View and Interpret Detection Logs (JSONL)#
Every pipeline can write its detection results as newline-delimited JSON
(json-lines) to a file inside the container — a durable, machine-readable
record of what the app detected, independent of the RTSP video stream.
Prerequisites#
A pipeline started by hand with a
destination.metadatablock. The camera-file manager (make start-rtsp) does not add one — its pipelines publish no per-frame log. Start the pipeline through the REST API as shown in Manage Pipelines — Start a pipeline.
How it’s configured#
Add this to the start request’s destination:
"destination": {
"metadata": {
"type": "file",
"path": "/tmp/results.jsonl",
"format": "json-lines"
},
"frame": {"type": "rtsp", "path": "uav-my-camera", "encoding": "h264"}
}
Several pipelines may write to the same path (lines interleave) or each
to its own, e.g., /tmp/results_front.jsonl. The file lives inside the
container; copy it out with docker cp as shown below.
Steps#
1. Tail the log live#
docker exec uav-va-dlstreamer-pipeline-server tail -f /tmp/results.jsonl
2. Pretty-print and filter with jq#
Each line is one frame’s metadata: a timestamp, resolution, and (if any
objects were detected) an objects array with bounding boxes, class labels,
and confidence scores.
# Pretty-print the most recent line
docker exec uav-va-dlstreamer-pipeline-server tail -n 1 /tmp/results.jsonl | jq .
# Show only frames where at least one object was detected
docker exec uav-va-dlstreamer-pipeline-server tail -f /tmp/results.jsonl \
| jq -c 'select(.objects | length > 0) | {ts: .timestamp, objects: [.objects[].detection.label]}'
Note
jq runs inside the container, which already has it available. To run it
from the host instead, see
Troubleshooting — jq: command not found.
3. Copy the log out for offline analysis#
docker cp uav-va-dlstreamer-pipeline-server:/tmp/results.jsonl ./results.jsonl
4. Correlate a log entry with a pipeline instance#
The metadata file itself does not include the pipeline instance_id. To
confirm which running pipeline is currently writing to a given path, check
GET /pipelines/status (see
Manage Pipelines — Check status)
and match on the destination.metadata.path you started it with.
Verify#
The file grows while a pipeline is
RUNNING(ls -la /tmp/results.jsonlinside the container shows an increasing size / mtime).jqparses every line without error (jq . < results.jsonl > /dev/nullreports no parse errors).
Troubleshooting#
A missing or empty log file is covered in
Troubleshooting — Detection log (JSONL) is missing or empty;
jq on the host in jq: command not found.
The JSONL carries no pipeline or camera name (the overlay label is drawn on
the video only) — give each pipeline its own path to tell them apart.