droboticflow/docs/poc-current-status.md
2026-06-29 02:58:23 +08:00

98 lines
4.4 KiB
Markdown

# POC Current Status
Last checked: 2026-06-26 Asia/Shanghai.
## Deployment
- Namespace: `droboticflow`
- Public API: `http://120.48.164.19`
- API deployment: `1/1`, image `ccr-29eug8s3-pub.cnc.bj.baidubce.com/dataflow-poc/droboticflow-api:fc420be-poc-20260626220856`
- Worker deployment: `1/1`, image `ccr-29eug8s3-pub.cnc.bj.baidubce.com/dataflow-poc/droboticflow-worker:fc420be-poc-20260626220856`
- API health: `{"service":"droboticflow-api","status":"ok"}`
## Artifact Preview Streaming
The real POC workflow run `run-178a79a2fa2c235880861abfef07d3d8` has video artifacts whose preview URLs point to `/api/runs/:id/artifacts/:artifactId/content` instead of inline base64 data URLs.
Verified content request:
```text
HTTP/1.1 206 Partial Content
Content-Length: 32
Content-Range: bytes 0-31/5415
Content-Type: video/mp4
```
## GPU Status
POC now has one schedulable GPU node with allocatable `nvidia.com/gpu=8`.
Checked nodes:
```text
192.168.80.136 nvidia.com/gpu=8 128 CPU / 1024 GiB / A800
192.168.80.249 nvidia.com/gpu=0
192.168.81.47 nvidia.com/gpu=0
```
Direct runtime acceptance on 2026-06-26 created a temporary Pod on `192.168.80.136` with `limits.nvidia.com/gpu=1` and ran `nvidia-smi` successfully:
```text
NVIDIA-SMI 570.195.03 Driver Version: 570.195.03 CUDA Version: 12.8
0, NVIDIA A800-SXM4-80GB, 570.195.03, 81920 MiB
```
`POC_GPU_CAPACITY_EXPECT_AVAILABLE=true npm run acceptance:poc-gpu-capacity` covers the deployed worker capacity endpoint without a POC user session. It starts a temporary Pod in the POC namespace with the deployed operator-runner image, reads `http://droboticflow-worker:3002/worker-api/capacity`, and passed with:
```text
gpu.available=true resource=nvidia.com/gpu schedulableAllocatable=8 maxAllocatable=8 nodes=3
node 192.168.80.136 nvidia.com/gpu=8 selectorMatched=true
```
`POC_GPU_WORKFLOW_EXPECT_AVAILABLE=true npm run acceptance:poc-gpu-workflow` covers the workflow-level GPU path after hardware is available. With `DROBOTICFLOW_ACCEPTANCE_EMAIL` and `DROBOTICFLOW_ACCEPTANCE_PASSWORD`, it exercises normal POC project/workflow writes. Without those credentials, it uses `DROBOTICFLOW_API_TOKEN` from the environment or `droboticflow-auth-secret` only against the bounded `/api/acceptance/poc-gpu-workflow` route, then port-forwards the deployed worker and triggers a real `process-once`. With GPU expected, the run must complete and produce artifacts; without GPU expected, the same script still verifies the real `gpu request cannot be scheduled` failure path.
Latest GPU workflow acceptance run:
```text
runId=run-008e6c48e55dfea3d645722247119ea0
status=completed
artifacts=9
lastLog="Produced 5 artifact(s)"
```
## Artifact Lineage Audit
The deployed API exposes `GET /api/artifacts/:artifactId/lineage`.
Verified against real run `run-178a79a2fa2c235880861abfef07d3d8` and video artifact `run-178a79a2fa2c235880861abfef07d3d8-slice-out-1-DJI_20260318112834_0001_D_clip001.mp4`:
```text
artifactName DJI_20260318112834_0001_D_clip001.mp4
producer slice 视频切片
consumers annotate 视频标注
upstreamEdges 1
downstreamEdges 1
nodes 3 edges 2
downstream slice DJI_20260318112834_0001_D_clip001.mp4 -> annotate in-1
```
## Strict POC Acceptance
`npm run acceptance:poc-readiness` currently expects:
- `deployment:droboticflow-api`: passed
- `deployment:droboticflow-worker`: passed
- `operator-runner-image`: passed
- `gpu-capacity`: passed when run with `POC_GPU_CAPACITY_EXPECT_AVAILABLE=true` (`gpu.available=true`, `maxAllocatable=8`)
- `platform-gpu-workflow-preflight`: passed, latest run `run-008e6c48e55dfea3d645722247119ea0` completed
- `fixture-manifest`: passed
`npm run acceptance:poc-strict` now uses `POC_ACCEPTANCE_REQUIRED_EXTERNALS=bos,mongo` by default. D-Cloud FAT API operators have been removed from the builtin catalog, so missing `droboticflow-dcloud-secret` is no longer a strict POC blocker.
If strict acceptance fails before creating the Job, the expected blocker is now `fixture-readiness` or a BOS/Mongo readiness issue. A final pass still requires real fixtures, real BOS evidence, real Mongo evidence, complete operator coverage, and clean repository policy evidence.
## Next Acceptance Gates
- Run `POC_GPU_WORKFLOW_EXPECT_AVAILABLE=true npm run acceptance:poc-gpu-workflow` and then `POC_GPU_CAPACITY_EXPECT_AVAILABLE=true npm run acceptance:poc-readiness`.
- Continue video semantic processing validation with real detection/tracking operators rather than heuristic motion-only annotations.