feat: collect model performance metrics (#4635)

web/default/src/features/system-settings/maintenance/config.ts

@@ -75,6 +75,10 @@ export const DEFAULT_MAINTENANCE_SETTINGS: MaintenanceSettings = {
   'performance_setting.monitor_cpu_threshold': 90,
   'performance_setting.monitor_memory_threshold': 90,
   'performance_setting.monitor_disk_threshold': 95,
+  'perf_metrics_setting.enabled': true,
+  'perf_metrics_setting.flush_interval': 5,
+  'perf_metrics_setting.bucket_time': 'hour',
+  'perf_metrics_setting.retention_days': 0,
 }
 
 const toBoolean = (value: unknown, fallback: boolean): boolean => {

web/default/src/features/system-settings/maintenance/performance-section.tsx

@@ -59,6 +59,10 @@ const perfSchema = z.object({
     .number()
     .min(0)
     .max(100),
+  'perf_metrics_setting.enabled': z.boolean(),
+  'perf_metrics_setting.flush_interval': z.coerce.number().min(1),
+  'perf_metrics_setting.bucket_time': z.enum(['minute', '5min', 'hour']),
+  'perf_metrics_setting.retention_days': z.coerce.number().min(0),
 })
 
 type PerfFormValues = z.infer<typeof perfSchema>
@@ -248,6 +252,7 @@ export function PerformanceSection(props: Props) {
 
   const diskEnabled = form.watch('performance_setting.disk_cache_enabled')
   const monitorEnabled = form.watch('performance_setting.monitor_enabled')
+  const perfMetricsEnabled = form.watch('perf_metrics_setting.enabled')
   const maxCacheSizeMb = form.watch(
     'performance_setting.disk_cache_max_size_mb'
   )
@@ -452,6 +457,97 @@ export function PerformanceSection(props: Props) {
             />
           </div>
 
+          <Separator />
+
+          <div>
+            <h4 className='font-medium'>{t('Model performance metrics')}</h4>
+            <p className='text-muted-foreground mt-1 text-xs'>
+              {t(
+                'Collect relay latency and success-rate metrics for the model square.'
+              )}
+            </p>
+          </div>
+
+          <div className='grid grid-cols-1 gap-4 md:grid-cols-4'>
+            <FormField
+              control={form.control}
+              name='perf_metrics_setting.enabled'
+              render={({ field }) => (
+                <FormItem className='flex items-center gap-2'>
+                  <FormControl>
+                    <Switch
+                      checked={field.value}
+                      onCheckedChange={field.onChange}
+                    />
+                  </FormControl>
+                  <FormLabel>{t('Enable model performance metrics')}</FormLabel>
+                </FormItem>
+              )}
+            />
+            <FormField
+              control={form.control}
+              name='perf_metrics_setting.flush_interval'
+              render={({ field }) => (
+                <FormItem>
+                  <FormLabel>{t('Flush interval (minutes)')}</FormLabel>
+                  <FormControl>
+                    <Input
+                      type='number'
+                      min={1}
+                      {...field}
+                      disabled={!perfMetricsEnabled}
+                    />
+                  </FormControl>
+                </FormItem>
+              )}
+            />
+            <FormField
+              control={form.control}
+              name='perf_metrics_setting.bucket_time'
+              render={({ field }) => (
+                <FormItem>
+                  <FormLabel>{t('Aggregation bucket')}</FormLabel>
+                  <Select
+                    value={field.value}
+                    onValueChange={field.onChange}
+                    disabled={!perfMetricsEnabled}
+                  >
+                    <FormControl>
+                      <SelectTrigger>
+                        <SelectValue />
+                      </SelectTrigger>
+                    </FormControl>
+                    <SelectContent>
+                      <SelectItem value='minute'>{t('1 minute')}</SelectItem>
+                      <SelectItem value='5min'>{t('5 minutes')}</SelectItem>
+                      <SelectItem value='hour'>{t('1 hour')}</SelectItem>
+                    </SelectContent>
+                  </Select>
+                </FormItem>
+              )}
+            />
+            <FormField
+              control={form.control}
+              name='perf_metrics_setting.retention_days'
+              render={({ field }) => (
+                <FormItem>
+                  <FormLabel>{t('Retention days')}</FormLabel>
+                  <FormControl>
+                    <Input
+                      type='number'
+                      min={0}
+                      {...field}
+                      disabled={!perfMetricsEnabled}
+                    />
+                  </FormControl>
+                  <FormDescription>
+                    {t('0 means data is kept permanently')}
+                  </FormDescription>
+                </FormItem>
+              )}
+            />
+          </div>
+
           <Button type='submit' disabled={updateOption.isPending}>
             {updateOption.isPending ? t('Saving...') : t('Save Changes')}
           </Button>

web/default/src/features/system-settings/maintenance/section-registry.tsx

@@ -102,6 +102,14 @@ const MAINTENANCE_SECTIONS = [
             settings['performance_setting.monitor_memory_threshold'] ?? 90,
           'performance_setting.monitor_disk_threshold':
             settings['performance_setting.monitor_disk_threshold'] ?? 95,
+          'perf_metrics_setting.enabled':
+            settings['perf_metrics_setting.enabled'] ?? true,
+          'perf_metrics_setting.flush_interval':
+            settings['perf_metrics_setting.flush_interval'] ?? 5,
+          'perf_metrics_setting.bucket_time':
+            settings['perf_metrics_setting.bucket_time'] ?? 'hour',
+          'perf_metrics_setting.retention_days':
+            settings['perf_metrics_setting.retention_days'] ?? 0,
         }}
       />
     ),

web/default/src/features/system-settings/types.ts

@@ -254,6 +254,10 @@ export type MaintenanceSettings = {
   'performance_setting.monitor_cpu_threshold': number
   'performance_setting.monitor_memory_threshold': number
   'performance_setting.monitor_disk_threshold': number
+  'perf_metrics_setting.enabled': boolean
+  'perf_metrics_setting.flush_interval': number
+  'perf_metrics_setting.bucket_time': 'hour' | 'minute' | '5min'
+  'perf_metrics_setting.retention_days': number
 }
 
 export type RequestLimitsSettings = {