Detect POLLHUP/POLLRDHUP while running queries.

Provide a new GUC check_client_connection_interval that can be used to
check whether the client connection has gone away, while running very
long queries.  It is disabled by default.

For now this uses a non-standard Linux extension (also adopted by at
least one other OS).  POLLRDHUP is not defined by POSIX, and other OSes
don't have a reliable way to know if a connection was closed without
actually trying to read or write.

In future we might consider trying to send a no-op/heartbeat message
instead, but that could require protocol changes.

Author: Sergey Cherkashin <s.cherkashin@postgrespro.ru>
Author: Thomas Munro <thomas.munro@gmail.com>
Reviewed-by: Thomas Munro <thomas.munro@gmail.com>
Reviewed-by: Tatsuo Ishii <ishii@sraoss.co.jp>
Reviewed-by: Konstantin Knizhnik <k.knizhnik@postgrespro.ru>
Reviewed-by: Zhihong Yu <zyu@yugabyte.com>
Reviewed-by: Andres Freund <andres@anarazel.de>
Reviewed-by: Maksim Milyutin <milyutinma@gmail.com>
Reviewed-by: Tsunakawa, Takayuki/綱川 貴之 <tsunakawa.takay@fujitsu.com>
Reviewed-by: Tom Lane <tgl@sss.pgh.pa.us> (much earlier version)
Discussion: https://postgr.es/m/77def86b27e41f0efcba411460e929ae%40postgrespro.ru

1 files changed, 37 insertions, 0 deletions

diff --git a/doc/src/sgml/config.sgml b/doc/src/sgml/config.sgml
index 9d87b5097af..0c9128a55d0 100644
--- a/doc/src/sgml/config.sgml
+++ b/doc/src/sgml/config.sgml
@@ -998,6 +998,43 @@ include_dir 'conf.d'
       </listitem>
      </varlistentry>
 
+     <varlistentry id="guc-client-connection-check-interval" xreflabel="client_connection_check_interval">
+      <term><varname>client_connection_check_interval</varname> (<type>integer</type>)
+      <indexterm>
+       <primary><varname>client_connection_check_interval</varname> configuration parameter</primary>
+      </indexterm>
+      </term>
+      <listitem>
+       <para>
+        Sets the time interval between optional checks that the client is still
+        connected, while running queries.  The check is performed by polling
+        the socket, and allows long running queries to be aborted sooner if
+        the kernel reports that the connection is closed.
+       </para>
+       <para>
+        This option is currently available only on systems that support the
+        non-standard <symbol>POLLRDHUP</symbol> extension to the
+        <symbol>poll</symbol> system call, including Linux.
+       </para>
+       <para>
+        If the value is specified without units, it is taken as milliseconds.
+        The default value is <literal>0</literal>, which disables connection
+        checks.  Without connection checks, the server will detect the loss of
+        the connection only at the next interaction with the socket, when it
+        waits for, receives or sends data.
+       </para>
+       <para>
+        For the kernel itself to detect lost TCP connections reliably and within
+        a known timeframe in all scenarios including network failure, it may
+        also be necessary to adjust the TCP keepalive settings of the operating
+        system, or the <xref linkend="guc-tcp-keepalives-idle"/>,
+        <xref linkend="guc-tcp-keepalives-interval"/> and
+        <xref linkend="guc-tcp-keepalives-count"/> settings of
+        <productname>PostgreSQL</productname>.
+       </para>
+      </listitem>
+     </varlistentry>
+
      </variablelist>
      </sect2>