fix(agent): avoid RAID health warnings during healthy scrubs (#2109)

agent/mdraid_linux.go

@@ -201,12 +201,9 @@ func mdraidSmartStatus(health mdraidHealth) string {
 	if health.mismatchCnt > 0 {
 		return "WARNING"
 	}
-	// "check" scans for consistency problems without repairing mismatches.
-	// With no mismatches, keep it green while reporting progress attributes.
-	switch syncAction {
-	case "repair":
-		return "WARNING"
-	}
+	// "check" and "repair" are requested consistency scans, not evidence of
+	// array failure. With no health issues above, keep scrubbing green while
+	// reporting the sync action and progress attributes.
 	switch state {
 	case "clean", "active", "active-idle", "write-pending", "read-auto", "readonly":
 		return "PASSED"

agent/mdraid_linux_test.go

@@ -174,8 +174,25 @@ func TestMdraidSmartStatus(t *testing.T) {
 	if got := mdraidSmartStatus(mdraidHealth{arrayState: "clean", mismatchCnt: 1}); got != "WARNING" {
 		t.Fatalf("mdraidSmartStatus(clean+mismatch) = %q, want WARNING", got)
 	}
-	if got := mdraidSmartStatus(mdraidHealth{arrayState: "clean", syncAction: "repair"}); got != "WARNING" {
-		t.Fatalf("mdraidSmartStatus(repair) = %q, want WARNING", got)
+	for _, tc := range []struct {
+		name   string
+		health mdraidHealth
+		want   string
+	}{
+		{"clean", mdraidHealth{arrayState: "clean"}, "PASSED"},
+		{"active", mdraidHealth{arrayState: "active"}, "PASSED"},
+		{"mismatch", mdraidHealth{arrayState: "active", mismatchCnt: 1}, "WARNING"},
+		{"degraded", mdraidHealth{arrayState: "active", degraded: 1}, "FAILED"},
+		{"faulty member", mdraidHealth{arrayState: "active", faultyDisks: 1}, "FAILED"},
+		{"inactive", mdraidHealth{arrayState: "inactive"}, "FAILED"},
+		{"unknown", mdraidHealth{arrayState: "unknown"}, "UNKNOWN"},
+	} {
+		t.Run("repair/"+tc.name, func(t *testing.T) {
+			tc.health.syncAction = "repair"
+			if got := mdraidSmartStatus(tc.health); got != tc.want {
+				t.Fatalf("mdraidSmartStatus(%+v) = %q, want %s", tc.health, got, tc.want)
+			}
+		})
 	}
 	if got := mdraidSmartStatus(mdraidHealth{arrayState: "clean"}); got != "PASSED" {
 		t.Fatalf("mdraidSmartStatus(clean) = %q, want PASSED", got)

supplemental/CHANGELOG.md

@@ -20,6 +20,8 @@
 
 - Fix missing root CA certificates in base agent image (#2291)
 
+- Fix false RAID health warnings during healthy data scrubbing (#2109)
+
 - Fix ZFS monitoring when /dev/zfs is unavailable (#2325)
 
 - Fix spurious `HUB_URL` warning in SSH-only mode (#2316)