fix(sync): tolerate non-numeric shard and replica names in stale replica cleanup

internal/controller/clickhouse/commands.go

@@ -23,6 +23,10 @@ WHERE
 	engine NOT IN ('Atomic', 'Lazy', 'SQLite', 'Ordinary', 'Memory')
 SETTINGS
 	format_display_secrets_in_show_and_select=1`
+	// system.clusters also lists static clusters from remote_servers and Replicated databases with
+	// names the operator did not choose; their shard and replica names are empty or non-numeric.
+	// toInt32OrNull keeps the cast from throwing on those rows, and the NULL filter drops them, so
+	// only replicas that the operator itself numbered are considered for cleanup.
 	listStaleDatabaseReplicasQuery = `SELECT
 	database,
 	shard_id,
@@ -31,11 +35,11 @@ SETTINGS
 FROM (
 	SELECT
 		cluster as database,
-		toInt32(database_shard_name) AS shard_id,
-		toInt32(database_replica_name) AS replica_id,
+		toInt32OrNull(database_shard_name) AS shard_id,
+		toInt32OrNull(database_replica_name) AS replica_id,
 		is_active
 	FROM clusterAllReplicas(default, system.clusters)
-	WHERE database_replica_name != '' AND (shard_id >= ? OR replica_id >= ?)
+	WHERE shard_id IS NOT NULL AND replica_id IS NOT NULL AND (shard_id >= ? OR replica_id >= ?)
 )
 GROUP BY database, shard_id, replica_id
 SETTINGS skip_unavailable_shards=1`

internal/controller/clickhouse/commands_test.go

@@ -431,6 +431,22 @@ var _ = Describe("commander", Ordered, Label("integration"), func() {
 			Expect(cmd.CleanupDatabaseReplicas(ctx, cmd.log, map[v1.ClickHouseReplicaID]struct{}{})).To(Succeed())
 		})
 
+		It("should ignore replicated databases whose shard and replica names the operator did not choose", func(ctx context.Context) {
+			// system.clusters lists every Replicated database on the server, including ones created by
+			// users with their own shard and replica names, next to the static clusters from
+			// remote_servers whose names are empty. None of those parse as replica numbers.
+			id0 := v1.ClickHouseReplicaID{ShardID: 0, Index: 0}
+			conn, err := cmd.getConn(id0)
+			Expect(err).NotTo(HaveOccurred())
+			Expect(conn.Exec(ctx, "CREATE DATABASE named ENGINE = Replicated('/clickhouse/databases/named', 'shard-a', 'replica-b')")).To(Succeed())
+
+			DeferCleanup(func(ctx context.Context) {
+				Expect(conn.Exec(ctx, "DROP DATABASE named SYNC")).To(Succeed())
+			})
+
+			Expect(cmd.CleanupDatabaseReplicas(ctx, cmd.log, map[v1.ClickHouseReplicaID]struct{}{})).To(Succeed())
+		})
+
 		It("should do nothing if still active", func(ctx context.Context) {
 			*cmd.cluster.Spec.Replicas = testReplicas - 1
 			err := cmd.CleanupDatabaseReplicas(ctx, cmd.log, map[v1.ClickHouseReplicaID]struct{}{})