pgsql/doc/src/sgml/html/kernel-resources.html - master - Ao/MaxReport

pgsql/doc/src/sgml/html/kernel-resources.html
537 строк40 KB
AoAnima
first_commit
13 июл 2026, 17:47
13 июл 2026, 17:47f1a670a
<?xml version="1.0" encoding="UTF-8" standalone="no"?>
<!DOCTYPE html PUBLIC "-//W3C//DTD XHTML 1.0 Transitional//EN" "http://www.w3.org/TR/xhtml1/DTD/xhtml1-transitional.dtd"><html xmlns="http://www.w3.org/1999/xhtml"><head><meta http-equiv="Content-Type" content="text/html; charset=UTF-8" /><title>18.4. Managing Kernel Resources</title><link rel="stylesheet" type="text/css" href="stylesheet.css" /><link rev="made" href="pgsql-docs@lists.postgresql.org" /><meta name="generator" content="DocBook XSL Stylesheets Vsnapshot" /><link rel="prev" href="server-start.html" title="18.3. Starting the Database Server" /><link rel="next" href="server-shutdown.html" title="18.5. Shutting Down the Server" /></head><body id="docContent" class="container-fluid col-10"><div class="navheader"><table width="100%" summary="Navigation header"><tr><th colspan="5" align="center">18.4. Managing Kernel Resources</th></tr><tr><td width="10%" align="left"><a accesskey="p" href="server-start.html" title="18.3. Starting the Database Server">Prev</a> </td><td width="10%" align="left"><a accesskey="u" href="runtime.html" title="Chapter 18. Server Setup and Operation">Up</a></td><th width="60%" align="center">Chapter 18. Server Setup and Operation</th><td width="10%" align="right"><a accesskey="h" href="index.html" title="PostgreSQL 18.4 Documentation">Home</a></td><td width="10%" align="right"> <a accesskey="n" href="server-shutdown.html" title="18.5. Shutting Down the Server">Next</a></td></tr></table><hr /></div><div class="sect1" id="KERNEL-RESOURCES"><div class="titlepage"><div><div><h2 class="title" style="clear: both">18.4. Managing Kernel Resources <a href="#KERNEL-RESOURCES" class="id_link">#</a></h2></div></div></div><div class="toc"><dl class="toc"><dt><span class="sect2"><a href="kernel-resources.html#SYSVIPC">18.4.1. Shared Memory and Semaphores</a></span></dt><dt><span class="sect2"><a href="kernel-resources.html#SYSTEMD-REMOVEIPC">18.4.2. systemd RemoveIPC</a></span></dt><dt><span class="sect2"><a href="kernel-resources.html#KERNEL-RESOURCES-LIMITS">18.4.3. Resource Limits</a></span></dt><dt><span class="sect2"><a href="kernel-resources.html#LINUX-MEMORY-OVERCOMMIT">18.4.4. Linux Memory Overcommit</a></span></dt><dt><span class="sect2"><a href="kernel-resources.html#LINUX-HUGE-PAGES">18.4.5. Linux Huge Pages</a></span></dt></dl></div><p>
   <span class="productname">PostgreSQL</span> can sometimes exhaust various operating system
   resource limits, especially when multiple copies of the server are running
   on the same system, or in very large installations.  This section explains
   the kernel resources used by <span class="productname">PostgreSQL</span> and the steps you
   can take to resolve problems related to kernel resource consumption.
  </p><div class="sect2" id="SYSVIPC"><div class="titlepage"><div><div><h3 class="title">18.4.1. Shared Memory and Semaphores <a href="#SYSVIPC" class="id_link">#</a></h3></div></div></div><a id="id-1.6.5.7.3.2" class="indexterm"></a><a id="id-1.6.5.7.3.3" class="indexterm"></a><p>
    <span class="productname">PostgreSQL</span> requires the operating system to provide
    inter-process communication (<acronym class="acronym">IPC</acronym>) features, specifically
    shared memory and semaphores.  Unix-derived systems typically provide
    <span class="quote">“<span class="quote"><span class="systemitem">System V</span></span>”</span> <acronym class="acronym">IPC</acronym>,
    <span class="quote">“<span class="quote"><span class="systemitem">POSIX</span></span>”</span> <acronym class="acronym">IPC</acronym>, or both.
    <span class="systemitem">Windows</span> has its own implementation of
    these features and is not discussed here.
   </p><p>
    By default, <span class="productname">PostgreSQL</span> allocates
    a very small amount of System V shared memory, as well as a much larger
    amount of anonymous <code class="function">mmap</code> shared memory.
    Alternatively, a single large System V shared memory region can be used
    (see <a class="xref" href="runtime-config-resource.html#GUC-SHARED-MEMORY-TYPE">shared_memory_type</a>).

    In addition a significant number of semaphores, which can be either
    System V or POSIX style, are created at server startup.  Currently,
    POSIX semaphores are used on Linux and FreeBSD systems while other
    platforms use System V semaphores.
   </p><p>
    System V <acronym class="acronym">IPC</acronym> features are typically constrained by
    system-wide allocation limits.
    When <span class="productname">PostgreSQL</span> exceeds one of these limits,
    the server will refuse to start and
    should leave an instructive error message describing the problem
    and what to do about it. (See also <a class="xref" href="server-start.html#SERVER-START-FAILURES" title="18.3.1. Server Start-up Failures">Section 18.3.1</a>.) The relevant kernel
    parameters are named consistently across different systems; <a class="xref" href="kernel-resources.html#SYSVIPC-PARAMETERS" title="Table 18.1. System V IPC Parameters">Table 18.1</a> gives an overview. The methods to set
    them, however, vary. Suggestions for some platforms are given below.
   </p><div class="table" id="SYSVIPC-PARAMETERS"><p class="title"><strong>Table 18.1. <span class="systemitem">System V</span> <acronym class="acronym">IPC</acronym> Parameters</strong></p><div class="table-contents"><table class="table" summary="System V IPC Parameters" border="1"><colgroup><col class="col1" /><col class="col2" /><col class="col3" /></colgroup><thead><tr><th>Name</th><th>Description</th><th>Values needed to run one <span class="productname">PostgreSQL</span> instance</th></tr></thead><tbody><tr><td><code class="varname">SHMMAX</code></td><td>Maximum size of shared memory segment (bytes)</td><td>at least 1kB, but the default is usually much higher</td></tr><tr><td><code class="varname">SHMMIN</code></td><td>Minimum size of shared memory segment (bytes)</td><td>1</td></tr><tr><td><code class="varname">SHMALL</code></td><td>Total amount of shared memory available (bytes or pages)</td><td>same as <code class="varname">SHMMAX</code> if bytes,
        or <code class="literal">ceil(SHMMAX/PAGE_SIZE)</code> if pages,
        plus room for other applications</td></tr><tr><td><code class="varname">SHMSEG</code></td><td>Maximum number of shared memory segments per process</td><td>only 1 segment is needed, but the default is much higher</td></tr><tr><td><code class="varname">SHMMNI</code></td><td>Maximum number of shared memory segments system-wide</td><td>like <code class="varname">SHMSEG</code> plus room for other applications</td></tr><tr><td><code class="varname">SEMMNI</code></td><td>Maximum number of semaphore identifiers (i.e., sets)</td><td>at least <code class="literal">ceil(num_os_semaphores / 16)</code> plus room for other applications</td></tr><tr><td><code class="varname">SEMMNS</code></td><td>Maximum number of semaphores system-wide</td><td><code class="literal">ceil(num_os_semaphores / 16) * 17</code> plus room for other applications</td></tr><tr><td><code class="varname">SEMMSL</code></td><td>Maximum number of semaphores per set</td><td>at least 17</td></tr><tr><td><code class="varname">SEMMAP</code></td><td>Number of entries in semaphore map</td><td>see text</td></tr><tr><td><code class="varname">SEMVMX</code></td><td>Maximum value of semaphore</td><td>at least 1000 (The default is often 32767; do not change unless necessary)</td></tr></tbody></table></div></div><br class="table-break" /><p>
    <span class="productname">PostgreSQL</span> requires a few bytes of System V shared memory
    (typically 48 bytes, on 64-bit platforms) for each copy of the server.
    On most modern operating systems, this amount can easily be allocated.
    However, if you are running many copies of the server or you explicitly
    configure the server to use large amounts of System V shared memory (see
    <a class="xref" href="runtime-config-resource.html#GUC-SHARED-MEMORY-TYPE">shared_memory_type</a> and <a class="xref" href="runtime-config-resource.html#GUC-DYNAMIC-SHARED-MEMORY-TYPE">dynamic_shared_memory_type</a>), it may be necessary to
    increase <code class="varname">SHMALL</code>, which is the total amount of System V shared
    memory system-wide.  Note that <code class="varname">SHMALL</code> is measured in pages
    rather than bytes on many systems.
   </p><p>
    Less likely to cause problems is the minimum size for shared
    memory segments (<code class="varname">SHMMIN</code>), which should be at most
    approximately 32 bytes for <span class="productname">PostgreSQL</span> (it is
    usually just 1). The maximum number of segments system-wide
    (<code class="varname">SHMMNI</code>) or per-process (<code class="varname">SHMSEG</code>) are unlikely
    to cause a problem unless your system has them set to zero.
   </p><p>
    When using System V semaphores,
    <span class="productname">PostgreSQL</span> uses one semaphore per allowed connection
    (<a class="xref" href="runtime-config-connection.html#GUC-MAX-CONNECTIONS">max_connections</a>), allowed autovacuum worker process
    (<a class="xref" href="runtime-config-vacuum.html#GUC-AUTOVACUUM-WORKER-SLOTS">autovacuum_worker_slots</a>), allowed WAL sender process
    (<a class="xref" href="runtime-config-replication.html#GUC-MAX-WAL-SENDERS">max_wal_senders</a>), allowed background
    process (<a class="xref" href="runtime-config-resource.html#GUC-MAX-WORKER-PROCESSES">max_worker_processes</a>), etc., in sets of 16.
    The runtime-computed parameter <a class="xref" href="runtime-config-preset.html#GUC-NUM-OS-SEMAPHORES">num_os_semaphores</a>
    reports the number of semaphores required.  This parameter can be viewed
    before starting the server with a <code class="command">postgres</code> command like:
</p><pre class="programlisting">
$ <strong class="userinput"><code>postgres -D $PGDATA -C num_os_semaphores</code></strong>
</pre><p>
   </p><p>
    Each set of 16 semaphores will
    also contain a 17th semaphore which contains a <span class="quote">“<span class="quote">magic
    number</span>”</span>, to detect collision with semaphore sets used by
    other applications. The maximum number of semaphores in the system
    is set by <code class="varname">SEMMNS</code>, which consequently must be at least
    as high as <code class="literal">num_os_semaphores</code> plus one extra for
    each set of 16 required semaphores (see the formula in <a class="xref" href="kernel-resources.html#SYSVIPC-PARAMETERS" title="Table 18.1. System V IPC Parameters">Table 18.1</a>).  The parameter <code class="varname">SEMMNI</code>
    determines the limit on the number of semaphore sets that can
    exist on the system at one time.  Hence this parameter must be at
    least <code class="literal">ceil(num_os_semaphores / 16)</code>.
    Lowering the number
    of allowed connections is a temporary workaround for failures,
    which are usually confusingly worded <span class="quote">“<span class="quote">No space
    left on device</span>”</span>, from the function <code class="function">semget</code>.
   </p><p>
    In some cases it might also be necessary to increase
    <code class="varname">SEMMAP</code> to be at least on the order of
    <code class="varname">SEMMNS</code>.  If the system has this parameter
    (many do not), it defines the size of the semaphore
    resource map, in which each contiguous block of available semaphores
    needs an entry. When a semaphore set is freed it is either added to
    an existing entry that is adjacent to the freed block or it is
    registered under a new map entry. If the map is full, the freed
    semaphores get lost (until reboot). Fragmentation of the semaphore
    space could over time lead to fewer available semaphores than there
    should be.
   </p><p>
    Various other settings related to <span class="quote">“<span class="quote">semaphore undo</span>”</span>, such as
    <code class="varname">SEMMNU</code> and <code class="varname">SEMUME</code>, do not affect
    <span class="productname">PostgreSQL</span>.
   </p><p>
    When using POSIX semaphores, the number of semaphores needed is the
    same as for System V, that is one semaphore per allowed connection
    (<a class="xref" href="runtime-config-connection.html#GUC-MAX-CONNECTIONS">max_connections</a>), allowed autovacuum worker process
    (<a class="xref" href="runtime-config-vacuum.html#GUC-AUTOVACUUM-WORKER-SLOTS">autovacuum_worker_slots</a>), allowed WAL sender process
    (<a class="xref" href="runtime-config-replication.html#GUC-MAX-WAL-SENDERS">max_wal_senders</a>), allowed background
    process (<a class="xref" href="runtime-config-resource.html#GUC-MAX-WORKER-PROCESSES">max_worker_processes</a>), etc.
    On the platforms where this option is preferred, there is no specific
    kernel limit on the number of POSIX semaphores.
   </p><div class="variablelist"><dl class="variablelist"><dt><span class="term"><span class="systemitem">FreeBSD</span>
      <a id="id-1.6.5.7.3.15.1.1.2" class="indexterm"></a>
      </span></dt><dd><p>
        The default shared memory settings are usually good enough, unless
        you have set <code class="literal">shared_memory_type</code> to <code class="literal">sysv</code>.
        System V semaphores are not used on this platform.
       </p><p>
        The default IPC settings can be changed using
        the <code class="command">sysctl</code> or
        <code class="command">loader</code> interfaces.  The following
        parameters can be set using <code class="command">sysctl</code>:
</p><pre class="screen">
<code class="prompt">#</code> <strong class="userinput"><code>sysctl kern.ipc.shmall=32768</code></strong>
<code class="prompt">#</code> <strong class="userinput"><code>sysctl kern.ipc.shmmax=134217728</code></strong>
</pre><p>
        To make these settings persist over reboots, modify
        <code class="filename">/etc/sysctl.conf</code>.
       </p><p>
        If you have set <code class="literal">shared_memory_type</code> to
        <code class="literal">sysv</code>, you might also want to configure your kernel
        to lock System V shared memory into RAM and prevent it from being paged
        out to swap.  This can be accomplished using the <code class="command">sysctl</code>
        setting <code class="literal">kern.ipc.shm_use_phys</code>.
       </p><p>
        If running in a FreeBSD jail, you should set its
        <code class="literal">sysvshm</code> parameter to <code class="literal">new</code>, so that
        it has its own separate System V shared memory namespace.
        (Before FreeBSD 11.0, it was necessary to enable shared access to
        the host's IPC namespace from jails, and take measures to avoid
        collisions.)
       </p></dd><dt><span class="term"><span class="systemitem">NetBSD</span>
      <a id="id-1.6.5.7.3.15.2.1.2" class="indexterm"></a>
      </span></dt><dd><p>
        The default shared memory settings are usually good enough, unless
        you have set <code class="literal">shared_memory_type</code> to <code class="literal">sysv</code>.
        However, you will need to increase <code class="literal">kern.ipc.semmni</code>
        and <code class="literal">kern.ipc.semmns</code>,
        as <span class="systemitem">NetBSD</span>'s default settings
        for these are unworkably small.
       </p><p>
        IPC parameters can be adjusted using <code class="command">sysctl</code>,
        for example:
</p><pre class="screen">
<code class="prompt">#</code> <strong class="userinput"><code>sysctl -w kern.ipc.semmni=100</code></strong>
</pre><p>
        To make these settings persist over reboots, modify
        <code class="filename">/etc/sysctl.conf</code>.
       </p><p>
        If you have set <code class="literal">shared_memory_type</code> to
        <code class="literal">sysv</code>, you might also want to configure your kernel
        to lock System V shared memory into RAM and prevent it from being paged
        out to swap.  This can be accomplished using the <code class="command">sysctl</code>
        setting <code class="literal">kern.ipc.shm_use_phys</code>.
       </p></dd><dt><span class="term"><span class="systemitem">OpenBSD</span>
      <a id="id-1.6.5.7.3.15.3.1.2" class="indexterm"></a>
      </span></dt><dd><p>
        The default shared memory settings are usually good enough, unless
        you have set <code class="literal">shared_memory_type</code> to <code class="literal">sysv</code>.
        However, you will need to
        increase <code class="literal">kern.seminfo.semmni</code>
        and <code class="literal">kern.seminfo.semmns</code>,
        as <span class="systemitem">OpenBSD</span>'s default settings
        for these are unworkably small.
       </p><p>
        IPC parameters can be adjusted using <code class="command">sysctl</code>,
        for example:
</p><pre class="screen">
<code class="prompt">#</code> <strong class="userinput"><code>sysctl kern.seminfo.semmni=100</code></strong>
</pre><p>
        To make these settings persist over reboots, modify
        <code class="filename">/etc/sysctl.conf</code>.
       </p></dd><dt><span class="term"><span class="systemitem">Linux</span>
      <a id="id-1.6.5.7.3.15.4.1.2" class="indexterm"></a>
      </span></dt><dd><p>
        The default shared memory settings are usually good enough, unless
        you have set <code class="literal">shared_memory_type</code> to <code class="literal">sysv</code>,
        and even then only on older kernel versions that shipped with low defaults.
        System V semaphores are not used on this platform.
       </p><p>
        The shared memory size settings can be changed via the
        <code class="command">sysctl</code> interface.  For example, to allow 16 GB:
</p><pre class="screen">
<code class="prompt">$</code> <strong class="userinput"><code>sysctl -w kernel.shmmax=17179869184</code></strong>
<code class="prompt">$</code> <strong class="userinput"><code>sysctl -w kernel.shmall=4194304</code></strong>
</pre><p>
        To make these settings persist over reboots, see
        <code class="filename">/etc/sysctl.conf</code>.
       </p></dd><dt><span class="term"><span class="systemitem">macOS</span>
      <a id="id-1.6.5.7.3.15.5.1.2" class="indexterm"></a>
      </span></dt><dd><p>
        The default shared memory and semaphore settings are usually good enough, unless
        you have set <code class="literal">shared_memory_type</code> to <code class="literal">sysv</code>.
       </p><p>
        The recommended method for configuring shared memory in macOS
        is to create a file named <code class="filename">/etc/sysctl.conf</code>,
        containing variable assignments such as:
</p><pre class="programlisting">
kern.sysv.shmmax=4194304
kern.sysv.shmmin=1
kern.sysv.shmmni=32
kern.sysv.shmseg=8
kern.sysv.shmall=1024
</pre><p>
        Note that in some macOS versions,
        <span class="emphasis"><em>all five</em></span> shared-memory parameters must be set in
        <code class="filename">/etc/sysctl.conf</code>, else the values will be ignored.
       </p><p>
        <code class="varname">SHMMAX</code> can only be set to a multiple of 4096.
       </p><p>
        <code class="varname">SHMALL</code> is measured in 4 kB pages on this platform.
       </p><p>
        It is possible to change all but <code class="varname">SHMMNI</code> on the fly, using
        <span class="application">sysctl</span>.  But it's still best to set up your preferred
        values via <code class="filename">/etc/sysctl.conf</code>, so that the values will be
        kept across reboots.
       </p></dd><dt><span class="term"><span class="systemitem">Solaris</span><br /></span><span class="term"><span class="systemitem">illumos</span></span></dt><dd><p>
        The default shared memory and semaphore settings are usually good enough for most
        <span class="productname">PostgreSQL</span> applications.  Solaris defaults
        to a <code class="varname">SHMMAX</code> of one-quarter of system <acronym class="acronym">RAM</acronym>.
        To further adjust this setting, use a project setting associated
        with the <code class="literal">postgres</code> user.  For example, run the
        following as <code class="literal">root</code>:
</p><pre class="programlisting">
projadd -c "PostgreSQL DB User" -K "project.max-shm-memory=(privileged,8GB,deny)" -U postgres -G postgres user.postgres
</pre><p>
       </p><p>
        This command adds the <code class="literal">user.postgres</code> project and
        sets the shared memory maximum for the <code class="literal">postgres</code>
        user to 8GB, and takes effect the next time that user logs
        in, or when you restart <span class="productname">PostgreSQL</span> (not reload).
        The above assumes that <span class="productname">PostgreSQL</span> is run by
        the <code class="literal">postgres</code> user in the <code class="literal">postgres</code>
        group.  No server reboot is required.
       </p><p>
        Other recommended kernel setting changes for database servers which will
        have a large number of connections are:
</p><pre class="programlisting">
project.max-shm-ids=(priv,32768,deny)
project.max-sem-ids=(priv,4096,deny)
project.max-msg-ids=(priv,4096,deny)
</pre><p>
       </p><p>
        Additionally, if you are running <span class="productname">PostgreSQL</span>
        inside a zone, you may need to raise the zone resource usage
        limits as well.  See "Chapter2:  Projects and Tasks" in the
        <em class="citetitle">System Administrator's Guide</em> for more
        information on <code class="literal">projects</code> and <code class="command">prctl</code>.
       </p></dd></dl></div></div><div class="sect2" id="SYSTEMD-REMOVEIPC"><div class="titlepage"><div><div><h3 class="title">18.4.2. systemd RemoveIPC <a href="#SYSTEMD-REMOVEIPC" class="id_link">#</a></h3></div></div></div><a id="id-1.6.5.7.4.2" class="indexterm"></a><p>
    If <span class="productname">systemd</span> is in use, some care must be taken
    that IPC resources (including shared memory) are not prematurely
    removed by the operating system.  This is especially of concern when
    installing PostgreSQL from source.  Users of distribution packages of
    PostgreSQL are less likely to be affected, as
    the <code class="literal">postgres</code> user is then normally created as a system
    user.
   </p><p>
    The setting <code class="literal">RemoveIPC</code>
    in <code class="filename">logind.conf</code> controls whether IPC objects are
    removed when a user fully logs out.  System users are exempt.  This
    setting defaults to on in stock <span class="productname">systemd</span>, but
    some operating system distributions default it to off.
   </p><p>
    A typical observed effect when this setting is on is that shared memory
    objects used for parallel query execution are removed at apparently random
    times, leading to errors and warnings while attempting to open and remove
    them, like
</p><pre class="screen">
WARNING:  could not remove shared memory segment "/PostgreSQL.1450751626": No such file or directory
</pre><p>
    Different types of IPC objects (shared memory vs. semaphores, System V
    vs. POSIX) are treated slightly differently
    by <span class="productname">systemd</span>, so one might observe that some IPC
    resources are not removed in the same way as others.  But it is not
    advisable to rely on these subtle differences.
   </p><p>
    A <span class="quote">“<span class="quote">user logging out</span>”</span> might happen as part of a maintenance
    job or manually when an administrator logs in as
    the <code class="literal">postgres</code> user or something similar, so it is hard
    to prevent in general.
   </p><p>
    What is a <span class="quote">“<span class="quote">system user</span>”</span> is determined
    at <span class="productname">systemd</span> compile time from
    the <code class="symbol">SYS_UID_MAX</code> setting
    in <code class="filename">/etc/login.defs</code>.
   </p><p>
    Packaging and deployment scripts should be careful to create
    the <code class="literal">postgres</code> user as a system user by
    using <code class="literal">useradd -r</code>, <code class="literal">adduser --system</code>,
    or equivalent.
   </p><p>
    Alternatively, if the user account was created incorrectly or cannot be
    changed, it is recommended to set
</p><pre class="programlisting">
RemoveIPC=no
</pre><p>
    in <code class="filename">/etc/systemd/logind.conf</code> or another appropriate
    configuration file.
   </p><div class="caution"><h3 class="title">Caution</h3><p>
     At least one of these two things has to be ensured, or the PostgreSQL
     server will be very unreliable.
    </p></div></div><div class="sect2" id="KERNEL-RESOURCES-LIMITS"><div class="titlepage"><div><div><h3 class="title">18.4.3. Resource Limits <a href="#KERNEL-RESOURCES-LIMITS" class="id_link">#</a></h3></div></div></div><p>
    Unix-like operating systems enforce various kinds of resource limits
    that might interfere with the operation of your
    <span class="productname">PostgreSQL</span> server. Of particular
    importance are limits on the number of processes per user, the
    number of open files per process, and the amount of memory available
    to each process. Each of these have a <span class="quote">“<span class="quote">hard</span>”</span> and a
    <span class="quote">“<span class="quote">soft</span>”</span> limit. The soft limit is what actually counts
    but it can be changed by the user up to the hard limit. The hard
    limit can only be changed by the root user. The system call
    <code class="function">setrlimit</code> is responsible for setting these
    parameters. The shell's built-in command <code class="command">ulimit</code>
    (Bourne shells) or <code class="command">limit</code> (<span class="application">csh</span>) is
    used to control the resource limits from the command line. On
    BSD-derived systems the file <code class="filename">/etc/login.conf</code>
    controls the various resource limits set during login. See the
    operating system documentation for details. The relevant
    parameters are <code class="varname">maxproc</code>,
    <code class="varname">openfiles</code>, and <code class="varname">datasize</code>. For
    example:
</p><pre class="programlisting">
default:\
...
        :datasize-cur=256M:\
        :maxproc-cur=256:\
        :openfiles-cur=256:\
...
</pre><p>
    (<code class="literal">-cur</code> is the soft limit.  Append
    <code class="literal">-max</code> to set the hard limit.)
   </p><p>
    Kernels can also have system-wide limits on some resources.
    </p><div class="itemizedlist"><ul class="itemizedlist" style="list-style-type: disc; "><li class="listitem"><p>
      On <span class="productname">Linux</span> the kernel parameter
      <code class="varname">fs.file-max</code> determines the maximum number of open
      files that the kernel will support.  It can be changed with
      <code class="literal">sysctl -w fs.file-max=<em class="replaceable"><code>N</code></em></code>.
      To make the setting persist across reboots, add an assignment
      in <code class="filename">/etc/sysctl.conf</code>.
      The maximum limit of files per process is fixed at the time the
      kernel is compiled; see
      <code class="filename">/usr/src/linux/Documentation/proc.txt</code> for
      more information.
      </p></li></ul></div><p>
   </p><p>
    The <span class="productname">PostgreSQL</span> server uses one process
    per connection so you should provide for at least as many processes
    as allowed connections, in addition to what you need for the rest
    of your system.  This is usually not a problem but if you run
    several servers on one machine things might get tight.
   </p><p>
    The factory default limit on open files is often set to
    <span class="quote">“<span class="quote">socially friendly</span>”</span> values that allow many users to
    coexist on a machine without using an inappropriate fraction of
    the system resources.  If you run many servers on a machine this
    is perhaps what you want, but on dedicated servers you might want to
    raise this limit.
   </p><p>
    On the other side of the coin, some systems allow individual
    processes to open large numbers of files; if more than a few
    processes do so then the system-wide limit can easily be exceeded.
    If you find this happening, and you do not want to alter the
    system-wide limit, you can set <span class="productname">PostgreSQL</span>'s <a class="xref" href="runtime-config-resource.html#GUC-MAX-FILES-PER-PROCESS">max_files_per_process</a> configuration parameter to
    limit the consumption of open files.
   </p><p>
    Another kernel limit that may be of concern when supporting large
    numbers of client connections is the maximum socket connection queue
    length.  If more than that many connection requests arrive within a very
    short period, some may get rejected before the <span class="productname">PostgreSQL</span> server can service
    the requests, with those clients receiving unhelpful connection failure
    errors such as <span class="quote">“<span class="quote">Resource temporarily unavailable</span>”</span> or
    <span class="quote">“<span class="quote">Connection refused</span>”</span>.  The default queue length limit is 128
    on many platforms.  To raise it, adjust the appropriate kernel parameter
    via <span class="application">sysctl</span>, then restart the <span class="productname">PostgreSQL</span> server.
    The parameter is variously named <code class="varname">net.core.somaxconn</code>
    on Linux, <code class="varname">kern.ipc.soacceptqueue</code> on newer FreeBSD,
    and <code class="varname">kern.ipc.somaxconn</code> on macOS and other BSD
    variants.
   </p></div><div class="sect2" id="LINUX-MEMORY-OVERCOMMIT"><div class="titlepage"><div><div><h3 class="title">18.4.4. Linux Memory Overcommit <a href="#LINUX-MEMORY-OVERCOMMIT" class="id_link">#</a></h3></div></div></div><a id="id-1.6.5.7.6.2" class="indexterm"></a><a id="id-1.6.5.7.6.3" class="indexterm"></a><a id="id-1.6.5.7.6.4" class="indexterm"></a><p>
    The default virtual memory behavior on Linux is not
    optimal for <span class="productname">PostgreSQL</span>. Because of the
    way that the kernel implements memory overcommit, the kernel might
    terminate the <span class="productname">PostgreSQL</span> postmaster (the
    supervisor server process) if the memory demands of either
    <span class="productname">PostgreSQL</span> or another process cause the
    system to run out of virtual memory.
   </p><p>
    If this happens, you will see a kernel message that looks like
    this (consult your system documentation and configuration on where
    to look for such a message):
</p><pre class="programlisting">
Out of Memory: Killed process 12345 (postgres).
</pre><p>
    This indicates that the <code class="filename">postgres</code> process
    has been terminated due to memory pressure.
    Although existing database connections will continue to function
    normally, no new connections will be accepted.  To recover,
    <span class="productname">PostgreSQL</span> will need to be restarted.
   </p><p>
    One way to avoid this problem is to run
    <span class="productname">PostgreSQL</span> on a machine where you can
    be sure that other processes will not run the machine out of
    memory.  If memory is tight, increasing the swap space of the
    operating system can help avoid the problem, because the
    out-of-memory (OOM) killer is invoked only when physical memory and
    swap space are exhausted.
   </p><p>
    If <span class="productname">PostgreSQL</span> itself is the cause of the
    system running out of memory, you can avoid the problem by changing
    your configuration.  In some cases, it may help to lower memory-related
    configuration parameters, particularly
    <a class="link" href="runtime-config-resource.html#GUC-SHARED-BUFFERS"><code class="varname">shared_buffers</code></a>,
    <a class="link" href="runtime-config-resource.html#GUC-WORK-MEM"><code class="varname">work_mem</code></a>, and
    <a class="link" href="runtime-config-resource.html#GUC-HASH-MEM-MULTIPLIER"><code class="varname">hash_mem_multiplier</code></a>.
    In other cases, the problem may be caused by allowing too many
    connections to the database server itself.  In many cases, it may
    be better to reduce
    <a class="link" href="runtime-config-connection.html#GUC-MAX-CONNECTIONS"><code class="varname">max_connections</code></a>
    and instead make use of external connection-pooling software.
   </p><p>
    It is possible to modify the
    kernel's behavior so that it will not <span class="quote">“<span class="quote">overcommit</span>”</span> memory.
    Although this setting will not prevent the <a class="ulink" href="https://lwn.net/Articles/104179/" target="_top">OOM killer</a> from being invoked
    altogether, it will lower the chances significantly and will therefore
    lead to more robust system behavior.  This is done by selecting strict
    overcommit mode via <code class="command">sysctl</code>:
</p><pre class="programlisting">
sysctl -w vm.overcommit_memory=2
</pre><p>
    or placing an equivalent entry in <code class="filename">/etc/sysctl.conf</code>.
    You might also wish to modify the related setting
    <code class="varname">vm.overcommit_ratio</code>.  For details see the kernel documentation
    file <a class="ulink" href="https://www.kernel.org/doc/Documentation/vm/overcommit-accounting" target="_top">https://www.kernel.org/doc/Documentation/vm/overcommit-accounting</a>.
   </p><p>
    Another approach, which can be used with or without altering
    <code class="varname">vm.overcommit_memory</code>, is to set the process-specific
    <em class="firstterm">OOM score adjustment</em> value for the postmaster process to
    <code class="literal">-1000</code>, thereby guaranteeing it will not be targeted by the OOM
    killer.  The simplest way to do this is to execute
</p><pre class="programlisting">
echo -1000 &gt; /proc/self/oom_score_adj
</pre><p>
    in the <span class="productname">PostgreSQL</span> startup script just before
    invoking <code class="filename">postgres</code>.
    Note that this action must be done as root, or it will have no effect;
    so a root-owned startup script is the easiest place to do it.  If you
    do this, you should also set these environment variables in the startup
    script before invoking <code class="filename">postgres</code>:
</p><pre class="programlisting">
export PG_OOM_ADJUST_FILE=/proc/self/oom_score_adj
export PG_OOM_ADJUST_VALUE=0
</pre><p>
    These settings will cause postmaster child processes to run with the
    normal OOM score adjustment of zero, so that the OOM killer can still
    target them at need.  You could use some other value for
    <code class="envar">PG_OOM_ADJUST_VALUE</code> if you want the child processes to run
    with some other OOM score adjustment.  (<code class="envar">PG_OOM_ADJUST_VALUE</code>
    can also be omitted, in which case it defaults to zero.)  If you do not
    set <code class="envar">PG_OOM_ADJUST_FILE</code>, the child processes will run with the
    same OOM score adjustment as the postmaster, which is unwise since the
    whole point is to ensure that the postmaster has a preferential setting.
   </p></div><div class="sect2" id="LINUX-HUGE-PAGES"><div class="titlepage"><div><div><h3 class="title">18.4.5. Linux Huge Pages <a href="#LINUX-HUGE-PAGES" class="id_link">#</a></h3></div></div></div><p>
    Using huge pages reduces overhead when using large contiguous chunks of
    memory, as <span class="productname">PostgreSQL</span> does, particularly when
    using large values of <a class="xref" href="runtime-config-resource.html#GUC-SHARED-BUFFERS">shared_buffers</a>.  To use this
    feature in <span class="productname">PostgreSQL</span> you need a kernel
    with <code class="varname">CONFIG_HUGETLBFS=y</code> and
    <code class="varname">CONFIG_HUGETLB_PAGE=y</code>. You will also have to configure
    the operating system to provide enough huge pages of the desired size.
    The runtime-computed parameter
    <a class="xref" href="runtime-config-preset.html#GUC-SHARED-MEMORY-SIZE-IN-HUGE-PAGES">shared_memory_size_in_huge_pages</a> reports the number
    of huge pages required.  This parameter can be viewed before starting the
    server with a <code class="command">postgres</code> command like:
</p><pre class="programlisting">
$ <strong class="userinput"><code>postgres -D $PGDATA -C shared_memory_size_in_huge_pages</code></strong>
3170
$ <strong class="userinput"><code>grep ^Hugepagesize /proc/meminfo</code></strong>
Hugepagesize:       2048 kB
$ <strong class="userinput"><code>ls /sys/kernel/mm/hugepages</code></strong>
hugepages-1048576kB  hugepages-2048kB
</pre><p>

     In this example the default is 2MB, but you can also explicitly request
     either 2MB or 1GB with <a class="xref" href="runtime-config-resource.html#GUC-HUGE-PAGE-SIZE">huge_page_size</a> to adapt
     the number of pages calculated by
     <code class="varname">shared_memory_size_in_huge_pages</code>.

     While we need at least <code class="literal">3170</code> huge pages in this example,
     a larger setting would be appropriate if other programs on the machine
     also need huge pages.
     We can set this with:
</p><pre class="programlisting">
# <strong class="userinput"><code>sysctl -w vm.nr_hugepages=3170</code></strong>
</pre><p>
     Don't forget to add this setting to <code class="filename">/etc/sysctl.conf</code>
     so that it is reapplied after reboots.  For non-default huge page sizes,
     we can instead use:
</p><pre class="programlisting">
# <strong class="userinput"><code>echo 3170 &gt; /sys/kernel/mm/hugepages/hugepages-2048kB/nr_hugepages</code></strong>
</pre><p>
    It is also possible to provide these settings at boot time using
    kernel parameters such as <code class="literal">hugepagesz=2M hugepages=3170</code>.
   </p><p>
    Sometimes the kernel is not able to allocate the desired number of huge
    pages immediately due to fragmentation, so it might be necessary
    to repeat the command or to reboot.  (Immediately after a reboot, most of
    the machine's memory should be available to convert into huge pages.)
    To verify the huge page allocation situation for a given size, use:
</p><pre class="programlisting">
$ <strong class="userinput"><code>cat /sys/kernel/mm/hugepages/hugepages-2048kB/nr_hugepages</code></strong>
</pre><p>
   </p><p>
    It may also be necessary to give the database server's operating system
    user permission to use huge pages by setting
    <code class="varname">vm.hugetlb_shm_group</code> via <span class="application">sysctl</span>, and/or
    give permission to lock memory with <code class="command">ulimit -l</code>.
   </p><p>
    The default behavior for huge pages in
    <span class="productname">PostgreSQL</span> is to use them when possible, with
    the system's default huge page size, and
    to fall back to normal pages on failure. To enforce the use of huge
    pages, you can set <a class="xref" href="runtime-config-resource.html#GUC-HUGE-PAGES">huge_pages</a>
    to <code class="literal">on</code> in <code class="filename">postgresql.conf</code>.
    Note that with this setting <span class="productname">PostgreSQL</span> will fail to
    start if not enough huge pages are available.
   </p><p>
    For a detailed description of the <span class="productname">Linux</span> huge
    pages feature have a look
    at <a class="ulink" href="https://www.kernel.org/doc/Documentation/vm/hugetlbpage.txt" target="_top">https://www.kernel.org/doc/Documentation/vm/hugetlbpage.txt</a>.
   </p></div></div><div class="navfooter"><hr /><table width="100%" summary="Navigation footer"><tr><td width="40%" align="left"><a accesskey="p" href="server-start.html" title="18.3. Starting the Database Server">Prev</a> </td><td width="20%" align="center"><a accesskey="u" href="runtime.html" title="Chapter 18. Server Setup and Operation">Up</a></td><td width="40%" align="right"> <a accesskey="n" href="server-shutdown.html" title="18.5. Shutting Down the Server">Next</a></td></tr><tr><td width="40%" align="left" valign="top">18.3. Starting the Database Server </td><td width="20%" align="center"><a accesskey="h" href="index.html" title="PostgreSQL 18.4 Documentation">Home</a></td><td width="40%" align="right" valign="top"> 18.5. Shutting Down the Server</td></tr></table></div></body></html>