Baoding-FenShuaJiang
/
Green_GoodERP18_FSJ_Standard
關註 1
收藏 0
Wiki 捐赠
gooderp18绿色标准版
You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
目錄樹: 0f75e31513
分支列表 標籤列表
${ item.name }
Create branch ${ searchTerm }
from '0f75e31513'
${ noResults }
Green_GoodERP18_FSJ_Standard/PostgreSQL/doc/postgresql/html/kernel-resources.html

639 line
45KB
原始文件 Blame 文件歷史 

  1. <?xml version="1.0" encoding="UTF-8" standalone="no"?>

  2. <!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 V1.79.1" /><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><div xmlns="http://www.w3.org/TR/xhtml1/transitional" 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 12.4 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></hr></div><div class="sect1" id="KERNEL-RESOURCES"><div class="titlepage"><div><div><h2 class="title" style="clear: both">18.4. Managing Kernel Resources</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#id-1.6.5.6.5">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>

  3.    <span class="productname">PostgreSQL</span> can sometimes exhaust various operating system

  4.    resource limits, especially when multiple copies of the server are running

  5.    on the same system, or in very large installations.  This section explains

  6.    the kernel resources used by <span class="productname">PostgreSQL</span> and the steps you

  7.    can take to resolve problems related to kernel resource consumption.

  8.   </p><div class="sect2" id="SYSVIPC"><div class="titlepage"><div><div><h3 class="title">18.4.1. Shared Memory and Semaphores</h3></div></div></div><a id="id-1.6.5.6.3.2" class="indexterm"></a><a id="id-1.6.5.6.3.3" class="indexterm"></a><p>

  9.     <span class="productname">PostgreSQL</span> requires the operating system to provide

  10.     inter-process communication (<acronym class="acronym">IPC</acronym>) features, specifically

  11.     shared memory and semaphores.  Unix-derived systems typically provide

  12.     <span class="quote">“<span class="quote"><span class="systemitem">System V</span></span>”</span> <acronym class="acronym">IPC</acronym>,

  13.     <span class="quote">“<span class="quote"><span class="systemitem">POSIX</span></span>”</span> <acronym class="acronym">IPC</acronym>, or both.

  14.     <span class="systemitem">Windows</span> has its own implementation of

  15.     these features and is not discussed here.

  16.    </p><p>

  17.     The complete lack of these facilities is usually manifested by an

  18.     <span class="quote">“<span class="quote"><span class="errorname">Illegal system call</span></span>”</span> error upon server

  19.     start.  In that case there is no alternative but to reconfigure your

  20.     kernel.  <span class="productname">PostgreSQL</span> won't work without them.

  21.     This situation is rare, however, among modern operating systems.

  22.    </p><p>

  23.     By default, <span class="productname">PostgreSQL</span> allocates

  24.     a very small amount of System V shared memory, as well as a much larger

  25.     amount of anonymous <code class="function">mmap</code> shared memory.

  26.     Alternatively, a single large System V shared memory region can be used

  27.     (see <a class="xref" href="runtime-config-resource.html#GUC-SHARED-MEMORY-TYPE">shared_memory_type</a>).

  28. 

  29.     In addition a significant number of semaphores, which can be either

  30.     System V or POSIX style, are created at server startup.  Currently,

  31.     POSIX semaphores are used on Linux and FreeBSD systems while other

  32.     platforms use System V semaphores.

  33.    </p><div class="note"><h3 class="title">Note</h3><p>

  34.      Prior to <span class="productname">PostgreSQL</span> 9.3, only System V shared memory

  35.      was used, so the amount of System V shared memory required to start the

  36.      server was much larger.  If you are running an older version of the

  37.      server, please consult the documentation for your server version.

  38.     </p></div><p>

  39.     System V <acronym class="acronym">IPC</acronym> features are typically constrained by

  40.     system-wide allocation limits.

  41.     When <span class="productname">PostgreSQL</span> exceeds one of these limits,

  42.     the server will refuse to start and

  43.     should leave an instructive error message describing the problem

  44.     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

  45.     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

  46.     them, however, vary. Suggestions for some platforms are given below.

  47.    </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 /><col /><col /></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,

  48.         or <code class="literal">ceil(SHMMAX/PAGE_SIZE)</code> if pages,

  49.         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((max_connections + autovacuum_max_workers + max_wal_senders + max_worker_processes + 5) / 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((max_connections + autovacuum_max_workers + max_wal_senders + max_worker_processes + 5) / 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>

  50.     <span class="productname">PostgreSQL</span> requires a few bytes of System V shared memory

  51.     (typically 48 bytes, on 64-bit platforms) for each copy of the server.

  52.     On most modern operating systems, this amount can easily be allocated.

  53.     However, if you are running many copies of the server or you explicitly

  54.     configure the server to use large amounts of System V shared memory (see

  55.     <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

  56.     increase <code class="varname">SHMALL</code>, which is the total amount of System V shared

  57.     memory system-wide.  Note that <code class="varname">SHMALL</code> is measured in pages

  58.     rather than bytes on many systems.

  59.    </p><p>

  60.     Less likely to cause problems is the minimum size for shared

  61.     memory segments (<code class="varname">SHMMIN</code>), which should be at most

  62.     approximately 32 bytes for <span class="productname">PostgreSQL</span> (it is

  63.     usually just 1). The maximum number of segments system-wide

  64.     (<code class="varname">SHMMNI</code>) or per-process (<code class="varname">SHMSEG</code>) are unlikely

  65.     to cause a problem unless your system has them set to zero.

  66.    </p><p>

  67.     When using System V semaphores,

  68.     <span class="productname">PostgreSQL</span> uses one semaphore per allowed connection

  69.     (<a class="xref" href="runtime-config-connection.html#GUC-MAX-CONNECTIONS">max_connections</a>), allowed autovacuum worker process

  70.     (<a class="xref" href="runtime-config-autovacuum.html#GUC-AUTOVACUUM-MAX-WORKERS">autovacuum_max_workers</a>) and allowed background

  71.     process (<a class="xref" href="runtime-config-resource.html#GUC-MAX-WORKER-PROCESSES">max_worker_processes</a>), in sets of 16.

  72.     Each such set will

  73.     also contain a 17th semaphore which contains a <span class="quote">“<span class="quote">magic

  74.     number</span>”</span>, to detect collision with semaphore sets used by

  75.     other applications. The maximum number of semaphores in the system

  76.     is set by <code class="varname">SEMMNS</code>, which consequently must be at least

  77.     as high as <code class="varname">max_connections</code> plus

  78.     <code class="varname">autovacuum_max_workers</code> plus <code class="varname">max_wal_senders</code>,

  79.     plus <code class="varname">max_worker_processes</code>, plus one extra for each 16

  80.     allowed connections plus workers (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>

  81.     determines the limit on the number of semaphore sets that can

  82.     exist on the system at one time.  Hence this parameter must be at

  83.     least <code class="literal">ceil((max_connections + autovacuum_max_workers + max_wal_senders + max_worker_processes + 5) / 16)</code>.

  84.     Lowering the number

  85.     of allowed connections is a temporary workaround for failures,

  86.     which are usually confusingly worded <span class="quote">“<span class="quote">No space

  87.     left on device</span>”</span>, from the function <code class="function">semget</code>.

  88.    </p><p>

  89.     In some cases it might also be necessary to increase

  90.     <code class="varname">SEMMAP</code> to be at least on the order of

  91.     <code class="varname">SEMMNS</code>.  If the system has this parameter

  92.     (many do not), it defines the size of the semaphore

  93.     resource map, in which each contiguous block of available semaphores

  94.     needs an entry. When a semaphore set is freed it is either added to

  95.     an existing entry that is adjacent to the freed block or it is

  96.     registered under a new map entry. If the map is full, the freed

  97.     semaphores get lost (until reboot). Fragmentation of the semaphore

  98.     space could over time lead to fewer available semaphores than there

  99.     should be.

  100.    </p><p>

  101.     Various other settings related to <span class="quote">“<span class="quote">semaphore undo</span>”</span>, such as

  102.     <code class="varname">SEMMNU</code> and <code class="varname">SEMUME</code>, do not affect

  103.     <span class="productname">PostgreSQL</span>.

  104.    </p><p>

  105.     When using POSIX semaphores, the number of semaphores needed is the

  106.     same as for System V, that is one semaphore per allowed connection

  107.     (<a class="xref" href="runtime-config-connection.html#GUC-MAX-CONNECTIONS">max_connections</a>), allowed autovacuum worker process

  108.     (<a class="xref" href="runtime-config-autovacuum.html#GUC-AUTOVACUUM-MAX-WORKERS">autovacuum_max_workers</a>) and allowed background

  109.     process (<a class="xref" href="runtime-config-resource.html#GUC-MAX-WORKER-PROCESSES">max_worker_processes</a>).

  110.     On the platforms where this option is preferred, there is no specific

  111.     kernel limit on the number of POSIX semaphores.

  112.    </p><div class="variablelist"><dl class="variablelist"><dt><span class="term"><span class="systemitem">AIX</span>

  113.       <a id="id-1.6.5.6.3.16.1.1.2" class="indexterm"></a>

  114.       </span></dt><dd><p>

  115.         At least as of version 5.1, it should not be necessary to do

  116.         any special configuration for such parameters as

  117.         <code class="varname">SHMMAX</code>, as it appears this is configured to

  118.         allow all memory to be used as shared memory.  That is the

  119.         sort of configuration commonly used for other databases such

  120.         as <span class="application">DB/2</span>.</p><p> It might, however, be necessary to modify the global

  121.        <code class="command">ulimit</code> information in

  122.        <code class="filename">/etc/security/limits</code>, as the default hard

  123.        limits for file sizes (<code class="varname">fsize</code>) and numbers of

  124.        files (<code class="varname">nofiles</code>) might be too low.

  125.        </p></dd><dt><span class="term"><span class="systemitem">FreeBSD</span>

  126.       <a id="id-1.6.5.6.3.16.2.1.2" class="indexterm"></a>

  127.       </span></dt><dd><p>

  128.         The default IPC settings can be changed using

  129.         the <code class="command">sysctl</code> or

  130.         <code class="command">loader</code> interfaces.  The following

  131.         parameters can be set using <code class="command">sysctl</code>:

  132. </p><pre class="screen">

  133. <code class="prompt">#</code> <strong class="userinput"><code>sysctl kern.ipc.shmall=32768</code></strong>

  134. <code class="prompt">#</code> <strong class="userinput"><code>sysctl kern.ipc.shmmax=134217728</code></strong>

  135. </pre><p>

  136.         To make these settings persist over reboots, modify

  137.         <code class="filename">/etc/sysctl.conf</code>.

  138.        </p><p>

  139.         These semaphore-related settings are read-only as far as

  140.         <code class="command">sysctl</code> is concerned, but can be set in

  141.         <code class="filename">/boot/loader.conf</code>:

  142. </p><pre class="programlisting">

  143. kern.ipc.semmni=256

  144. kern.ipc.semmns=512

  145. </pre><p>

  146.         After modifying that file, a reboot is required for the new

  147.         settings to take effect.

  148.        </p><p>

  149.         You might also want to configure your kernel to lock System V shared

  150.         memory into RAM and prevent it from being paged out to swap.

  151.         This can be accomplished using the <code class="command">sysctl</code>

  152.         setting <code class="literal">kern.ipc.shm_use_phys</code>.

  153.        </p><p>

  154.         If running in FreeBSD jails by enabling <span class="application">sysctl</span>'s

  155.         <code class="literal">security.jail.sysvipc_allowed</code>, <span class="application">postmaster</span>s

  156.         running in different jails should be run by different operating system

  157.         users.  This improves security because it prevents non-root users

  158.         from interfering with shared memory or semaphores in different jails,

  159.         and it allows the PostgreSQL IPC cleanup code to function properly.

  160.         (In FreeBSD 6.0 and later the IPC cleanup code does not properly detect

  161.         processes in other jails, preventing the running of postmasters on the

  162.         same port in different jails.)

  163.        </p><p>

  164.         <span class="systemitem">FreeBSD</span> versions before 4.0 work like

  165.         old <span class="systemitem">OpenBSD</span> (see below).

  166.        </p></dd><dt><span class="term"><span class="systemitem">NetBSD</span>

  167.       <a id="id-1.6.5.6.3.16.3.1.2" class="indexterm"></a>

  168.       </span></dt><dd><p>

  169.         In <span class="systemitem">NetBSD</span> 5.0 and later,

  170.         IPC parameters can be adjusted using <code class="command">sysctl</code>,

  171.         for example:

  172. </p><pre class="screen">

  173. <code class="prompt">#</code> <strong class="userinput"><code>sysctl -w kern.ipc.semmni=100</code></strong>

  174. </pre><p>

  175.         To make these settings persist over reboots, modify

  176.         <code class="filename">/etc/sysctl.conf</code>.

  177.        </p><p>

  178.         You will usually want to increase <code class="literal">kern.ipc.semmni</code>

  179.         and <code class="literal">kern.ipc.semmns</code>,

  180.         as <span class="systemitem">NetBSD</span>'s default settings

  181.         for these are uncomfortably small.

  182.        </p><p>

  183.         You might also want to configure your kernel to lock System V shared

  184.         memory into RAM and prevent it from being paged out to swap.

  185.         This can be accomplished using the <code class="command">sysctl</code>

  186.         setting <code class="literal">kern.ipc.shm_use_phys</code>.

  187.        </p><p>

  188.         <span class="systemitem">NetBSD</span> versions before 5.0

  189.         work like old <span class="systemitem">OpenBSD</span>

  190.         (see below), except that kernel parameters should be set with the

  191.         keyword <code class="literal">options</code> not <code class="literal">option</code>.

  192.        </p></dd><dt><span class="term"><span class="systemitem">OpenBSD</span>

  193.       <a id="id-1.6.5.6.3.16.4.1.2" class="indexterm"></a>

  194.       </span></dt><dd><p>

  195.         In <span class="systemitem">OpenBSD</span> 3.3 and later,

  196.         IPC parameters can be adjusted using <code class="command">sysctl</code>,

  197.         for example:

  198. </p><pre class="screen">

  199. <code class="prompt">#</code> <strong class="userinput"><code>sysctl kern.seminfo.semmni=100</code></strong>

  200. </pre><p>

  201.         To make these settings persist over reboots, modify

  202.         <code class="filename">/etc/sysctl.conf</code>.

  203.        </p><p>

  204.         You will usually want to

  205.         increase <code class="literal">kern.seminfo.semmni</code>

  206.         and <code class="literal">kern.seminfo.semmns</code>,

  207.         as <span class="systemitem">OpenBSD</span>'s default settings

  208.         for these are uncomfortably small.

  209.        </p><p>

  210.         In older <span class="systemitem">OpenBSD</span> versions,

  211.         you will need to build a custom kernel to change the IPC parameters.

  212.         Make sure that the options <code class="varname">SYSVSHM</code>

  213.         and <code class="varname">SYSVSEM</code> are enabled, too.  (They are by

  214.         default.)  The following shows an example of how to set the various

  215.         parameters in the kernel configuration file:

  216. </p><pre class="programlisting">

  217. option        SYSVSHM

  218. option        SHMMAXPGS=4096

  219. option        SHMSEG=256

  220. 

  221. option        SYSVSEM

  222. option        SEMMNI=256

  223. option        SEMMNS=512

  224. option        SEMMNU=256

  225. </pre><p>

  226.        </p></dd><dt><span class="term"><span class="systemitem">HP-UX</span>

  227.       <a id="id-1.6.5.6.3.16.5.1.2" class="indexterm"></a>

  228.       </span></dt><dd><p>

  229.         The default settings tend to suffice for normal installations.

  230.         On <span class="productname">HP-UX</span> 10, the factory default for

  231.         <code class="varname">SEMMNS</code> is 128, which might be too low for larger

  232.         database sites.

  233.        </p><p>

  234.         <acronym class="acronym">IPC</acronym> parameters can be set in the <span class="application">System

  235.         Administration Manager</span> (<acronym class="acronym">SAM</acronym>) under

  236.         <span class="guimenu">Kernel

  237.         Configuration</span> → <span class="guimenuitem">Configurable Parameters</span>. Choose

  238.         <span class="guibutton">Create A New Kernel</span> when you're done.

  239.        </p></dd><dt><span class="term"><span class="systemitem">Linux</span>

  240.       <a id="id-1.6.5.6.3.16.6.1.2" class="indexterm"></a>

  241.       </span></dt><dd><p>

  242.         The default maximum segment size is 32 MB, and the

  243.         default maximum total size is 2097152

  244.         pages.  A page is almost always 4096 bytes except in unusual

  245.         kernel configurations with <span class="quote">“<span class="quote">huge pages</span>”</span>

  246.         (use <code class="literal">getconf PAGE_SIZE</code> to verify).

  247.        </p><p>

  248.         The shared memory size settings can be changed via the

  249.         <code class="command">sysctl</code> interface.  For example, to allow 16 GB:

  250. </p><pre class="screen">

  251. <code class="prompt">$</code> <strong class="userinput"><code>sysctl -w kernel.shmmax=17179869184</code></strong>

  252. <code class="prompt">$</code> <strong class="userinput"><code>sysctl -w kernel.shmall=4194304</code></strong>

  253. </pre><p>

  254.         In addition these settings can be preserved between reboots in

  255.         the file <code class="filename">/etc/sysctl.conf</code>.  Doing that is

  256.         highly recommended.

  257.        </p><p>

  258.         Ancient distributions might not have the <code class="command">sysctl</code> program,

  259.         but equivalent changes can be made by manipulating the

  260.         <code class="filename">/proc</code> file system:

  261. </p><pre class="screen">

  262. <code class="prompt">$</code> <strong class="userinput"><code>echo 17179869184 &gt;/proc/sys/kernel/shmmax</code></strong>

  263. <code class="prompt">$</code> <strong class="userinput"><code>echo 4194304 &gt;/proc/sys/kernel/shmall</code></strong>

  264. </pre><p>

  265.        </p><p>

  266.         The remaining defaults are quite generously sized, and usually

  267.         do not require changes.

  268.        </p></dd><dt><span class="term"><span class="systemitem">macOS</span>

  269.       <a id="id-1.6.5.6.3.16.7.1.2" class="indexterm"></a>

  270.       </span></dt><dd><p>

  271.         The recommended method for configuring shared memory in macOS

  272.         is to create a file named <code class="filename">/etc/sysctl.conf</code>,

  273.         containing variable assignments such as:

  274. </p><pre class="programlisting">

  275. kern.sysv.shmmax=4194304

  276. kern.sysv.shmmin=1

  277. kern.sysv.shmmni=32

  278. kern.sysv.shmseg=8

  279. kern.sysv.shmall=1024

  280. </pre><p>

  281.         Note that in some macOS versions,

  282.         <span class="emphasis"><em>all five</em></span> shared-memory parameters must be set in

  283.         <code class="filename">/etc/sysctl.conf</code>, else the values will be ignored.

  284.        </p><p>

  285.         Beware that recent releases of macOS ignore attempts to set

  286.         <code class="varname">SHMMAX</code> to a value that isn't an exact multiple of 4096.

  287.        </p><p>

  288.         <code class="varname">SHMALL</code> is measured in 4 kB pages on this platform.

  289.        </p><p>

  290.         In older macOS versions, you will need to reboot to have changes in the

  291.         shared memory parameters take effect.  As of 10.5 it is possible to

  292.         change all but <code class="varname">SHMMNI</code> on the fly, using

  293.         <span class="application">sysctl</span>.  But it's still best to set up your preferred

  294.         values via <code class="filename">/etc/sysctl.conf</code>, so that the values will be

  295.         kept across reboots.

  296.        </p><p>

  297.         The file <code class="filename">/etc/sysctl.conf</code> is only honored in macOS

  298.         10.3.9 and later.  If you are running a previous 10.3.x release,

  299.         you must edit the file <code class="filename">/etc/rc</code>

  300.         and change the values in the following commands:

  301. </p><pre class="programlisting">

  302. sysctl -w kern.sysv.shmmax

  303. sysctl -w kern.sysv.shmmin

  304. sysctl -w kern.sysv.shmmni

  305. sysctl -w kern.sysv.shmseg

  306. sysctl -w kern.sysv.shmall

  307. </pre><p>

  308.         Note that

  309.         <code class="filename">/etc/rc</code> is usually overwritten by macOS system updates,

  310.         so you should expect to have to redo these edits after each update.

  311.        </p><p>

  312.         In macOS 10.2 and earlier, instead edit these commands in the file

  313.         <code class="filename">/System/Library/StartupItems/SystemTuning/SystemTuning</code>.

  314.        </p></dd><dt><span class="term"><span class="systemitem">Solaris</span> 2.6 to 2.9 (Solaris

  315.       6 to Solaris 9)

  316.       <a id="id-1.6.5.6.3.16.8.1.2" class="indexterm"></a>

  317.       </span></dt><dd><p>

  318.         The relevant settings can be changed in

  319.         <code class="filename">/etc/system</code>, for example:

  320. </p><pre class="programlisting">

  321. set shmsys:shminfo_shmmax=0x2000000

  322. set shmsys:shminfo_shmmin=1

  323. set shmsys:shminfo_shmmni=256

  324. set shmsys:shminfo_shmseg=256

  325. 

  326. set semsys:seminfo_semmap=256

  327. set semsys:seminfo_semmni=512

  328. set semsys:seminfo_semmns=512

  329. set semsys:seminfo_semmsl=32

  330. </pre><p>

  331.         You need to reboot for the changes to take effect.  See also

  332.         <a class="ulink" href="http://sunsite.uakom.sk/sunworldonline/swol-09-1997/swol-09-insidesolaris.html" target="_top">http://sunsite.uakom.sk/sunworldonline/swol-09-1997/swol-09-insidesolaris.html</a>

  333.         for information on shared memory under older versions of Solaris.

  334.        </p></dd><dt><span class="term"><span class="systemitem">Solaris</span> 2.10 (Solaris

  335.       10) and later<br /></span><span class="term"><span class="systemitem">OpenSolaris</span></span></dt><dd><p>

  336.         In Solaris 10 and later, and OpenSolaris, the default shared memory and

  337.         semaphore settings are good enough for most

  338.         <span class="productname">PostgreSQL</span> applications.  Solaris now defaults

  339.         to a <code class="varname">SHMMAX</code> of one-quarter of system <acronym class="acronym">RAM</acronym>.

  340.         To further adjust this setting, use a project setting associated

  341.         with the <code class="literal">postgres</code> user.  For example, run the

  342.         following as <code class="literal">root</code>:

  343. </p><pre class="programlisting">

  344. projadd -c "PostgreSQL DB User" -K "project.max-shm-memory=(privileged,8GB,deny)" -U postgres -G postgres user.postgres

  345. </pre><p>

  346.        </p><p>

  347.         This command adds the <code class="literal">user.postgres</code> project and

  348.         sets the shared memory maximum for the <code class="literal">postgres</code>

  349.         user to 8GB, and takes effect the next time that user logs

  350.         in, or when you restart <span class="productname">PostgreSQL</span> (not reload).

  351.         The above assumes that <span class="productname">PostgreSQL</span> is run by

  352.         the <code class="literal">postgres</code> user in the <code class="literal">postgres</code>

  353.         group.  No server reboot is required.

  354.        </p><p>

  355.         Other recommended kernel setting changes for database servers which will

  356.         have a large number of connections are:

  357. </p><pre class="programlisting">

  358. project.max-shm-ids=(priv,32768,deny)

  359. project.max-sem-ids=(priv,4096,deny)

  360. project.max-msg-ids=(priv,4096,deny)

  361. </pre><p>

  362.        </p><p>

  363.         Additionally, if you are running <span class="productname">PostgreSQL</span>

  364.         inside a zone, you may need to raise the zone resource usage

  365.         limits as well.  See "Chapter2:  Projects and Tasks" in the

  366.         <em class="citetitle">System Administrator's Guide</em> for more

  367.         information on <code class="literal">projects</code> and <code class="command">prctl</code>.

  368.        </p></dd></dl></div></div><div class="sect2" id="SYSTEMD-REMOVEIPC"><div class="titlepage"><div><div><h3 class="title">18.4.2. systemd RemoveIPC</h3></div></div></div><a id="id-1.6.5.6.4.2" class="indexterm"></a><p>

  369.     If <span class="productname">systemd</span> is in use, some care must be taken

  370.     that IPC resources (including shared memory) are not prematurely

  371.     removed by the operating system.  This is especially of concern when

  372.     installing PostgreSQL from source.  Users of distribution packages of

  373.     PostgreSQL are less likely to be affected, as

  374.     the <code class="literal">postgres</code> user is then normally created as a system

  375.     user.

  376.    </p><p>

  377.     The setting <code class="literal">RemoveIPC</code>

  378.     in <code class="filename">logind.conf</code> controls whether IPC objects are

  379.     removed when a user fully logs out.  System users are exempt.  This

  380.     setting defaults to on in stock <span class="productname">systemd</span>, but

  381.     some operating system distributions default it to off.

  382.    </p><p>

  383.     A typical observed effect when this setting is on is that shared memory

  384.     objects used for parallel query execution are removed at apparently random

  385.     times, leading to errors and warnings while attempting to open and remove

  386.     them, like

  387. </p><pre class="screen">

  388. WARNING:  could not remove shared memory segment "/PostgreSQL.1450751626": No such file or directory

  389. </pre><p>

  390.     Different types of IPC objects (shared memory vs. semaphores, System V

  391.     vs. POSIX) are treated slightly differently

  392.     by <span class="productname">systemd</span>, so one might observe that some IPC

  393.     resources are not removed in the same way as others.  But it is not

  394.     advisable to rely on these subtle differences.

  395.    </p><p>

  396.     A <span class="quote">“<span class="quote">user logging out</span>”</span> might happen as part of a maintenance

  397.     job or manually when an administrator logs in as

  398.     the <code class="literal">postgres</code> user or something similar, so it is hard

  399.     to prevent in general.

  400.    </p><p>

  401.     What is a <span class="quote">“<span class="quote">system user</span>”</span> is determined

  402.     at <span class="productname">systemd</span> compile time from

  403.     the <code class="symbol">SYS_UID_MAX</code> setting

  404.     in <code class="filename">/etc/login.defs</code>.

  405.    </p><p>

  406.     Packaging and deployment scripts should be careful to create

  407.     the <code class="literal">postgres</code> user as a system user by

  408.     using <code class="literal">useradd -r</code>, <code class="literal">adduser --system</code>,

  409.     or equivalent.

  410.    </p><p>

  411.     Alternatively, if the user account was created incorrectly or cannot be

  412.     changed, it is recommended to set

  413. </p><pre class="programlisting">

  414. RemoveIPC=no

  415. </pre><p>

  416.     in <code class="filename">/etc/systemd/logind.conf</code> or another appropriate

  417.     configuration file.

  418.    </p><div class="caution"><h3 class="title">Caution</h3><p>

  419.      At least one of these two things has to be ensured, or the PostgreSQL

  420.      server will be very unreliable.

  421.     </p></div></div><div class="sect2" id="id-1.6.5.6.5"><div class="titlepage"><div><div><h3 class="title">18.4.3. Resource Limits</h3></div></div></div><p>

  422.     Unix-like operating systems enforce various kinds of resource limits

  423.     that might interfere with the operation of your

  424.     <span class="productname">PostgreSQL</span> server. Of particular

  425.     importance are limits on the number of processes per user, the

  426.     number of open files per process, and the amount of memory available

  427.     to each process. Each of these have a <span class="quote">“<span class="quote">hard</span>”</span> and a

  428.     <span class="quote">“<span class="quote">soft</span>”</span> limit. The soft limit is what actually counts

  429.     but it can be changed by the user up to the hard limit. The hard

  430.     limit can only be changed by the root user. The system call

  431.     <code class="function">setrlimit</code> is responsible for setting these

  432.     parameters. The shell's built-in command <code class="command">ulimit</code>

  433.     (Bourne shells) or <code class="command">limit</code> (<span class="application">csh</span>) is

  434.     used to control the resource limits from the command line. On

  435.     BSD-derived systems the file <code class="filename">/etc/login.conf</code>

  436.     controls the various resource limits set during login. See the

  437.     operating system documentation for details. The relevant

  438.     parameters are <code class="varname">maxproc</code>,

  439.     <code class="varname">openfiles</code>, and <code class="varname">datasize</code>. For

  440.     example:

  441. </p><pre class="programlisting">

  442. default:\

  443. ...

  444.         :datasize-cur=256M:\

  445.         :maxproc-cur=256:\

  446.         :openfiles-cur=256:\

  447. ...

  448. </pre><p>

  449.     (<code class="literal">-cur</code> is the soft limit.  Append

  450.     <code class="literal">-max</code> to set the hard limit.)

  451.    </p><p>

  452.     Kernels can also have system-wide limits on some resources.

  453.     </p><div class="itemizedlist"><ul class="itemizedlist" style="list-style-type: disc; "><li class="listitem"><p>

  454.       On <span class="productname">Linux</span>

  455.       <code class="filename">/proc/sys/fs/file-max</code> determines the

  456.       maximum number of open files that the kernel will support.  It can

  457.       be changed by writing a different number into the file or by

  458.       adding an assignment in <code class="filename">/etc/sysctl.conf</code>.

  459.       The maximum limit of files per process is fixed at the time the

  460.       kernel is compiled; see

  461.       <code class="filename">/usr/src/linux/Documentation/proc.txt</code> for

  462.       more information.

  463.       </p></li></ul></div><p>

  464.    </p><p>

  465.     The <span class="productname">PostgreSQL</span> server uses one process

  466.     per connection so you should provide for at least as many processes

  467.     as allowed connections, in addition to what you need for the rest

  468.     of your system.  This is usually not a problem but if you run

  469.     several servers on one machine things might get tight.

  470.    </p><p>

  471.     The factory default limit on open files is often set to

  472.     <span class="quote">“<span class="quote">socially friendly</span>”</span> values that allow many users to

  473.     coexist on a machine without using an inappropriate fraction of

  474.     the system resources.  If you run many servers on a machine this

  475.     is perhaps what you want, but on dedicated servers you might want to

  476.     raise this limit.

  477.    </p><p>

  478.     On the other side of the coin, some systems allow individual

  479.     processes to open large numbers of files; if more than a few

  480.     processes do so then the system-wide limit can easily be exceeded.

  481.     If you find this happening, and you do not want to alter the

  482.     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

  483.     limit the consumption of open files.

  484.    </p></div><div class="sect2" id="LINUX-MEMORY-OVERCOMMIT"><div class="titlepage"><div><div><h3 class="title">18.4.4. Linux Memory Overcommit</h3></div></div></div><a id="id-1.6.5.6.6.2" class="indexterm"></a><a id="id-1.6.5.6.6.3" class="indexterm"></a><a id="id-1.6.5.6.6.4" class="indexterm"></a><p>

  485.     In Linux 2.4 and later, the default virtual memory behavior is not

  486.     optimal for <span class="productname">PostgreSQL</span>. Because of the

  487.     way that the kernel implements memory overcommit, the kernel might

  488.     terminate the <span class="productname">PostgreSQL</span> postmaster (the

  489.     master server process) if the memory demands of either

  490.     <span class="productname">PostgreSQL</span> or another process cause the

  491.     system to run out of virtual memory.

  492.    </p><p>

  493.     If this happens, you will see a kernel message that looks like

  494.     this (consult your system documentation and configuration on where

  495.     to look for such a message):

  496. </p><pre class="programlisting">

  497. Out of Memory: Killed process 12345 (postgres).

  498. </pre><p>

  499.     This indicates that the <code class="filename">postgres</code> process

  500.     has been terminated due to memory pressure.

  501.     Although existing database connections will continue to function

  502.     normally, no new connections will be accepted.  To recover,

  503.     <span class="productname">PostgreSQL</span> will need to be restarted.

  504.    </p><p>

  505.     One way to avoid this problem is to run

  506.     <span class="productname">PostgreSQL</span> on a machine where you can

  507.     be sure that other processes will not run the machine out of

  508.     memory.  If memory is tight, increasing the swap space of the

  509.     operating system can help avoid the problem, because the

  510.     out-of-memory (OOM) killer is invoked only when physical memory and

  511.     swap space are exhausted.

  512.    </p><p>

  513.     If <span class="productname">PostgreSQL</span> itself is the cause of the

  514.     system running out of memory, you can avoid the problem by changing

  515.     your configuration.  In some cases, it may help to lower memory-related

  516.     configuration parameters, particularly

  517.     <a class="link" href="runtime-config-resource.html#GUC-SHARED-BUFFERS"><code class="varname">shared_buffers</code></a>

  518.     and <a class="link" href="runtime-config-resource.html#GUC-WORK-MEM"><code class="varname">work_mem</code></a>.  In

  519.     other cases, the problem may be caused by allowing too many connections

  520.     to the database server itself.  In many cases, it may be better to reduce

  521.     <a class="link" href="runtime-config-connection.html#GUC-MAX-CONNECTIONS"><code class="varname">max_connections</code></a>

  522.     and instead make use of external connection-pooling software.

  523.    </p><p>

  524.     On Linux 2.6 and later, it is possible to modify the

  525.     kernel's behavior so that it will not <span class="quote">“<span class="quote">overcommit</span>”</span> memory.

  526.     Although this setting will not prevent the <a class="ulink" href="https://lwn.net/Articles/104179/" target="_top">OOM killer</a> from being invoked

  527.     altogether, it will lower the chances significantly and will therefore

  528.     lead to more robust system behavior.  This is done by selecting strict

  529.     overcommit mode via <code class="command">sysctl</code>:

  530. </p><pre class="programlisting">

  531. sysctl -w vm.overcommit_memory=2

  532. </pre><p>

  533.     or placing an equivalent entry in <code class="filename">/etc/sysctl.conf</code>.

  534.     You might also wish to modify the related setting

  535.     <code class="varname">vm.overcommit_ratio</code>.  For details see the kernel documentation

  536.     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>.

  537.    </p><p>

  538.     Another approach, which can be used with or without altering

  539.     <code class="varname">vm.overcommit_memory</code>, is to set the process-specific

  540.     <em class="firstterm">OOM score adjustment</em> value for the postmaster process to

  541.     <code class="literal">-1000</code>, thereby guaranteeing it will not be targeted by the OOM

  542.     killer.  The simplest way to do this is to execute

  543. </p><pre class="programlisting">

  544. echo -1000 &gt; /proc/self/oom_score_adj

  545. </pre><p>

  546.     in the postmaster's startup script just before invoking the postmaster.

  547.     Note that this action must be done as root, or it will have no effect;

  548.     so a root-owned startup script is the easiest place to do it.  If you

  549.     do this, you should also set these environment variables in the startup

  550.     script before invoking the postmaster:

  551. </p><pre class="programlisting">

  552. export PG_OOM_ADJUST_FILE=/proc/self/oom_score_adj

  553. export PG_OOM_ADJUST_VALUE=0

  554. </pre><p>

  555.     These settings will cause postmaster child processes to run with the

  556.     normal OOM score adjustment of zero, so that the OOM killer can still

  557.     target them at need.  You could use some other value for

  558.     <code class="envar">PG_OOM_ADJUST_VALUE</code> if you want the child processes to run

  559.     with some other OOM score adjustment.  (<code class="envar">PG_OOM_ADJUST_VALUE</code>

  560.     can also be omitted, in which case it defaults to zero.)  If you do not

  561.     set <code class="envar">PG_OOM_ADJUST_FILE</code>, the child processes will run with the

  562.     same OOM score adjustment as the postmaster, which is unwise since the

  563.     whole point is to ensure that the postmaster has a preferential setting.

  564.    </p><p>

  565.     Older Linux kernels do not offer <code class="filename">/proc/self/oom_score_adj</code>,

  566.     but may have a previous version of the same functionality called

  567.     <code class="filename">/proc/self/oom_adj</code>.  This works the same except the disable

  568.     value is <code class="literal">-17</code> not <code class="literal">-1000</code>.

  569.    </p><div class="note"><h3 class="title">Note</h3><p>

  570.     Some vendors' Linux 2.4 kernels are reported to have early versions

  571.     of the 2.6 overcommit <code class="command">sysctl</code> parameter.  However, setting

  572.     <code class="literal">vm.overcommit_memory</code> to 2

  573.     on a 2.4 kernel that does not have the relevant code will make

  574.     things worse, not better.  It is recommended that you inspect

  575.     the actual kernel source code (see the function

  576.     <code class="function">vm_enough_memory</code> in the file <code class="filename">mm/mmap.c</code>)

  577.     to verify what is supported in your kernel before you try this in a 2.4

  578.     installation.  The presence of the <code class="filename">overcommit-accounting</code>

  579.     documentation file should <span class="emphasis"><em>not</em></span> be taken as evidence that the

  580.     feature is there.  If in any doubt, consult a kernel expert or your

  581.     kernel vendor.

  582.    </p></div></div><div class="sect2" id="LINUX-HUGE-PAGES"><div class="titlepage"><div><div><h3 class="title">18.4.5. Linux Huge Pages</h3></div></div></div><p>

  583.     Using huge pages reduces overhead when using large contiguous chunks of

  584.     memory, as <span class="productname">PostgreSQL</span> does, particularly when

  585.     using large values of <a class="xref" href="runtime-config-resource.html#GUC-SHARED-BUFFERS">shared_buffers</a>.  To use this

  586.     feature in <span class="productname">PostgreSQL</span> you need a kernel

  587.     with <code class="varname">CONFIG_HUGETLBFS=y</code> and

  588.     <code class="varname">CONFIG_HUGETLB_PAGE=y</code>. You will also have to adjust

  589.     the kernel setting <code class="varname">vm.nr_hugepages</code>. To estimate the

  590.     number of huge pages needed, start <span class="productname">PostgreSQL</span>

  591.     without huge pages enabled and check the

  592.     postmaster's anonymous shared memory segment size, as well as the system's

  593.     huge page size, using the <code class="filename">/proc</code> file system.  This might

  594.     look like:

  595. </p><pre class="programlisting">

  596. $ <strong class="userinput"><code>head -1 $PGDATA/postmaster.pid</code></strong>

  597. 4170

  598. $ <strong class="userinput"><code>pmap 4170 | awk '/rw-s/ &amp;&amp; /zero/ {print $2}'</code></strong>

  599. 6490428K

  600. $ <strong class="userinput"><code>grep ^Hugepagesize /proc/meminfo</code></strong>

  601. Hugepagesize:       2048 kB

  602. </pre><p>

  603.      <code class="literal">6490428</code> / <code class="literal">2048</code> gives approximately

  604.      <code class="literal">3169.154</code>, so in this example we need at

  605.      least <code class="literal">3170</code> huge pages, which we can set with:

  606. </p><pre class="programlisting">

  607. $ <strong class="userinput"><code>sysctl -w vm.nr_hugepages=3170</code></strong>

  608. </pre><p>

  609.     A larger setting would be appropriate if other programs on the machine

  610.     also need huge pages.  Don't forget to add this setting

  611.     to <code class="filename">/etc/sysctl.conf</code> so that it will be reapplied

  612.     after reboots.

  613.    </p><p>

  614.     Sometimes the kernel is not able to allocate the desired number of huge

  615.     pages immediately, so it might be necessary to repeat the command or to

  616.     reboot.  (Immediately after a reboot, most of the machine's memory

  617.     should be available to convert into huge pages.)  To verify the huge

  618.     page allocation situation, use:

  619. </p><pre class="programlisting">

  620. $ <strong class="userinput"><code>grep Huge /proc/meminfo</code></strong>

  621. </pre><p>

  622.    </p><p>

  623.     It may also be necessary to give the database server's operating system

  624.     user permission to use huge pages by setting

  625.     <code class="varname">vm.hugetlb_shm_group</code> via <span class="application">sysctl</span>, and/or

  626.     give permission to lock memory with <code class="command">ulimit -l</code>.

  627.    </p><p>

  628.     The default behavior for huge pages in

  629.     <span class="productname">PostgreSQL</span> is to use them when possible and

  630.     to fall back to normal pages when failing. To enforce the use of huge

  631.     pages, you can set <a class="xref" href="runtime-config-resource.html#GUC-HUGE-PAGES">huge_pages</a>

  632.     to <code class="literal">on</code> in <code class="filename">postgresql.conf</code>.

  633.     Note that with this setting <span class="productname">PostgreSQL</span> will fail to

  634.     start if not enough huge pages are available.

  635.    </p><p>

  636.     For a detailed description of the <span class="productname">Linux</span> huge

  637.     pages feature have a look

  638.     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>.

  639.    </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">Prev</a> </td><td width="20%" align="center"><a accesskey="u" href="runtime.html">Up</a></td><td width="40%" align="right"> <a accesskey="n" href="server-shutdown.html">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">Home</a></td><td width="40%" align="right" valign="top"> 18.5. Shutting Down the Server</td></tr></table></div></body></html>
上海开阖软件有限公司 沪ICP备12045867号-1