Skip to content

docs(architecture): add ARCHITECTURE.md with three-plane component overview#13

Open
saiyam1814 wants to merge 1 commit intoNVIDIA:mainfrom
saiyam1814:saiyam/docs/architecture-overview
Open

docs(architecture): add ARCHITECTURE.md with three-plane component overview#13
saiyam1814 wants to merge 1 commit intoNVIDIA:mainfrom
saiyam1814:saiyam/docs/architecture-overview

Conversation

@saiyam1814
Copy link
Copy Markdown

TL;DR

Adds ARCHITECTURE.md at the repository root covering the three-plane topology, component responsibilities, single-request lifecycle, scale-to-zero mechanism, multi-cluster routing, workload types, and CRD hierarchy.

Additional Details

New contributors currently reconstruct the full architecture picture by reading across multiple AGENTS.md files, the helmfile, proto definitions, and source code. A single canonical document at the repo root lowers the onboarding barrier and gives a shared reference for design discussions.

Every claim in the document is directly derived from source:

  • NATS subjects from src/compute-plane-services/nvca/pkg/queue/nats/client.go
  • CRD names from src/compute-plane-services/nvca/pkg/apis/
  • Request lifecycle from nvca/internal/miniservice/ and http-invocation crates
  • Scale-to-zero description from autoscaler Rust loop + NATS JetStream durable consumer behavior

For the Reviewer

The ASCII diagram is intentionally text-based (no external image dependency). If the team prefers Mermaid, happy to convert.

For QA

Documentation only - no behavior change.

Issues

NO-REF

Checklist

  • I am familiar with the Contributing Guidelines.
  • I have signed off my commits for Developer Certificate of Origin (DCO) compliance.
  • New or existing tests cover these changes.
  • The documentation is up to date with these changes.

@saiyam1814 saiyam1814 requested a review from a team as a code owner May 11, 2026 06:15
…erview

Adds ARCHITECTURE.md at the repository root covering:
- ASCII diagram of the control/invocation/compute plane topology
  with all five invocation-plane services including nats-auth-callout
- Component responsibility table for all 13 services
- Step-by-step single request lifecycle (12 steps, sync and async)
- Scale-to-zero mechanism via NATS JetStream durable buffering
- Multi-cluster routing using per-cluster NATS JetStream subjects
  (source: nvca/pkg/queue/nats/client.go)
- Function workload types (container, Helm, task variants)
- CRD hierarchy with verified API groups (nvcf/v1, nvca/v2beta1,
  nvca/v1alpha1)
- Links to all 10 AGENTS.md files that exist in the repository

New contributors currently reconstruct this picture by reading across
multiple AGENTS.md files, the helmfile, proto definitions, and source
code. A single canonical document at the repo root lowers the
onboarding barrier and provides a shared reference for design
discussions.

JIRA: NO-REF
NVBug: none
Signed-off-by: Saiyam Pathak <saiyam911@gmail.com>
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