Skip to content

feat(canary): support LLMService targets and protect experiment arms - #108

Open
primowenjoy wants to merge 4 commits into
modelsphere:release-canaryfrom
primowenjoy:feat/canary-llmservice-target-pr
Open

primowenjoy wants to merge 4 commits into
modelsphere:release-canaryfrom
primowenjoy:feat/canary-llmservice-target-pr

Conversation

@primowenjoy

@primowenjoy primowenjoy commented Oct 29, 2026 •

Copy link
Copy Markdown

What type of PR is this?

  • feature

What this PR does and why

Canary experiments can select LLMService production targets while preserving HelmRelease support. Model service list/detail now identify experiment arms, link to the owning experiment using namespace/name/UID, and disable independent upgrade, rollback, SLO edits and uninstall. Canary arm details link back to the actual child service. Missing or inconsistent ownership remains protected and never links to a replacement experiment of the same name.

The deployment-plan view includes final canary-system and autoscale layers, matching Swiss effective-value precedence.

The experiment overview uses readable English/Chinese field labels and lifecycle descriptions instead of raw API field names. Chinese users see labels such as 服务状态, 访问路径 and 全部组就绪时间; generation explicitly identifies the applied/latest configuration versions. Historical experiments retain their last-observed-status explanation.

Design and coordinated changes

Final migration design. Backend enforcement requires Swiss #39; experiment UID and target checks require canaryd #4. Operator lifecycle and target guard: operator #8.

How it was tested

npm run typecheck, npm test (64 files, 544 tests), npm run build, and npm run build:swiss passed. New cases cover disabled arm write actions, preserved read actions and UID-based linking after same-name recreation. Tests also retain ordinary service and uninstalling service behavior.

Manual browser validation with two CPU Mock arms confirmed owner links, disabled arm mutations, enabled ordinary-service upgrade/uninstall actions, and token/TTFT metrics after 80 successful streaming requests.

The disposable experiment was stopped through the UI; child resources and the experiment route were cleaned up while its metric snapshot and the production Mock service were retained.

The final branch passed npm run typecheck, the full frontend suite (544 tests), and npm run build. The overview change was deployed to the test environment; both Console replicas became ready. Chrome validation on an existing stopped experiment confirmed translated fields and phase values, immediate English/Chinese switching, unchanged recorded values, and the final Chinese label layout. The browser language was restored to Chinese.

Which issue(s) this PR fixes

No separate issue for the overview wording correction; coordinated design and implementation PRs are linked above.

Checks

  • npm run typecheck && npm test && npm run build in web/
  • No internal hostnames, IPs, registries or credentials in the changes
  • Go and chart code are unchanged by the overview wording correction; existing PR CI checks cover the coordinated feature.

Notes for reviewers

Only the final design in the Canary plugin repository is maintained; intermediate development documents are excluded. Kubernetes admission and SLO/configuration drift tracking are separate follow-ups.

Release note

Canary supports LLMService production targets. Experiment arm services show their owner and must be managed through the experiment lifecycle. Experiment overview fields and service status descriptions follow the selected English or Chinese language.

@primowenjoy primowenjoy changed the title feat(web): support LLMService Canary targets feat(canary): support LLMService targets and protect experiment arms Oct 29, 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