feat: add a runtime diagnostics page
CI / checks (pull_request) Failing after 13m18s

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.
This commit is contained in:
Zakaria
2026-10-23 11:29:48 -04:00
parent 432d9d4ce1
commit 953ba951db
13 changed files with 793 additions and 2 deletions
+5
View File
@@ -4,6 +4,7 @@ import {SecretsManagerPage} from '@/features/secretsmanager/SecretsManagerPage'
import {CloudExplorerPage} from '@/pages/CloudExplorerPage'
import {CloudConsoleHomePage} from '@/pages/CloudConsoleHomePage'
import {DatabaseDataPage} from '@/pages/DatabaseDataPage'
import {DiagnosticsPage} from '@/pages/DiagnosticsPage'
import {DynamoDbItemEditorPage} from '@/pages/DynamoDbItemEditorPage'
import {WorkflowStudioPage} from '@/pages/WorkflowStudioPage'
@@ -18,6 +19,10 @@ export default function App() {
<Route path="/console/azure" element={<Navigate to="/console/aws" replace/>}/>
<Route path="/console/gcp" element={<Navigate to="/console/aws" replace/>}/>
<Route path="/console/:cloud" element={<CloudConsoleHomePage/>}/>
<Route path="/diagnostics" element={<Navigate to="/diagnostics/aws" replace/>}/>
<Route path="/diagnostics/azure" element={<Navigate to="/diagnostics/aws" replace/>}/>
<Route path="/diagnostics/gcp" element={<Navigate to="/diagnostics/aws" replace/>}/>
<Route path="/diagnostics/:cloud" element={<DiagnosticsPage/>}/>
<Route path="/cloud-explorer" element={<Navigate to="/cloud-explorer/aws/storage" replace/>}/>
<Route path="/cloud-explorer/azure/*" element={<Navigate to="/cloud-explorer/aws/storage" replace/>}/>
<Route path="/cloud-explorer/gcp/*" element={<Navigate to="/cloud-explorer/aws/storage" replace/>}/>
+8
View File
@@ -13,6 +13,7 @@ export const apiEndpointKeys = {
list: "clouds.list",
services: "clouds.services.list",
status: "clouds.status.get",
diagnostics: "clouds.diagnostics.get",
serviceStatus: "clouds.services.status.get",
schema: "clouds.services.schema.get",
dependencies: "clouds.services.dependencies.get",
@@ -323,6 +324,13 @@ export const endpointRegistry: EndpointRegistry = new Map([
method: "GET",
},
],
[
apiEndpointKeys.clouds.diagnostics,
{
path: "/clouds/:cloud/diagnostics",
method: "GET",
},
],
[
apiEndpointKeys.clouds.serviceStatus,
{
@@ -6,6 +6,7 @@ import type {
CloudServiceStatus,
CloudServiceType,
CloudStatus,
RuntimeDiagnostics,
} from "@/types/cloud";
import type {
AppConfigConfigurationProfile,
@@ -86,6 +87,18 @@ export async function getCloudStatus(
return res.data;
}
export async function getCloudDiagnostics(
cloud: CloudProvider,
signal?: AbortSignal,
): Promise<RuntimeDiagnostics> {
const res = await apiClient.call<RuntimeDiagnostics>(
apiEndpointKeys.clouds.diagnostics,
requestOptions(cloud, "cloud-proxy", { signal }),
{ cloud },
);
return res.data;
}
export async function getCloudServiceStatus(
cloud: CloudProvider,
service: CloudServiceType,
@@ -1,5 +1,6 @@
import {useQuery} from '@tanstack/react-query'
import {
getCloudDiagnostics,
getCloudResource,
getCloudServiceStatus,
getCloudStatus,
@@ -20,6 +21,7 @@ export const cloudQueryKeys = {
clouds: ['clouds'] as const,
services: (cloud: CloudProvider) => ['cloud-services', cloud] as const,
status: (cloud: CloudProvider) => ['cloud-status', cloud] as const,
diagnostics: (cloud: CloudProvider) => ['cloud-diagnostics', cloud] as const,
serviceStatus: (cloud: CloudProvider, service: CloudServiceType) =>
['cloud-service-status', cloud, service] as const,
resources: (cloud: CloudProvider, service: CloudServiceType) =>
@@ -53,6 +55,15 @@ export function useCloudStatusQuery(cloud: CloudProvider, options: {refetchInter
})
}
/** Full runtime report; polled slowly because it probes every console service. */
export function useCloudDiagnosticsQuery(cloud: CloudProvider) {
return useQuery({
queryKey: cloudQueryKeys.diagnostics(cloud),
queryFn: ({signal}) => getCloudDiagnostics(cloud, signal),
refetchInterval: 30_000,
})
}
/** Health of the single service being viewed, rather than of the whole cloud. */
export function useCloudServiceStatusQuery(
cloud: CloudProvider,
+3 -2
View File
@@ -1,5 +1,5 @@
import {NavLink, Outlet, useLocation, useSearchParams} from 'react-router-dom'
import {AlertTriangle, ChevronsLeft, ChevronsRight, LayoutDashboard, LogOut, Moon, Search, Sun, Terminal} from 'lucide-react'
import {Activity, AlertTriangle, ChevronsLeft, ChevronsRight, LayoutDashboard, LogOut, Moon, Search, Sun, Terminal} from 'lucide-react'
import {useCallback, useEffect, useRef, useState} from 'react'
import theCloud from '../../../../TheCloud.png'
import {useTheme} from '@/lib/useTheme'
@@ -184,6 +184,7 @@ export function Layout() {
<div className="nav-section">
<span className="nav-label">General</span>
<NavItem to={`/console/${activeCloud}`} icon={<LayoutDashboard size={18} aria-hidden="true"/>} label="Console Home" collapsed={collapsed}/>
<NavItem to={`/diagnostics/${activeCloud}`} icon={<Activity size={18} aria-hidden="true"/>} label="Diagnostics" collapsed={collapsed}/>
</div>
<CloudServiceNav collapsed={collapsed}/>
</nav>
@@ -356,6 +357,6 @@ function TopbarSearch() {
}
function activeCloudFromPath(pathname: string): 'aws' | 'azure' | 'gcp' {
const match = pathname.match(/^\/(?:cloud-explorer|console)\/(aws|azure|gcp)(?:\/|$)/)
const match = pathname.match(/^\/(?:cloud-explorer|console|diagnostics)\/(aws|azure|gcp)(?:\/|$)/)
return (match?.[1] ?? 'aws') as 'aws' | 'azure' | 'gcp'
}
+98
View File
@@ -4261,6 +4261,104 @@ pre,
gap: 14px;
}
.diagnostics {
display: grid;
gap: 14px;
}
.diagnostics .table-panel {
overflow-x: auto;
}
.diagnostics-section-header {
display: flex;
flex-wrap: wrap;
align-items: center;
justify-content: space-between;
gap: 10px;
padding: 12px 16px;
border-bottom: 1px solid var(--border);
}
.diagnostics-section-header h2 {
margin: 0;
color: var(--text-bright);
font-size: 14px;
}
.diagnostics-filters,
.diagnostics-filter-group {
display: flex;
flex-wrap: wrap;
align-items: center;
gap: 6px;
}
.diagnostics-filters .input {
width: 180px;
}
.diagnostics-service-grid {
display: grid;
grid-template-columns: repeat(auto-fill, minmax(220px, 1fr));
margin: 0;
padding: 0;
list-style: none;
}
.diagnostics-service {
display: flex;
flex-wrap: wrap;
align-items: center;
gap: 6px 8px;
min-width: 0;
padding: 9px 16px;
border-bottom: 1px solid var(--border-faint);
font-size: 12px;
}
.diagnostics-page-link {
padding: 1px 7px;
border: 1px solid var(--border);
border-radius: 999px;
color: var(--link);
font-size: 11px;
text-decoration: none;
}
.diagnostics-page-link:hover {
border-color: var(--link);
}
.diagnostics-facts {
display: grid;
grid-template-columns: repeat(auto-fit, minmax(160px, 1fr));
margin: 0;
}
.diagnostics-facts div {
padding: 12px 16px;
}
.diagnostics-facts dt {
color: var(--text-2);
font-size: 11px;
font-weight: 700;
letter-spacing: 0.08em;
text-transform: uppercase;
}
.diagnostics-facts dd {
margin: 4px 0 0;
color: var(--text-bright);
}
@media (max-width: 520px) {
.diagnostics-filters .input {
width: 100%;
}
}
.console-provider-banner {
display: flex;
align-items: center;
@@ -0,0 +1,95 @@
import {QueryClient, QueryClientProvider} from '@tanstack/react-query'
import {render, screen, within} from '@testing-library/react'
import userEvent from '@testing-library/user-event'
import {MemoryRouter, Route, Routes} from 'react-router-dom'
import {expect, test, vi} from 'vitest'
import {DiagnosticsPage} from './DiagnosticsPage'
import type {RuntimeDiagnostics} from '@/types/cloud'
const api = vi.hoisted(() => ({getCloudDiagnostics: vi.fn()}))
vi.mock('@/api/cloudProxyClient', () => api)
const healthy: RuntimeDiagnostics = {
cloud: 'aws',
checkedAt: new Date().toISOString(),
runtime: {status: 'reachable', endpoint: 'http://floci:4566', latencyMs: 12, error: null, version: '1.9.0', edition: 'community'},
emulatedServices: [
{name: 'route53', state: 'running', consolePages: []},
{name: 's3', state: 'running', consolePages: [{service: 'storage', displayName: 'S3', route: '/cloud-explorer/aws/storage'}]},
{name: 'sqs', state: 'stopped', consolePages: [{service: 'messaging', displayName: 'SQS', route: '/cloud-explorer/aws/messaging'}]},
],
consoleServices: [
{cloud: 'aws', service: 'storage', adapterRegistered: true, runtime: 'reachable', endpoint: 'http://floci:4566', checkedAt: '', latencyMs: 4, error: null, errorCode: null},
{cloud: 'aws', service: 'messaging', adapterRegistered: true, runtime: 'unavailable', endpoint: 'http://floci:4566', checkedAt: '', latencyMs: 9, error: 'Service is not running', errorCode: 'RUNTIME_UNAVAILABLE'},
],
containers: [{kind: 'ec2-instance', resourceId: 'i-0abc', status: 'running', sshHost: 'localhost', sshPort: 2201}],
console: {startedAt: '2026-10-23T08:00:00.000Z', uptimeSeconds: 9_015, authMode: 'oidc', activeSessions: 2},
}
function renderPage(path = '/diagnostics/aws') {
const client = new QueryClient({defaultOptions: {queries: {retry: false}}})
return render(
<QueryClientProvider client={client}>
<MemoryRouter initialEntries={[path]}>
<Routes><Route path="/diagnostics/:cloud" element={<DiagnosticsPage/>}/></Routes>
</MemoryRouter>
</QueryClientProvider>,
)
}
test('summarises the runtime and lists failing console services first', async () => {
api.getCloudDiagnostics.mockResolvedValue(healthy)
renderPage()
const summary = await screen.findByRole('region', {name: 'Runtime summary'})
expect(within(summary).getByText('Reachable')).toBeInTheDocument()
expect(screen.getByText('http://floci:4566 · 12 ms')).toBeInTheDocument()
expect(screen.getByText('1.9.0')).toBeInTheDocument()
expect(screen.getByText('1 not running')).toBeInTheDocument()
expect(screen.getByText('2 / 3')).toBeInTheDocument()
const rows = within(screen.getByRole('region', {name: 'Console services'})).getAllByRole('row')
expect(rows[1]).toHaveTextContent('messaging')
expect(rows[1]).toHaveTextContent('Service is not running (RUNTIME_UNAVAILABLE)')
expect(screen.getByRole('link', {name: 'i-0abc'})).toHaveAttribute('href', '/cloud-explorer/aws/compute?resource=i-0abc')
expect(screen.getByText('localhost:2201')).toBeInTheDocument()
expect(screen.getByText('2 h 30 min')).toBeInTheDocument()
expect(screen.getByText('OIDC single sign-on')).toBeInTheDocument()
})
test('filters emulated services by console coverage and state', async () => {
const user = userEvent.setup()
api.getCloudDiagnostics.mockResolvedValue(healthy)
renderPage()
const section = await screen.findByRole('region', {name: 'Emulated services'})
await user.click(within(section).getByRole('button', {name: 'No console page'}))
expect(within(section).getAllByRole('listitem').map((item) => item.textContent)).toEqual(['route53'])
await user.click(within(section).getByRole('button', {name: 'Not running'}))
expect(within(section).getByRole('link', {name: 'SQS'})).toHaveAttribute('href', '/cloud-explorer/aws/messaging')
expect(within(section).getAllByRole('listitem')).toHaveLength(1)
await user.click(within(section).getByRole('button', {name: 'All'}))
await user.type(within(section).getByRole('searchbox'), 's3')
expect(within(section).getAllByRole('listitem').map((item) => item.textContent)).toEqual(['s3S3'])
})
test('shows what is known when the runtime is down', async () => {
api.getCloudDiagnostics.mockResolvedValue({
...healthy,
runtime: {status: 'unavailable', endpoint: 'http://floci:4566', latencyMs: null, error: 'Cannot reach Floci core at http://floci:4566', version: null, edition: null},
emulatedServices: null,
consoleServices: [],
containers: null,
} satisfies RuntimeDiagnostics)
renderPage()
expect(await screen.findByText('Unavailable')).toBeInTheDocument()
expect(screen.getByText('Cannot reach Floci core at http://floci:4566')).toBeInTheDocument()
expect(screen.getAllByText('Not reported by the runtime')).toHaveLength(2)
expect(screen.queryByRole('region', {name: 'Emulated services'})).not.toBeInTheDocument()
expect(screen.getByText('This runtime does not publish container diagnostics.')).toBeInTheDocument()
})
@@ -0,0 +1,221 @@
import {useMemo, useState} from 'react'
import {Activity, RefreshCw} from 'lucide-react'
import {Link, Navigate, useParams} from 'react-router-dom'
import {useCloudDiagnosticsQuery} from '@/api/queries/cloudQueries'
import {formatDateTime, formatRelativeTime} from '@/lib/format'
import type {CloudProvider, CloudServiceStatus, EmulatedServiceStatus, RuntimeDiagnostics} from '@/types/cloud'
type ServiceFilter = 'all' | 'not-running' | 'with-page' | 'without-page'
const SERVICE_FILTERS: Array<{id: ServiceFilter; label: string}> = [
{id: 'all', label: 'All'},
{id: 'not-running', label: 'Not running'},
{id: 'with-page', label: 'Has console page'},
{id: 'without-page', label: 'No console page'},
]
export function DiagnosticsPage() {
const params = useParams()
const cloud = normalizeCloud(params.cloud)
const query = useCloudDiagnosticsQuery(cloud ?? 'aws')
if (!cloud) return <Navigate to="/diagnostics/aws" replace/>
const report = query.data
return (
<>
<div className="page-header cloud-explorer-header">
<div className="page-title">
<Activity size={20}/>
<div>
<h1>Diagnostics</h1>
<p className="muted">{report ? `Checked ${formatRelativeTime(report.checkedAt) ?? 'just now'}` : 'Runtime, services and console health'}</p>
</div>
</div>
<button className="button" type="button" disabled={query.isFetching} onClick={() => void query.refetch()}>
<RefreshCw size={14}/>{query.isFetching ? 'Checking' : 'Re-check'}
</button>
</div>
<div className="content diagnostics">
{query.isLoading && <p className="muted padded">Running diagnostics</p>}
{query.error && <p className="error-text">Could not load diagnostics: {query.error.message}</p>}
{report && <DiagnosticsReport report={report}/>}
</div>
</>
)
}
function DiagnosticsReport({report}: {report: RuntimeDiagnostics}) {
const emulated = report.emulatedServices
const notRunning = emulated?.filter((service) => service.state !== 'running').length ?? 0
const covered = emulated?.filter((service) => service.consolePages.length > 0).length ?? 0
const reachable = report.runtime.status === 'reachable'
return (
<>
<section className="console-summary" aria-label="Runtime summary">
<div className={`summary-tile ${reachable ? 'ready' : 'unavailable'}`}>
<span>Runtime</span>
<strong>{reachable ? 'Reachable' : 'Unavailable'}</strong>
<small>{report.runtime.endpoint ?? 'No endpoint'}{report.runtime.latencyMs !== null ? ` · ${report.runtime.latencyMs} ms` : ''}</small>
</div>
<div className="summary-tile">
<span>Version</span>
<strong>{report.runtime.version ?? 'Unknown'}</strong>
<small>{report.runtime.edition ? `${report.runtime.edition} edition` : 'Not reported by the runtime'}</small>
</div>
<div className={`summary-tile ${notRunning > 0 ? 'pending' : ''}`}>
<span>Emulated services</span>
<strong>{emulated ? emulated.length : '—'}</strong>
<small>{emulated ? (notRunning > 0 ? `${notRunning} not running` : 'All running') : 'Not reported by the runtime'}</small>
</div>
<div className="summary-tile">
<span>Console coverage</span>
<strong>{emulated ? `${covered} / ${emulated.length}` : `${report.consoleServices.length}`}</strong>
<small>{emulated ? 'Emulated services with a console page' : 'Console services'}</small>
</div>
</section>
{report.runtime.error && <p className="error-text">{report.runtime.error}</p>}
<ConsoleServicesTable services={report.consoleServices}/>
{emulated && <EmulatedServices services={emulated}/>}
<ContainersTable report={report}/>
<ConsoleProcess report={report}/>
</>
)
}
function ConsoleServicesTable({services}: {services: CloudServiceStatus[]}) {
const failing = services.filter((service) => service.runtime !== 'reachable')
const ordered = [...failing, ...services.filter((service) => service.runtime === 'reachable')]
return (
<section className="table-panel" aria-labelledby="diagnostics-console-services">
<header className="diagnostics-section-header">
<h2 id="diagnostics-console-services">Console services</h2>
<span className="muted">{failing.length ? `${failing.length} of ${services.length} failing` : `${services.length} responding`}</span>
</header>
{services.length === 0
? <p className="muted padded">No console services are registered for this cloud.</p>
: <table className="table">
<thead><tr><th>Service</th><th>Status</th><th>Latency</th><th>Detail</th></tr></thead>
<tbody>{ordered.map((service) => (
<tr key={service.service}>
<td><code>{service.service}</code></td>
<td><span className={`status ${statusClass(service.runtime)}`}>{statusLabel(service.runtime)}</span></td>
<td>{service.latencyMs === null ? '—' : `${service.latencyMs} ms`}</td>
<td>{service.error ? <span className="error-text compact-text">{service.error}{service.errorCode ? ` (${service.errorCode})` : ''}</span> : <span className="muted">—</span>}</td>
</tr>
))}</tbody>
</table>}
</section>
)
}
function EmulatedServices({services}: {services: EmulatedServiceStatus[]}) {
const [filter, setFilter] = useState<ServiceFilter>('all')
const [search, setSearch] = useState('')
const visible = useMemo(() => {
const term = search.trim().toLowerCase()
return services.filter((service) => {
if (term && !service.name.includes(term) && !service.consolePages.some((page) => page.displayName.toLowerCase().includes(term))) return false
if (filter === 'not-running') return service.state !== 'running'
if (filter === 'with-page') return service.consolePages.length > 0
if (filter === 'without-page') return service.consolePages.length === 0
return true
})
}, [services, filter, search])
return (
<section className="table-panel" aria-labelledby="diagnostics-emulated-services">
<header className="diagnostics-section-header">
<h2 id="diagnostics-emulated-services">Emulated services</h2>
<div className="diagnostics-filters">
<div className="diagnostics-filter-group" role="group" aria-label="Filter emulated services">
{SERVICE_FILTERS.map((option) => (
<button key={option.id} className={`button compact ${filter === option.id ? 'primary' : ''}`} type="button" aria-pressed={filter === option.id} onClick={() => setFilter(option.id)}>{option.label}</button>
))}
</div>
<input className="input" type="search" value={search} onChange={(event) => setSearch(event.target.value)} placeholder="Filter services" aria-label="Filter emulated services by name"/>
</div>
</header>
{visible.length === 0
? <p className="muted padded">No emulated services match this filter.</p>
: <ul className="diagnostics-service-grid">{visible.map((service) => (
<li key={service.name} className="diagnostics-service">
<span className={`dot ${service.state === 'running' ? 'healthy' : 'unavailable'}`} aria-hidden="true"/>
<code>{service.name}</code>
{service.state !== 'running' && <span className="muted">{service.state}</span>}
{service.consolePages.map((page) => <Link key={page.service} className="diagnostics-page-link" to={page.route}>{page.displayName}</Link>)}
</li>
))}</ul>}
</section>
)
}
function ContainersTable({report}: {report: RuntimeDiagnostics}) {
const containers = report.containers
return (
<section className="table-panel" aria-labelledby="diagnostics-containers">
<header className="diagnostics-section-header">
<h2 id="diagnostics-containers">Runtime containers</h2>
<span className="muted">Containers started for EC2 instances</span>
</header>
{containers === null && <p className="muted padded">This runtime does not publish container diagnostics.</p>}
{containers?.length === 0 && <p className="muted padded">No EC2 instance containers are running.</p>}
{containers && containers.length > 0 && <table className="table">
<thead><tr><th>Instance</th><th>Container</th><th>SSH</th></tr></thead>
<tbody>{containers.map((container) => (
<tr key={container.resourceId}>
<td><Link to={`/cloud-explorer/${report.cloud}/compute?resource=${encodeURIComponent(container.resourceId)}`}><code>{container.resourceId}</code></Link></td>
<td>{container.status ?? '—'}</td>
<td>{container.sshPort ? <code>{container.sshHost ?? 'localhost'}:{container.sshPort}</code> : '—'}</td>
</tr>
))}</tbody>
</table>}
</section>
)
}
function ConsoleProcess({report}: {report: RuntimeDiagnostics}) {
return (
<section className="table-panel" aria-labelledby="diagnostics-console">
<header className="diagnostics-section-header">
<h2 id="diagnostics-console">Console</h2>
</header>
<dl className="diagnostics-facts">
<div><dt>Uptime</dt><dd>{formatUptime(report.console.uptimeSeconds)}</dd></div>
<div><dt>Started</dt><dd>{formatDateTime(report.console.startedAt) ?? report.console.startedAt}</dd></div>
<div><dt>Sign-in</dt><dd>{report.console.authMode === 'oidc' ? 'OIDC single sign-on' : 'Local administrator'}</dd></div>
<div><dt>Active sessions</dt><dd>{report.console.activeSessions}</dd></div>
</dl>
</section>
)
}
function formatUptime(totalSeconds: number): string {
const days = Math.floor(totalSeconds / 86_400)
const hours = Math.floor((totalSeconds % 86_400) / 3_600)
const minutes = Math.floor((totalSeconds % 3_600) / 60)
if (days > 0) return `${days} d ${hours} h`
if (hours > 0) return `${hours} h ${minutes} min`
if (minutes > 0) return `${minutes} min`
return `${totalSeconds} s`
}
function statusClass(runtime: CloudServiceStatus['runtime']): string {
if (runtime === 'reachable') return 'healthy'
if (runtime === 'coming_soon') return 'degraded'
return 'unavailable'
}
function statusLabel(runtime: CloudServiceStatus['runtime']): string {
if (runtime === 'reachable') return 'Reachable'
if (runtime === 'coming_soon') return 'Not registered'
return 'Unavailable'
}
function normalizeCloud(value?: string): CloudProvider | null {
return value === 'aws' || value === 'azure' || value === 'gcp' ? value : null
}
+36
View File
@@ -83,3 +83,39 @@ export interface CloudServiceStatus {
/** Mapped error code, so "not implemented" reads differently from "down". */
errorCode: string | null
}
export interface EmulatedServiceStatus {
name: string
state: string
consolePages: Array<{service: CloudServiceType; displayName: string; route: string}>
}
export interface RuntimeContainer {
kind: 'ec2-instance'
resourceId: string
status: string | null
sshHost: string | null
sshPort: number | null
}
export interface RuntimeDiagnostics {
cloud: CloudProvider
checkedAt: string
runtime: {
status: RuntimeReachability
endpoint: string | null
latencyMs: number | null
error: string | null
version: string | null
edition: string | null
}
emulatedServices: EmulatedServiceStatus[] | null
consoleServices: CloudServiceStatus[]
containers: RuntimeContainer[] | null
console: {
startedAt: string
uptimeSeconds: number
authMode: 'local' | 'oidc'
activeSessions: number
}
}