Make settings controlling connection drop on overloaded CPU hot-reloadable

Author: Alex Katsman

programs/server/Server.cpp

@@ -1951,6 +1951,8 @@ try
             global_context->setMaxPendingMutationsExecutionTimeToWarn(new_server_settings[ServerSetting::max_pending_mutations_execution_time_to_warn]);
             global_context->getAccessControl().setAllowTierSettings(new_server_settings[ServerSetting::allow_feature_tier]);
 
+            global_context->setOSCPUOverloadSettings(new_server_settings[ServerSetting::min_os_cpu_wait_time_ratio_to_drop_connection], new_server_settings[ServerSetting::max_os_cpu_wait_time_ratio_to_drop_connection]);
+
             size_t read_bandwidth = new_server_settings[ServerSetting::max_remote_read_network_bandwidth_for_server];
             size_t write_bandwidth = new_server_settings[ServerSetting::max_remote_write_network_bandwidth_for_server];
 
@@ -2859,10 +2861,9 @@ void Server::createServers(
 
     const TCPServerConnectionFilter::Ptr & connection_filter = new TCPServerConnectionFilter{[&]()
     {
-        const auto & server_settings = global_context->getServerSettings();
-        return !ProfileEvents::checkCPUOverload(server_settings[ServerSetting::os_cpu_busy_time_threshold],
-                server_settings[ServerSetting::min_os_cpu_wait_time_ratio_to_drop_connection],
-                server_settings[ServerSetting::max_os_cpu_wait_time_ratio_to_drop_connection],
+        return !ProfileEvents::checkCPUOverload(global_context->getServerSettings()[ServerSetting::os_cpu_busy_time_threshold],
+                global_context->getMinOSCPUWaitTimeRatioToDropConnection(),
+                global_context->getMaxOSCPUWaitTimeRatioToDropConnection(),
                 /*should_throw*/ false);
     }};
 

src/Core/ServerSettings.cpp

@@ -1173,6 +1173,9 @@ void ServerSettings::dumpToSystemServerSettingsColumns(ServerSettingColumnsParam
             {"max_pending_mutations_execution_time_to_warn", {std::to_string(context->getMaxPendingMutationsExecutionTimeToWarn()), ChangeableWithoutRestart::Yes}},
             {"max_partition_size_to_drop", {std::to_string(context->getMaxPartitionSizeToDrop()), ChangeableWithoutRestart::Yes}},
 
+            {"min_os_cpu_wait_time_ratio_to_drop_connection", {std::to_string(context->getMinOSCPUWaitTimeRatioToDropConnection()), ChangeableWithoutRestart::Yes}},
+            {"max_os_cpu_wait_time_ratio_to_drop_connection", {std::to_string(context->getMaxOSCPUWaitTimeRatioToDropConnection()), ChangeableWithoutRestart::Yes}},
+
             {"max_concurrent_queries", {std::to_string(context->getProcessList().getMaxSize()), ChangeableWithoutRestart::Yes}},
             {"max_concurrent_insert_queries",
             {std::to_string(context->getProcessList().getMaxInsertQueriesAmount()), ChangeableWithoutRestart::Yes}},

src/Interpreters/Context.cpp

@@ -530,6 +530,9 @@ struct ContextSharedPart : boost::noncopyable
     /// Only for system.server_settings, actually value stored in reloader itself
     std::atomic_size_t config_reload_interval_ms = ConfigReloader::DEFAULT_RELOAD_INTERVAL.count();
 
+    double min_os_cpu_wait_time_ratio_to_drop_connection = 10.0;
+    double max_os_cpu_wait_time_ratio_to_drop_connection = 20.0;
+
     String format_schema_path;                              /// Path to a directory that contains schema files used by input formats.
     String google_protos_path; /// Path to a directory that contains the proto files for the well-known Protobuf types.
     mutable OnceFlag action_locks_manager_initialized;
@@ -4561,6 +4564,25 @@ void Context::setMaxDatabaseNumToWarn(size_t max_database_to_warn)
     shared->max_database_num_to_warn = max_database_to_warn;
 }
 
+double Context::getMinOSCPUWaitTimeRatioToDropConnection() const
+{
+    SharedLockGuard lock(shared->mutex);
+    return shared->min_os_cpu_wait_time_ratio_to_drop_connection;
+}
+
+double Context::getMaxOSCPUWaitTimeRatioToDropConnection() const
+{
+    SharedLockGuard lock(shared->mutex);
+    return shared->max_os_cpu_wait_time_ratio_to_drop_connection;
+}
+
+void Context::setOSCPUOverloadSettings(double min_os_cpu_wait_time_ratio_to_drop_connection, double max_os_cpu_wait_time_ratio_to_drop_connection)
+{
+    SharedLockGuard lock(shared->mutex);
+    shared->min_os_cpu_wait_time_ratio_to_drop_connection = min_os_cpu_wait_time_ratio_to_drop_connection;
+    shared->max_os_cpu_wait_time_ratio_to_drop_connection = max_os_cpu_wait_time_ratio_to_drop_connection;
+}
+
 std::shared_ptr<Cluster> Context::getCluster(const std::string & cluster_name) const
 {
     if (auto res = tryGetCluster(cluster_name))

src/Interpreters/Context.h

@@ -1035,6 +1035,10 @@ class Context: public ContextData, public std::enable_shared_from_this<Context>
     void setMaxPendingMutationsToWarn(size_t max_pending_mutations_to_warn);
     void setMaxPendingMutationsExecutionTimeToWarn(size_t max_pending_mutations_execution_time_to_warn);
 
+    double getMinOSCPUWaitTimeRatioToDropConnection() const;
+    double getMaxOSCPUWaitTimeRatioToDropConnection() const;
+    void setOSCPUOverloadSettings(double min_os_cpu_wait_time_ratio_to_drop_connection, double max_os_cpu_wait_time_ratio_to_drop_connection);
+
     /// The port that the server listens for executing SQL queries.
     UInt16 getTCPPort() const;