mirror of
https://github.com/pacnpal/thrilltrack-explorer.git
synced 2025-12-20 17:31:13 -05:00
Migrate 3 Phase 2 monitoring functions (collect-metrics, detect-anomalies, monitor-rate-limits) to use wrapEdgeFunction with smaller batch updates, replacing manual handlers, adding shared logging/tracing, and standardizing error handling.
275 lines
8.0 KiB
TypeScript
275 lines
8.0 KiB
TypeScript
/**
|
|
* Rate Limit Monitor
|
|
*
|
|
* Periodically checks rate limit metrics against configured thresholds
|
|
* and triggers alerts when limits are exceeded.
|
|
*
|
|
* Designed to run as a cron job every 5 minutes.
|
|
*/
|
|
|
|
import { createClient } from 'jsr:@supabase/supabase-js@2';
|
|
import { createEdgeFunction } from '../_shared/edgeFunctionWrapper.ts';
|
|
import { edgeLogger } from '../_shared/logger.ts';
|
|
import { getMetricsStats } from '../_shared/rateLimitMetrics.ts';
|
|
|
|
interface AlertConfig {
|
|
id: string;
|
|
metric_type: 'block_rate' | 'total_requests' | 'unique_ips' | 'function_specific';
|
|
threshold_value: number;
|
|
time_window_ms: number;
|
|
function_name?: string;
|
|
enabled: boolean;
|
|
}
|
|
|
|
interface AlertCheck {
|
|
configId: string;
|
|
metricType: string;
|
|
metricValue: number;
|
|
thresholdValue: number;
|
|
timeWindowMs: number;
|
|
functionName?: string;
|
|
exceeded: boolean;
|
|
message: string;
|
|
}
|
|
|
|
async function checkAlertConditions(configs: AlertConfig[]): Promise<AlertCheck[]> {
|
|
const checks: AlertCheck[] = [];
|
|
|
|
for (const config of configs) {
|
|
if (!config.enabled) continue;
|
|
|
|
const stats = getMetricsStats(config.time_window_ms);
|
|
let metricValue = 0;
|
|
let exceeded = false;
|
|
let message = '';
|
|
|
|
switch (config.metric_type) {
|
|
case 'block_rate':
|
|
metricValue = stats.blockRate;
|
|
exceeded = metricValue > config.threshold_value;
|
|
message = `Rate limit block rate (${(metricValue * 100).toFixed(1)}%) exceeded threshold (${(config.threshold_value * 100).toFixed(1)}%) in last ${config.time_window_ms / 1000}s. ${stats.blockedRequests} of ${stats.totalRequests} requests blocked.`;
|
|
break;
|
|
|
|
case 'total_requests':
|
|
metricValue = stats.totalRequests;
|
|
exceeded = metricValue > config.threshold_value;
|
|
message = `Total requests (${metricValue}) exceeded threshold (${config.threshold_value}) in last ${config.time_window_ms / 1000}s.`;
|
|
break;
|
|
|
|
case 'unique_ips':
|
|
metricValue = stats.uniqueIPs;
|
|
exceeded = metricValue > config.threshold_value;
|
|
message = `Unique IPs (${metricValue}) exceeded threshold (${config.threshold_value}) in last ${config.time_window_ms / 1000}s. Possible DDoS attack.`;
|
|
break;
|
|
|
|
case 'function_specific':
|
|
// For function-specific alerts, we'd need to track metrics per function
|
|
// This would require enhancing the metrics system
|
|
console.log('Function-specific alerts not yet implemented');
|
|
continue;
|
|
}
|
|
|
|
checks.push({
|
|
configId: config.id,
|
|
metricType: config.metric_type,
|
|
metricValue,
|
|
thresholdValue: config.threshold_value,
|
|
timeWindowMs: config.time_window_ms,
|
|
functionName: config.function_name,
|
|
exceeded,
|
|
message,
|
|
});
|
|
}
|
|
|
|
return checks;
|
|
}
|
|
|
|
async function recordAlert(
|
|
supabase: any,
|
|
check: AlertCheck
|
|
): Promise<{ success: boolean; error?: string }> {
|
|
try {
|
|
const { error } = await supabase
|
|
.from('rate_limit_alerts')
|
|
.insert({
|
|
config_id: check.configId,
|
|
metric_type: check.metricType,
|
|
metric_value: check.metricValue,
|
|
threshold_value: check.thresholdValue,
|
|
time_window_ms: check.timeWindowMs,
|
|
function_name: check.functionName,
|
|
alert_message: check.message,
|
|
});
|
|
|
|
if (error) {
|
|
console.error('Failed to record alert:', error);
|
|
return { success: false, error: error.message };
|
|
}
|
|
|
|
return { success: true };
|
|
} catch (error) {
|
|
console.error('Exception recording alert:', error);
|
|
return {
|
|
success: false,
|
|
error: error instanceof Error ? error.message : 'Unknown error'
|
|
};
|
|
}
|
|
}
|
|
|
|
async function sendNotification(
|
|
supabase: any,
|
|
check: AlertCheck
|
|
): Promise<{ success: boolean; error?: string }> {
|
|
try {
|
|
// Send notification to moderators via the moderator topic
|
|
const { data, error } = await supabase.functions.invoke('trigger-notification', {
|
|
body: {
|
|
workflowId: 'rate-limit-alert',
|
|
topicKey: 'moderators',
|
|
payload: {
|
|
message: check.message,
|
|
metricType: check.metricType,
|
|
metricValue: check.metricValue,
|
|
thresholdValue: check.thresholdValue,
|
|
functionName: check.functionName || 'all',
|
|
},
|
|
overrides: {
|
|
email: {
|
|
subject: '🚨 Rate Limit Alert',
|
|
},
|
|
},
|
|
},
|
|
});
|
|
|
|
if (error) {
|
|
console.error('Failed to send notification:', error);
|
|
return { success: false, error: error.message };
|
|
}
|
|
|
|
return { success: true };
|
|
} catch (error) {
|
|
console.error('Exception sending notification:', error);
|
|
return {
|
|
success: false,
|
|
error: error instanceof Error ? error.message : 'Unknown error'
|
|
};
|
|
}
|
|
}
|
|
|
|
export default createEdgeFunction(
|
|
{
|
|
name: 'monitor-rate-limits',
|
|
requireAuth: false,
|
|
},
|
|
async (req, context, supabase) => {
|
|
const startTime = Date.now();
|
|
edgeLogger.info('Rate limit monitor starting', { requestId: context.requestId });
|
|
|
|
// Fetch enabled alert configurations
|
|
const { data: configs, error: configError } = await supabase
|
|
.from('rate_limit_alert_config')
|
|
.select('*')
|
|
.eq('enabled', true);
|
|
|
|
if (configError) {
|
|
edgeLogger.error('Failed to fetch alert configs', {
|
|
error: configError,
|
|
requestId: context.requestId
|
|
});
|
|
throw configError;
|
|
}
|
|
|
|
if (!configs || configs.length === 0) {
|
|
edgeLogger.info('No enabled alert configurations found', {
|
|
requestId: context.requestId
|
|
});
|
|
return new Response(
|
|
JSON.stringify({
|
|
success: true,
|
|
message: 'No enabled alert configurations',
|
|
checked: 0
|
|
}),
|
|
{ headers: { 'Content-Type': 'application/json' } }
|
|
);
|
|
}
|
|
|
|
edgeLogger.info('Checking alert configurations', {
|
|
count: configs.length,
|
|
requestId: context.requestId
|
|
});
|
|
|
|
// Check all alert conditions
|
|
const checks = await checkAlertConditions(configs);
|
|
const exceededChecks = checks.filter(c => c.exceeded);
|
|
|
|
edgeLogger.info('Threshold violations found', {
|
|
count: exceededChecks.length,
|
|
requestId: context.requestId
|
|
});
|
|
|
|
// Process exceeded thresholds
|
|
const alertResults = [];
|
|
for (const check of exceededChecks) {
|
|
edgeLogger.info('Processing alert', {
|
|
message: check.message,
|
|
requestId: context.requestId
|
|
});
|
|
|
|
// Check if we've already sent a recent alert for this config
|
|
const { data: recentAlerts } = await supabase
|
|
.from('rate_limit_alerts')
|
|
.select('created_at')
|
|
.eq('config_id', check.configId)
|
|
.is('resolved_at', null)
|
|
.gte('created_at', new Date(Date.now() - 15 * 60 * 1000).toISOString()) // Last 15 minutes
|
|
.order('created_at', { ascending: false })
|
|
.limit(1);
|
|
|
|
if (recentAlerts && recentAlerts.length > 0) {
|
|
edgeLogger.info('Skipping alert - recent unresolved alert exists', {
|
|
configId: check.configId,
|
|
requestId: context.requestId
|
|
});
|
|
alertResults.push({
|
|
configId: check.configId,
|
|
skipped: true,
|
|
reason: 'Recent alert exists',
|
|
});
|
|
continue;
|
|
}
|
|
|
|
// Record the alert
|
|
const recordResult = await recordAlert(supabase, check);
|
|
|
|
// Send notification
|
|
const notifyResult = await sendNotification(supabase, check);
|
|
|
|
alertResults.push({
|
|
configId: check.configId,
|
|
metricType: check.metricType,
|
|
recorded: recordResult.success,
|
|
notified: notifyResult.success,
|
|
recordError: recordResult.error,
|
|
notifyError: notifyResult.error,
|
|
});
|
|
}
|
|
|
|
const duration = Date.now() - startTime;
|
|
edgeLogger.info('Monitor completed', {
|
|
duration,
|
|
requestId: context.requestId
|
|
});
|
|
|
|
return new Response(
|
|
JSON.stringify({
|
|
success: true,
|
|
checked: configs.length,
|
|
exceeded: exceededChecks.length,
|
|
alerts: alertResults,
|
|
duration_ms: duration,
|
|
}),
|
|
{ headers: { 'Content-Type': 'application/json' } }
|
|
);
|
|
}
|
|
);
|