feat(dashboard): enhance risk summary and peak node display in dashboard overview

- Added risk summary metrics including critical alerts, high CPU/memory nodes, and lagging nodes to the dashboard overview.
- Introduced peak node metrics for busiest and riskiest nodes with detailed display.
- Updated types for dashboard overview to include risk and peak summaries.
- Implemented new components for displaying risk signals and peak cards.
- Enhanced unit tests to cover new risk and peak metrics in the dashboard overview.

fix(update): improve error messages for server upgrade processes

- Translated and clarified error messages related to server upgrades and binary uploads.
- Ensured consistent error handling and messaging throughout the upgrade process.

test(update): refactor test utilities for server upgrade

- Renamed roundTripFunc to serverUpdateRoundTripFunc for clarity in test code.
- Updated test cases to utilize the new naming convention.

feat(node-detail): add traffic structure distribution and health event timeline

- Implemented traffic breakdown visualization for status codes and top domains in node detail page.
- Added health event timeline to display active and resolved events for better observability.

feat(rank-chart): create reusable rank chart component

- Developed a new RankChart component for visualizing ranked data distributions.
- Integrated RankChart into node detail page for displaying status code and domain distributions.
This commit is contained in:
ryan
2026-03-14 12:12:38 +08:00
parent 16bb9e5879
commit 497289b462
9 changed files with 689 additions and 57 deletions
@@ -13,6 +13,7 @@ import { getDashboardOverview } from '@/features/dashboard/api/overview';
import type {
DashboardAlert,
DashboardNodeHealth,
DashboardPeakNode,
} from '@/features/dashboard/types';
import {
getNodeStatusLabel,
@@ -56,6 +57,84 @@ function getAlertVariant(
return 'info';
}
function formatPeakHour(value: string) {
if (!value) {
return '暂无';
}
const date = new Date(value);
if (Number.isNaN(date.getTime())) {
return '暂无';
}
return `${formatTrendHour(value)} - ${date
.getMinutes()
.toString()
.padStart(2, '0')}`;
}
function RiskSignal({
label,
value,
tone,
hint,
}: {
label: string;
value: number;
tone: 'danger' | 'warning' | 'info' | 'success';
hint: string;
}) {
const toneClass =
tone === 'danger'
? 'border-rose-400/30 bg-rose-500/10 text-rose-100'
: tone === 'warning'
? 'border-amber-400/30 bg-amber-500/10 text-amber-100'
: tone === 'success'
? 'border-emerald-400/30 bg-emerald-500/10 text-emerald-100'
: 'border-sky-400/30 bg-sky-500/10 text-sky-100';
return (
<div className={`rounded-3xl border px-4 py-4 ${toneClass}`}>
<p className="text-xs tracking-[0.22em] uppercase opacity-75">{label}</p>
<p className="mt-3 text-3xl font-semibold">{value}</p>
<p className="mt-2 text-sm opacity-80">{hint}</p>
</div>
);
}
function PeakCard({
label,
value,
hint,
}: {
label: string;
value: string;
hint: string;
}) {
return (
<div className="rounded-3xl border border-[var(--border-default)] bg-[var(--surface-elevated)] px-4 py-4">
<p className="text-xs tracking-[0.22em] text-[var(--foreground-muted)] uppercase">
{label}
</p>
<p className="mt-3 text-xl font-semibold text-[var(--foreground-primary)]">
{value}
</p>
<p className="mt-2 text-sm text-[var(--foreground-secondary)]">{hint}</p>
</div>
);
}
function formatPeakNode(node: DashboardPeakNode | null) {
if (!node) {
return {
title: '暂无',
hint: '当前没有可用节点数据',
};
}
return {
title: node.node_name,
hint: `请求 ${node.request_count} · 错误 ${node.error_count} · CPU ${formatPercent(node.cpu_usage_percent)}`,
};
}
function OverviewMetric({
label,
value,
@@ -224,6 +303,86 @@ export function DashboardOverview() {
</div>
</AppCard>
<div className="grid gap-6 xl:grid-cols-[1.15fr_0.85fr]">
<AppCard
title="风险态势"
description="把异常、容量压力和配置落后聚到同一层,方便先确定今天该盯哪里。"
>
<div className="grid gap-4 md:grid-cols-2 xl:grid-cols-3">
<RiskSignal
label="Critical"
value={overview.risk.critical_alerts}
tone="danger"
hint="当前仍在触发中的严重异常"
/>
<RiskSignal
label="Warning"
value={overview.risk.warning_alerts}
tone="warning"
hint="需要尽快介入但尚未到故障级"
/>
<RiskSignal
label="配置落后"
value={overview.risk.lagging_nodes}
tone="info"
hint="当前版本未追平全局激活配置"
/>
<RiskSignal
label="OpenResty 异常"
value={overview.risk.unhealthy_nodes}
tone="danger"
hint="运行态已出现不健康节点"
/>
<RiskSignal
label="高 CPU / 内存"
value={
overview.risk.high_cpu_nodes + overview.risk.high_memory_nodes
}
tone="warning"
hint={`${overview.risk.high_cpu_nodes} 个高 CPU · ${overview.risk.high_memory_nodes} 个高内存`}
/>
<RiskSignal
label="高存储压力"
value={overview.risk.high_storage_nodes}
tone="warning"
hint={`${overview.risk.offline_nodes} 个离线节点`}
/>
</div>
</AppCard>
<AppCard
title="峰值摘要"
description="快速回答什么时候最忙、什么时候最危险,以及该先看哪台节点。"
>
<div className="grid gap-4 md:grid-cols-2">
<PeakCard
label="请求峰值时段"
value={formatPeakHour(
overview.peaks.peak_request_hour.bucket_started_at,
)}
hint={`峰值请求 ${overview.peaks.peak_request_hour.request_count}`}
/>
<PeakCard
label="错误峰值时段"
value={formatPeakHour(
overview.peaks.peak_error_hour.bucket_started_at,
)}
hint={`峰值错误 ${overview.peaks.peak_error_hour.error_count}`}
/>
<PeakCard
label="最忙节点"
value={formatPeakNode(overview.peaks.busiest_node).title}
hint={formatPeakNode(overview.peaks.busiest_node).hint}
/>
<PeakCard
label="优先排查节点"
value={formatPeakNode(overview.peaks.riskiest_node).title}
hint={formatPeakNode(overview.peaks.riskiest_node).hint}
/>
</div>
</AppCard>
</div>
<div className="grid gap-6 xl:grid-cols-2">
<AppCard
title="24 小时请求趋势"
@@ -30,6 +30,42 @@ export interface DashboardConfig {
pending_nodes: number;
}
export interface DashboardRiskSummary {
critical_alerts: number;
warning_alerts: number;
info_alerts: number;
offline_nodes: number;
unhealthy_nodes: number;
lagging_nodes: number;
high_cpu_nodes: number;
high_memory_nodes: number;
high_storage_nodes: number;
}
export interface DashboardPeakHour {
bucket_started_at: string;
request_count: number;
error_count: number;
}
export interface DashboardPeakNode {
node_id: string;
node_name: string;
request_count: number;
error_count: number;
cpu_usage_percent: number;
active_event_count: number;
openresty_status: 'healthy' | 'unhealthy' | 'unknown';
storage_usage_percent: number;
}
export interface DashboardPeakSummary {
peak_request_hour: DashboardPeakHour;
peak_error_hour: DashboardPeakHour;
busiest_node: DashboardPeakNode | null;
riskiest_node: DashboardPeakNode | null;
}
export interface TrafficTrendPoint {
bucket_started_at: string;
request_count: number;
@@ -82,6 +118,8 @@ export interface DashboardOverview {
traffic: DashboardTraffic;
capacity: DashboardCapacity;
config: DashboardConfig;
risk: DashboardRiskSummary;
peaks: DashboardPeakSummary;
trends: DashboardTrends;
nodes: DashboardNodeHealth[];
active_alerts: DashboardAlert[];
@@ -8,6 +8,7 @@ import { useEffect, useMemo, useState } from 'react';
import { useForm, useWatch } from 'react-hook-form';
import { z } from 'zod';
import { RankChart } from '@/components/data/rank-chart';
import { TrendChart } from '@/components/data/trend-chart';
import { EmptyState } from '@/components/feedback/empty-state';
import { ErrorState } from '@/components/feedback/error-state';
@@ -152,6 +153,48 @@ function formatTrendHour(value: string) {
return `${date.getHours().toString().padStart(2, '0')}:00`;
}
function parseTrafficMap(value?: string | null) {
if (!value) {
return {} as Record<string, number>;
}
try {
const parsed = JSON.parse(value) as Record<string, number>;
return Object.entries(parsed).reduce<Record<string, number>>(
(result, [key, count]) => {
if (typeof count === 'number' && Number.isFinite(count)) {
result[key] = count;
}
return result;
},
{},
);
} catch {
return {} as Record<string, number>;
}
}
function aggregateTrafficBreakdown(
reports: NodeObservability['traffic_reports'],
field: 'status_codes_json' | 'top_domains_json',
) {
const summary = new Map<string, number>();
for (const report of reports) {
const parsed = parseTrafficMap(report[field]);
for (const [key, value] of Object.entries(parsed)) {
summary.set(key, (summary.get(key) ?? 0) + value);
}
}
return Array.from(summary.entries())
.sort((left, right) => {
if (right[1] === left[1]) {
return left[0].localeCompare(right[0]);
}
return right[1] - left[1];
})
.slice(0, 6)
.map(([label, value]) => ({ label, value }));
}
function getHealthEventVariant(
event: NodeObservability['health_events'][number],
): 'success' | 'warning' | 'danger' | 'info' {
@@ -467,6 +510,22 @@ export function NodeDetailPage({ nodeId }: { nodeId: string }) {
const activeHealthEvents =
observability?.health_events.filter((event) => event.status === 'active') ??
[];
const statusCodeDistribution = useMemo(
() =>
aggregateTrafficBreakdown(
observability?.traffic_reports ?? [],
'status_codes_json',
),
[observability?.traffic_reports],
);
const topDomains = useMemo(
() =>
aggregateTrafficBreakdown(
observability?.traffic_reports ?? [],
'top_domains_json',
),
[observability?.traffic_reports],
);
const memoryUsageRatio = formatUsageRatio(
latestMetricSnapshot?.memory_used_bytes,
latestMetricSnapshot?.memory_total_bytes,
@@ -908,6 +967,93 @@ export function NodeDetailPage({ nodeId }: { nodeId: string }) {
</AppCard>
</div>
<div className="grid gap-6 xl:grid-cols-[0.95fr_1.05fr]">
<AppCard
title="请求结构分布"
description="聚合最近 24 小时窗口上报,帮助判断错误集中在哪些状态码、流量集中在哪些域名。"
>
<div className="grid gap-6 xl:grid-cols-2">
<div>
<p className="mb-4 text-xs tracking-[0.2em] text-[var(--foreground-muted)] uppercase">
状态码分布
</p>
<RankChart
items={statusCodeDistribution}
color="#f59e0b"
emptyMessage="暂无状态码分布"
/>
</div>
<div>
<p className="mb-4 text-xs tracking-[0.2em] text-[var(--foreground-muted)] uppercase">
Top Domain
</p>
<RankChart
items={topDomains}
color="#2563eb"
emptyMessage="暂无域名分布"
/>
</div>
</div>
</AppCard>
<AppCard
title="健康事件时间线"
description="保留活动与已恢复事件,帮助判断问题是持续中、间歇性还是已经恢复。"
>
{observability?.health_events.length ? (
<div className="space-y-4">
{observability.health_events.slice(0, 8).map((event) => (
<div
key={`${event.event_type}-${event.last_triggered_at}-${event.status}`}
className="rounded-2xl border border-[var(--border-default)] bg-[var(--surface-elevated)] px-4 py-4"
>
<div className="flex flex-wrap items-center gap-2">
<StatusBadge
label={getHealthEventLabel(event)}
variant={getHealthEventVariant(event)}
/>
<StatusBadge
label={event.status === 'active' ? '活动中' : '已恢复'}
variant={
event.status === 'active' ? 'warning' : 'success'
}
/>
</div>
<p className="mt-3 text-sm text-[var(--foreground-secondary)]">
{event.message || '暂无详细消息'}
</p>
<div className="mt-3 grid gap-2 text-xs text-[var(--foreground-muted)] md:grid-cols-3">
<p>
首次触发:
{isMeaningfulTime(event.first_triggered_at)
? ` ${formatDateTime(event.first_triggered_at)}`
: ' —'}
</p>
<p>
最近触发:
{isMeaningfulTime(event.last_triggered_at)
? ` ${formatDateTime(event.last_triggered_at)}`
: ' —'}
</p>
<p>
恢复时间:
{isMeaningfulTime(event.resolved_at)
? ` ${formatDateTime(event.resolved_at)}`
: ' —'}
</p>
</div>
</div>
))}
</div>
) : (
<EmptyState
title="暂无健康事件时间线"
description="节点当前还没有上报可展示的健康事件记录。"
/>
)}
</AppCard>
</div>
<div className="grid gap-4 xl:grid-cols-3">
<AppCard title="更新模式">
<div className="space-y-3">