Add support TCP user timeout in libpq and the backend server

Similarly to the set of parameters for keepalive, a connection parameter
for libpq is added as well as a backend GUC, called tcp_user_timeout.

Increasing the TCP user timeout is useful to allow a connection to
survive extended periods without end-to-end connection, and decreasing
it allows application to fail faster.  By default, the parameter is 0,
which makes the connection use the system default, and follows a logic
close to the keepalive parameters in its handling.  When connecting
through a Unix-socket domain, the parameters have no effect.

Author: Ryohei Nagaura
Reviewed-by: Fabien Coelho, Robert Haas, Kyotaro Horiguchi, Kirk
Jamison, Mikalai Keida, Takayuki Tsunakawa, Andrei Yahorau
Discussion: https://postgr.es/m/EDA4195584F5064680D8130B1CA91C45367328@G01JPEXMBYT04

Author: michaelpq

contrib/postgres_fdw/expected/postgres_fdw.out

@@ -151,6 +151,7 @@ ALTER SERVER testserver1 OPTIONS (
 	keepalives 'value',
 	keepalives_idle 'value',
 	keepalives_interval 'value',
+	tcp_user_timeout 'value',
 	-- requiressl 'value',
 	sslcompression 'value',
 	sslmode 'value',

contrib/postgres_fdw/sql/postgres_fdw.sql

@@ -164,6 +164,7 @@ ALTER SERVER testserver1 OPTIONS (
 	keepalives 'value',
 	keepalives_idle 'value',
 	keepalives_interval 'value',
+	tcp_user_timeout 'value',
 	-- requiressl 'value',
 	sslcompression 'value',
 	sslmode 'value',

doc/src/sgml/config.sgml

@@ -939,6 +939,31 @@ include_dir 'conf.d'
       </listitem>
      </varlistentry>
 
+     <varlistentry id="guc-tcp-user-timeout" xreflabel="tcp_user_timeout">
+      <term><varname>tcp_user_timeout</varname> (<type>integer</type>)
+      <indexterm>
+       <primary><varname>tcp_user_timeout</varname> configuration parameter</primary>
+      </indexterm>
+      </term>
+      <listitem>
+       <para>
+        Specifies the number of milliseconds that transmitted data may
+        remain unacknowledged before a connection is forcibly closed.
+        A value of 0 uses the system default.
+        This parameter is supported only on systems that support
+        <symbol>TCP_USER_TIMEOUT</symbol>; on other systems, it must be zero.
+        In sessions connected via a Unix-domain socket, this parameter is
+        ignored and always reads as zero. 
+       </para>
+       <note>
+        <para>
+         This parameter is not supported on Windows and on Linux version
+         2.6.36 or older.
+        </para>
+       </note>
+      </listitem>
+     </varlistentry>
+
      </variablelist>
      </sect2>
 

doc/src/sgml/libpq.sgml

@@ -1249,6 +1249,20 @@ postgresql://%2Fvar%2Flib%2Fpostgresql/dbname
       </listitem>
      </varlistentry>
 
+     <varlistentry id="libpq-tcp-user-timeout" xreflabel="libpq_tcp_user_timeout">
+      <term><literal>tcp_user_timeout</literal></term>
+      <listitem>
+       <para>
+        Controls the number of milliseconds that transmitted data may
+        remain unacknowledged before a connection is forcibly closed.
+        A value of zero uses the system default. This parameter is
+        ignored for connections made via a Unix-domain socket.
+        It is only supported on systems where <symbol>TCP_USER_TIMEOUT</symbol>
+        is available; on other systems, it has no effect.
+       </para>
+      </listitem>
+     </varlistentry>
+
      <varlistentry id="libpq-connect-tty" xreflabel="tty">
       <term><literal>tty</literal></term>
       <listitem>

src/backend/libpq/pqcomm.c

@@ -825,6 +825,7 @@ StreamConnection(pgsocket server_fd, Port *port)
 		(void) pq_setkeepalivesidle(tcp_keepalives_idle, port);
 		(void) pq_setkeepalivesinterval(tcp_keepalives_interval, port);
 		(void) pq_setkeepalivescount(tcp_keepalives_count, port);
+		(void) pq_settcpusertimeout(tcp_user_timeout, port);
 	}
 
 	return STATUS_OK;
@@ -1926,3 +1927,75 @@ pq_setkeepalivescount(int count, Port *port)
 
 	return STATUS_OK;
 }
+
+int
+pq_gettcpusertimeout(Port *port)
+{
+#ifdef TCP_USER_TIMEOUT
+	if (port == NULL || IS_AF_UNIX(port->laddr.addr.ss_family))
+		return 0;
+
+	if (port->tcp_user_timeout != 0)
+		return port->tcp_user_timeout;
+
+	if (port->default_tcp_user_timeout == 0)
+	{
+		ACCEPT_TYPE_ARG3 size = sizeof(port->default_tcp_user_timeout);
+
+		if (getsockopt(port->sock, IPPROTO_TCP, TCP_USER_TIMEOUT,
+					   (char *) &port->default_tcp_user_timeout,
+					   &size) < 0)
+		{
+			elog(LOG, "getsockopt(%s) failed: %m", "TCP_USER_TIMEOUT");
+			port->default_tcp_user_timeout = -1;	/* don't know */
+		}
+	}
+
+	return port->default_tcp_user_timeout;
+#else
+	return 0;
+#endif
+}
+
+int
+pq_settcpusertimeout(int timeout, Port *port)
+{
+	if (port == NULL || IS_AF_UNIX(port->laddr.addr.ss_family))
+		return STATUS_OK;
+
+#ifdef TCP_USER_TIMEOUT
+	if (timeout == port->tcp_user_timeout)
+		return STATUS_OK;
+
+	if (port->default_tcp_user_timeout <= 0)
+	{
+		if (pq_gettcpusertimeout(port) < 0)
+		{
+			if (timeout == 0)
+				return STATUS_OK;	/* default is set but unknown */
+			else
+				return STATUS_ERROR;
+		}
+	}
+
+	if (timeout == 0)
+		timeout = port->default_tcp_user_timeout;
+
+	if (setsockopt(port->sock, IPPROTO_TCP, TCP_USER_TIMEOUT,
+				   (char *) &timeout, sizeof(timeout)) < 0)
+	{
+		elog(LOG, "setsockopt(%s) failed: %m", "TCP_USER_TIMEOUT");
+		return STATUS_ERROR;
+	}
+
+	port->tcp_user_timeout = timeout;
+#else
+	if (timeout != 0)
+	{
+		elog(LOG, "setsockopt(%s) not supported", "TCP_USER_TIMEOUT");
+		return STATUS_ERROR;
+	}
+#endif
+
+	return STATUS_OK;
+}

src/backend/utils/misc/guc.c

@@ -182,9 +182,11 @@ static const char *show_archive_command(void);
 static void assign_tcp_keepalives_idle(int newval, void *extra);
 static void assign_tcp_keepalives_interval(int newval, void *extra);
 static void assign_tcp_keepalives_count(int newval, void *extra);
+static void assign_tcp_user_timeout(int newval, void *extra);
 static const char *show_tcp_keepalives_idle(void);
 static const char *show_tcp_keepalives_interval(void);
 static const char *show_tcp_keepalives_count(void);
+static const char *show_tcp_user_timeout(void);
 static bool check_maxconnections(int *newval, void **extra, GucSource source);
 static bool check_max_worker_processes(int *newval, void **extra, GucSource source);
 static bool check_autovacuum_max_workers(int *newval, void **extra, GucSource source);
@@ -530,6 +532,7 @@ char	   *application_name;
 int			tcp_keepalives_idle;
 int			tcp_keepalives_interval;
 int			tcp_keepalives_count;
+int			tcp_user_timeout;
 
 /*
  * SSL renegotiation was been removed in PostgreSQL 9.5, but we tolerate it
@@ -3182,6 +3185,17 @@ static struct config_int ConfigureNamesInt[] =
 		NULL, NULL, NULL
 	},
 
+	{
+		{"tcp_user_timeout", PGC_USERSET, CLIENT_CONN_OTHER,
+			gettext_noop("TCP user timeout."),
+			gettext_noop("A value of 0 uses the system default."),
+			GUC_UNIT_MS
+		},
+		&tcp_user_timeout,
+		0, 0, INT_MAX,
+		NULL, assign_tcp_user_timeout, show_tcp_user_timeout
+	},
+
 	/* End-of-list marker */
 	{
 		{NULL, 0, 0, NULL, NULL}, NULL, 0, 0, 0, NULL, NULL, NULL
@@ -11238,6 +11252,23 @@ show_tcp_keepalives_count(void)
 	return nbuf;
 }
 
+static void
+assign_tcp_user_timeout(int newval, void *extra)
+{
+	/* See comments in assign_tcp_keepalives_idle */
+	(void) pq_settcpusertimeout(newval, MyProcPort);
+}
+
+static const char *
+show_tcp_user_timeout(void)
+{
+	/* See comments in assign_tcp_keepalives_idle */
+	static char nbuf[16];
+
+	snprintf(nbuf, sizeof(nbuf), "%d", pq_gettcpusertimeout(MyProcPort));
+	return nbuf;
+}
+
 static bool
 check_maxconnections(int *newval, void **extra, GucSource source)
 {

src/backend/utils/misc/postgresql.conf.sample

@@ -73,7 +73,7 @@
 #bonjour_name = ''			# defaults to the computer name
 					# (change requires restart)
 
-# - TCP Keepalives -
+# - TCP settings -
 # see "man 7 tcp" for details
 
 #tcp_keepalives_idle = 0		# TCP_KEEPIDLE, in seconds;
@@ -82,6 +82,8 @@
 					# 0 selects the system default
 #tcp_keepalives_count = 0		# TCP_KEEPCNT;
 					# 0 selects the system default
+#tcp_user_timeout = 0			# TCP_USER_TIMEOUT, in milliseconds;
+					# 0 selects the system default
 
 # - Authentication -
 

src/include/libpq/libpq-be.h

@@ -155,7 +155,7 @@ typedef struct Port
 	HbaLine    *hba;
 
 	/*
-	 * TCP keepalive settings.
+	 * TCP keepalive and user timeout settings.
 	 *
 	 * default values are 0 if AF_UNIX or not yet known; current values are 0
 	 * if AF_UNIX or using the default. Also, -1 in a default value means we
@@ -164,9 +164,11 @@ typedef struct Port
 	int			default_keepalives_idle;
 	int			default_keepalives_interval;
 	int			default_keepalives_count;
+	int			default_tcp_user_timeout;
 	int			keepalives_idle;
 	int			keepalives_interval;
 	int			keepalives_count;
+	int			tcp_user_timeout;
 
 	/*
 	 * GSSAPI structures.
@@ -306,9 +308,11 @@ extern ProtocolVersion FrontendProtocol;
 extern int	pq_getkeepalivesidle(Port *port);
 extern int	pq_getkeepalivesinterval(Port *port);
 extern int	pq_getkeepalivescount(Port *port);
+extern int	pq_gettcpusertimeout(Port *port);
 
 extern int	pq_setkeepalivesidle(int idle, Port *port);
 extern int	pq_setkeepalivesinterval(int interval, Port *port);
 extern int	pq_setkeepalivescount(int count, Port *port);
+extern int	pq_settcpusertimeout(int timeout, Port *port);
 
 #endif							/* LIBPQ_BE_H */

src/include/utils/guc.h

@@ -271,6 +271,7 @@ extern PGDLLIMPORT char *application_name;
 extern int	tcp_keepalives_idle;
 extern int	tcp_keepalives_interval;
 extern int	tcp_keepalives_count;
+extern int	tcp_user_timeout;
 
 #ifdef TRACE_SORT
 extern bool trace_sort;

src/interfaces/libpq/fe-connect.c

@@ -270,6 +270,10 @@ static const internalPQconninfoOption PQconninfoOptions[] = {
 		"TCP-Keepalives-Count", "", 10, /* strlen(INT32_MAX) == 10 */
 	offsetof(struct pg_conn, keepalives_count)},
 
+	{"tcp_user_timeout", NULL, NULL, NULL,
+		"TCP-User-Timeout", "", 10, /* strlen(INT32_MAX) == 10 */
+	offsetof(struct pg_conn, pgtcp_user_timeout)},
+
 	/*
 	 * ssl options are allowed even without client SSL support because the
 	 * client can still handle SSL modes "disable" and "allow". Other
@@ -1833,6 +1837,41 @@ setKeepalivesWin32(PGconn *conn)
 #endif							/* SIO_KEEPALIVE_VALS */
 #endif							/* WIN32 */
 
+/*
+ * Set the TCP user timeout.
+ */
+static int
+setTCPUserTimeout(PGconn *conn)
+{
+	int			timeout;
+
+	if (conn->pgtcp_user_timeout == NULL)
+		return 1;
+
+	if (!parse_int_param(conn->pgtcp_user_timeout, &timeout, conn,
+						 "tcp_user_timeout"))
+		return 0;
+
+	if (timeout < 0)
+		timeout = 0;
+
+#ifdef TCP_USER_TIMEOUT
+	if (setsockopt(conn->sock, IPPROTO_TCP, TCP_USER_TIMEOUT,
+				   (char *) &timeout, sizeof(timeout)) < 0)
+	{
+		char		sebuf[256];
+
+		appendPQExpBuffer(&conn->errorMessage,
+						  libpq_gettext("setsockopt(%s) failed: %s\n"),
+						  "TCP_USER_TIMEOUT",
+						  SOCK_STRERROR(SOCK_ERRNO, sebuf, sizeof(sebuf)));
+		return 0;
+	}
+#endif
+
+	return 1;
+}
+
 /* ----------
  * connectDBStart -
  *		Begin the process of making a connection to the backend.
@@ -2480,6 +2519,8 @@ PQconnectPoll(PGconn *conn)
 							err = 1;
 #endif							/* SIO_KEEPALIVE_VALS */
 #endif							/* WIN32 */
+						else if (!setTCPUserTimeout(conn))
+							err = 1;
 
 						if (err)
 						{
@@ -3863,6 +3904,8 @@ freePGconn(PGconn *conn)
 		free(conn->pgtty);
 	if (conn->connect_timeout)
 		free(conn->connect_timeout);
+	if (conn->pgtcp_user_timeout)
+		free(conn->pgtcp_user_timeout);
 	if (conn->pgoptions)
 		free(conn->pgoptions);
 	if (conn->appname)

src/interfaces/libpq/libpq-int.h

@@ -336,6 +336,7 @@ struct pg_conn
 	char	   *pgtty;			/* tty on which the backend messages is
 								 * displayed (OBSOLETE, NOT USED) */
 	char	   *connect_timeout;	/* connection timeout (numeric string) */
+	char	   *pgtcp_user_timeout; /* tcp user timeout (numeric string) */
 	char	   *client_encoding_initial;	/* encoding to use */
 	char	   *pgoptions;		/* options to start the backend with */
 	char	   *appname;		/* application name */