Files
cameleer-server/ui/src/pages/AgentHealth/AgentHealth.tsx

320 lines
12 KiB
TypeScript
Raw Normal View History

import { useMemo, useState } from 'react';
import { useParams, useNavigate } from 'react-router';
import {
StatCard, StatusDot, Badge, MonoText,
GroupCard, EventFeed, Breadcrumb, Alert,
DetailPanel, ProgressBar, LineChart,
} from '@cameleer/design-system';
import styles from './AgentHealth.module.css';
import { useAgents, useAgentEvents } from '../../api/queries/agents';
import { useRouteCatalog } from '../../api/queries/catalog';
import { useAgentMetrics } from '../../api/queries/agent-metrics';
function formatUptime(seconds?: number): string {
if (!seconds) return '—';
const days = Math.floor(seconds / 86400);
const hours = Math.floor((seconds % 86400) / 3600);
const mins = Math.floor((seconds % 3600) / 60);
if (days > 0) return `${days}d ${hours}h`;
if (hours > 0) return `${hours}h ${mins}m`;
return `${mins}m`;
}
function formatRelativeTime(iso?: string): string {
if (!iso) return '—';
const diff = Date.now() - new Date(iso).getTime();
const mins = Math.floor(diff / 60000);
if (mins < 1) return 'just now';
if (mins < 60) return `${mins}m ago`;
const hours = Math.floor(mins / 60);
if (hours < 24) return `${hours}h ago`;
return `${Math.floor(hours / 24)}d ago`;
}
function AgentOverviewContent({ agent }: { agent: any }) {
const { data: memMetrics } = useAgentMetrics(
agent.id,
['jvm.memory.heap.used', 'jvm.memory.heap.max'],
1,
);
const { data: cpuMetrics } = useAgentMetrics(agent.id, ['jvm.cpu.process'], 1);
const cpuValue = cpuMetrics?.metrics?.['jvm.cpu.process']?.[0]?.value;
const heapUsed = memMetrics?.metrics?.['jvm.memory.heap.used']?.[0]?.value;
const heapMax = memMetrics?.metrics?.['jvm.memory.heap.max']?.[0]?.value;
const heapPercent = heapUsed != null && heapMax != null && heapMax > 0
? Math.round((heapUsed / heapMax) * 100)
: undefined;
const cpuPercent = cpuValue != null ? Math.round(cpuValue * 100) : undefined;
const statusVariant: 'live' | 'stale' | 'dead' =
agent.status === 'LIVE' ? 'live' : agent.status === 'STALE' ? 'stale' : 'dead';
const statusColor: 'success' | 'warning' | 'error' =
agent.status === 'LIVE' ? 'success' : agent.status === 'STALE' ? 'warning' : 'error';
return (
<div className={styles.overviewContent}>
<div className={styles.overviewRow}>
<StatusDot variant={statusVariant} />
<Badge label={agent.status} color={statusColor} />
</div>
<dl className={styles.detailList}>
<div className={styles.detailRow}>
<dt>Application</dt>
<dd><MonoText>{agent.group ?? '—'}</MonoText></dd>
</div>
<div className={styles.detailRow}>
<dt>Version</dt>
<dd><MonoText>{agent.version ?? '—'}</MonoText></dd>
</div>
<div className={styles.detailRow}>
<dt>Uptime</dt>
<dd>{formatUptime(agent.uptimeSeconds)}</dd>
</div>
<div className={styles.detailRow}>
<dt>Last Heartbeat</dt>
<dd>{formatRelativeTime(agent.lastHeartbeat)}</dd>
</div>
<div className={styles.detailRow}>
<dt>TPS</dt>
<dd>{agent.tps != null ? (agent.tps as number).toFixed(2) : '—'}</dd>
</div>
<div className={styles.detailRow}>
<dt>Error Rate</dt>
<dd>{agent.errorRate != null ? `${((agent.errorRate as number) * 100).toFixed(1)}%` : '—'}</dd>
</div>
<div className={styles.detailRow}>
<dt>Routes</dt>
<dd>{agent.activeRoutes ?? '—'} active / {agent.totalRoutes ?? '—'} total</dd>
</div>
</dl>
<div className={styles.metricsSection}>
<div className={styles.metricLabel}>
Heap Memory{heapUsed != null && heapMax != null
? `${Math.round(heapUsed / 1024 / 1024)}MB / ${Math.round(heapMax / 1024 / 1024)}MB`
: ''}
</div>
<ProgressBar
value={heapPercent}
variant={heapPercent == null ? 'primary' : heapPercent > 85 ? 'error' : heapPercent > 70 ? 'warning' : 'success'}
indeterminate={heapPercent == null}
size="sm"
/>
</div>
<div className={styles.metricsSection}>
<div className={styles.metricLabel}>
CPU Usage{cpuPercent != null ? `${cpuPercent}%` : ''}
</div>
<ProgressBar
value={cpuPercent}
variant={cpuPercent == null ? 'primary' : cpuPercent > 80 ? 'error' : cpuPercent > 60 ? 'warning' : 'success'}
indeterminate={cpuPercent == null}
size="sm"
/>
</div>
</div>
);
}
function AgentPerformanceContent({ agent }: { agent: any }) {
const { data: tpsMetrics } = useAgentMetrics(agent.id, ['cameleer.tps'], 60);
const { data: errMetrics } = useAgentMetrics(agent.id, ['cameleer.error.rate'], 60);
const tpsSeries = useMemo(() => {
const raw = tpsMetrics?.metrics?.['cameleer.tps'] ?? [];
return [{
label: 'TPS',
data: raw.map((p) => ({ x: new Date(p.time), y: p.value })),
}];
}, [tpsMetrics]);
const errSeries = useMemo(() => {
const raw = errMetrics?.metrics?.['cameleer.error.rate'] ?? [];
return [{
label: 'Error Rate',
data: raw.map((p) => ({ x: new Date(p.time), y: p.value * 100 })),
}];
}, [errMetrics]);
return (
<div className={styles.performanceContent}>
<div className={styles.chartSection}>
<div className={styles.chartLabel}>Throughput (TPS)</div>
{tpsSeries[0].data.length > 0 ? (
<LineChart series={tpsSeries} yLabel="req/s" height={160} />
) : (
<div className={styles.emptyChart}>No data available</div>
)}
</div>
<div className={styles.chartSection}>
<div className={styles.chartLabel}>Error Rate (%)</div>
{errSeries[0].data.length > 0 ? (
<LineChart series={errSeries} yLabel="%" height={160} />
) : (
<div className={styles.emptyChart}>No data available</div>
)}
</div>
</div>
);
}
export default function AgentHealth() {
const { appId } = useParams();
const navigate = useNavigate();
const { data: agents } = useAgents(undefined, appId);
const { data: catalog } = useRouteCatalog();
const { data: events } = useAgentEvents(appId);
const [selectedAgent, setSelectedAgent] = useState<any>(null);
const agentsByApp = useMemo(() => {
const map: Record<string, any[]> = {};
(agents || []).forEach((a: any) => {
const g = a.group;
if (!map[g]) map[g] = [];
map[g].push(a);
});
return map;
}, [agents]);
const liveCount = (agents || []).filter((a: any) => a.status === 'LIVE').length;
const staleCount = (agents || []).filter((a: any) => a.status === 'STALE').length;
const deadCount = (agents || []).filter((a: any) => a.status === 'DEAD').length;
const uniqueApps = new Set((agents || []).map((a: any) => a.group)).size;
const activeRoutes = (agents || []).filter((a: any) => a.status === 'LIVE').reduce((sum: number, a: any) => sum + (a.activeRoutes || 0), 0);
const totalTps = (agents || []).filter((a: any) => a.status === 'LIVE').reduce((sum: number, a: any) => sum + (a.tps || 0), 0);
const groupHealth: 'live' | 'stale' | 'dead' = useMemo(() => {
if (!appId) return 'live';
const groupAgents = agentsByApp[appId] || [];
if (groupAgents.some((a: any) => a.status === 'DEAD')) return 'dead';
if (groupAgents.some((a: any) => a.status === 'STALE')) return 'stale';
return 'live';
}, [appId, agentsByApp]);
const scopeItems = useMemo(() => {
const items: { label: string; href?: string }[] = [
{ label: 'Agent Health', href: '/agents' },
];
if (appId) {
items.push({ label: appId });
}
return items;
}, [appId]);
const feedEvents = useMemo(() =>
(events || []).map((e: any) => ({
id: String(e.id),
severity: e.eventType === 'WENT_DEAD' ? 'error' as const
: e.eventType === 'WENT_STALE' ? 'warning' as const
: e.eventType === 'RECOVERED' ? 'success' as const
: 'running' as const,
message: `${e.agentId}: ${e.eventType}${e.detail ? ' — ' + e.detail : ''}`,
timestamp: new Date(e.timestamp),
})),
[events],
);
const apps = appId ? { [appId]: agentsByApp[appId] || [] } : agentsByApp;
return (
<div>
<div className={styles.statStrip}>
<StatCard label="Total Agents" value={(agents || []).length} detail={`${liveCount} live / ${staleCount} stale / ${deadCount} dead`} />
<StatCard label="Applications" value={uniqueApps} />
<StatCard label="Active Routes" value={activeRoutes} />
<StatCard label="Total TPS" value={totalTps.toFixed(1)} />
<StatCard label="Dead" value={deadCount} accent={deadCount > 0 ? 'error' : undefined} />
</div>
<div className={styles.scopeTrail}>
<Breadcrumb items={scopeItems} />
{!appId && <Badge label={`${liveCount} live`} variant="outlined" />}
{appId && (
<Badge
label={groupHealth}
color={groupHealth === 'live' ? 'success' : groupHealth === 'stale' ? 'warning' : 'error'}
/>
)}
</div>
<div className={styles.groupGrid}>
{Object.entries(apps).map(([group, groupAgents]) => {
const deadInGroup = (groupAgents || []).filter((a: any) => a.status === 'DEAD');
return (
<GroupCard
key={group}
title={group}
headerRight={<Badge label={`${groupAgents?.length ?? 0} instances`} />}
accent={
groupAgents?.some((a: any) => a.status === 'DEAD') ? 'error'
: groupAgents?.some((a: any) => a.status === 'STALE') ? 'warning'
: 'success'
}
onClick={() => navigate(`/agents/${group}`)}
>
{deadInGroup.length > 0 && (
<Alert variant="error">{deadInGroup.length} instance(s) unreachable</Alert>
)}
{(groupAgents || []).map((agent: any) => (
<div
key={agent.id}
className={styles.instanceRow}
onClick={(e) => {
e.stopPropagation();
setSelectedAgent(agent);
navigate(`/agents/${group}/${agent.id}`);
}}
>
<StatusDot variant={agent.status === 'LIVE' ? 'live' : agent.status === 'STALE' ? 'stale' : 'dead'} />
<span className={styles.instanceName}>{agent.name}</span>
<Badge label={agent.status} color={agent.status === 'LIVE' ? 'success' : agent.status === 'STALE' ? 'warning' : 'error'} />
<span className={styles.instanceMeta}>{formatUptime(agent.uptimeSeconds)}</span>
{agent.tps != null && <span className={styles.instanceMeta}>{(agent.tps || 0).toFixed(1)} tps</span>}
{agent.errorRate != null && (
<span className={styles.instanceMeta}>{(agent.errorRate * 100).toFixed(1)}% err</span>
)}
<span className={styles.instanceMeta}>{formatRelativeTime(agent.lastHeartbeat)}</span>
<span className={styles.instanceLink} aria-label="View instance"></span>
</div>
))}
</GroupCard>
);
})}
</div>
{feedEvents.length > 0 && (
<div className={styles.eventCard}>
<div className={styles.eventCardHeader}>Event Log</div>
<EventFeed events={feedEvents} maxItems={100} />
</div>
)}
{selectedAgent && (
<DetailPanel
open={!!selectedAgent}
title={selectedAgent.name ?? selectedAgent.id}
onClose={() => setSelectedAgent(null)}
tabs={[
{
label: 'Overview',
value: 'overview',
content: <AgentOverviewContent agent={selectedAgent} />,
},
{
label: 'Performance',
value: 'performance',
content: <AgentPerformanceContent agent={selectedAgent} />,
},
]}
/>
)}
</div>
);
}