Memory footprint reduction and pdfium isolation

docker-compose.yaml

@@ -191,6 +191,7 @@ services:
       - sys_nice
     environment:
       - CUDA_VISIBLE_DEVICES=-1
+      - DISABLE_FAST_API_ACCESS_LOGGING=true
       - EMBEDDING_NIM_MODEL_NAME=${EMBEDDING_NIM_MODEL_NAME:-nvidia/llama-3.2-nv-embedqa-1b-v2}
       - INGEST_LOG_LEVEL=DEFAULT
       # Message client for development
@@ -210,27 +211,28 @@ services:
       - NVIDIA_BUILD_API_KEY=${NVIDIA_BUILD_API_KEY:-${NGC_API_KEY:-ngcapikey}}
       - OTEL_EXPORTER_OTLP_ENDPOINT=otel-collector:4317
       # Self-hosted paddle endpoints.
-      - PADDLE_GRPC_ENDPOINT=paddle:8001
+      #- PADDLE_GRPC_ENDPOINT=paddle:8001
       - PADDLE_HTTP_ENDPOINT=http://paddle:8000/v1/infer
-      - PADDLE_INFER_PROTOCOL=grpc
+      - PADDLE_INFER_PROTOCOL=http
       # build.nvidia.com hosted paddle endpoints.
       #- PADDLE_HTTP_ENDPOINT=https://ai.api.nvidia.com/v1/cv/baidu/paddleocr
       #- PADDLE_INFER_PROTOCOL=http
       - READY_CHECK_ALL_COMPONENTS=True
       - REDIS_MORPHEUS_TASK_QUEUE=morpheus_task_queue
       # Self-hosted redis endpoints.
-      - YOLOX_GRPC_ENDPOINT=yolox:8001
+      #- YOLOX_GRPC_ENDPOINT=yolox:8001
       - YOLOX_HTTP_ENDPOINT=http://yolox:8000/v1/infer
-      - YOLOX_INFER_PROTOCOL=grpc
+      - YOLOX_INFER_PROTOCOL=http
       # build.nvidia.com hosted yolox endpoints.
       #- YOLOX_HTTP_ENDPOINT=https://ai.api.nvidia.com/v1/cv/nvidia/nv-yolox-page-elements-v1
       #- YOLOX_INFER_PROTOCOL=http
-      - YOLOX_GRAPHIC_ELEMENTS_GRPC_ENDPOINT=yolox-graphic-elements:8001
+      #- YOLOX_GRAPHIC_ELEMENTS_GRPC_ENDPOINT=yolox-graphic-elements:8001
       - YOLOX_GRAPHIC_ELEMENTS_HTTP_ENDPOINT=http://yolox-graphic-elements:8000/v1/infer
-      - YOLOX_GRAPHIC_ELEMENTS_INFER_PROTOCOL=grpc
-      - YOLOX_TABLE_STRUCTURE_GRPC_ENDPOINT=yolox-table-structure:8001
+      - YOLOX_GRAPHIC_ELEMENTS_INFER_PROTOCOL=http
+      #- YOLOX_TABLE_STRUCTURE_GRPC_ENDPOINT=yolox-table-structure:8001
       - YOLOX_TABLE_STRUCTURE_HTTP_ENDPOINT=http://yolox-table-structure:8000/v1/infer
-      - YOLOX_TABLE_STRUCTURE_INFER_PROTOCOL=grpc
+      - YOLOX_TABLE_STRUCTURE_INFER_PROTOCOL=http
+      - VLM_CAPTION_ENDPOINT=https://ai.api.nvidia.com/v1/gr/meta/llama-3.2-11b-vision-instruct/chat/completions
       - VLM_CAPTION_ENDPOINT=http://vlm:8000/v1/chat/completions
       - VLM_CAPTION_MODEL_NAME=meta/llama-3.2-11b-vision-instruct
     healthcheck:

docker/scripts/entrypoint.sh

@@ -27,14 +27,29 @@ SRC_FILE="/opt/docker/bin/entrypoint_source"
 
 # Check if user supplied a command
 if [ "$#" -gt 0 ]; then
-    # If a command is provided, run it
+    # If a command is provided, run it.
     exec "$@"
 else
-    # If no command is provided, run the default startup launch
+    # If no command is provided, run the default startup launch.
     if [ "${MESSAGE_CLIENT_TYPE}" != "simple" ]; then
-      # Start uvicorn if MESSAGE_CLIENT_TYPE is not 'simple'
-      uvicorn nv_ingest.main:app --workers 32 --host 0.0.0.0 --port 7670 &
+        # Determine the log level for uvicorn.
+        log_level=$(echo "${INGEST_LOG_LEVEL:-default}" | tr '[:upper:]' '[:lower:]')
+        if [ "$log_level" = "default" ]; then
+            log_level="info"
+        fi
+
+        # Build the uvicorn command with the specified log level.
+        uvicorn_cmd="uvicorn nv_ingest.main:app --workers 32 --host 0.0.0.0 --port 7670 --log-level ${log_level}"
+
+        # If DISABLE_FAST_API_ACCESS_LOGGING is true, disable access logs.
+        if [ "${DISABLE_FAST_API_ACCESS_LOGGING}" == "true" ]; then
+            uvicorn_cmd="${uvicorn_cmd} --no-access-log"
+        fi
+
+        # Start uvicorn in the background.
+        $uvicorn_cmd &
     fi
 
+    # Start the microservice entrypoint.
     python /workspace/microservice_entrypoint.py
 fi