merge feat(split): register child partition XiaoMi#391

src/dist/replication/common/replication_common.h

@@ -38,6 +38,7 @@ typedef std::unordered_map<::dsn::rpc_address, dsn::task_ptr> node_tasks;
 typedef rpc_holder<query_split_request, query_split_response> query_split_rpc;
 typedef rpc_holder<stop_split_request, stop_split_response> stop_split_rpc;
 typedef rpc_holder<control_split_request, control_split_response> control_split_rpc;
+typedef rpc_holder<register_child_request, register_child_response> register_child_rpc;
 
 class replication_options
 {

src/dist/replication/lib/replica.h

@@ -378,11 +378,16 @@ class replica : public serverlet<replica>, public ref_counter, public replica_ba
         rpc_address finish_update_address,
         bool is_update_child);
 
-    // all replicas update partition count, primary will register children on meta
-    virtual void register_child_on_meta(ballot b);
-    virtual void on_register_child_on_meta_reply(dsn::error_code ec,
-                                                 std::shared_ptr<register_child_request> request,
-                                                 std::shared_ptr<register_child_response> response);
+    // primary parent register children on meta_server
+    void register_child_on_meta(ballot b);
+    void on_register_child_on_meta_reply(dsn::error_code ec,
+                                         const register_child_request &request,
+                                         const register_child_response &response);
+    // primary sends register request to meta_server
+    void parent_send_register_request(const register_child_request &request);
+
+    // child partition has been registered on meta_server, could be active
+    void child_partition_active(const partition_configuration &config);
 
     // meta <=> replica configuration sync through on_config_sync
     // called by primary replica to check if partition count changed and partition flag changed to
@@ -414,9 +419,6 @@ class replica : public serverlet<replica>, public ref_counter, public replica_ba
     // parent replica handle child ack when child copy mutation synchronously
     void on_copy_mutation_reply(dsn::error_code ec, ballot b, decree d);
 
-    // child partitions have been registered on meta, could be active
-    void child_partition_active(const partition_configuration &config);
-
     // child copy parent prepare list and call child_learn_states
     void child_copy_prepare_list(learn_state lstate,
                                  std::vector<mutation_ptr> mutation_list,

src/dist/replication/lib/replica_config.cpp

@@ -362,6 +362,13 @@ void replica::update_configuration_on_meta_server(config_type::type type,
                                                   ::dsn::rpc_address node,
                                                   partition_configuration &newConfig)
 {
+    // type should never be `CT_REGISTER_CHILD`
+    // if this happens, it means serious mistake happened during partition split
+    // assert here to stop split and avoid splitting wrong
+    if (type == config_type::CT_REGISTER_CHILD) {
+        dassert_replica(false, "invalid config_type, type = {}", enum_to_string(type));
+    }
+
     newConfig.last_committed_decree = last_committed_decree();
 
     if (type == config_type::CT_PRIMARY_FORCE_UPDATE_BALLOT) {
@@ -370,8 +377,6 @@ void replica::update_configuration_on_meta_server(config_type::type type,
                 "");
         dassert(
             newConfig.primary == node, "%s VS %s", newConfig.primary.to_string(), node.to_string());
-    } else if (type == config_type::CT_REGISTER_CHILD) {
-        dassert(false, "invalid config_type, type = %s", enum_to_string(type));
     } else if (type != config_type::CT_ASSIGN_PRIMARY &&
                type != config_type::CT_UPGRADE_TO_PRIMARY) {
         dassert(status() == partition_status::PS_PRIMARY,
@@ -720,12 +725,11 @@ bool replica::update_local_configuration(const replica_configuration &config,
         break;
     case partition_status::PS_PARTITION_SPLIT:
         if (config.status == partition_status::PS_INACTIVE) {
-            dwarn("%s: status change from %s @ %" PRId64 " to %s @ %" PRId64 " is not allowed",
-                  name(),
-                  enum_to_string(old_status),
-                  old_ballot,
-                  enum_to_string(config.status),
-                  config.ballot);
+            dwarn_replica("status change from {} @ {} to {} @ {} is not allowed",
+                          enum_to_string(old_status),
+                          old_ballot,
+                          enum_to_string(config.status),
+                          config.ballot);
             return false;
         }
         break;
@@ -1321,15 +1325,5 @@ void replica::on_query_child_state_reply(error_code ec,
     }
 }
 
-void replica::child_partition_active(const partition_configuration &config)
-{
-    ddebug_f("{} finish partition split and become active", name());
-    _stub->_counter_replicas_splitting_recent_split_succ_count->increment();
-    // TODO(hyc): should set is false
-    _primary_states.sync_send_write_request = false;
-    _primary_states.last_prepare_decree_on_new_primary = _prepare_list->max_decree();
-    update_configuration(config);
-}
-
 } // namespace replication
 } // namespace dsn

src/dist/replication/lib/replica_context.h

@@ -119,7 +119,8 @@ class primary_context
     // otherwise, not during partition split
     bool sync_send_write_request{false};
 
-    // replica->meta register child on meta server and remote storage
+    // Used for partition split
+    // primary parent register child on meta_server task
     dsn::task_ptr register_child_task;
 
     // replica-> meta query child partition configuration