[v23.2.x] c/topics_frontend: break the loop when dispatching to current leader

src/v/cluster/service.cc

@@ -206,7 +206,9 @@ service::do_finish_partition_update(finish_partition_update_request&& req) {
         req.ntp,
         req.new_replica_set,
         config::shard_local_cfg().replicate_append_timeout_ms()
-          + model::timeout_clock::now());
+          + model::timeout_clock::now(),
+        topics_frontend::dispatch_to_leader::no);
+
     finish_partition_update_reply reply{.result = errc::success};
     if (ec) {
         if (ec.category() == cluster::error_category()) {

src/v/cluster/topics_frontend.cc

@@ -856,27 +856,35 @@ ss::future<std::error_code> topics_frontend::abort_moving_partition_replicas(
 ss::future<std::error_code> topics_frontend::finish_moving_partition_replicas(
   model::ntp ntp,
   std::vector<model::broker_shard> new_replica_set,
-  model::timeout_clock::time_point tout) {
+  model::timeout_clock::time_point tout,
+  dispatch_to_leader dispatch) {
     auto leader = _leaders.local().get_leader(model::controller_ntp);
 
     // no leader available
     if (!leader) {
         return ss::make_ready_future<std::error_code>(
           errc::no_leader_controller);
     }
-    // optimization: if update is not in progress return early
-    if (!_topics.local().is_update_in_progress(ntp)) {
-        return ss::make_ready_future<std::error_code>(
-          errc::no_update_in_progress);
-    }
+
     // current node is a leader, just replicate
     if (leader == _self) {
+        // optimization: if update is not in progress return early
+        if (!_topics.local().is_update_in_progress(ntp)) {
+            return ss::make_ready_future<std::error_code>(
+              errc::no_update_in_progress);
+        }
+
         finish_moving_partition_replicas_cmd cmd(
           std::move(ntp), std::move(new_replica_set));
 
         return replicate_and_wait(_stm, _features, _as, std::move(cmd), tout);
     }
 
+    if (!dispatch) {
+        return ss::make_ready_future<std::error_code>(
+          errc::not_leader_controller);
+    }
+
     return _connections.local()
       .with_node_client<controller_client_protocol>(
         _self,

src/v/cluster/topics_frontend.h

@@ -28,6 +28,7 @@
 #include <seastar/core/abort_source.hh>
 #include <seastar/core/chunked_fifo.hh>
 #include <seastar/core/sharded.hh>
+#include <seastar/util/bool_class.hh>
 
 #include <absl/container/flat_hash_map.h>
 
@@ -38,6 +39,13 @@ namespace cluster {
 // on every core
 class topics_frontend {
 public:
+    /**
+     * A boolean that may be used by the caller to request redirecting a request
+     * to the leader. This is useful as topic operations must be executed on
+     * `redpanda/controller` partition leader.
+     */
+    using dispatch_to_leader = ss::bool_class<struct dispatch_to_leader_tag>;
+
     struct capacity_info {
         absl::flat_hash_map<model::node_id, node_disk_space> node_disk_reports;
         absl::flat_hash_map<model::partition_id, int64_t> ntp_sizes;
@@ -103,7 +111,8 @@ class topics_frontend {
     ss::future<std::error_code> finish_moving_partition_replicas(
       model::ntp,
       std::vector<model::broker_shard>,
-      model::timeout_clock::time_point);
+      model::timeout_clock::time_point,
+      dispatch_to_leader = dispatch_to_leader::yes);
 
     ss::future<std::error_code> revert_cancel_partition_move(
       model::ntp, model::timeout_clock::time_point);