWBl5dIcmIpJKNe8_TLcTUuN2vcKyHeAwa-eSo,1589
 nv_ingest/framework/util/telemetry/__init__.py,sha256=wQSlVx3T14ZgQAt-EPzEczQusXVW0W8yynnUaFFGE3s,143
 nv_ingest/framework/util/telemetry/global_stats.py,sha256=nq65pEEdiwjAfGiqsxG1CeQMC96O3CfQxsZuGFCY-ds,4554
 nv_ingest/pipeline/__init__.py,sha256=wQSlVx3T14ZgQAt-EPzEczQusXVW0W8yynnUaFFGE3s,143
-nv_ingest/pipeline/default_libmode_pipeline_impl.py,sha256=MiyKe8RS18PNYwEVvrASiHFpynR_BavOe0hhVnUdbEc,15618
-nv_ingest/pipeline/default_pipeline_impl.py,sha256=irVm_wmJW5a7a3xTJd18AFZfwLheERkhCty-0XZrIMY,15288
+nv_ingest/pipeline/default_libmode_pipeline_impl.py,sha256=yNJtjfHQyxtasGa1hQrvgX7UrPa7BAd0oog8EIN8Y_w,15592
+nv_ingest/pipeline/default_pipeline_impl.py,sha256=DhClC17lWUvtBIi2mCC4WkLWT0lxY-CFY0n6nriAxas,16017
 nv_ingest/pipeline/ingest_pipeline.py,sha256=wHAJhqAM2s8nbY-8itVogmSU-yVN4PZONGWcKnhzgfg,17794
 nv_ingest/pipeline/pipeline_schema.py,sha256=rLZZz2It2o2hVNWrZUJU8CarrqRei1fho3ZEMkkoBcg,17940
 nv_ingest/pipeline/config/__init__.py,sha256=wQSlVx3T14ZgQAt-EPzEczQusXVW0W8yynnUaFFGE3s,143
 nv_ingest/pipeline/config/loaders.py,sha256=75Yr9WYO7j7ghvKTnYLfZXQZEH3J3VEZo5J4TunC_Us,7590
 nv_ingest/pipeline/config/replica_resolver.py,sha256=3zjh8gmepEYORFZRM4inq7GoBW0YL3gzUDiixUugjzQ,8899
-nv_ingest-2025.10.22.dev20251022.dist-info/licenses/LICENSE,sha256=xx0jnfkXJvxRnG63LTGOxlggYnIysveWIZ6H3PNdCrQ,11357
-nv_ingest-2025.10.22.dev20251022.dist-info/METADATA,sha256=fBAiUkJijOoKO-QsdNYEpDF9X1ovQ2BBSBBhLP-Yykw,15122
-nv_ingest-2025.10.22.dev20251022.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
-nv_ingest-2025.10.22.dev20251022.dist-info/top_level.txt,sha256=sjb0ajIsgn3YgftSjZHlYO0HjYAIIhNuXG_AmywCvaU,10
-nv_ingest-2025.10.22.dev20251022.dist-info/RECORD,,
+nv_ingest-2025.11.19.dev20251119.dist-info/licenses/LICENSE,sha256=xx0jnfkXJvxRnG63LTGOxlggYnIysveWIZ6H3PNdCrQ,11357
+nv_ingest-2025.11.19.dev20251119.dist-info/METADATA,sha256=arJTf3Axy2qKAFDlP4lsKCftTw4vnJp3EECP6hmylYU,15092
+nv_ingest-2025.11.19.dev20251119.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
+nv_ingest-2025.11.19.dev20251119.dist-info/top_level.txt,sha256=sjb0ajIsgn3YgftSjZHlYO0HjYAIIhNuXG_AmywCvaU,10
+nv_ingest-2025.11.19.dev20251119.dist-info/RECORD,,