some DHT work

examples/client_test.cpp

@@ -1391,8 +1391,9 @@ int main(int argc, char* argv[])
 			for (std::vector<dht_lookup>::iterator i = sess_stat.active_requests.begin()
 				, end(sess_stat.active_requests.end()); i != end; ++i)
 			{
-				snprintf(str, sizeof(str), "  %s %d (%d) ( timeouts %d responses %d)\n"
-					, i->type, i->outstanding_requests, i->branch_factor, i->timeouts, i->responses);
+				snprintf(str, sizeof(str), "  %s in flight: %d [limit: %d] timeouts %d responses %d\n"
+					, i->type, i->outstanding_requests, i->branch_factor, i->timeouts
+					, i->responses);
 				out += str;
 			}
 		}

include/libtorrent/kademlia/dht_tracker.hpp

@@ -165,8 +165,6 @@ namespace libtorrent { namespace dht
 		int m_replies_bytes_sent[5];
 		int m_queries_bytes_received[5];
 		int m_counter;
-		int m_announces;
-		int m_failed_announces;
 
 		int m_total_message_input;
 		int m_total_in_bytes;

include/libtorrent/kademlia/find_data.hpp

@@ -101,6 +101,7 @@ public:
 		, m_self(self)
 	{}
 	~find_data_observer();
+	void short_timeout();
 	void timeout();
 	void reply(msg const&);
 	void abort() { m_algorithm = 0; }

include/libtorrent/kademlia/node.hpp

@@ -164,6 +164,7 @@ public:
 		, m_token(write_token)
 	{}
 
+	void short_timeout() {}
 	void timeout() {}
 	void reply(msg const&) {}
 	void abort() {}

include/libtorrent/kademlia/observer.hpp

@@ -82,6 +82,10 @@ struct observer : boost::noncopyable
 	// this is called when a reply is received
 	virtual void reply(msg const& m) = 0;
 
+	// this is called if no response has been received after
+	// a few seconds, before the request has timed out
+	virtual void short_timeout() = 0;
+
 	// this is called when no reply has been received within
 	// some timeout
 	virtual void timeout() = 0;

include/libtorrent/kademlia/rpc_manager.hpp

@@ -54,8 +54,6 @@ POSSIBILITY OF SUCH DAMAGE.
 namespace libtorrent { namespace dht
 {
 
-struct observer;
-
 #ifdef TORRENT_DHT_VERBOSE_LOGGING
 TORRENT_DECLARE_LOG(rpc);
 #endif
@@ -64,6 +62,7 @@ struct null_observer : public observer
 {
 	null_observer(boost::pool<>& allocator): observer(allocator) {}
 	virtual void reply(msg const&) {}
+	virtual void short_timeout() {}
 	virtual void timeout() {}
 	void abort() {}
 };

include/libtorrent/kademlia/traversal_algorithm.hpp

@@ -61,7 +61,9 @@ class traversal_algorithm : boost::noncopyable
 public:
 	void traverse(node_id const& id, udp::endpoint addr);
 	void finished(node_id const& id);
-	void failed(node_id const& id, bool prevent_request = false);
+
+	enum flags_t { prevent_request = 1, short_timeout = 2 };
+	void failed(node_id const& id, int flags = 0);
 	virtual ~traversal_algorithm();
 	boost::pool<>& allocator() const;
 	void status(dht_lookup& l);
@@ -81,7 +83,7 @@ public:
 		node_id id;
 		// TODO: replace with union of address_v4 and address_v6 and a port
 		udp::endpoint addr;
-		enum { queried = 1, initial = 2, no_id = 4 };
+		enum { queried = 1, initial = 2, no_id = 4, short_timeout = 8 };
 		unsigned char flags;
 	};
 

src/kademlia/dht_tracker.cpp

@@ -110,6 +110,9 @@ namespace libtorrent { namespace dht
 	int g_gr_message_input = 0;
 	int g_mo_message_input = 0;
 	int g_unknown_message_input = 0;
+
+	int g_announces = 0;
+	int g_failed_announces = 0;
 #endif
 		
 	void intrusive_ptr_add_ref(dht_tracker const* c)
@@ -230,8 +233,8 @@ namespace libtorrent { namespace dht
 		std::fill_n(m_queries_bytes_received, 5, 0);
 		std::fill_n(m_replies_sent, 5, 0);
 		std::fill_n(m_queries_received, 5, 0);
-		m_announces = 0;
-		m_failed_announces = 0;
+		g_announces = 0;
+		g_failed_announces = 0;
 		m_total_message_input = 0;
 		m_total_in_bytes = 0;
 		m_total_out_bytes = 0;
@@ -396,8 +399,8 @@ namespace libtorrent { namespace dht
 		
 		pc << "\t" << torrents
 			<< "\t" << peers
-			<< "\t" << m_announces / float(tick_period)
-			<< "\t" << m_failed_announces / float(tick_period)
+			<< "\t" << g_announces / float(tick_period)
+			<< "\t" << g_failed_announces / float(tick_period)
 			<< "\t" << (m_total_message_input / float(tick_period))
 			<< "\t" << (g_az_message_input / float(tick_period))
 			<< "\t" << (g_ut_message_input / float(tick_period))
@@ -414,8 +417,8 @@ namespace libtorrent { namespace dht
 		std::fill_n(m_queries_bytes_received, 5, 0);
 		std::fill_n(m_replies_sent, 5, 0);
 		std::fill_n(m_queries_received, 5, 0);
-		m_announces = 0;
-		m_failed_announces = 0;
+		g_announces = 0;
+		g_failed_announces = 0;
 		m_total_message_input = 0;
 		g_az_message_input = 0;
 		g_ut_message_input = 0;

src/kademlia/find_data.cpp

@@ -177,6 +177,12 @@ void find_data_observer::reply(msg const& m)
 #endif
 }
 
+void find_data_observer::short_timeout()
+{
+	if (!m_algorithm) return;
+	m_algorithm->failed(m_self, traversal_algorithm::short_timeout);
+}
+
 void find_data_observer::timeout()
 {
 	if (!m_algorithm) return;
@@ -184,7 +190,6 @@ void find_data_observer::timeout()
 	m_algorithm = 0;
 }
 
-
 find_data::find_data(
 	node_impl& node
 	, node_id target

src/kademlia/node.cpp

@@ -794,6 +794,7 @@ void node_impl::incoming_request(msg const& m, entry& e)
 	}
 	else if (strcmp(query, "announce_peer") == 0)
 	{
+		extern int g_failed_announces;
 		key_desc_t msg_desc[] = {
 			{"info_hash", lazy_entry::string_t, 20, 0},
 			{"port", lazy_entry::int_t, 0, 0},
@@ -803,6 +804,7 @@ void node_impl::incoming_request(msg const& m, entry& e)
 		lazy_entry const* msg_keys[3];
 		if (!verify_message(arg_ent, msg_desc, msg_keys, 3, error_string, sizeof(error_string)))
 		{
+			++g_failed_announces;
 			incoming_error(e, error_string);
 			return;
 		}
@@ -810,6 +812,7 @@ void node_impl::incoming_request(msg const& m, entry& e)
 		int port = msg_keys[1]->int_value();
 		if (port < 0 || port >= 65536)
 		{
+			++g_failed_announces;
 			incoming_error(e, "invalid 'port' in announce");
 			return;
 		}
@@ -822,6 +825,7 @@ void node_impl::incoming_request(msg const& m, entry& e)
 
 		if (!verify_token(msg_keys[2]->string_value(), msg_keys[0]->string_ptr(), m.addr))
 		{
+			++g_failed_announces;
 			incoming_error(e, "invalid token in announce");
 			return;
 		}
@@ -838,6 +842,8 @@ void node_impl::incoming_request(msg const& m, entry& e)
 		std::set<peer_entry>::iterator i = v.peers.find(e);
 		if (i != v.peers.end()) v.peers.erase(i++);
 		v.peers.insert(i, e);
+		extern int g_announces;
+		++g_announces;
 	}
 	else if (strcmp(query, "find_torrent") == 0)
 	{

src/kademlia/rpc_manager.cpp

@@ -338,15 +338,17 @@ time_duration rpc_manager::tick()
 {
 	INVARIANT_CHECK;
 
-	const int timeout_ms = 10 * 1000;
+	const static int short_timeout = 2;
+	const static int timeout = 10;
 
-	//	look for observers that has timed out
+	//	look for observers that have timed out
 
-	if (m_next_transaction_id == m_oldest_transaction_id) return milliseconds(timeout_ms);
+	if (m_next_transaction_id == m_oldest_transaction_id) return seconds(short_timeout);
 
 	std::vector<observer_ptr> timeouts;
 
-	time_duration ret = milliseconds(timeout_ms);
+	time_duration ret = seconds(short_timeout);
+	ptime now = time_now();
 
 	for (;m_next_transaction_id != m_oldest_transaction_id;
 		m_oldest_transaction_id = (m_oldest_transaction_id + 1) % max_transactions)
@@ -357,19 +359,14 @@ time_duration rpc_manager::tick()
 		observer_ptr o = m_transactions[m_oldest_transaction_id];
 		if (!o) continue;
 
-		time_duration diff = o->sent() + milliseconds(timeout_ms) - time_now();
-		if (diff > seconds(0))
+		// if we reach an observer that hasn't timed out
+		// break, because every observer after this one will
+		// also not have timed out yet
+		time_duration diff = now - o->sent();
+		if (diff < seconds(timeout))
 		{
-			if (diff < seconds(1))
-			{
-				ret = seconds(1);
-				break;
-			}
-			else
-			{
-				ret = diff;
-				break;	
-			}
+			ret = seconds(timeout) - diff;
+			break;
 		}
 		
 #ifndef BOOST_NO_EXCEPTIONS
@@ -389,11 +386,35 @@ time_duration rpc_manager::tick()
 	
 	std::for_each(timeouts.begin(), timeouts.end(), bind(&observer::timeout, _1));
 	timeouts.clear();
-	
+
 	// clear the aborted transactions, will likely
 	// generate new requests. We need to swap, since the
 	// destrutors may add more observers to the m_aborted_transactions
 	std::vector<observer_ptr>().swap(m_aborted_transactions);
+
+	for (int i = m_oldest_transaction_id; i != m_next_transaction_id;
+		i = (i + 1) % max_transactions)
+	{
+		observer_ptr o = m_transactions[i];
+		if (!o) continue;
+
+		// if we reach an observer that hasn't timed out
+		// break, because every observer after this one will
+		// also not have timed out yet
+		time_duration diff = now - o->sent();
+		if (diff < seconds(short_timeout))
+		{
+			ret = seconds(short_timeout) - diff;
+			break;
+		}
+		
+		// TODO: don't call short_timeout() again if we've
+		// already called it once
+		timeouts.push_back(o);
+	}
+
+	std::for_each(timeouts.begin(), timeouts.end(), bind(&observer::short_timeout, _1));
+	
 	return ret;
 }
 

src/kademlia/traversal_algorithm.cpp

@@ -108,8 +108,29 @@ void traversal_algorithm::traverse(node_id const& id, udp::endpoint addr)
 
 void traversal_algorithm::finished(node_id const& id)
 {
+	std::vector<result>::iterator i = std::find_if(
+		m_results.begin()
+		, m_results.end()
+		, bind(
+			std::equal_to<node_id>()
+			, bind(&result::id, _1)
+			, id
+		)
+	);
+
+	TORRENT_ASSERT(i != m_results.end());
+
+	if (i != m_results.end())
+	{
+		// if this flag is set, it means we increased the
+		// branch factor for it, and we should restore it
+		if (i->flags & result::short_timeout)
+			--m_branch_factor;
+	}
+
 	++m_responses;
 	--m_invoke_count;
+	TORRENT_ASSERT(m_invoke_count >= 0);
 	add_requests();
 	if (m_invoke_count == 0) done();
 }
@@ -117,9 +138,9 @@ void traversal_algorithm::finished(node_id const& id)
 // prevent request means that the total number of requests has
 // overflown. This query failed because it was the oldest one.
 // So, if this is true, don't make another request
-void traversal_algorithm::failed(node_id const& id, bool prevent_request)
+void traversal_algorithm::failed(node_id const& id, int flags)
 {
-	--m_invoke_count;
+	TORRENT_ASSERT(m_invoke_count >= 0);
 
 	TORRENT_ASSERT(!id.is_all_zeros());
 	std::vector<result>::iterator i = std::find_if(
@@ -137,18 +158,44 @@ void traversal_algorithm::failed(node_id const& id, bool prevent_request)
 	if (i != m_results.end())
 	{
 		TORRENT_ASSERT(i->flags & result::queried);
-		m_failed.insert(i->addr);
+		if (flags & short_timeout)
+		{
+			// short timeout means that it has been more than
+			// two seconds since we sent the request, and that
+			// we'll most likely not get a response. But, in case
+			// we do get a late response, keep the handler
+			// around for some more, but open up the slot
+			// by increasing the branch factor
+			if ((i->flags & result::short_timeout) == 0)
+				++m_branch_factor;
+			i->flags |= result::short_timeout;
+		}
+		else
+		{
+			m_failed.insert(i->addr);
 #ifdef TORRENT_DHT_VERBOSE_LOGGING
-		TORRENT_LOG(traversal) << "failed: " << i->id << " " << i->addr;
+			TORRENT_LOG(traversal) << "failed: " << i->id << " " << i->addr;
 #endif
-		// don't tell the routing table about
-		// node ids that we just generated ourself
-		if ((i->flags & result::no_id) == 0)
-			m_node.m_table.node_failed(id);
-		m_results.erase(i);
-		++m_timeouts;
+			// if this flag is set, it means we increased the
+			// branch factor for it, and we should restore it
+			if (i->flags & result::short_timeout)
+				--m_branch_factor;
+
+			// don't tell the routing table about
+			// node ids that we just generated ourself
+			if ((i->flags & result::no_id) == 0)
+				m_node.m_table.node_failed(id);
+			m_results.erase(i);
+			++m_timeouts;
+			--m_invoke_count;
+		}
 	}
-	if (prevent_request)
+	else
+	{
+		--m_invoke_count;
+	}
+
+	if (flags & prevent_request)
 	{
 		--m_branch_factor;
 		if (m_branch_factor <= 0) m_branch_factor = 1;
@@ -186,13 +233,17 @@ void traversal_algorithm::add_requests()
 
 		if (i == last_iterator()) break;
 
+#ifndef BOOST_NO_EXCEPTIONS
 		try
 		{
+#endif
 			invoke(i->id, i->addr);
 			++m_invoke_count;
 			i->flags |= result::queried;
+#ifndef BOOST_NO_EXCEPTIONS
 		}
 		catch (std::exception& e) {}
+#endif
 	}
 }