feat: add a runtime diagnostics page

Adds GET /api/clouds/:cloud/diagnostics and a Diagnostics page in the General nav. The report combines, in one call: - runtime reachability, latency, version and edition from /_floci/health - every service the runtime emulates, with the console pages backed by it - per-console-service probes (the existing status check) - EC2 instance containers from /_floci/diagnose - console uptime, sign-in mode and active session count Each source is optional, so a down runtime still yields a partial report.
2026-10-23 11:29:48 -04:00
parent 432d9d4ce1
commit 953ba951db
13 changed files with 793 additions and 2 deletions
@@ -0,0 +1,152 @@
+import {authStateSizes} from '../auth'
+import type {
+    CloudProvider,
+    CloudServiceDescriptor,
+    CloudServiceType,
+    EmulatedServiceStatus,
+    RuntimeContainer,
+    RuntimeDiagnostics,
+} from '../cloud-spi/types'
+import type {CloudProxyService} from './CloudProxyService'
+import {awsEndpoint} from './runtimeProbe'
+
+const DIAGNOSTICS_TIMEOUT_MS = 3_000
+const processStartedAt = new Date()
+
+/**
+ * Floci core service ids mapped to the console services they back. Only the
+ * runtime's own ids are listed; whether a console page is actually available is
+ * still decided by the adapter registry through `services()`.
+ */
+const AWS_CONSOLE_SERVICES: Record<string, CloudServiceType[]> = {
+    apigateway: ['apigateway'],
+    appconfig: ['configuration'],
+    cloudformation: ['iac'],
+    codebuild: ['codebuild'],
+    codedeploy: ['codedeploy'],
+    codepipeline: ['codepipeline'],
+    dynamodb: ['nosql'],
+    ec2: ['compute', 'networking'],
+    ecr: ['ecr'],
+    ecs: ['ecs', 'fargate'],
+    eks: ['k8s'],
+    elasticache: ['elasticache'],
+    elasticloadbalancing: ['loadbalancing', 'targetgroups'],
+    email: ['email'],
+    events: ['events'],
+    iam: ['identity'],
+    kafka: ['msk'],
+    kafkaconnect: ['msk-connect'],
+    kinesis: ['kinesis'],
+    kms: ['kms'],
+    lambda: ['serverless'],
+    lightsail: ['lightsail'],
+    logs: ['logs'],
+    rds: ['database'],
+    'rds-data': ['database'],
+    s3: ['storage'],
+    secretsmanager: ['secrets'],
+    sns: ['notifications'],
+    sqs: ['messaging'],
+    ssm: ['parameters'],
+    states: ['workflows'],
+}
+
+type FlociHealth = {services?: Record<string, unknown>; version?: unknown; edition?: unknown}
+type FlociDiagnose = {ec2?: {instances?: Array<{instanceId?: unknown; containerStatus?: unknown; sshHost?: unknown; sshPort?: unknown}>}}
+
+export type DiagnosticsDependencies = {
+    fetch?: typeof fetch
+    now?: () => Date
+    startedAt?: Date
+    authMode?: 'local' | 'oidc'
+    activeSessions?: () => number
+}
+
+/**
+ * Everything the console can learn about a runtime in one call. Each source is
+ * optional: a runtime that is down, or one that does not publish a service list
+ * or container diagnostics, still yields a partial report rather than an error.
+ */
+export async function runtimeDiagnostics(
+    cloud: CloudProvider,
+    proxy: Pick<CloudProxyService, 'status' | 'services'>,
+    dependencies: DiagnosticsDependencies = {},
+): Promise<RuntimeDiagnostics> {
+    const fetcher = dependencies.fetch ?? globalThis.fetch
+    const now = dependencies.now?.() ?? new Date()
+    const startedAt = dependencies.startedAt ?? processStartedAt
+    const flociCore = cloud === 'aws'
+
+    const [status, health, diagnose] = await Promise.all([
+        proxy.status(cloud, {includeServices: true}),
+        flociCore ? fetchJson<FlociHealth>(fetcher, `${awsEndpoint()}/_floci/health`) : null,
+        flociCore ? fetchJson<FlociDiagnose>(fetcher, `${awsEndpoint()}/_floci/diagnose`) : null,
+    ])
+
+    return {
+        cloud,
+        checkedAt: now.toISOString(),
+        runtime: {
+            status: status.runtime,
+            endpoint: status.endpoint,
+            latencyMs: health?.latencyMs ?? null,
+            error: status.error,
+            version: stringOrNull(health?.body.version),
+            edition: stringOrNull(health?.body.edition),
+        },
+        emulatedServices: emulatedServices(health?.body.services, proxy.services(cloud)),
+        consoleServices: status.services ?? [],
+        containers: containers(diagnose?.body),
+        console: {
+            startedAt: startedAt.toISOString(),
+            uptimeSeconds: Math.max(0, Math.floor((now.getTime() - startedAt.getTime()) / 1000)),
+            authMode: dependencies.authMode ?? (process.env.FLOCI_UI_AUTH_MODE === 'oidc' ? 'oidc' : 'local'),
+            activeSessions: (dependencies.activeSessions ?? (() => authStateSizes().sessions))(),
+        },
+    }
+}
+
+function emulatedServices(reported: Record<string, unknown> | undefined, descriptors: CloudServiceDescriptor[]): EmulatedServiceStatus[] | null {
+    if (!reported || typeof reported !== 'object') return null
+    const available = new Map(descriptors.filter((descriptor) => descriptor.availability === 'available').map((descriptor) => [descriptor.service, descriptor]))
+    return Object.entries(reported)
+        .map(([name, state]) => ({
+            name,
+            state: typeof state === 'string' ? state : 'unknown',
+            consolePages: (AWS_CONSOLE_SERVICES[name] ?? [])
+                .map((service) => available.get(service))
+                .filter((descriptor): descriptor is CloudServiceDescriptor => Boolean(descriptor))
+                .map(({service, displayName, route}) => ({service, displayName, route})),
+        }))
+        .sort((left, right) => left.name.localeCompare(right.name))
+}
+
+function containers(diagnose: FlociDiagnose | undefined): RuntimeContainer[] | null {
+    const instances = diagnose?.ec2?.instances
+    if (!Array.isArray(instances)) return null
+    return instances
+        .filter((instance) => typeof instance.instanceId === 'string')
+        .map((instance) => ({
+            kind: 'ec2-instance' as const,
+            resourceId: instance.instanceId as string,
+            status: stringOrNull(instance.containerStatus),
+            sshHost: stringOrNull(instance.sshHost),
+            sshPort: typeof instance.sshPort === 'number' ? instance.sshPort : null,
+        }))
+}
+
+async function fetchJson<T>(fetcher: typeof fetch, url: string): Promise<{body: T; latencyMs: number} | null> {
+    const startedAt = performance.now()
+    try {
+        const response = await fetcher(url, {signal: AbortSignal.timeout(DIAGNOSTICS_TIMEOUT_MS)})
+        if (!response.ok) return null
+        return {body: await response.json() as T, latencyMs: Math.round(performance.now() - startedAt)}
+    } catch {
+        return null
+    }
+}
+
+function stringOrNull(value: unknown): string | null {
+    return typeof value === 'string' && value ? value : null
+}