fix(validation-harness): dynamic RTSP live-proxy port + stale sensor residue gotchas#1079
Open
ary1111 wants to merge 3 commits into
Open
fix(validation-harness): dynamic RTSP live-proxy port + stale sensor residue gotchas#1079ary1111 wants to merge 3 commits into
ary1111 wants to merge 3 commits into
Conversation
…residue gotchas - Section 4 step 4: replace hardcoded rtsp://HOST:30554/live/SID with a retry loop that reads the actual port from vss-vios-streamprocessing logs (RtspLoadBalancer pool is dynamic; landed on 30556 in last run). VIOS GET /sensor/<id>/streams returns empty .url for Rtsp sensors so the log line is the only discovery source (Finding B, 2026-06-17). - Gotchas: add bullet documenting the dynamic live-proxy port and the correct grep command to extract it (Finding B). - Gotchas: add live registration residue on re-run bullet (Finding C, 2026-06-17) -- vst_data is a host bind-mount that survives down/-v, so a prior NvStreamer sensor stays in Postgres and causes HTTP 400 on re-run. Documents DELETE preferred for iterative runs vs full rm -rf vst_data for clean-slate runs. Signed-off-by: Adam Ryason <aryason@nvidia.com>
Collaborator
Author
|
/ok to test a57dfab |
Contributor
vss-playbook-complianceStep 1 - playbook compliancePASS: no errors - errors=0; warnings=1; skills_checked=1
vss-playbook-compliance - commit |
Contributor
Greptile SummaryThis PR fixes the hardcoded RTSP proxy port in the NvStreamer smoke-test sequence — VIOS's
Confidence Score: 5/5Safe to merge — all changes are documentation and runbook content; no production code paths are modified. The core RTSP port fix is correct and well-evidenced. The new patch/deploy/integrate reference files are thorough and internally consistent. The only new finding is a workflow gap in the Alert Bridge .env template where the runtime model ID is hardcoded, but the hf-1208 tag appears stable for the 3.2.0 release and the fix path is documented. skills/vss-build-vision-agent/references/patch-alerts.md — the generated .env template hardcodes the RT-VLM model tag; the deploy workflow should include a step to resolve it dynamically after RT-VLM starts. Important Files Changed
Sequence Diagram%%{init: {'theme': 'neutral'}}%%
sequenceDiagram
participant Script as Smoke Script
participant NV as NvStreamer :31000
participant VIOS as VIOS :30888
participant SPLogs as vss-vios-streamprocessing logs
participant RTVLM as RT-VLM :8018
Script->>NV: GET /sensor/list (retry until NVSID found)
NV-->>Script: sensorId
Script->>NV: GET /sensor/NVSID/streams
NV-->>Script: URL rtsp HOST 315xx nvstream
Script->>VIOS: POST /sensor/add sensorUrl URL
VIOS-->>Script: SID
loop up to 6x every 5s
Script->>SPLogs: docker logs grep Live proxy url grep SID
SPLogs-->>Script: rtsp HOST DYNAMIC_PORT live SID
Note over Script: PROXY found break
end
Note over Script: Port is dynamic e.g. 30556 not 30554
Script->>RTVLM: GET /v1/models
RTVLM-->>Script: MODEL_ID nim_nvidia_cosmos-reason2-8b_hf-1208
Script->>RTVLM: POST /v1/streams/add liveStreamUrl PROXY
RTVLM-->>Script: STREAM_ID
Script->>RTVLM: POST /v1/generate_captions id STREAM_ID model MODEL_ID
RTVLM-->>Script: SSE caption stream
%%{init: {'theme': 'base', 'themeVariables': {"darkMode": true, "background": "#0d1117", "primaryColor": "#21262d", "primaryTextColor": "#e6edf3", "primaryBorderColor": "#8b949e", "lineColor": "#8b949e", "textColor": "#e6edf3", "edgeLabelBackground": "#161b22", "actorBkg": "#21262d", "actorBorder": "#8b949e", "actorTextColor": "#e6edf3", "actorLineColor": "#8b949e", "signalColor": "#8b949e", "signalTextColor": "#e6edf3", "noteBkgColor": "#373320", "noteBorderColor": "#d4a72c", "noteTextColor": "#f0e6c0", "labelBoxBkgColor": "#21262d", "labelBoxBorderColor": "#8b949e", "labelTextColor": "#e6edf3", "loopTextColor": "#e6edf3", "activationBkgColor": "#30363d", "activationBorderColor": "#8b949e"}}}%%
sequenceDiagram
participant Script as Smoke Script
participant NV as NvStreamer :31000
participant VIOS as VIOS :30888
participant SPLogs as vss-vios-streamprocessing logs
participant RTVLM as RT-VLM :8018
Script->>NV: GET /sensor/list (retry until NVSID found)
NV-->>Script: sensorId
Script->>NV: GET /sensor/NVSID/streams
NV-->>Script: URL rtsp HOST 315xx nvstream
Script->>VIOS: POST /sensor/add sensorUrl URL
VIOS-->>Script: SID
loop up to 6x every 5s
Script->>SPLogs: docker logs grep Live proxy url grep SID
SPLogs-->>Script: rtsp HOST DYNAMIC_PORT live SID
Note over Script: PROXY found break
end
Note over Script: Port is dynamic e.g. 30556 not 30554
Script->>RTVLM: GET /v1/models
RTVLM-->>Script: MODEL_ID nim_nvidia_cosmos-reason2-8b_hf-1208
Script->>RTVLM: POST /v1/streams/add liveStreamUrl PROXY
RTVLM-->>Script: STREAM_ID
Script->>RTVLM: POST /v1/generate_captions id STREAM_ID model MODEL_ID
RTVLM-->>Script: SSE caption stream
Reviews (3): Last reviewed commit: "feat(skill): VS + Alerting additive prof..." | Re-trigger Greptile |
![greptile-apps[bot]](https://avatars.githubusercontent.com/in/867647?s=60&v=4){kind=small}
Comment on lines
+143
to
+145
| for i in 1 2 3 4 5 6; do | ||
| PROXY=$(docker logs vss-vios-streamprocessing 2>&1 | grep "Live proxy url" | grep "$SID" | tail -1 \ | ||
| | sed -E 's#.*(rtsp://[^[:space:]]+).*#\1#') |
Contributor
There was a problem hiding this comment.
$SID turns grep into a wildcard
If POST /sensor/add fails and SID evaluates to empty or the literal string null (from jq -r), then grep "$SID" matches every "Live proxy url" line in the container logs, and tail -1 will return the proxy URL of whichever sensor was registered last — potentially one from a previous run. The extracted PROXY will then point to the wrong sensor. Adding a check for SID being non-empty and not equal to null before entering the loop would prevent silently feeding the wrong stream to RT-VLM.
Contributor
🔬 Skills Review — 6-paradigm consolidation19 findings across 1 skill(s) · 🔴 critical=1 · 🟠 high=5 · 🟡 medium=5 · 🔵 low=8 🔴 Critical / high (act now)
|
…, .env guard standalone-compose-patches.md: - Patch 0: use sudo mkdir/chmod for root-owned bind-mount dirs - Patch 0: busybox named-volume pre-chmod (mdx-kafka, mdx-elastic-data/logs) before first `up`; PROJ=mdx not basename of BUILD_DIR - Patch 0: busybox rm -f stale redis *.log before up (dnsmasq:bootstrap 0600 ownership blocks Redis even after chmod 777 on the directory) - Patch 5 (new): set network_mode: host on wait-for-redis in patched SDRC compose to fix 120s bridge->host timeout on iptables FORWARD DROP hosts validation-harness.md: - Section 2: nvstaging -> nvidia image registry for nvstreamer-validation - Section 4: runtime model ID resolution via GET /v1/models before generate_captions (hardcoded cosmos-reason2-8b returns BadParameters); renumber duplicate step 5 -> steps 5/6/7 - Gotchas: VOD clip_storage split-brain residue (file on disk, absent from Postgres) -- busybox rm -f through bind-mount to unblock re-upload patch-rt-vlm.md: - Add FILE_URL_ALLOWED_DIRS env var injection (RTVI_VLM_FILE_URL_ALLOWED_DIRS in .env + FILE_URL_ALLOWED_DIRS in rtvi-vlm environment block) - Add VOD path design section: VIOS vodUrl -> RT-VLM shared clip_storage mount flow (upload -> resolve -> file:// register -> caption) - Document runtime model ID resolution requirement and error it prevents env-file-enumeration.md: - Add "Required additions for IN-1" table: RTVI_VLM_FILE_URL_ALLOWED_DIRS and SAMPLE_VIDEO_DATASET (both absent from all upstream .env files) SKILL.md: - Step 6: .env overwrite guard -- write to .env.new when existing .env contains production values (NGC_CLI_API_KEY/HOST_IP sentinel check) All surfaced live 2026-06-18, IN-1 expanded eval. Signed-off-by: Adam Ryason <aryason@nvidia.com>
…improvements New pair files for two IN-1 extension profiles validated E2E on 2026-06-18: Video Summarization (VS) — IN-1-1 (flag bp_developer_in_1_lvs, 23 services): - skills/vss-summarize-video/references/integrate-lvs.md: VS integration contract; corrected ES output to default_<file_id> (not lvs-events); local NIM documented as upstream default (dev-profile-lvs local_shared) - skills/vss-summarize-video/references/deploy-lvs-service.md: VS deployment contract (CPU-only, port 38111) - skills/vss-build-vision-agent/references/patch-lvs.md: component_services block; GPU-slot check (host_gpu_count - profile_gpu_count >= 1 -> local NIM default); cloud inference (integrate.api.nvidia.com) documented as fallback only with cloud-entitlement-vs-NGC-key warning; Patch 2 keep/strip rule for nvidia-nemotron-nano-9b-v2 when defined vs undefined VLM Real-time Alerting — IN-1-2 (flag bp_developer_an_1, 22 services): - skills/vss-manage-alerts/references/integrate-alerts.md: alert-bridge contract; POST /api/v1/realtime corrected to live_stream_url/alert_type/ prompt (not name/model/schedule); always_on:false default; ES index mdx-vlm-incidents-YYYY-MM-DD (date-suffixed); CPU-only, no 2nd GPU - skills/vss-manage-alerts/references/deploy-alerts-service.md: alert deployment contract - skills/vss-build-vision-agent/references/patch-alerts.md: component_services block; Patch 3 materializes realtime-config.yml; Patch 2 strips 9 undefined depends_on peers SKILL.md Step 4 improvements: - GPU-slot check for VS profiles: propose local NIM when spare GPU available - Terminology: Long Video Summarization (LVS) -> Video Summarization (VS) in prose; protected names unchanged (lvs-server, LVS_* env vars, mdx-lvs, etc.) All surfaced live 2026-06-18, IN-1-1 + IN-1-2 E2E validation. deploy/docker byte-identical. Signed-off-by: Adam Ryason <aryason@nvidia.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Description
Section 4 step 4: replace hardcoded rtsp://HOST:30554/live/SID with a retry loop that reads the actual port from vss-vios-streamprocessing logs (RtspLoadBalancer pool is dynamic; landed on 30556 in last run). VIOS GET /sensor//streams returns empty .url for Rtsp sensors so the log line is the only discovery source (Finding B, 2026-06-17).
Gotchas: add bullet documenting the dynamic live-proxy port and the correct grep command to extract it (Finding B).
Gotchas: add live registration residue on re-run bullet (Finding C, 2026-06-17) -- vst_data is a host bind-mount that survives down/-v, so a prior NvStreamer sensor stays in Postgres and causes HTTP 400 on re-run. Documents DELETE preferred for iterative runs vs full rm -rf vst_data for clean-slate runs.
Checklist
uv run pre-commit installonce, then hooks run on everygit commit).git commit -sadds aSigned-off-bytrailer that certifies you have the right to submit the change under Apache-2.0).