feat: Add retry option for monitor checks

1 files changed, 30 insertions, 1 deletions

diff --git a/internal/monitor/scheduler.go b/internal/monitor/scheduler.go
index 7a06131..5a7e817 100644
--- a/internal/monitor/scheduler.go
+++ b/internal/monitor/scheduler.go
@@ -100,7 +100,36 @@ func (s *Scheduler) executeCheck(mon Monitor) {
 	checkCtx, cancel := context.WithTimeout(s.ctx, mon.Interval())
 	defer cancel()
 
-	result := mon.Check(checkCtx)
+	var result *Result
+	retries := mon.Retries()
+
+	// Try the check, with retries if configured
+	for attempt := 0; attempt <= retries; attempt++ {
+		result = mon.Check(checkCtx)
+
+		// If check succeeded (up or degraded), no need to retry
+		if result.Status == StatusUp || result.Status == StatusDegraded {
+			break
+		}
+
+		// If this wasn't the last attempt, log and retry
+		if attempt < retries {
+			s.logger.Debug("check failed, retrying",
+				"name", mon.Name(),
+				"attempt", attempt+1,
+				"max_retries", retries,
+				"error", result.Error)
+
+			// Small delay before retry (500ms)
+			select {
+			case <-checkCtx.Done():
+				// Context cancelled, abort retries
+				break
+			case <-time.After(500 * time.Millisecond):
+				// Continue to next retry
+			}
+		}
+	}
 
 	// Log the result
 	logAttrs := []any{