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