mirror of https://github.com/amir20/dozzle.git synced 2026-06-23 04:10:12 +00:00

Files

T

Amir Raminfar a79ffdaf50 feat: add metric-based alerts for container CPU/memory thresholds (#4454 )

Co-authored-by: Dhaval Patel <dhavu262@gmail.com>
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>

2026-02-16 20:55:35 +00:00

Bug Hunter Agent Memory

Codebase Patterns

notification.Manager owns subscriptions (xsync.Map) and dispatchers (xsync.Map)
Two processing goroutines: processLogEvents and processStatEvents started in NewManager
Stats listener has start/stop lifecycle; log listener is always-on once started
MultiHostService wraps Manager and handles config persistence + agent broadcast

broadcastNotificationConfig: Has historically missed fields when converting between internal and types packages (APIKey, Prefix, ExpiresAt were missed)
TriggeredContainerIDs lazy init: Race condition risk - initialized lazily in AddTriggeredContainer without sync
Channel close handling: enrich() in stats_listener doesn't check for closed channel, can hot-spin
LogAlertFields canSave: Allows empty logExpression (no error = valid), creates dead subscriptions

container.ContainerStat -> types.NotificationStat: field names differ (CPUPercent vs cpu expr tag)
notification.DispatcherConfig -> types.DispatcherConfig: must copy ALL fields including APIKey, Prefix, ExpiresAt
Frontend NotificationRule.cooldown is optional, backend defaults to 300 via GetCooldownSeconds()

xsync.Map used throughout for concurrent access (subscriptions, dispatchers, activeStreams, containers)
Subscription fields use atomic types: TriggerCount (atomic.Int64), LastTriggeredAt (atomic.Pointer)
MetricCooldowns uses xsync.Map for per-container cooldown tracking
sendSem (semaphore.Weighted=5) limits concurrent notification sends