Skip to content

feat(deploy): replace Traefik with Edge and Node Proxy - #56

Open
WenYuLuo wants to merge 2 commits into
inclusionAI:mainfrom
WenYuLuo:feat/edge-node-proxy-cn4
Open

feat(deploy): replace Traefik with Edge and Node Proxy#56
WenYuLuo wants to merge 2 commits into
inclusionAI:mainfrom
WenYuLuo:feat/edge-node-proxy-cn4

Conversation

@WenYuLuo

@WenYuLuo WenYuLuo commented Sep 9, 2026

Copy link
Copy Markdown
Collaborator

Replace the Traefik ingress with Edge and Node Proxy in the default standalone, Helm and Terraform deployments. Standalone now starts one all-in-one container; the Go CLI starts Edge beside Frontend and a Node Proxy on each runtime node. Remove the Traefik chart resources, image/configuration inputs, provider setup and CI dependency.

Package checksum-pinned Core and data-plane source from YuanRong 2b54c26885c6, including the process scripts and Node Proxy address registration. Keep RRT and the sandbox SDK on 0.10.2rc2 to preserve their command protocol. Advance sandboxd to main b5d25eede1729, which includes the sandbox IP returned from Start.

Expose HTTP/WS and HTTPS/WSS through the Edge Service with configurable TLS, IAM validation and CIDRs. Support retaining an existing LoadBalancer identity during migration. Route /internal-stats and optional /grafana through Edge; use the advertised listener ports for SDK internal URLs. Update the environment helper, cloud profiles and deployment documentation.

The entrypoint and node Bash supervisor forward termination and wait for the CLI. KillMode=mixed lets the supervisor drive shutdown, while sandboxd retains its separate service and ordering dependency.

Validation

  • Linux SDK: 270 unit tests passed with the pinned openyuanrong-sandbox==0.10.2rc2; Ruff and mypy passed.
  • Bash supervisor regression and deployment script checks passed.
  • Helm lint/rendering passed for default ingress, custom TLS, and both cloud profiles. Alibaba Cloud and Huawei Cloud Terraform validation passed.
  • Built the pinned data-plane binaries, the default RRT root filesystem, and sandboxd main; verified the Core/source artifact checksums.
  • Standalone SDK: five maintained integration tests passed without skips; HTTP/HTTPS public and internal port forwarding, WS/WSS reverse tunnels, gateway metadata and Grafana routing passed.

Validation boundary

Standalone validation uses a prepared all-in-one base overlaid with the selected Core, rebuilt RRT root filesystem, data-plane binaries, sandboxd and deployment scripts. The data-plane build used the available Linux compile image override. TLS routing checks used the deployment certificate with verification disabled. A complete clean all-in-one build and ARM runtime validation have not been rerun locally.

The full graceful-shutdown contract also requires the Go CLI signal-wait and deploy.sh shutdown-order fixes maintained outside this AKernel PR. Earlier cluster shutdown measurements used those runtime fixes; they do not validate every final default deployment setting.

Configure Edge and Node Proxy through the Go CLI, expose their Helm
service and readiness settings, and discover the gateway for SDK exports.

Keep deployment supervisors alive while the CLI performs ordered cleanup.
Translate the shared image stop signal for the control plane and forward
node termination through Bash with mixed service cleanup. Add regression
coverage for signal forwarding, cleanup waiting, and exit-code propagation.

Signed-off-by: robbluo <luo1442@gmail.com>
Make the Edge ingress the default for standalone, Helm and cloud profiles,
and remove the Traefik deployment, routing and image configuration. Package
the matching Core and data-plane binaries while preserving the RRT and SDK
command protocol. Update sandboxd to main for its Start response IP.

Serve gateway metadata and optional Grafana routes through Edge, resolve
internal SDK ports from the gateway listeners, and update CI and deployment
documentation for the single-container standalone topology.

Signed-off-by: robbluo <luo1442@gmail.com>
@WenYuLuo WenYuLuo changed the title feat(deploy): add Edge ingress and supervised shutdown feat(deploy): replace Traefik with Edge and Node Proxy Sep 9, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant