Design Deployment Dashboard - System Design | WinJob

Problem Statement: Unified Deployment Visibility

Enterprises run Spinnaker, Argo CD, GitHub Actions, and bespoke scripts in parallel. During an incident, on-call engineers waste minutes opening three UIs to learn what version is live in prod. Your design is a deployment dashboard: an aggregation control plane that normalizes CD events, materializes authoritative current state per (application, environment), and exposes guarded promote/rollback actions—without replacing underlying CD engines.

Who uses it and why now

Release engineers need a matrix: service × environment × cluster/region with git SHA, image digest, deploy status, and last change time.
Service owners want self-service visibility into their team's queue without paging platform for every question.
Incident commanders need a correlated timeline: commit → build → artifact → deploy stage → post-deploy health within one click path.

Core capabilities (in scope)

Live deployment matrix with staleness indicators when a source stops emitting events.
Immutable deployment timeline per application/environment (append-only events, not destructive updates).
Multi-source ingestion from Spinnaker pipeline notifications, Argo CD application status, and CI workflow webhooks.
Action façade that calls existing CD APIs (promote, rollback, cancel) behind RBAC and approval policies.

Explicit non-goals

Replacing Spinnaker Deck or Argo CD UI for deep manifest debugging.
Owning build systems or artifact storage (link out by digest/URL).
Running kubectl exec or shell access from the dashboard.

Scale anchors to state aloud

Assume 350 microservices, 8 clusters, 4 environments (dev/stage/canary/prod) → ~11,200 matrix cells. Ingest ~25 webhook events/sec peak (Friday deploy window), ~180k events/day retained 90 days (~16M rows). Dashboard read API target p95 < 250 ms for matrix slice; WebSocket fan-out < 2 s from event commit to UI.

javaOne Dark Pro

1 public final class DeploymentEventNormalizer {
2   public NormalizedEvent fromArgoNotification(JsonNode payload) {
3     String app = payload.path("application").asText();
4     String phase = payload.path("status").path("operationState").path("phase").asText();
5     DeploymentStatus status = "Succeeded".equals(phase) ? DeploymentStatus.SUCCEEDED : DeploymentStatus.RUNNING;
6     return new NormalizedEvent(app, status, payload.path("commitSha").asText(), payload.path("id").asText());
7   }
8 }

pythonOne Dark Pro

1 from dataclasses import dataclass
2 from typing import Literal
3  
4 @dataclass
5 class CurrentDeployment:
6     application_id: str
7     environment_id: str
8     version: str
9     git_sha: str
10     artifact_digest: str
11     status: Literal["pending", "running", "succeeded", "failed", "rolled_back"]
12  
13 def is_stale(last_event_ts: float, now: float, threshold_sec: float = 300.0) -> bool:
14     return (now - last_event_ts) > threshold_sec

typescriptOne Dark Pro

1 export interface DeploymentMatrixCell {
2   applicationId: string;
3   environmentId: string;
4   version: string;
5   gitSha: string;
6   artifactDigest: string;
7   status: "pending" | "running" | "succeeded" | "failed" | "rolled_back";
8   lastUpdatedAt: string;
9   sourceStale: boolean;
10 }
11  
12 export function canRollback(role: string, env: string): boolean {
13   if (env === "prod") return role === "deploy-admin";
14   return role === "deploy-operator" || role === "deploy-admin";
15 }

Opening line for the whiteboard

"I will optimize for trustworthy current state and fast incident triage: event-sourced timelines, idempotent webhook ingestion, and RBAC-wrapped actions on top of Spinnaker/Argo—not another CI runner."

How to open this one

The framing that signals depth on a deployment dashboard is a single pane that answers what is deployed where, right now, across every service and environment — sourced from the deploy controllers via event-sourced ingestion, not hand-maintained. Lead with the read model aggregated from many sources with explicit staleness indicators, and the failure story that proves it: during an incident a source stops emitting, the matrix goes stale, and on-call rolls back the wrong version. That shows you understand the dashboard is an operational tool whose value is correctness under pressure, not pretty charts.

Key Highlights

•Idempotent ingestion on pipeline execution_id
•Separate observed health from declared deploy success
•Staleness banners when source CD API unreachable

1	public final class DeploymentEventNormalizer {
2	public NormalizedEvent fromArgoNotification(JsonNode payload) {
3	String app = payload.path("application").asText();
4	String phase = payload.path("status").path("operationState").path("phase").asText();
5	DeploymentStatus status = "Succeeded".equals(phase) ? DeploymentStatus.SUCCEEDED : DeploymentStatus.RUNNING;
6	return new NormalizedEvent(app, status, payload.path("commitSha").asText(), payload.path("id").asText());
7	}
8	}

1	from dataclasses import dataclass
2	from typing import Literal
3
4	@dataclass
5	class CurrentDeployment:
6	application_id: str
7	environment_id: str
8	version: str
9	git_sha: str
10	artifact_digest: str
11	status: Literal["pending", "running", "succeeded", "failed", "rolled_back"]
12
13	def is_stale(last_event_ts: float, now: float, threshold_sec: float = 300.0) -> bool:
14	return (now - last_event_ts) > threshold_sec

1	export interface DeploymentMatrixCell {
2	applicationId: string;
3	environmentId: string;
4	version: string;
5	gitSha: string;
6	artifactDigest: string;
7	status: "pending" \| "running" \| "succeeded" \| "failed" \| "rolled_back";
8	lastUpdatedAt: string;
9	sourceStale: boolean;
10	}
11
12	export function canRollback(role: string, env: string): boolean {
13	if (env === "prod") return role === "deploy-admin";
14	return role === "deploy-operator" \|\| role === "deploy-admin";
15	}