1 <?xml version="1.0" encoding="ISO-8859-1"?>
2 <!DOCTYPE html PUBLIC "-//W3C//DTD XHTML 1.0 Strict//EN" "http://www.w3.org/TR/xhtml1/DTD/xhtml1-strict.dtd">
3 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head><!--
4 XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX
5 This file is generated from xml source: DO NOT EDIT
6 XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX
8 <title>mod_unique_id - Apache HTTP Server</title>
9 <link href="../style/css/manual.css" rel="stylesheet" media="all" type="text/css" title="Main stylesheet" />
10 <link href="../style/css/manual-loose-100pc.css" rel="alternate stylesheet" media="all" type="text/css" title="No Sidebar - Default font size" />
11 <link href="../style/css/manual-print.css" rel="stylesheet" media="print" type="text/css" />
12 <link href="../images/favicon.ico" rel="shortcut icon" /></head>
14 <div id="page-header">
15 <p class="menu"><a href="../mod/">Modules</a> | <a href="../mod/directives.html">Directives</a> | <a href="../faq/">FAQ</a> | <a href="../glossary.html">Glossary</a> | <a href="../sitemap.html">Sitemap</a></p>
16 <p class="apache">Apache HTTP Server Version 2.1</p>
17 <img alt="" src="../images/feather.gif" /></div>
18 <div class="up"><a href="./"><img title="<-" alt="<-" src="../images/left.gif" /></a></div>
20 <a href="http://www.apache.org/">Apache</a> > <a href="http://httpd.apache.org/">HTTP Server</a> > <a href="http://httpd.apache.org/docs-project/">Documentation</a> > <a href="../">Version 2.1</a> > <a href="./">Modules</a></div>
21 <div id="page-content">
22 <div id="preamble"><h1>Apache Module mod_unique_id</h1>
24 <p><span>Available Languages: </span><a href="../en/mod/mod_unique_id.html" title="English"> en </a> |
25 <a href="../ja/mod/mod_unique_id.html" hreflang="ja" rel="alternate" title="Japanese"> ja </a> |
26 <a href="../ko/mod/mod_unique_id.html" hreflang="ko" rel="alternate" title="Korean"> ko </a></p>
28 <table class="module"><tr><th><a href="module-dict.html#Description">Description:</a></th><td>Provides an environment variable with a unique
29 identifier for each request</td></tr>
30 <tr><th><a href="module-dict.html#Status">Status:</a></th><td>Extension</td></tr>
31 <tr><th><a href="module-dict.html#ModuleIdentifier">Module Identifier:</a></th><td>unique_id_module</td></tr>
32 <tr><th><a href="module-dict.html#SourceFile">Source File:</a></th><td>mod_unique_id.c</td></tr></table>
36 <p>This module provides a magic token for each request which is
37 guaranteed to be unique across "all" requests under very
38 specific conditions. The unique identifier is even unique
39 across multiple machines in a properly configured cluster of
40 machines. The environment variable <code>UNIQUE_ID</code> is
41 set to the identifier for each request. Unique identifiers are
42 useful for various reasons which are beyond the scope of this
45 <div id="quickview"><h3 class="directives">Directives</h3>
46 <p>This module provides no
50 <li><img alt="" src="../images/down.gif" /> <a href="#theory">Theory</a></li>
52 <div class="top"><a href="#page-header"><img alt="top" src="../images/up.gif" /></a></div>
54 <h2><a name="theory" id="theory">Theory</a></h2>
57 <p>First a brief recap of how the Apache server works on Unix
58 machines. This feature currently isn't supported on Windows NT.
59 On Unix machines, Apache creates several children, the children
60 process requests one at a time. Each child can serve multiple
61 requests in its lifetime. For the purpose of this discussion,
62 the children don't share any data with each other. We'll refer
63 to the children as httpd processes.</p>
65 <p>Your website has one or more machines under your
66 administrative control, together we'll call them a cluster of
67 machines. Each machine can possibly run multiple instances of
68 Apache. All of these collectively are considered "the
69 universe", and with certain assumptions we'll show that in this
70 universe we can generate unique identifiers for each request,
71 without extensive communication between machines in the
74 <p>The machines in your cluster should satisfy these
75 requirements. (Even if you have only one machine you should
76 synchronize its clock with NTP.)</p>
79 <li>The machines' times are synchronized via NTP or other
80 network time protocol.</li>
82 <li>The machines' hostnames all differ, such that the module
83 can do a hostname lookup on the hostname and receive a
84 different IP address for each machine in the cluster.</li>
87 <p>As far as operating system assumptions go, we assume that
88 pids (process ids) fit in 32-bits. If the operating system uses
89 more than 32-bits for a pid, the fix is trivial but must be
90 performed in the code.</p>
92 <p>Given those assumptions, at a single point in time we can
93 identify any httpd process on any machine in the cluster from
94 all other httpd processes. The machine's IP address and the pid
95 of the httpd process are sufficient to do this. So in order to
96 generate unique identifiers for requests we need only
97 distinguish between different points in time.</p>
99 <p>To distinguish time we will use a Unix timestamp (seconds
100 since January 1, 1970 UTC), and a 16-bit counter. The timestamp
101 has only one second granularity, so the counter is used to
102 represent up to 65536 values during a single second. The
103 quadruple <em>( ip_addr, pid, time_stamp, counter )</em> is
104 sufficient to enumerate 65536 requests per second per httpd
105 process. There are issues however with pid reuse over time, and
106 the counter is used to alleviate this issue.</p>
108 <p>When an httpd child is created, the counter is initialized
109 with ( current microseconds divided by 10 ) modulo 65536 (this
110 formula was chosen to eliminate some variance problems with the
111 low order bits of the microsecond timers on some systems). When
112 a unique identifier is generated, the time stamp used is the
113 time the request arrived at the web server. The counter is
114 incremented every time an identifier is generated (and allowed
117 <p>The kernel generates a pid for each process as it forks the
118 process, and pids are allowed to roll over (they're 16-bits on
119 many Unixes, but newer systems have expanded to 32-bits). So
120 over time the same pid will be reused. However unless it is
121 reused within the same second, it does not destroy the
122 uniqueness of our quadruple. That is, we assume the system does
123 not spawn 65536 processes in a one second interval (it may even
124 be 32768 processes on some Unixes, but even this isn't likely
127 <p>Suppose that time repeats itself for some reason. That is,
128 suppose that the system's clock is screwed up and it revisits a
129 past time (or it is too far forward, is reset correctly, and
130 then revisits the future time). In this case we can easily show
131 that we can get pid and time stamp reuse. The choice of
132 initializer for the counter is intended to help defeat this.
133 Note that we really want a random number to initialize the
134 counter, but there aren't any readily available numbers on most
135 systems (<em>i.e.</em>, you can't use rand() because you need
136 to seed the generator, and can't seed it with the time because
137 time, at least at one second resolution, has repeated itself).
138 This is not a perfect defense.</p>
140 <p>How good a defense is it? Suppose that one of your machines
141 serves at most 500 requests per second (which is a very
142 reasonable upper bound at this writing, because systems
143 generally do more than just shovel out static files). To do
144 that it will require a number of children which depends on how
145 many concurrent clients you have. But we'll be pessimistic and
146 suppose that a single child is able to serve 500 requests per
147 second. There are 1000 possible starting counter values such
148 that two sequences of 500 requests overlap. So there is a 1.5%
149 chance that if time (at one second resolution) repeats itself
150 this child will repeat a counter value, and uniqueness will be
151 broken. This was a very pessimistic example, and with real
152 world values it's even less likely to occur. If your system is
153 such that it's still likely to occur, then perhaps you should
154 make the counter 32 bits (by editing the code).</p>
156 <p>You may be concerned about the clock being "set back" during
157 summer daylight savings. However this isn't an issue because
158 the times used here are UTC, which "always" go forward. Note
159 that x86 based Unixes may need proper configuration for this to
160 be true -- they should be configured to assume that the
161 motherboard clock is on UTC and compensate appropriately. But
162 even still, if you're running NTP then your UTC time will be
163 correct very shortly after reboot.</p>
165 <p>The <code>UNIQUE_ID</code> environment variable is
166 constructed by encoding the 112-bit (32-bit IP address, 32 bit
167 pid, 32 bit time stamp, 16 bit counter) quadruple using the
168 alphabet <code>[A-Za-z0-9@-]</code> in a manner similar to MIME
169 base64 encoding, producing 19 characters. The MIME base64
170 alphabet is actually <code>[A-Za-z0-9+/]</code> however
171 <code>+</code> and <code>/</code> need to be specially encoded
172 in URLs, which makes them less desirable. All values are
173 encoded in network byte ordering so that the encoding is
174 comparable across architectures of different byte ordering. The
175 actual ordering of the encoding is: time stamp, IP address,
176 pid, counter. This ordering has a purpose, but it should be
177 emphasized that applications should not dissect the encoding.
178 Applications should treat the entire encoded
179 <code>UNIQUE_ID</code> as an opaque token, which can be
180 compared against other <code>UNIQUE_ID</code>s for equality
183 <p>The ordering was chosen such that it's possible to change
184 the encoding in the future without worrying about collision
185 with an existing database of <code>UNIQUE_ID</code>s. The new
186 encodings should also keep the time stamp as the first element,
187 and can otherwise use the same alphabet and bit length. Since
188 the time stamps are essentially an increasing sequence, it's
189 sufficient to have a <em>flag second</em> in which all machines
190 in the cluster stop serving and request, and stop using the old
191 encoding format. Afterwards they can resume requests and begin
192 issuing the new encodings.</p>
194 <p>This we believe is a relatively portable solution to this
195 problem. It can be extended to multithreaded systems like
196 Windows NT, and can grow with future needs. The identifiers
197 generated have essentially an infinite life-time because future
198 identifiers can be made longer as required. Essentially no
199 communication is required between machines in the cluster (only
200 NTP synchronization is required, which is low overhead), and no
201 communication between httpd processes is required (the
202 communication is implicit in the pid value assigned by the
203 kernel). In very specific situations the identifier can be
204 shortened, but more information needs to be assumed (for
205 example the 32-bit IP address is overkill for any site, but
206 there is no portable shorter replacement for it). </p>
209 <div class="bottomlang">
210 <p><span>Available Languages: </span><a href="../en/mod/mod_unique_id.html" title="English"> en </a> |
211 <a href="../ja/mod/mod_unique_id.html" hreflang="ja" rel="alternate" title="Japanese"> ja </a> |
212 <a href="../ko/mod/mod_unique_id.html" hreflang="ko" rel="alternate" title="Korean"> ko </a></p>
213 </div><div id="footer">
214 <p class="apache">Copyright 1999-2004 The Apache Software Foundation.<br />Licensed under the <a href="http://www.apache.org/licenses/LICENSE-2.0">Apache License, Version 2.0</a>.</p>
215 <p class="menu"><a href="../mod/">Modules</a> | <a href="../mod/directives.html">Directives</a> | <a href="../faq/">FAQ</a> | <a href="../glossary.html">Glossary</a> | <a href="../sitemap.html">Sitemap</a></p></div>