Pipeline failure in some multi-GPU systems

[whbruce]

Pipelines can sometimes fail in multi GPU systems due to video decode incompatibility. Pipeline will return an error when starting and service log will contain an entry like this

{"levelname": "INFO", "asctime": "2021-11-30 22:48:40,620", "message": "Creating Instance of Pipeline object_detection/person_vehicle_bike", "module": "pipeline_manager"}
{"levelname": "ERROR", "asctime": "2021-11-30 22:48:47,369", "message": "Error on Pipeline 1: gst-stream-error-quark: No valid frames decoded before end of stream (7): ../gst-libs/gst/video/gstvideodecoder.c(1246): gst_video_decoder_sink_event_default (): /GstPipeline:pipeline11/GstURIDecodeBin:source/GstDecodeBin:decodebin2/GstVaapiDecodeBin:vaapidecodebin2/GstVaapiDecode:vaapidecode2:\nno valid frames found", "module": "gstreamer_pipeline"}

Workaround is to disable GPU decode by clearing environment variable LIBVA_DRIVER_NAME
e.g. if using run script to start service:

./run.sh -v /tmp:/tmp -e LIBVA_DRIVER_NAME=

[whbruce]

Fixed in 2022.2.0 (v1.0). See https://github.com/dlstreamer/pipeline-server/blob/main/docs/running_pipeline_server.md#gpu.