Multithreaded request aggregator (#4469)

* Run aggregator threaded * Configurable number of aggregator threads * Default to max 4 threads * Added unit test for request_aggregator_threads toml * Improve the description of request_aggregator_threads --------- Co-authored-by: Dimitrios Siganos <[email protected]>
nanocurrency · Mar 11, 2024 · d595655 · d595655
1 parent 8dd0bf9
commit d595655
Show file tree

Hide file tree

Showing 5 changed files with 21 additions and 5 deletions.
diff --git a/nano/core_test/toml.cpp b/nano/core_test/toml.cpp
@@ -193,6 +193,7 @@ TEST (toml, daemon_config_deserialize_defaults)
 	ASSERT_EQ (conf.node.work_peers, defaults.node.work_peers);
 	ASSERT_EQ (conf.node.work_threads, defaults.node.work_threads);
 	ASSERT_EQ (conf.node.max_queued_requests, defaults.node.max_queued_requests);
+	ASSERT_EQ (conf.node.request_aggregator_threads, defaults.node.request_aggregator_threads);
 	ASSERT_EQ (conf.node.max_unchecked_blocks, defaults.node.max_unchecked_blocks);
 	ASSERT_EQ (conf.node.backlog_scan_batch_size, defaults.node.backlog_scan_batch_size);
 	ASSERT_EQ (conf.node.backlog_scan_frequency, defaults.node.backlog_scan_frequency);
@@ -422,6 +423,7 @@ TEST (toml, daemon_config_deserialize_no_defaults)
 	work_threads = 999
 	max_work_generate_multiplier = 1.0
 	max_queued_requests = 999
+	request_aggregator_threads = 999
 	max_unchecked_blocks = 999
 	frontiers_confirmation = "always"
 	backlog_scan_batch_size = 999
@@ -613,6 +615,7 @@ TEST (toml, daemon_config_deserialize_no_defaults)
 	ASSERT_NE (conf.node.work_peers, defaults.node.work_peers);
 	ASSERT_NE (conf.node.work_threads, defaults.node.work_threads);
 	ASSERT_NE (conf.node.max_queued_requests, defaults.node.max_queued_requests);
+	ASSERT_NE (conf.node.request_aggregator_threads, defaults.node.request_aggregator_threads);
 	ASSERT_NE (conf.node.backlog_scan_batch_size, defaults.node.backlog_scan_batch_size);
 	ASSERT_NE (conf.node.backlog_scan_frequency, defaults.node.backlog_scan_frequency);
 

diff --git a/nano/node/nodeconfig.cpp b/nano/node/nodeconfig.cpp
@@ -130,6 +130,7 @@ nano::error nano::node_config::serialize_toml (nano::tomlconfig & toml) const
 	toml.put ("max_work_generate_multiplier", max_work_generate_multiplier, "Maximum allowed difficulty multiplier for work generation.\ntype:double,[1..]");
 	toml.put ("frontiers_confirmation", serialize_frontiers_confirmation (frontiers_confirmation), "Mode controlling frontier confirmation rate.\ntype:string,{auto,always,disabled}");
 	toml.put ("max_queued_requests", max_queued_requests, "Limit for number of queued confirmation requests for one channel, after which new requests are dropped until the queue drops below this value.\ntype:uint32");
+	toml.put ("request_aggregator_threads", request_aggregator_threads, "Number of threads to dedicate to request aggregator. The default value is the minimum of 4 or the number returned by nano::hardware_concurency(), which is the number of hardware threads or the value of the environment variable NANO_HARDWARE_CONCURRENCY.");
 	toml.put ("max_unchecked_blocks", max_unchecked_blocks, "Maximum number of unchecked blocks to store in memory. Defaults to 65536. \ntype:uint64,[0..]");
 	toml.put ("rep_crawler_weight_minimum", rep_crawler_weight_minimum.to_string_dec (), "Rep crawler minimum weight, if this is less than minimum principal weight then this is taken as the minimum weight a rep must have to be tracked. If you want to track all reps set this to 0. If you do not want this to influence anything then set it to max value. This is only useful for debugging or for people who really know what they are doing.\ntype:string,amount,raw");
 	toml.put ("backlog_scan_batch_size", backlog_scan_batch_size, "Number of accounts per second to process when doing backlog population scan. Increasing this value will help unconfirmed frontiers get into election prioritization queue faster, however it will also increase resource usage. \ntype:uint");
@@ -427,6 +428,7 @@ nano::error nano::node_config::deserialize_toml (nano::tomlconfig & toml)
 		toml.get<double> ("max_work_generate_multiplier", max_work_generate_multiplier);
 
 		toml.get<uint32_t> ("max_queued_requests", max_queued_requests);
+		toml.get<uint32_t> ("request_aggregator_threads", request_aggregator_threads);
 
 		toml.get<unsigned> ("max_unchecked_blocks", max_unchecked_blocks);
 

diff --git a/nano/node/nodeconfig.hpp b/nano/node/nodeconfig.hpp
@@ -118,6 +118,7 @@ class node_config
 	bool backup_before_upgrade{ false };
 	double max_work_generate_multiplier{ 64. };
 	uint32_t max_queued_requests{ 512 };
+	unsigned request_aggregator_threads{ std::min (nano::hardware_concurrency (), 4u) }; // Max 4 threads if available
 	unsigned max_unchecked_blocks{ 65536 };
 	std::chrono::seconds max_pruning_age{ !network_params.network.is_beta_network () ? std::chrono::seconds (24 * 60 * 60) : std::chrono::seconds (5 * 60) }; // 1 day; 5 minutes for beta network
 	uint64_t max_pruning_depth{ 0 };

diff --git a/nano/node/request_aggregator.cpp b/nano/node/request_aggregator.cpp
@@ -15,15 +15,20 @@ nano::request_aggregator::request_aggregator (nano::node_config const & config_a
 	max_delay (config_a.network_params.network.is_dev_network () ? 50 : 300),
 	small_delay (config_a.network_params.network.is_dev_network () ? 10 : 50),
 	max_channel_requests (config_a.max_queued_requests),
+	request_aggregator_threads (config_a.request_aggregator_threads),
 	stats (stats_a),
 	local_votes (history_a),
 	ledger (ledger_a),
 	wallets (wallets_a),
 	active (active_a),
 	generator (generator_a),
-	final_generator (final_generator_a),
-	thread ([this] () { run (); })
+	final_generator (final_generator_a)
 {
+	for (auto i = 0; i < request_aggregator_threads; ++i)
+	{
+		threads.emplace_back ([this] () { run (); });
+	}
+
 	generator.set_reply_action ([this] (std::shared_ptr<nano::vote> const & vote_a, std::shared_ptr<nano::transport::channel> const & channel_a) {
 		this->reply_action (vote_a, channel_a);
 	});
@@ -132,9 +137,12 @@ void nano::request_aggregator::stop ()
 		stopped = true;
 	}
 	condition.notify_all ();
-	if (thread.joinable ())
+	for (auto & thread : threads)
 	{
-		thread.join ();
+		if (thread.joinable ())
+		{
+			thread.join ();
+		}
 	}
 }
 

diff --git a/nano/node/request_aggregator.hpp b/nano/node/request_aggregator.hpp
@@ -13,6 +13,7 @@
 #include <condition_variable>
 #include <thread>
 #include <unordered_map>
+#include <vector>
 
 namespace mi = boost::multi_index;
 
@@ -74,6 +75,7 @@ class request_aggregator final
 	std::chrono::milliseconds const max_delay;
 	std::chrono::milliseconds const small_delay;
 	std::size_t const max_channel_requests;
+	std::size_t const request_aggregator_threads;
 
 private:
 	void run ();
@@ -105,7 +107,7 @@ class request_aggregator final
 	bool started{ false };
 	nano::condition_variable condition;
 	nano::mutex mutex{ mutex_identifier (mutexes::request_aggregator) };
-	std::thread thread;
+	std::vector<std::thread> threads;
 
 	friend std::unique_ptr<container_info_component> collect_container_info (request_aggregator &, std::string const &);
 };