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>
23 <table class="module"><tr><th><a href="module-dict.html#Description">Description:</a></th><td>Provides an environment variable with a unique
24 identifier for each request</td></tr>
25 <tr><th><a href="module-dict.html#Status">Status:</a></th><td>Extension</td></tr>
26 <tr><th><a href="module-dict.html#ModuleIdentifier">Module Identifier:</a></th><td>unique_id_module</td></tr>
27 <tr><th><a href="module-dict.html#SourceFile">Source File:</a></th><td>mod_unique_id.c</td></tr></table>
31 <p>This module provides a magic token for each request which is
32 guaranteed to be unique across "all" requests under very
33 specific conditions. The unique identifier is even unique
34 across multiple machines in a properly configured cluster of
35 machines. The environment variable <code>UNIQUE_ID</code> is
36 set to the identifier for each request. Unique identifiers are
37 useful for various reasons which are beyond the scope of this
40 <div id="quickview"><h3 class="directives">Directives</h3>
41 <p>This module provides no directives.</p>
44 <li><img alt="" src="../images/down.gif" /> Theory</li>
46 <div class="top"><a href="#page-header"><img alt="top" src="../images/up.gif" /></a></div>
51 <p>First a brief recap of how the Apache server works on Unix
52 machines. This feature currently isn't supported on Windows NT.
53 On Unix machines, Apache creates several children, the children
54 process requests one at a time. Each child can serve multiple
55 requests in its lifetime. For the purpose of this discussion,
56 the children don't share any data with each other. We'll refer
57 to the children as httpd processes.</p>
59 <p>Your website has one or more machines under your
60 administrative control, together we'll call them a cluster of
61 machines. Each machine can possibly run multiple instances of
62 Apache. All of these collectively are considered "the
63 universe", and with certain assumptions we'll show that in this
64 universe we can generate unique identifiers for each request,
65 without extensive communication between machines in the
68 <p>The machines in your cluster should satisfy these
69 requirements. (Even if you have only one machine you should
70 synchronize its clock with NTP.)</p>
73 <li>The machines' times are synchronized via NTP or other
74 network time protocol.</li>
76 <li>The machines' hostnames all differ, such that the module
77 can do a hostname lookup on the hostname and receive a
78 different IP address for each machine in the cluster.</li>
81 <p>As far as operating system assumptions go, we assume that
82 pids (process ids) fit in 32-bits. If the operating system uses
83 more than 32-bits for a pid, the fix is trivial but must be
84 performed in the code.</p>
86 <p>Given those assumptions, at a single point in time we can
87 identify any httpd process on any machine in the cluster from
88 all other httpd processes. The machine's IP address and the pid
89 of the httpd process are sufficient to do this. So in order to
90 generate unique identifiers for requests we need only
91 distinguish between different points in time.</p>
93 <p>To distinguish time we will use a Unix timestamp (seconds
94 since January 1, 1970 UTC), and a 16-bit counter. The timestamp
95 has only one second granularity, so the counter is used to
96 represent up to 65536 values during a single second. The
97 quadruple <em>( ip_addr, pid, time_stamp, counter )</em> is
98 sufficient to enumerate 65536 requests per second per httpd
99 process. There are issues however with pid reuse over time, and
100 the counter is used to alleviate this issue.</p>
102 <p>When an httpd child is created, the counter is initialized
103 with ( current microseconds divided by 10 ) modulo 65536 (this
104 formula was chosen to eliminate some variance problems with the
105 low order bits of the microsecond timers on some systems). When
106 a unique identifier is generated, the time stamp used is the
107 time the request arrived at the web server. The counter is
108 incremented every time an identifier is generated (and allowed
111 <p>The kernel generates a pid for each process as it forks the
112 process, and pids are allowed to roll over (they're 16-bits on
113 many Unixes, but newer systems have expanded to 32-bits). So
114 over time the same pid will be reused. However unless it is
115 reused within the same second, it does not destroy the
116 uniqueness of our quadruple. That is, we assume the system does
117 not spawn 65536 processes in a one second interval (it may even
118 be 32768 processes on some Unixes, but even this isn't likely
121 <p>Suppose that time repeats itself for some reason. That is,
122 suppose that the system's clock is screwed up and it revisits a
123 past time (or it is too far forward, is reset correctly, and
124 then revisits the future time). In this case we can easily show
125 that we can get pid and time stamp reuse. The choice of
126 initializer for the counter is intended to help defeat this.
127 Note that we really want a random number to initialize the
128 counter, but there aren't any readily available numbers on most
129 systems (<em>i.e.</em>, you can't use rand() because you need
130 to seed the generator, and can't seed it with the time because
131 time, at least at one second resolution, has repeated itself).
132 This is not a perfect defense.</p>
134 <p>How good a defense is it? Suppose that one of your machines
135 serves at most 500 requests per second (which is a very
136 reasonable upper bound at this writing, because systems
137 generally do more than just shovel out static files). To do
138 that it will require a number of children which depends on how
139 many concurrent clients you have. But we'll be pessimistic and
140 suppose that a single child is able to serve 500 requests per
141 second. There are 1000 possible starting counter values such
142 that two sequences of 500 requests overlap. So there is a 1.5%
143 chance that if time (at one second resolution) repeats itself
144 this child will repeat a counter value, and uniqueness will be
145 broken. This was a very pessimistic example, and with real
146 world values it's even less likely to occur. If your system is
147 such that it's still likely to occur, then perhaps you should
148 make the counter 32 bits (by editing the code).</p>
150 <p>You may be concerned about the clock being "set back" during
151 summer daylight savings. However this isn't an issue because
152 the times used here are UTC, which "always" go forward. Note
153 that x86 based Unixes may need proper configuration for this to
154 be true -- they should be configured to assume that the
155 motherboard clock is on UTC and compensate appropriately. But
156 even still, if you're running NTP then your UTC time will be
157 correct very shortly after reboot.</p>
159 <p>The <code>UNIQUE_ID</code> environment variable is
160 constructed by encoding the 112-bit (32-bit IP address, 32 bit
161 pid, 32 bit time stamp, 16 bit counter) quadruple using the
162 alphabet <code>[A-Za-z0-9@-]</code> in a manner similar to MIME
163 base64 encoding, producing 19 characters. The MIME base64
164 alphabet is actually <code>[A-Za-z0-9+/]</code> however
165 <code>+</code> and <code>/</code> need to be specially encoded
166 in URLs, which makes them less desirable. All values are
167 encoded in network byte ordering so that the encoding is
168 comparable across architectures of different byte ordering. The
169 actual ordering of the encoding is: time stamp, IP address,
170 pid, counter. This ordering has a purpose, but it should be
171 emphasized that applications should not dissect the encoding.
172 Applications should treat the entire encoded
173 <code>UNIQUE_ID</code> as an opaque token, which can be
174 compared against other <code>UNIQUE_ID</code>s for equality
177 <p>The ordering was chosen such that it's possible to change
178 the encoding in the future without worrying about collision
179 with an existing database of <code>UNIQUE_ID</code>s. The new
180 encodings should also keep the time stamp as the first element,
181 and can otherwise use the same alphabet and bit length. Since
182 the time stamps are essentially an increasing sequence, it's
183 sufficient to have a <em>flag second</em> in which all machines
184 in the cluster stop serving and request, and stop using the old
185 encoding format. Afterwards they can resume requests and begin
186 issuing the new encodings.</p>
188 <p>This we believe is a relatively portable solution to this
189 problem. It can be extended to multithreaded systems like
190 Windows NT, and can grow with future needs. The identifiers
191 generated have essentially an infinite life-time because future
192 identifiers can be made longer as required. Essentially no
193 communication is required between machines in the cluster (only
194 NTP synchronization is required, which is low overhead), and no
195 communication between httpd processes is required (the
196 communication is implicit in the pid value assigned by the
197 kernel). In very specific situations the identifier can be
198 shortened, but more information needs to be assumed (for
199 example the 32-bit IP address is overkill for any site, but
200 there is no portable shorter replacement for it). </p>
204 <p class="apache">Maintained by the <a href="http://httpd.apache.org/docs-project/">Apache HTTP Server Documentation Project</a></p>
205 <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>