vdms-dataprep-user

name

vdms-dataprep-user

description

Deploy and consume the VDMS DataPrep video-ingestion stack (dataprep + VDMS vector DB + MinIO) — bring it up with setup.sh + docker compose (from a repo clone, or by fetching those same files from GitHub when no clone exists) using the prebuilt intel/vdms-dataprep image, then upload/ingest MP4s, add text-summary embeddings, and list/download/delete videos through the REST API at http://localhost:6007/v1/dataprep. Ingestion only: it does not answer search queries. Not for modifying the service's source — that is vdms-dataprep-dev.

VDMS DataPrep — User

Run the ingestion stack and feed it videos. Run commands yourself and relay output. API base: http://localhost:6007/v1/dataprep. Scope note: this service ingests (video → embeddings → VDMS); querying/searching those embeddings is a different component (e.g. a video-search app reading the same VDMS).

When to Use

• Bring up the dataprep + VDMS + MinIO stack and confirm health
• Upload/ingest MP4s (direct upload or from MinIO) with embeddings
• Add text-summary embeddings for a video time range
• List, download, or delete ingested videos
• Diagnose 413 uploads, dimension mismatches, or MinIO credential errors

Example Prompts

Sample Problem-solving scenarios this skill handles end-to-end:

Example	Problem it solves
manufacturing-inspection-archive.md	Archive production-line clips with frame + object-crop metadata
object-aware-video-catalog.md	Catalog clips by detected objects with confidence thresholds and tags
edge-video-preprocessing-box.md	Preprocess camera MP4s near the source (MinIO + VDMS) before central sync

Docs & deploy files — with or without a clone

All paths below are relative to microservices/visual-data-preparation-for-retrieval/vdms/ in the edge-ai-libraries repo. No clone? Fetch any of them from GitHub raw:

https://raw.githubusercontent.com/open-edge-platform/edge-ai-libraries/main/microservices/visual-data-preparation-for-retrieval/vdms/<path>

Load these existing docs only when needed:

Resource	Load when…
docs/user-guide/api-reference.md + docs/user-guide/api-docs/openapi.yaml	building requests beyond a simple upload (minio ingest, summaries, delete/download, telemetry)
docs/user-guide/get-started.md	env-var tables, detection/ROI tuning, more curl examples, troubleshooting
docs/user-guide/Overview.md + docs/user-guide/overview-architecture.md	how the pipeline works (frames → detection → embeddings → VDMS)
setup.sh, docker/compose.yaml	the deploy artifacts used below

1. Context routing — repo clone or standalone?

[ -f setup.sh ] && grep -q 'name = "vdms-dataprep"' pyproject.toml 2>/dev/null \
  && echo REPO || echo STANDALONE

• REPO → run Step 2 from the microservice root.
• STANDALONE → fetch the two deploy files, then the exact same Step 2:

  RAW=https://raw.githubusercontent.com/open-edge-platform/edge-ai-libraries/main/microservices/visual-data-preparation-for-retrieval/vdms
  mkdir -p vdms-dataprep/docker && cd vdms-dataprep
  curl -fsSL $RAW/setup.sh -o setup.sh
  curl -fsSL $RAW/docker/compose.yaml -o docker/compose.yaml
  

• Already running (curl -sf http://localhost:6007/v1/dataprep/health) → Step 3.

2. Bring-up (identical in both contexts)

Credentials are never committed — export strong values in-shell and retain them securely if clients must keep using the same credentials across restarts. setup.sh must be sourced; --nosetup exports env without touching containers, REGISTRY_URL=intel selects the prebuilt image, and --no-build prevents a source build (building is the -dev skill's job). Run in the background — image pulls plus embedding/YOLOX model downloads take a while:

bash -c 'export MINIO_ROOT_USER="<existing-or-new-user>" MINIO_ROOT_PASSWORD="<existing-or-new-strong-password>" \
  EMBEDDING_MODEL_NAME="CLIP/clip-vit-b-32" REGISTRY_URL=intel TAG=latest \
  && source ./setup.sh --nosetup \
  && docker compose -f docker/compose.yaml up -d --no-build'

setup.sh fixes INDEX_NAME=video-rag, which compose passes as DB_COLLECTION; VS_INDEX_NAME is unused. Then wait for API readiness:

until curl -sf http://localhost:6007/v1/dataprep/health \
  | python3 -c 'import json,sys; d=json.load(sys.stdin); raise SystemExit(0 if d.get("embedding_mode") != "sdk" or d.get("sdk_client_status") == "preloaded" else 1)'
do
  sleep 10
done

Health shows embedding_mode (sdk default) and, in SDK mode, require sdk_client_status=preloaded. This endpoint does not probe MinIO or VDMS, so also check the Compose services and MinIO separately:

docker compose -f docker/compose.yaml ps
curl -sf http://localhost:6010/minio/health/live
timeout 2 bash -c '</dev/tcp/localhost/6020'

Teardown later with docker compose -f docker/compose.yaml down.

3. Ingest a video

Upload an MP4 — it is stored in MinIO and embedded in one step:

curl -s -X POST 'http://localhost:6007/v1/dataprep/videos/upload?frame_interval=15' \
  -F 'file=@/path/to/video.mp4;type=video/mp4'

201 → {"status":"success","message":"..."}. First ingestion also downloads the YOLOX detector (needs network; without it, object detection is silently skipped).

Other flows — ingest from MinIO (POST /videos/minio), text summaries (POST /summary), RTSP, detection tuning: docs/user-guide/api-reference.md.

The upload response contains only status/message. Obtain the generated dp_video_<timestamp> identifier from GET /videos, then attach a summary using that exact bucket and video_id:

curl -s -X POST 'http://localhost:6007/v1/dataprep/summary' \
  -H 'Content-Type: application/json' \
  -d '{
    "bucket_name": "vdms-bucket",
    "video_id": "dp_video_1730000000",
    "video_summary": "forklift narrowly misses pedestrian",
    "video_start_time": 33,
    "video_end_time": 41
  }'

4. Manage & observe

curl -s 'http://localhost:6007/v1/dataprep/videos'              # list (default bucket vdms-bucket)
curl -s 'http://localhost:6007/v1/dataprep/telemetry?limit=5'   # ingestion timings

Download: GET /videos/download?video_id=…. Delete: DELETE /videos/{bucket_name}/{video_id}[?video_name=…] — without video_name it deletes the whole video directory; destructive, confirm with the user first. MinIO console: http://localhost:6011.

Troubleshooting

Symptom	Likely cause → action
Health never responds on 6007	still pulling/downloading models → docker compose -f docker/compose.yaml logs -f vdms-dataprep
setup.sh errors about MINIO_ROOT_USER/PASSWORD	export them before sourcing
Startup error "model name must be provided"	EMBEDDING_MODEL_NAME unset → export and redeploy
Ingestion fails with "Dimensions mismatch"	collection was built with a different embedding model → override INDEX_NAME after source setup.sh --nosetup, or wipe (destructive — confirm), then re-ingest
413 on upload	the source endpoint has no implemented size check despite its docs; identify the rejecting proxy/server from headers and logs. For large files, avoid buffering through this endpoint: stage in MinIO and use POST /videos/minio
Detected objects missing from results	YOLOX download failed on first run (no network) → restart with network access
MinIO auth errors after re-deploy	inspect the effective container MINIO_ROOT_* environment and ensure clients use the same current credentials
MinIO bind-mount errors at start	setup.sh overwrites MINIO_MOUNT_PATH=/mnt/miniodata; after source setup.sh --nosetup, export a writable path, then invoke Compose manually