Skip to content

feat(deploy): add an opt-in production control-plane capacity profile - #157

Open
rcyw wants to merge 2 commits into
inclusionAI:developfrom
rcyw:dev/production-control-capacity
Open

rcyw wants to merge 2 commits into
inclusionAI:developfrom
rcyw:dev/production-control-capacity

Conversation

@rcyw

@rcyw rcyw commented Nov 9, 2026 •

Copy link
Copy Markdown
Collaborator

Add an opt-in core-chart capacity profile for larger installations while preserving lightweight defaults.

The profile configures one single-active Coordinator (4/8 CPU, 16/32Gi), four combined Ingress/API replicas (4/8 CPU, 8/16Gi each), and one persistent managed Redis instance (4/8 CPU, 32/64Gi) with AOF and a 200Gi PVC. API replicas require distinct hosts and can be expanded to eight after provisioning and qualifying the required capacity. Redis gains optional node selection, tolerations and affinity so it can use a stable control pool.

values-production.yaml owns the allocations. The deployment guide explains core/umbrella/Terraform consumption, host and memory budgets, explicit SSD StorageClass selection, maintenance and rollback. Existing StatefulSet claim-template fields must be retained; expanding a bound PVC or moving storage requires a separate reviewed procedure that preserves data and credentials.

These are initial allocations, not a measured sandbox-count or connection-count guarantee. Single-instance Redis and Coordinator remain availability boundaries. No live cluster, runtime pins, credentials, API endpoints or worker scaling policies are changed.

Validation:

  • All 165 core-chart tests passed. Three focused profile/placement compatibility tests are included in test_adx_chart.py, which existing CI already executes; no separate production test runner is added.
  • The 13 Terraform ADX contract tests passed against the unchanged cloud/profile inputs.
  • Production-profile Helm lint, deployment script syntax checks, changed documentation links and diff whitespace checks passed.
  • No target-scale load test, runtime SDK regression or live deployment was performed.

rcyw added 2 commits November 9, 2026 21:17
Provide a single-active Coordinator, four API replicas with required host
separation, and persistent single-instance Redis as a sizing starting point.
Keep small-installation defaults and existing worker/runtime behavior intact.

Allow explicit Redis control-pool placement. Document SSD selection, the
four-to-eight API capacity budget, safe existing-PVC expansion and rollback,
and qualification required before claiming target sandbox or connection scale.

Validate core and umbrella rendering, default compatibility and site overrides.

Signed-off-by: wayne <rcywjo@gmail.com>
Move the profile rendering, host separation and Redis placement coverage into
the chart contract suite already executed by CI. Remove the standalone test
file and duplicated Redis/default-value assertions.

Keep the capacity guide focused on reusable deployment and storage migration,
clarify preservation of all immutable claim-template fields, and leave sizing
values in the profile without implying an unmeasured scale guarantee.

Signed-off-by: wayne <rcywjo@gmail.com>
@rcyw
rcyw requested review from WenYuLuo and hky1999 November 9, 2026 13:48

@WenYuLuo WenYuLuo left a comment

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

LGTM

This branch has not been deployed

No deployments
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.

2 participants