From e5e5be37a36e5f7a222449504da982e6db574d75 Mon Sep 17 00:00:00 2001
From: Nathan Coad <nathan@thecoads.com>
Date: Thu, 15 Jan 2026 16:02:58 +1100
Subject: [PATCH] handle crashes better

---
 internal/tasks/cronstatus.go         | 9 +++++++++
 internal/tasks/inventorySnapshots.go | 7 ++++++-
 2 files changed, 15 insertions(+), 1 deletion(-)

diff --git a/internal/tasks/cronstatus.go b/internal/tasks/cronstatus.go
index 23a0c8a..cbd0c6f 100644
--- a/internal/tasks/cronstatus.go
+++ b/internal/tasks/cronstatus.go
@@ -21,6 +21,15 @@ func NewCronTracker(database db.Database) *CronTracker {
 	}
 }
 
+// ClearAllInProgress resets any stuck in-progress flags (e.g., after crashes).
+func (c *CronTracker) ClearAllInProgress(ctx context.Context) error {
+	if err := c.ensureTable(ctx); err != nil {
+		return err
+	}
+	_, err := c.db.DB().ExecContext(ctx, `UPDATE cron_status SET in_progress = FALSE`)
+	return err
+}
+
 func (c *CronTracker) ensureTable(ctx context.Context) error {
 	conn := c.db.DB()
 	driver := conn.DriverName()
diff --git a/internal/tasks/inventorySnapshots.go b/internal/tasks/inventorySnapshots.go
index 71de41b..e6eaacd 100644
--- a/internal/tasks/inventorySnapshots.go
+++ b/internal/tasks/inventorySnapshots.go
@@ -57,11 +57,16 @@ func (c *CronTask) RunVcenterSnapshotHourly(ctx context.Context, logger *slog.Lo
 		jobCtx, cancel = context.WithTimeout(ctx, jobTimeout)
 		defer cancel()
 	}
+	tracker := NewCronTracker(c.Database)
+	// Clear any stale in-progress markers (e.g., after a crash) before attempting the run.
+	if err := tracker.ClearAllInProgress(jobCtx); err != nil {
+		logger.Warn("failed to clear stale cron status", "error", err)
+	}
+
 	startedAt := time.Now()
 	defer func() {
 		logger.Info("Hourly snapshot job finished", "duration", time.Since(startedAt))
 	}()
-	tracker := NewCronTracker(c.Database)
 	done, skip, err := tracker.Start(jobCtx, "hourly_snapshot")
 	if err != nil {
 		return err