feat: enable default base dag retry policy

yottahmd · yottahmd · commit 6d729b55898e · 2026-03-15T00:59:01.000+09:00
diff --git a/internal/core/spec/loader_test.go b/internal/core/spec/loader_test.go
@@ -1435,6 +1435,26 @@ steps:
 		require.Equal(t, 600*time.Second, dag.Steps[0].Timeout)
 	})
 
+	t.Run("BaseConfigDAGRetryPolicy", func(t *testing.T) {
+		t.Parallel()
+
+		base := createTempYAMLFile(t, `
+retry_policy:
+  limit: 1
+  interval_sec: 60
+`)
+		child := createTempYAMLFile(t, `
+steps:
+  - name: step1
+    command: echo "test"
+`)
+		dag, err := spec.Load(context.Background(), child, spec.WithBaseConfig(base))
+		require.NoError(t, err)
+		require.NotNil(t, dag.RetryPolicy)
+		require.Equal(t, 1, dag.RetryPolicy.Limit)
+		require.Equal(t, 60*time.Second, dag.RetryPolicy.Interval)
+	})
+
 	t.Run("UnknownKeyInDefaults", func(t *testing.T) {
 		t.Parallel()
 
diff --git a/internal/persis/filebaseconfig/default_base_config.go b/internal/persis/filebaseconfig/default_base_config.go
@@ -46,6 +46,13 @@ max_output_size: 1048576  # 1MB (bytes)
 # Duration string: e.g. "6h", "24h", "2d12h". Empty = no catchup (missed runs discarded).
 catchup_window: "6h"
 
+# Retry the entire DAG after a terminal failure.
+# This absorbs transient infrastructure or dependency failures by default.
+# Override or disable per DAG if the workflow is intentionally non-idempotent.
+retry_policy:
+  limit: 1
+  interval_sec: 60
+
 # -- Shell --
 # Shell interpreter for command steps.
 # Default resolution order: $DAGU_DEFAULT_SHELL -> $SHELL -> sh
diff --git a/internal/persis/filebaseconfig/store_test.go b/internal/persis/filebaseconfig/store_test.go
@@ -126,6 +126,7 @@ func TestInitialize(t *testing.T) {
 		assert.Contains(t, string(data), "Base DAG Configuration")
 		assert.Contains(t, string(data), "catchup_window")
 		assert.Contains(t, string(data), "hist_retention_days")
+		assert.Contains(t, string(data), "\nretry_policy:\n  limit: 1\n  interval_sec: 60\n")
 	})
 
 	t.Run("skips when file already exists", func(t *testing.T) {
diff --git a/internal/service/scheduler/retry_scanner_test.go b/internal/service/scheduler/retry_scanner_test.go
@@ -6,6 +6,7 @@ package scheduler
 import (
 	"context"
 	"errors"
+	"slices"
 	"testing"
 	"time"
 
@@ -516,12 +517,7 @@ func cloneRetryStatus(status *exec.DAGRunStatus) *exec.DAGRunStatus {
 }
 
 func containsStatus(statuses []core.Status, want core.Status) bool {
-	for _, status := range statuses {
-		if status == want {
-			return true
-		}
-	}
-	return false
+	return slices.Contains(statuses, want)
 }
 
 func withRetryCount(status *exec.DAGRunStatus, retryCount int) *exec.DAGRunStatus {
