path: root/doc/src/sgml/runtime.sgml

<!-- doc/src/sgml/runtime.sgml -->

<chapter id="runtime">
 <title>Server Setup and Operation</title>

 <para>
  This chapter discusses how to set up and run the <productname>Pgpool-II</> server
  and its interactions with the operating system.
 </para>

 <sect1 id="kernel-resource">
  <title>Managing kernel resouces</title>

  <indexterm>
   <primary>kernel resouces</primary>
  </indexterm>

  <para>
   To run <productname>Pgpool-II</productname> System V shared memory
   and System V semaphore are necessary. Although on linux enough
   amount of both resources are provided, other platforms may need to
   configure the kernel parameters.
  </para>
  <para>
   The minimum requirement of System V shared memory is around 140 MB.
   If you plan to use more than default 64MB
   of <link linkend="runtime-in-memory-query-cache">query
   cache</link>, you may need more shared memory.  Number of necessary
   shared memory segments is 3.  Usually the kernel provides enough
   shared memory segments.
  </para>
  
  <para>
   The minimum requirement of System V semaphore is around 10.  Number
   of necessary semaphore sets is 3.  Usually the kernel provides
   enough semaphore sets.
  </para>
 </sect1>

 <sect1 id="pgpool-II-user">
  <title>The <productname>Pgpool-II</productname> User Account</title>

  <indexterm>
   <primary>Pgpool-II user</primary>
  </indexterm>

  <para>
   As with any server daemon that is accessible to the outside world,
   it is advisable to run <productname>Pgpool-II</productname> under a
   separate user account. This user account should only own the data
   that is managed by the server, and should not be shared with other
   daemons. (For example, using the user <literal>nobody</literal> is a bad
   idea.) It is not advisable to install executables owned by this
   user because compromised systems could then modify their own
   binaries.
  </para>

  <para>
   To add a Unix user account to your system, look for a command
   <command>useradd</command> or <command>adduser</command>. The user
   name <systemitem>pgpool</systemitem> is often used, and is assumed
   throughout this book, but you can use another name if you like.
  </para>
 </sect1>

 <sect1 id="configuring-pcp-conf">
  <title>Configuring pcp.conf</title>

  <indexterm>
   <primary>pcp configuration</primary>
  </indexterm>

  <para>
   <productname>Pgpool-II</productname> provides a interface
   for administrators to perform management operation, such as
   getting <productname>Pgpool-II</productname> status or terminating
   <productname>Pgpool-II</productname> processes
   remotely. <filename>pcp.conf</filename> is the user/password file
   used for authentication by this interface. All operation modes
   require the <filename>pcp.conf</filename> file to be set. A
   <filename>$prefix/etc/pcp.conf.sample</filename> file is created
   during the installation
   of <productname>Pgpool-II</productname>. Copy the file as 
   <filename>$prefix/etc/pcp.conf</filename> and add your user name and password
   to it.

   <screen>
    <prompt>$</> <userinput>cp $prefix/etc/pcp.conf.sample $prefix/etc/pcp.conf</userinput>
   </screen>

   An empty line or a line starting with <literal>#</literal> is treated as a
   comment and will be ignored. A user name and its associated
   password must be written as one line using the following format:

   <synopsis>
    <replaceable>username</>:<replaceable>[md5 encrypted password]</>
   </synopsis>

   <replaceable>[md5 encrypted password]</> can be produced with the <filename>$prefix/bin/pg_md5</> command.

   <screen>
    <prompt>$</> <userinput>pg_md5 your_password</userinput>
    1060b7b46a3bd36b3a0d66e0127d0517
   </screen>

   If you don't want pass the password as the argument, execute <command>pg_md5 -p</command>.

   <screen>
    <prompt>$</> <userinput>pg_md5 -p</userinput>
    password: <userinput>your_password</userinput>
   </screen>

   The <filename>pcp.conf</filename> file must be readable by the
   user who executes <productname>Pgpool-II</productname>.
  </para>

 </sect1>

 <sect1 id="configuring-pgpool">
  <title>Configuring Pgpool-II</title>

  <indexterm>
   <primary>Pgpool-II configuration</primary>
  </indexterm>

  <sect2 id="configuring-pgpool-conf">
   <title>Configuring pgpool.conf</title>

   <para>
    <filename>pgpool.conf</filename> is the main configuration file
    of <productname>Pgpool-II</productname>. You need to specify the
    path to the file when
    starting <productname>Pgpool-II</productname>
    using <option>-f</option> option.
    <filename>pgpool.conf</filename> is located
    at <filename>$prefix/etc/pgpool.conf</filename> by default,
    if it installed from source code.
   </para>

   <para>
    To specify the <productname>Pgpool-II</productname> clustering
    mode, set <xref linkend="guc-backend-clustering-mode"> parameter
    to the value explained below.
   </para>

   <table id="backend-clustering-mode-table">
    <title><xref linkend="guc-backend-clustering-mode"> value in pgpool.conf</title>

    <tgroup cols="2">
     <thead>
      <row>
       <entry>Clustering mode</entry>
       <entry>value</entry>
      </row>
     </thead>

     <tbody>
      <row>
       <entry>Streaming replication mode</entry>
       <entry><literal>streaming_replication</literal></entry>
      </row>
      <row>
       <entry>Replication mode</entry>
       <entry><literal>native_replication</literal></entry>
      </row>
      <row>
       <entry>Logical replication mode</entry>
       <entry><literal>logical_replication</literal> </entry>
      </row>
      <row>
       <entry>Slony mode</entry>
       <entry><literal>slony</literal></entry>
      </row>
      <row>
       <entry>Snapshot isolation mode</entry>
       <entry><literal>snapshot_isolation</literal></entry>
      </row>
      <row>
       <entry>Raw mode</entry>
       <entry><literal>raw</literal> </entry>
      </row>
     </tbody>
    </tgroup>
   </table>

   <para>
    These configuration files are located at /usr/local/etc with
    default installation from source code.
    You can copy one of them as <filename>pgpool.conf</filename>.
    (probably you need root privilege for this)
    <programlisting>
# cd /usr/local/etc
# cp pgpool.conf.sample pgpool.conf
    </programlisting>
   </para>
  </sect2>

  <sect2 id="running-mode">
   <title>Clustering mode of Pgpool-II</title>

   <indexterm zone="running-mode">
    <primary>streaming replication mode</primary>
   </indexterm>

   <indexterm zone="running-mode">
    <primary>native replication mode</primary>
   </indexterm>

   <indexterm zone="running-mode">
    <primary>main replica mode</primary>
   </indexterm>

   <indexterm zone="running-mode">
    <primary>logical replication mode</primary>
   </indexterm>

   <indexterm zone="running-mode">
    <primary>snapshot isolation mode</primary>
   </indexterm>

   <para>
    There are six different <link linkend="guc-backend-clustering-mode">clustering modes</link> in
    <productname>Pgpool-II</>: streaming replication mode, logical
    replication mode, main replica mode (slony mode), native
    replication mode, raw mode and snapshot isolation mode. In any
    mode, <productname>Pgpool-II</> provides connection pooling, and
    automatic fail over.  Online recovery can be used only with
    streaming replication mode, snapshot isolation mode and native
    replication mode.  See <xref linkend="runtime-online-recovery">
    for more details of online recovery.
   </para>

   <para>
    Those modes are exclusive each other and cannot be changed after
    starting the server. You should make a decision which to use in
    the early stage of designing the system. If you are not sure, it
    is recommended to use the streaming replication mode or the
    snapshot isolation mode.
   </para>

   <para>
    The <firstterm>streaming replication mode</firstterm> can be used
    with <productname>PostgreSQL</> servers operating streaming
    replication. In this mode, <productname>PostgreSQL</> is
    responsible for synchronizing databases. This mode is widely used
    and most recommended way to use <productname>Pgpool-II</>. Load
    balancing is possible in the mode. visibility consistency among
    nodes is not guaranteed.
   </para>

   <para>
    In the <firstterm>snapshot isolation mode</firstterm>
    <productname>Pgpool-II</> is responsible for synchronizing
    databases. The advantage for the mode is the synchronization is
    done in synchronous way: writing to the database does not return
    until all of <productname>PostgreSQL</> servers finish the write
    operation. Also it guarantees the visibility consistency among
    nodes. To put it simply, it means that the visibility rule of
    transactions on single server is applied to a cluster consisting
    of multiple servers as well.  This is a remarkable feature of the
    snapshot isolation mode in
    <productname>Pgpool-II</productname>. In fact, the snapshot
    isolation mode in <productname>Pgpool-II</productname> is the only
    system which guarantees the visibility consistency among nodes
    without modifications to <productname>PostgreSQL</> at the moment.
    Because of this, applications do not need to recognize that they
    are using a cluster consisting of
    <productname>PostgreSQL</productname> servers, rather than a
    single <productname>PostgreSQL</productname> system.  However in
    this mode the transaction isolation level <emphasis>must be
    REPEATABLE READ</emphasis>.  You need to set
    <filename>postgresql.conf</filename> like this:
    <programlisting>
default_transaction_isolation = 'repeatable read'
    </programlisting>
   </para>
   <para>
    Also you need to aware that performance in the mode may be worse
    than the streaming replication mode and native replication mode
    due to the overhead to keep the consistency in transactions.
   </para>
   
   <para>
    In the <firstterm>native replication mode</firstterm>,
    <productname>Pgpool-II</> is responsible for synchronizing
    databases. The advantage for the mode is the synchronization is
    done in synchronous way: writing to the database does not return
    until all of <productname>PostgreSQL</> servers finish the write
    operation. As visibility consistency among nodes is not
    guaranteed, it is recommended to use the snapshot isolation mode
    except you want to use other than REPEATABLE READ isolation mode.
    Load balancing is possible in the mode.
   </para>

   <para>
    The <firstterm>logical replication mode</firstterm> can be used
    with <productname>PostgreSQL</> servers operating logical
    replication. In this mode, <productname>PostgreSQL</> is
    responsible for synchronizing tables. Load balancing is possible
    in the mode. Since logical replication does not replicate all
    tables, it's user's responsibility to replicate the table which
    could be load balanced. <productname>Pgpool-II</> load balances
    all tables. This means that if a table is not
    replicated, <productname>Pgpool-II</> may lookup outdated tables
    in the subscriber side.
   </para>

   <para>
    The <firstterm>main replica mode</firstterm> (slony mode)
    can be used with <productname>PostgreSQL</> servers
    operating <productname>Slony</>. In this
    mode, <productname>Slony</>/<productname>PostgreSQL</> is
    responsible for synchronizing
    databases. Since <productname>Slony-I</> is being obsoleted by
    streaming replication, we do not recommend to use this mode
    unless you have specific reason to
    use <productname>Slony</>. Load balancing is possible in the
    mode.
   </para>

   <para>
    In the <firstterm>raw
     mode</firstterm>, <productname>Pgpool-II</> does not care about
    the database synchronization. It's user's responsibility to make
    the whole system does a meaningful thing. Load balancing
    is <emphasis>not</emphasis> possible in the mode.
   </para>
  </sect2>

  <sect2 id="process-management-mode">
   <title>Process management modes</title>

   <indexterm zone="running-mode">
    <primary>dynamic process management</primary>
   </indexterm>

   <indexterm zone="running-mode">
    <primary>static process management</primary>
   </indexterm>
     <para>
          <productname>Pgpool-II</> implements a multi-process architecture where
          each child process can handle exactly one client connection at any time.
          The total number of concurrent client connections <productname>Pgpool-II</>
          can handle is configured by the
           <link linkend="guc-num-init-children">num_init_children</link> config parameter.
          <productname>Pgpool-II</> supports two child process management modes.
          <firstterm>Dynamic</firstterm> and <firstterm>Static</firstterm>.
          In static process management mode, <productname>Pgpool-II</> pre-forks the
          num_init_children number of child
          process at startup, and each child process keeps listening for incoming
          client connection. While with dynamic process management mode,
          <productname>Pgpool-II</> keeps track of idle processes and forks or kills
          processes to keep this number within the specified boundaries.
     </para>
     <para>
       <xref linkend="guc-process-management-mode"> is not available prior to
       <productname>Pgpool-II </productname><emphasis>V4.4</emphasis>.
      </para>
  </sect2>


 </sect1>

 <sect1 id="configuring-backend-info">
  <title>Configuring backend information</title>

  <para>
   For <productname>Pgpool-II</productname> to recognize <productname>PostgreSQL</>
   backend servers, you need to configure <varname>backend*</varname>
   in <filename>pgpool.conf</filename>.  For starters, at
   least <xref linkend="guc-backend-hostname">
    and <xref linkend="guc-backend-port"> parameters are required to
     be set up to start <productname>Pgpool-II</> server.
  </para>

  <sect2 id="backend-settings">
   <title>Backend Settings</title>

   <para>
    Backend <productname>PostgreSQL</> used by <productname>Pgpool-II</> must be specified in <filename>pgpool.conf</filename>.
    See <xref linkend="runtime-config-backend-settings">
   </para>

   <!--
   <variablelist>

   <varlistentry id="guc-backend-hostname" xreflabel="backend_hostname">
   <term><varname>backend_hostname</varname> (<type>string</type>)
   <indexterm>
   <primary><varname>backend_hostname</varname> configuration parameter</primary>
  </indexterm>
  </term>

   <listitem>
   <para>
   <varname>backend_hostname</varname> specifies where to
   connect to the <productname>PostgreSQL</productname>
   backend. It is used
   by <productname>Pgpool-II</productname> to communicate
   with the server.
  </para>

   <para>
   For TCP/IP communication, this parameter can take a hostname
   or an IP address. If this begins with a slash, it specifies
   Unix-domain communication rather than TCP/IP; the value is
   the name of the directory in which the socket file is
   stored. The default behavior when backend_hostname is empty
   ('') is to connect to a Unix-domain socket in /tmp.
  </para>

   <para>
   Multiple backends can be specified by adding a number at the
   end of the parameter name (e.g.backend_hostname0). This
   number is referred to as "DB node ID", and it starts from
   0. The backend which was given the DB node ID of 0 will be
   called "Main DB". When multiple backends are defined, the
   service can be continued even if the Main DB is down (not
   true in some modes). In this case, the youngest DB node ID
   alive will be the new Main DB.
  </para>

   <para>
   Please note that the DB node which has id 0 has no special
   meaning if operated in streaming replication mode. Rather,
   you should care about if the DB node is the "primary node" or
   not. See Streaming Replication for more details.
  </para>

   <para>
   If you plan to use only one PostgreSQL server, specify it by
   backend_hostname0.
  </para>

   <para>
   New nodes can be added in this parameter by reloading a
   configuration file. However, values cannot be updated so
   you must restart <productname>Pgpool-II</productname> in
   that case.
  </para>

  </listitem>

  </varlistentry>

   <varlistentry id="guc-backend-port" xreflabel="backend_port">
   <term><varname>backend_port</varname> (<type>integer</type>)
   <indexterm>
   <primary><varname>backend_port</varname> configuration parameter</primary>
  </indexterm>
  </term>

   <listitem>
   <para>
   <varname>backend_port</varname> specifies the port number
   of the backends. Multiple backends can be specified by
   adding a number at the end of the parameter name
   (e.g. backend_port0). If you plan to use only one
   PostgreSQL server, specify it by backend_port0.
  </para>
   <para>
   New backend ports can be added in this parameter by
   reloading a configuration file. However, values cannot be
   updated so you must
   restart <productname>Pgpool-II</productname> in that case.
  </para>
  </listitem>
  </varlistentry>

   <varlistentry id="guc-backend-data-directory" xreflabel="backend_data_directory">
   <term><varname>backend_data_directory</varname> (<type>string</type>)
   <indexterm>
   <primary><varname>backend_data_directory</varname> configuration parameter</primary>
  </indexterm>
  </term>

   <listitem>
   <para>
   <varname>backend_data_directory</varname> specifies the
   database directory. Multiple backends can be
   specified by adding a number at the end of the parameter
   name (e.g. backend_data_directory0). If you plan to use
   only one PostgreSQL server, specify it by
   backend_data_directory0.
  </para>
   <para>
   New backend data_directories can be added in this parameter
   by reloading a configuration file. However, values cannot
   be updated so you must
   restart <productname>Pgpool-II</productname> in that case.
  </para>
  </listitem>
  </varlistentry>

   <varlistentry id="guc-backend-flag" xreflabel="backend_flag">
   <term><varname>backend_flag</varname> (<type>string</type>)
   <indexterm>
   <primary><varname>backend_flag</varname> configuration parameter</primary>
  </indexterm>
  </term>

   <listitem>
   <para>
   <varname>backend_flag</varname> controls various backend
   behavior. Multiple backends can be specified by adding a
   number at the end of the parameter name
   (e.g. backend_flag0). If you plan to use only one
   PostgreSQL server, specify it by backend_flag0.
  </para>
   <para>
   New backend flags can be added in this parameter by
   reloading a configuration file.  Currently followings are
   allowed. Multiple flags can be specified by using "|".
  </para>

   <table id="backend-flag-table">
   <title>Backend flags</title>
   <tgroup cols="2">
   <thead>
   <row>
   <entry>Flag</entry>
   <entry>Description</entry>
  </row>
  </thead>

   <tbody>
   <row>
   <entry><literal>ALLOW_TO_FAILOVER</literal></entry>
   <entry>Allow to failover or detaching backend. This
   is the default. You cannot specify with
   DISALLOW_TO_FAILOVER at a same time.</entry>
  </row>
   <row>
   <entry><literal>DISALLOW_TO_FAILOVER</literal></entry>
   <entry>This is useful when you protect backend by
   using HA (High Availability) softwares such as
   Heartbeat or Pacemaker. You cannot specify with
   ALLOW_TO_FAILOVER at a same time.
  </entry>
  </row>
  </tbody>
  </tgroup>
  </table>

  </listitem>
  </varlistentry>

  </variablelist>

   -->

  </sect2>

 </sect1>

 <sect1 id="server-start">
  <title>Starting Pgpool-II and PostgreSQL</title>

  <para>
   To start <productname>Pgpool-II</productname>, execute:
<screen>
$ <userinput>pgpool -f /usr/local/etc/pgpool.conf -F /usr/local/etc/pcp.conf</userinput>
</screen>
   which will start the server running in the background. "-f"
   specifies the path to the main pgpool configuration file and "-F"
   specifies the path to the configuration file of pcp server, which
   is the control server for <productname>Pgpool-II</productname>. For
   other options of the command please take a look at <xref
   linkend="pgpool"> manual.
  </para>

  <para>
   Before starting <productname>Pgpool-II</productname>, you must
   start <productname>PostgreSQL</productname> because if
   <productname>PostgreSQL</productname> has not started yet,
   <productname>Pgpool-II</productname> triggers failover process and
   makes <productname>PostgreSQL</productname> is in down status.
  </para>

  <para>
   If you have difficulty in controlling the startup sequence of
   <productname>PostgreSQL</productname>, for example
   <productname>Pgpool-II</productname> and
   <productname>PostgreSQL</productname> are installed on different
   servers, you can make <xref
   linkend="guc-search-primary-node-timeout"> longer (the default is 5
   minutes) so that <productname>Pgpool-II</productname> waits for
   <productname>PostgreSQL</productname> starts up until <xref
   linkend="guc-search-primary-node-timeout"> expires. If
   <productname>PostgreSQL</productname> starts up before <xref
   linkend="guc-search-primary-node-timeout"> expires,
   <productname>Pgpool-II</productname> should start up without
   problem. If <xref linkend="guc-search-primary-node-timeout">
   expires before <productname>PostgreSQL</productname> starts up, no
   primary node will be detected, which means you cannot execute
   DML/DDL. You need to restart <productname>Pgpool-II</productname>
   in this case. To confirm that the primary node exists you can use
   <xref linkend="sql-show-pool-nodes"> command.
  </para>

  <para>
   Please note <xref linkend="guc-search-primary-node-timeout"> can
   only be used in the streaming replication mode because the
   parameter is only valid in the mode. See <xref
   linkend="running-mode"> for more details about streaming
   replication mode. For other mode, tweak the health check (see <xref
   linkend="runtime-config-health-check">) parameters so that there's
   enough time before <productname>PostgreSQL</productname> becomes
   available.
  </para>

  <para>
   If health check detects that <productname>PostgreSQL</productname>
   is not available before <productname>Pgpool-II</productname> starts
   up, some or all <productname>PostgreSQL</productname> are
   recognized in "down" status. In this case you need to manually put
   the <productname>PostgreSQL</productname> server in "up" state by
   using <xref linkend="pcp-attach-node"> command.  If a client tries
   to connect to <productname>Pgpool-II</productname> before
   <productname>PostgreSQL</productname> is available, failover could
   be triggered. In this case you also need to execute <xref
   linkend="pcp-attach-node"> command to put the
   <productname>PostgreSQL</productname> server in "up" state.
  </para>
 </sect1>

 <sect1 id="server-stop">
  <title>Stopping Pgpool-II and PostgreSQL</title>

  <indexterm zone="server-stop">
   <primary>shutdown</primary>
  </indexterm>

  <para>
   To stop <productname>Pgpool-II</productname>, execute:
<screen>
$ <userinput>pgpool -f /usr/local/etc/pgpool.conf -F /usr/local/etc/pcp.conf -m fast stop</userinput>
</screen>
   "-m" option specifies how gently stops
   <productname>Pgpool-II</productname>. "fast" means shutdown
   <productname>Pgpool-II</productname> immediately even if there are
   existing connections from clients. You can specify "smart" to the
   option, which force <productname>Pgpool-II</productname> to wait
   until all clients are disconnected from
   <productname>Pgpool-II</productname>. But this could make
   <productname>Pgpool-II</productname> wait forever and this may
   result in sending SIGKILL signal from the operating system and
   leaving garbage, which will bring troubles next time when you start
   <productname>Pgpool-II</productname>.
  </para>

  <para>
   After shutting down <productname>Pgpool-II</productname>, you can
   shutdown <productname>PostgreSQL</productname>.
  </para>
 </sect1>

 <sect1 id="server-temporarily-shutdown">
  <title>Temporarily Shutting Down PostgreSQL</title>

  <indexterm zone="server-temporarily-shutdown">
   <primary>shutdown</primary>
  </indexterm>

  <para>
   Sometimes you want to temporarily stop or restart
   <productname>PostgreSQL</productname> to maintain or version up
   it. In this section how to perform the task with minimum downtime.
  </para>

  <sect2 id="using-pcp-detach-node">
   <title>Using pcp_detach_node command</title>
   <para>
    If you stop <productname>PostgreSQL</productname> by using
    <command>pg_ctl</command>, failover will not happen until
    <productname>Pgpool-II</productname> detects it by the health
    check depending on the health check settings and it will take
    sometime to detach <productname>PostgreSQL</productname>.
    Especially if <link linkend="tutorial-watchdog">Watchdog</link> is
    enabled and <xref linkend="guc-failover-require-consensus"> is on,
    <productname>Pgpool-II</productname> will not start failover until
    more than half of watchdog nodes agree that
    <productname>PostgreSQL</productname> is stopped. If you detach
    the node by using <xref linkend="pcp-detach-node">, failover will
    immediately start regardless the settings of health check. Please
    note that the detached <productname>PostgreSQL</productname> node
    is not actually stopped and if necessary, you need to manually
    stop it.
   </para>
  </sect2>

  <sect2 id="using-backend-flag">
   <title>Using backend_flag</title>
   <para>
    Stopping or restarting <productname>PostgreSQL</productname>
    causes failover. If the running mode is not streaming replication
    mode, or the server is a standby server in streaming replication
    mode, probably that's not a big deal because clients can always
    use other servers in the cluster. However if the server is primary
    server, it will cause change of primary server by promoting one of
    the standby servers. Moreover if there's only one server remains
    in the cluster, there are no alternative server or standby server
    which can be promoted.
   </para>

   <para>
    In this case you can use <xref linkend="guc-backend-flag"> to avoid
     failover.  By setting below in pgpool.conf will avoid failover of
     backend0.
     <programlisting>
backend_flag0 = DISALLOW_TO_FAILOVER
     </programlisting>
     This will take effect by reloading or restarting
     <productname>Pgpool-II</productname>. If this flag is set, failover
     will not happen if the backend is not available. While the backend
     is not available, clients will get error message:
     <programlisting>
psql: error: could not connect to server: FATAL:  failed to create a backend connection
DETAIL:  executing failover on backend
     </programlisting>
     After restarting the backend, clients can connect as usual.
     To allow failover on the backend again, you can set:
     <programlisting>
backend_flag0 = ALLOW_TO_FAILOVER
     </programlisting>
     and reload or restart <productname>Pgpool-II</productname>.
   </para>
  </sect2>

 </sect1>

 <sect1 id="backup">
  <title>Backing up PostgreSQL database</title>

  <indexterm zone="backup">
   <primary>backup</primary>
  </indexterm>

  <para>
   If you plan to backup <productname>PostgreSQL</productname>
   database
   using <command>pg_dump</command>, <command>pg_basebackup</command>
   or any other tools, we strongly recommend to run the commands
   against <productname>PostgreSQL</productname>
   directly. As <productname>Pgpool-II</productname> is a proxy
   software, it gives an overhead for relaying message packets. Since
   obtaining a backup tends to produce lots of packets, performing
   backup through <productname>Pgpool-II</productname> will be slow
   compared with directly
   connecting <productname>PostgreSQL</productname> unless the
   database is very small.
  </para>
  <para>
   Moreover, parallel <command>pg_dump</command> raises error if it is
   executed through <productname>Pgpool-II</productname> because the
   command handles snapshot id, which is a database depending object.
  </para>
  <para>
   In most cases you want to choose primary server as the backup
   target.  If you want backup standby server, you have to be very
   careful in choosing the right <productname>PostgreSQL</productname>
   server to obtain a backup because if the data is outdated, you are
   likely to have outdated database backup. You can
   use <xref linkend="SQL-SHOW-POOL-NODES">
   or <xref linkend="PCP-NODE-INFO"> to know how the standby server
   catches up the primary server.
  </para>
 </sect1>

</chapter>