How to achieve overlapping/parallel Smart Recording sessions in DeepStream without I/O blocking or Cache conflicts?

Accelerated Computing Intelligent Video Analytics DeepStream SDK
1

DeepStream 9.0 NVIDIA GPU Driver 595.80 RTX 3050

build in docker.

The Issue: Based on my observation, nvurisrcbin utilizes a single shared circular buffer (Cache) for the Smart Record module. When I attempt to trigger multiple start_recording calls, the system enters a “BLOCKING MODE,” which causes significant I/O latency and often leads to the Could not read from resource error, effectively dropping frames for the second recording task.

Specific Concerns:

  1. Cache Overwrite: In a high-concurrency scenario, if the smart-rec-cache size is limited, the time window for the second video (Video B) might be overwritten by the time I/O operations for the first video (Video A) are finalized.

  2. Blocking I/O: The current architecture seems to force a synchronous block on the pipeline when writing to the physical disk, preventing parallel recording sessions for the same source.

Questions:

  • Is there a recommended architectural pattern in DeepStream to handle multiple overlapping recording sessions from a single RTSP source without causing BLOCKING MODE or buffer contention?

  • Should I be implementing a custom “Tee” architecture where I split the stream into multiple recording-only pipelines, or is there a way to offload the raw buffer handling to a custom memory pool/callback before it triggers the I/O block?

  • Does the Smart Record module support concurrent Session IDs for the same source, or is this fundamentally a single-session-at-a-time limitation of the underlying sr_context?

I am looking for a scalable approach for traffic law enforcement applications where event detection occurs in rapid succession. Any insights or architectural best practices would be greatly appreciated.

3

The nvurisrcbin is open source in /opt/nvidia/deepstream/deepstream/sources/gst-plugins/gst-nvurisrcbin, the smart recording is implemented by the smart recording APIs. The smart recording is also open source /opt/nvidia/deepstream/deepstream/sources/libs/gst-nvdssr. There is only one smart recording context in nvurisrcbin, so there is only one smart recording session supported.

The so-called overlay recording means multiple smart recording sessions, you may implement it by yourself.

The other questions can be replied by the source codes.

4

You may also try the DeepStream Code Agent which provide the DeepStream skills. DeepStream/skills at main · NVIDIA/DeepStream, you can use Cursor or Claude CLI to generate sample app for overlapped smart recording.

5

There is no update from you for a period, assuming this is not an issue anymore. Hence we are closing this topic. If need further support, please open a new one. Thanks.