<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.0 Transitional//EN">

<html>
<head>
<title>How To Use Google Logging Library (glog)</title>

<meta http-equiv="Content-Type" content="text/html; charset=utf-8">
<link href="http://www.google.com/favicon.ico" type="image/x-icon"
      rel="shortcut icon">
<link href="designstyle.css" type="text/css" rel="stylesheet">
<style type="text/css">
<!--
  ol.bluelist li {
    color: #3366ff;
    font-family: sans-serif;
  }
  ol.bluelist li p {
    color: #000;
    font-family: "Times Roman", times, serif;
  }
  ul.blacklist li {
    color: #000;
    font-family: "Times Roman", times, serif;
  }
//-->
</style>
</head>

<body>

<h1>How To Use Google Logging Library (glog)</h1>
<small>(as of
<script type=text/javascript>
  var lm = new Date(document.lastModified);
  document.write(lm.toDateString());
</script>)
</small>
<br>

<h2> <A NAME=intro>Introduction</A> </h2>

<p><b>Google glog</b> is a library that implements application-level
logging.  This library provides logging APIs based on C++-style
streams and various helper macros.
You can log a message by simply streaming things to LOG(&lt;a
particular <a href="#severity">severity level</a>&gt;), e.g.

<pre>
   #include &lt;glog/logging.h&gt;

   int main(int argc, char* argv[]) {
     // Initialize Google's logging library.
     google::InitGoogleLogging(argv[0]);

     // ...
     LOG(INFO) &lt;&lt; "Found " &lt;&lt; num_cookies &lt;&lt; " cookies";
   }
</pre>

<p>Google glog defines a series of macros that simplify many common logging
tasks.  You can log messages by severity level, control logging
behavior from the command line, log based on conditionals, abort the
program when expected conditions are not met, introduce your own
verbose logging levels, and more.  This document describes the
functionality supported by glog.  Please note that this document
doesn't describe all features in this library, but the most useful
ones.  If you want to find less common features, please check
header files under <code>src/glog</code> directory.

<h2> <A NAME=severity>Severity Level</A> </h2>

<p>
You can specify one of the following severity levels (in
increasing order of severity): <code>INFO</code>, <code>WARNING</code>,
<code>ERROR</code>, and <code>FATAL</code>.
Logging a <code>FATAL</code> message terminates the program (after the
message is logged).
Note that messages of a given severity are logged not only in the
logfile for that severity, but also in all logfiles of lower severity.
E.g., a message of severity <code>FATAL</code> will be logged to the
logfiles of severity <code>FATAL</code>, <code>ERROR</code>,
<code>WARNING</code>, and <code>INFO</code>.

<p>
The <code>DFATAL</code> severity logs a <code>FATAL</code> error in
debug mode (i.e., there is no <code>NDEBUG</code> macro defined), but
avoids halting the program in production by automatically reducing the
severity to <code>ERROR</code>.

<p>Unless otherwise specified, glog writes to the filename
"/tmp/&lt;program name&gt;.&lt;hostname&gt;.&lt;user name&gt;.log.&lt;severity level&gt;.&lt;date&gt;.&lt;time&gt;.&lt;pid&gt;"
(e.g., "/tmp/hello_world.example.com.hamaji.log.INFO.20080709-222411.10474").
By default, glog copies the log messages of severity level
<code>ERROR</code> or <code>FATAL</code> to standard error (stderr)
in addition to log files.

<h2><A NAME=flags>Setting Flags</A></h2>

<p>Several flags influence glog's output behavior.
If the <a href="http://code.google.com/p/google-gflags/">Google
gflags library</a> is installed on your machine, the
<code>configure</code> script (see the INSTALL file in the package for
detail of this script) will automatically detect and use it,
allowing you to pass flags on the command line.  For example, if you
want to turn the flag <code>--logtostderr</code> on, you can start
your application with the following command line:

<pre>
   ./your_application --logtostderr=1
</pre>

If the Google gflags library isn't installed, you set flags via
environment variables, prefixing the flag name with "GLOG_", e.g.

<pre>
   GLOG_logtostderr=1 ./your_application
</pre>

<!-- TODO(hamaji): Fill the version number
<p>By glog version 0.x.x, you can use GLOG_* environment variables
even if you have gflags. If both an environment variable and a flag
are specified, the value specified by a flag wins. E.g., if GLOG_v=0
and --v=1, the verbosity will be 1, not 0.
-->

<p>The following flags are most commonly used:

<dl>
<dt><code>logtostderr</code> (<code>bool</code>, default=<code>false</code>)
<dd>Log messages to stderr instead of logfiles.<br>
Note: you can set binary flags to <code>true</code> by specifying
<code>1</code>, <code>true</code>, or <code>yes</code> (case
insensitive).
Also, you can set binary flags to <code>false</code> by specifying
<code>0</code>, <code>false</code>, or <code>no</code> (again, case
insensitive).
<dt><code>stderrthreshold</code> (<code>int</code>, default=2, which
is <code>ERROR</code>)
<dd>Copy log messages at or above this level to stderr in
addition to logfiles.  The numbers of severity levels
<code>INFO</code>, <code>WARNING</code>, <code>ERROR</code>, and
<code>FATAL</code> are 0, 1, 2, and 3, respectively.
<dt><code>minloglevel</code> (<code>int</code>, default=0, which
is <code>INFO</code>)
<dd>Log messages at or above this level.  Again, the numbers of
severity levels <code>INFO</code>, <code>WARNING</code>,
<code>ERROR</code>, and <code>FATAL</code> are 0, 1, 2, and 3,
respectively.
<dt><code>log_dir</code> (<code>string</code>, default="")
<dd>If specified, logfiles are written into this directory instead
of the default logging directory.
<dt><code>v</code> (<code>int</code>, default=0)
<dd>Show all <code>VLOG(m)</code> messages for <code>m</code> less or
equal the value of this flag.  Overridable by --vmodule.
See <a href="#verbose">the section about verbose logging</a> for more
detail.
<dt><code>vmodule</code> (<code>string</code>, default="")
<dd>Per-module verbose level.  The argument has to contain a
comma-separated list of &lt;module name&gt;=&lt;log level&gt;.
&lt;module name&gt;
is a glob pattern (e.g., <code>gfs*</code> for all modules whose name
starts with "gfs"), matched against the filename base
(that is, name ignoring .cc/.h./-inl.h).
&lt;log level&gt; overrides any value given by --v.
See also <a href="#verbose">the section about verbose logging</a>.
</dl>

<p>There are some other flags defined in logging.cc.  Please grep the
source code for "DEFINE_" to see a complete list of all flags.

<h2><A NAME=conditional>Conditional / Occasional Logging</A></h2>

<p>Sometimes, you may only want to log a message under certain
conditions. You can use the following macros to perform conditional
logging:

<pre>
   LOG_IF(INFO, num_cookies &gt; 10) &lt;&lt; "Got lots of cookies";
</pre>

The "Got lots of cookies" message is logged only when the variable
<code>num_cookies</code> exceeds 10.

If a line of code is executed many times, it may be useful to only log
a message at certain intervals.  This kind of logging is most useful
for informational messages.

<pre>
   LOG_EVERY_N(INFO, 10) &lt;&lt; "Got the " &lt;&lt; COUNTER &lt;&lt; "th cookie";
</pre>

<p>The above line outputs a log messages on the 1st, 11th,
21st, ... times it is executed.  Note that the special
<code>COUNTER</code> value is used to identify which repetition is
happening.

<p>You can combine conditional and occasional logging with the
following macro.

<pre>
   LOG_IF_EVERY_N(INFO, (size &gt; 1024), 10) &lt;&lt; "Got the " &lt;&lt; COUNTER
                                           &lt;&lt; "th big cookie";
</pre>

<p>Instead of outputting a message every nth time, you can also limit
the output to the first n occurrences:

<pre>
   LOG_FIRST_N(INFO, 20) &lt;&lt; "Got the " &lt;&lt; COUNTER &lt;&lt; "th cookie";
</pre>

<p>Outputs log messages for the first 20 times it is executed.  Again,
the <code>COUNTER</code> identifier indicates which repetition is
happening.

<h2><A NAME=debug>Debug Mode Support</A></h2>

<p>Special "debug mode" logging macros only have an effect in debug
mode and are compiled away to nothing for non-debug mode
compiles.  Use these macros to avoid slowing down your production
application due to excessive logging.

<pre>
   DLOG(INFO) &lt;&lt; "Found cookies";

   DLOG_IF(INFO, num_cookies &gt; 10) &lt;&lt; "Got lots of cookies";

   DLOG_EVERY_N(INFO, 10) &lt;&lt; "Got the " &lt;&lt; COUNTER &lt;&lt; "th cookie";
</pre>

<h2><A NAME=check>CHECK Macros</A></h2>

<p>It is a good practice to check expected conditions in your program
frequently to detect errors as early as possible. The
<code>CHECK</code> macro provides the ability to abort the application
when a condition is not met, similar to the <code>assert</code> macro
defined in the standard C library.

<p><code>CHECK</code> aborts the application if a condition is not
true.  Unlike <code>assert</code>, it is *not* controlled by
<code>NDEBUG</code>, so the check will be executed regardless of
compilation mode.  Therefore, <code>fp-&gt;Write(x)</code> in the
following example is always executed:

<pre>
   CHECK(fp-&gt;Write(x) == 4) &lt;&lt; "Write failed!";
</pre>

<p>There are various helper macros for
equality/inequality checks - <code>CHECK_EQ</code>,
<code>CHECK_NE</code>, <code>CHECK_LE</code>, <code>CHECK_LT</code>,
<code>CHECK_GE</code>, and <code>CHECK_GT</code>.
They compare two values, and log a
<code>FATAL</code> message including the two values when the result is
not as expected.  The values must have <code>operator&lt;&lt;(ostream,
...)</code> defined.

<p>You may append to the error message like so:

<pre>
   CHECK_NE(1, 2) &lt;&lt; ": The world must be ending!";
</pre>

<p>We are very careful to ensure that each argument is evaluated exactly
once, and that anything which is legal to pass as a function argument is
legal here.  In particular, the arguments may be temporary expressions
which will end up being destroyed at the end of the apparent statement,
for example:

<pre>
   CHECK_EQ(string("abc")[1], 'b');
</pre>

<p>The compiler reports an error if one of the arguments is a
pointer and the other is NULL. To work around this, simply static_cast
NULL to the type of the desired pointer.

<pre>
   CHECK_EQ(some_ptr, static_cast&lt;SomeType*&gt;(NULL));
</pre>

<p>Better yet, use the CHECK_NOTNULL macro:

<pre>
   CHECK_NOTNULL(some_ptr);
   some_ptr-&gt;DoSomething();
</pre>

<p>Since this macro returns the given pointer, this is very useful in
constructor initializer lists.

<pre>
   struct S {
     S(Something* ptr) : ptr_(CHECK_NOTNULL(ptr)) {}
     Something* ptr_;
   };
</pre>

<p>Note that you cannot use this macro as a C++ stream due to this
feature.  Please use <code>CHECK_EQ</code> described above to log a
custom message before aborting the application.

<p>If you are comparing C strings (char *), a handy set of macros
performs case sensitive as well as case insensitive comparisons -
<code>CHECK_STREQ</code>, <code>CHECK_STRNE</code>,
<code>CHECK_STRCASEEQ</code>, and <code>CHECK_STRCASENE</code>.  The
CASE versions are case-insensitive.  You can safely pass <code>NULL</code>
pointers for this macro.  They treat <code>NULL</code> and any
non-<code>NULL</code> string as not equal.  Two <code>NULL</code>s are
equal.

<p>Note that both arguments may be temporary strings which are
destructed at the end of the current "full expression"
(e.g., <code>CHECK_STREQ(Foo().c_str(), Bar().c_str())</code> where
<code>Foo</code> and <code>Bar</code> return C++'s
<code>std::string</code>).

<p>The <code>CHECK_DOUBLE_EQ</code> macro checks the equality of two
floating point values, accepting a small error margin.
<code>CHECK_NEAR</code> accepts a third floating point argument, which
specifies the acceptable error margin.

<h2><A NAME=verbose>Verbose Logging</A></h2>

<p>When you are chasing difficult bugs, thorough log messages are very
useful.  However, you may want to ignore too verbose messages in usual
development.  For such verbose logging, glog provides the
<code>VLOG</code> macro, which allows you to define your own numeric
logging levels.  The <code>--v</code> command line option controls
which verbose messages are logged:

<pre>
   VLOG(1) &lt;&lt; "I'm printed when you run the program with --v=1 or higher";
   VLOG(2) &lt;&lt; "I'm printed when you run the program with --v=2 or higher";
</pre>

<p>With <code>VLOG</code>, the lower the verbose level, the more
likely messages are to be logged.  For example, if
<code>--v==1</code>, <code>VLOG(1)</code> will log, but
<code>VLOG(2)</code> will not log.  This is opposite of the severity
level, where <code>INFO</code> is 0, and <code>ERROR</code> is 2.
<code>--minloglevel</code> of 1 will log <code>WARNING</code> and
above.  Though you can specify any integers for both <code>VLOG</code>
macro and <code>--v</code> flag, the common values for them are small
positive integers.  For example, if you write <code>VLOG(0)</code>,
you should specify <code>--v=-1</code> or lower to silence it.  This
is less useful since we may not want verbose logs by default in most
cases.  The <code>VLOG</code> macros always log at the
<code>INFO</code> log level (when they log at all).

<p>Verbose logging can be controlled from the command line on a
per-module basis:

<pre>
   --vmodule=mapreduce=2,file=1,gfs*=3 --v=0
</pre>

<p>will:

<ul>
  <li>a. Print VLOG(2) and lower messages from mapreduce.{h,cc}
  <li>b. Print VLOG(1) and lower messages from file.{h,cc}
  <li>c. Print VLOG(3) and lower messages from files prefixed with "gfs"
  <li>d. Print VLOG(0) and lower messages from elsewhere
</ul>

<p>The wildcarding functionality shown by (c) supports both '*'
(matches 0 or more characters) and '?' (matches any single character)
wildcards.  Please also check the section about <a
href="#flags">command line flags</a>.

<p>There's also <code>VLOG_IS_ON(n)</code> "verbose level" condition
macro.  This macro returns true when the <code>--v</code> is equal or
greater than <code>n</code>.  To be used as

<pre>
   if (VLOG_IS_ON(2)) {
     // do some logging preparation and logging
     // that can't be accomplished with just VLOG(2) &lt;&lt; ...;
   }
</pre>

<p>Verbose level condition macros <code>VLOG_IF</code>,
<code>VLOG_EVERY_N</code> and <code>VLOG_IF_EVERY_N</code> behave
analogous to <code>LOG_IF</code>, <code>LOG_EVERY_N</code>,
<code>LOF_IF_EVERY</code>, but accept a numeric verbosity level as
opposed to a severity level.

<pre>
   VLOG_IF(1, (size &gt; 1024))
      &lt;&lt; "I'm printed when size is more than 1024 and when you run the "
         "program with --v=1 or more";
   VLOG_EVERY_N(1, 10)
      &lt;&lt; "I'm printed every 10th occurrence, and when you run the program "
         "with --v=1 or more. Present occurence is " &lt;&lt; COUNTER;
   VLOG_IF_EVERY_N(1, (size &gt; 1024), 10)
      &lt;&lt; "I'm printed on every 10th occurence of case when size is more "
         " than 1024, when you run the program with --v=1 or more. ";
         "Present occurence is " &lt;&lt; COUNTER;
</pre>

<h2> <A name="signal">Failure Signal Handler</A> </h2>

<p>
The library provides a convenient signal handler that will dump useful
information when the program crashes on certain signals such as SIGSEGV.
The signal handler can be installed by
google::InstallFailureSignalHandler().  The following is an example of output
from the signal handler.

<pre>
*** Aborted at 1225095260 (unix time) try "date -d @1225095260" if you are using GNU date ***
*** SIGSEGV (@0x0) received by PID 17711 (TID 0x7f893090a6f0) from PID 0; stack trace: ***
PC: @           0x412eb1 TestWaitingLogSink::send()
    @     0x7f892fb417d0 (unknown)
    @           0x412eb1 TestWaitingLogSink::send()
    @     0x7f89304f7f06 google::LogMessage::SendToLog()
    @     0x7f89304f35af google::LogMessage::Flush()
    @     0x7f89304f3739 google::LogMessage::~LogMessage()
    @           0x408cf4 TestLogSinkWaitTillSent()
    @           0x4115de main
    @     0x7f892f7ef1c4 (unknown)
    @           0x4046f9 (unknown)
</pre>

<p>
By default, the signal handler writes the failure dump to the standard
error.  You can customize the destination by InstallFailureWriter().

<h2> <A name="misc">Miscellaneous Notes</A> </h2>

<h3><A NAME=message>Performance of Messages</A></h3>

<p>The conditional logging macros provided by glog (e.g.,
<code>CHECK</code>, <code>LOG_IF</code>, <code>VLOG</code>, ...) are
carefully implemented and don't execute the right hand side
expressions when the conditions are false.  So, the following check
may not sacrifice the performance of your application.

<pre>
   CHECK(obj.ok) &lt;&lt; obj.CreatePrettyFormattedStringButVerySlow();
</pre>

<h3><A NAME=failure>User-defined Failure Function</A></h3>

<p><code>FATAL</code> severity level messages or unsatisfied
<code>CHECK</code> condition terminate your program.  You can change
the behavior of the termination by
<code>InstallFailureFunction</code>.

<pre>
   void YourFailureFunction() {
     // Reports something...
     exit(1);
   }

   int main(int argc, char* argv[]) {
     google::InstallFailureFunction(&amp;YourFailureFunction);
   }
</pre>

<p>By default, glog tries to dump stacktrace and makes the program
exit with status 1.  The stacktrace is produced only when you run the
program on an architecture for which glog supports stack tracing (as
of September 2008, glog supports stack tracing for x86 and x86_64).

<h3><A NAME=raw>Raw Logging</A></h3>

<p>The header file <code>&lt;glog/raw_logging.h&gt;</code> can be
used for thread-safe logging, which does not allocate any memory or
acquire any locks.  Therefore, the macros defined in this
header file can be used by low-level memory allocation and
synchronization code.
Please check <code>src/glog/raw_logging.h.in</code> for detail.
</p>

<h3><A NAME=plog>Google Style perror()</A></h3>

<p><code>PLOG()</code> and <code>PLOG_IF()</code> and
<code>PCHECK()</code> behave exactly like their <code>LOG*</code> and
<code>CHECK</code> equivalents with the addition that they append a
description of the current state of errno to their output lines.
E.g.

<pre>
   PCHECK(write(1, NULL, 2) &gt;= 0) &lt;&lt; "Write NULL failed";
</pre>

<p>This check fails with the following error message.

<pre>
   F0825 185142 test.cc:22] Check failed: write(1, NULL, 2) &gt;= 0 Write NULL failed: Bad address [14]
</pre>

<h3><A NAME=syslog>Syslog</A></h3>

<p><code>SYSLOG</code>, <code>SYSLOG_IF</code>, and
<code>SYSLOG_EVERY_N</code> macros are available.
These log to syslog in addition to the normal logs.  Be aware that
logging to syslog can drastically impact performance, especially if
syslog is configured for remote logging!  Make sure you understand the
implications of outputting to syslog before you use these macros. In
general, it's wise to use these macros sparingly.

<h3><A NAME=strip>Strip Logging Messages</A></h3>

<p>Strings used in log messages can increase the size of your binary
and present a privacy concern.  You can therefore instruct glog to
remove all strings which fall below a certain severity level by using
the GOOGLE_STRIP_LOG macro:

<p>If your application has code like this:

<pre>
   #define GOOGLE_STRIP_LOG 1    // this must go before the #include!
   #include &lt;glog/logging.h&gt;
</pre>

<p>The compiler will remove the log messages whose severities are less
than the specified integer value.  Since
<code>VLOG</code> logs at the severity level <code>INFO</code>
(numeric value <code>0</code>),
setting <code>GOOGLE_STRIP_LOG</code> to 1 or greater removes
all log messages associated with <code>VLOG</code>s as well as
<code>INFO</code> log statements.

<h3><A NAME=windows>Notes for Windows users</A></h3>

<p>Google glog defines a severity level <code>ERROR</code>, which is
also defined in <code>windows.h</code>
There are two known workarounds to avoid this conflict:

<ul>
  <li>#define <code>WIN32_LEAN_AND_MEAN</code> or <code>NOGDI</code>
      <strong>before</strong> you #include <code>windows.h</code> .
  <li>#undef <code>ERROR</code> <strong>after</strong> you #include
      <code>windows.h</code> .
</ul>

<p>See <a href="http://code.google.com/p/google-glog/issues/detail?id=33">
this issue</a> for more detail.

<hr>
<address>
Shinichiro Hamaji<br>
Gregor Hohpe<br>
<script type=text/javascript>
  var lm = new Date(document.lastModified);
  document.write(lm.toDateString());
</script>
</address>

</body>
</html>