This chapter covers everything about administering a running mod_perl server. First, we will explain techniques for starting, restarting, and shutting down the server. As with Perl, there’s more than one way to do it, and each technique has different implications for the server itself and the code it runs. A few widely used techniques for operating a server are presented. You may choose to use one of the suggested techniques or develop your own.
Later in the chapter, we give instructions on upgrading and disabling scripts on a live server, using a three-tier scheme, and monitoring and maintaining a web server.
To start Apache manually, just run its executable. For example, on our machine, a mod_perl-enabled Apache executable is located at /home/httpd/httpd_perl/httpd_perl. So to start it, we simply execute:
panic% /home/httpd/httpd_perl/bin/httpd_perl
This executable accepts a number of optional arguments. To find out what they are (without starting the server), use the -h argument:
panic% /home/httpd/httpd_perl/bin/httpd_perl -h
The most interesting arguments will be covered in the following sections. Any other arguments will be introduced as needed.
When developing new code, it is often helpful to run the server in single-process mode. This is most often used to find bugs in code that seems to work fine when the server starts, but refuses to work correctly after a few requests have been made. It also helps to uncover problems related to collisions between module names.
Running in single-process mode inhibits the server from automatically running in the background. This allows it to more easily be run under the control of a debugger. The -X switch is used to enable this mode:
panic% /home/httpd/httpd_perl/bin/httpd_perl -X
With the -X switch, the server runs in the foreground of the shell, so it can be killed by typing Ctrl-C. You can run it in the background by appending an ampersand:
panic% /home/httpd/httpd_perl/bin/httpd_perl -X &
Note that in -X (single-process) mode, the server will run very slowly when fetching images. Because only one request can be served at a time, requests for images normally done in parallel by the browser will now be serialized, making the page display slower.
Also note that when running with -X, the control messages that the parent server normally writes to error_log (e.g., “server started”, “server stopped”, etc.) will not be written anywhere. httpd -X causes the server to handle all requests itself without forking any children, so there is no controlling parent to write the status messages.
Usually Ctrl-C is used to kill a server running in single process mode, but Ctrl-C doesn’t constitute a clean shutdown. httpd.pid doesn’t get removed, so the next time the server is started, the message:
[warn] pid file /home/httpd/httpd_perl/logs/httpd.pid overwritten -- Unclean shutdown of previous Apache run?
will appear in error_log. You can ignore this warning; there’s nothing to worry about.
Linux and other Unix-like operating systems support a form of interprocess communication called signals. The kill command is used to send a signal to a running process. How a process responds to a signal, if it responds at all, depends on the specific signal sent and on the handler set by the process. If you are familiar with Unix signal handling, you will find that Apache adheres to the usual conventions, and you can probably skip this section. This section describes the use of kill in relation to Apache for readers who aren’t accustomed to working with signals.
The name “kill” is a misnomer; it
sounds as if the command is inherently destructive, but
kill simply sends signals to programs. Only a
few signals will actually kill the process by default. Most signals
can be caught by the process, which may choose to either perform a
specific action or ignore the signal. When a process is in a zombie
or uninterruptible sleep( ) state, it might ignore
any signals.
The following example will help dispel any fear of using this command. Most people who are familiar with the command line know that pressing Ctrl-C will usually terminate a process running in a console. For example, it is common to execute:
panic% tail -f /home/httpd/httpd_perl/logs/error_log
to monitor the Apache server’s error_log file. The only way to stop tail is by pressing Ctrl-C in the console in which the process is running. The same result can be achieved by sending the INT (interrupt) signal to this process. For example:
panic% kill -INT 17084
When this command is run, the tail process is aborted, assuming that the process identifier (PID) of the tail process is 17084.
Every process running in the system has its own PID. kill identifies processes by their PIDs. If kill were to use process names and there were two tail processes running, it might send the signal to the wrong process. The most common way to determine the PID of a process is to use ps to display information about the current processes on the machine. The arguments to this utility vary depending on the operating system. For example, on BSD-family systems, the following command works:
panic% ps auxc | grep tail
On a System V Unix flavor such as Solaris, the following command may be used instead:
panic% ps -eaf | grep tail
In the first part of the command, ps prints information about all the current processes. This is then piped to a grep command that prints lines containing the text “tail”. Assuming only one such tail process is running, we get the following output:
root 17084 0.1 0.1 1112 408 pts/8 S 17:28 0:00 tail
The first column shows the username of the account running the process, the second column shows the PID, and the last column shows the name of the command. The other columns vary between operating systems.
Processes are free to ignore almost all signals they receive, and there are cases when they will. Let’s run the less command on the same error_log file:
panic% less /home/httpd/httpd_perl/logs/error_log
Neither pressing Ctrl-C nor sending the INT signal will kill the process, because the implementers of this utility chose to ignore that signal. The way to kill the process is to type q.
Sometimes numerical signal values are used instead of their symbolic names. For example, 2 is normally the numeric equivalent of the symbolic name INT. Hence, these two commands are equivalent on Linux:
panic% kill -2 17084 panic% kill -INT 17084
On Solaris, the -s option is used when working with symbolic signal names:
panic% kill -s INT 17084
To find the numerical equivalents, either refer to the signal(7) manpage, or ask Perl to help you:
panic% perl -MConfig -e 'printf "%6s %2d\n", $_, $sig++ \
for split / /, $Config{sig_name}'If you want to send a signal to all processes with the same name, you can use pkill on Solaris or killall on Linux.
Apache performs certain actions in response to the KILL, TERM, HUP, and USR1 signals (as arguments to kill). All Apache system administrators should be familiar with the use of these signals to control the Apache web server.
By referring to the signal.h file, we learn the numerical equivalents of these signals:
#define SIGHUP 1 /* hangup, generated when terminal disconnects */ #define SIGKILL 9 /* last resort */ #define SIGTERM 15 /* software termination signal */ #define SIGUSR1 30 /* user defined signal 1 */
The four types of signal are:
- KILL signal: forcefully shutdown
The KILL (9) signal should never be used unless absolutely necessary, because it will unconditionally kill Apache, without allowing it to clean up properly. For example, the httpd.pid file will not be deleted, and any existing requests will simply be terminated halfway through. Although failure to delete httpd.pid is harmless, if code was registered to run upon child exit but was not executed because Apache was sent the KILL signal, you may have problems. For example, a database connection may be closed incorrectly, leaving the database in an inconsistent state.
The three other signals have safe and legitimate uses, and the next sections will explain what happens when each of them is sent to an Apache server process.
It should be noted that these signals should be sent only to the parent process, not to any of the child processes. The parent process PID may be found either by using ps auxc | grep apache (where it will usually be the lowest-numbered Apache process) or by executing cat on the httpd.pid file. See Section 5.3.3, later in this chapter, for more information.
- TERM signal: stop now
Sending the TERM signal to the parent causes it to attempt to kill off all its children immediately. Any requests in progress are terminated, and no further requests are accepted. This operation may take tens of seconds to complete. To stop a child, the parent sends it an HUP signal. If the child does not die before a predetermined amount of time, the parent sends a second HUP signal. If the child fails to respond to the second HUP, the parent then sends a TERM signal, and if the child still does not die, the parent sends the KILL signal as a last resort. Each failed attempt to kill a child generates an entry in the error_log file.
Before each process is terminated, the Perl cleanup stage happens, in which Perl
ENDblocks and global objects’DESTROYmethods are run.When all child processes have been terminated, all open log files are closed and the parent itself exits.
Unless an explicit signal name is provided, kill sends the TERM signal by default. Therefore:
panic# kill -TERM 1640
and:
panic# kill 1640
will do the same thing.
- HUP signal: restart now
Sending the HUP signal to the parent causes it to kill off its children as if the TERM signal had been sent. That is, any requests in progress are terminated, but the parent does not exit. Instead, the parent rereads its configuration files, spawns a new set of child processes, and continues to serve requests. It is almost equivalent to stopping and then restarting the server.
If the configuration files contain errors when restart is signaled, the parent will exit, so it is important to check the configuration files for errors before issuing a restart. We’ll cover how to check for errors shortly.
Using this approach to restart mod_perl-enabled Apache may cause the processes’ memory consumption to grow after each restart. This happens when Perl code loaded in memory is not completely torn down, leading to a memory leak.
- USR1 signal: gracefully restart now
The USR1 signal causes the parent process to advise the children to exit after serving their current requests, or to exit immediately if they are not serving a request. The parent rereads its configuration files and reopens its log files. As each child dies off, the parent replaces it with a child from the new generation (the new children use the new configuration) and the new child processes begin serving new requests immediately.
The only difference between USR1 and HUP is that USR1 allows the children to complete any current requests prior to terminating. There is no interruption in the service, unlike with the HUP signal, where service is interrupted for the few (and sometimes more) seconds it takes for a restart to complete.
By default, if a server is restarted using the USR1 or the HUP signal
and mod_perl is not compiled as a DSO, Perl scripts and modules are
not reloaded. To reload modules pulled in via
PerlRequire, PerlModule, or
use, and to flush the
Apache::Registry cache, either completely stop the
server and then start it again, or use this directive in
httpd.conf:
PerlFreshRestart On
(This directive is not always recommended. See Chapter 22 for further details.)
Restart or
termination
of a mod_perl server may sometimes take quite a long time, perhaps
even tens of seconds. The reason for this is a call to the
perl_destruct( )
function during the child exit
phase, which is also known as the cleanup phase. In this phase, the
Perl END blocks are run and the
DESTROY
method is called on any global objects
that are still around.
Sometimes this will produce a series of messages in the
error_log file, warning that certain child
processes did not exit as expected. This happens when a child
process, after a few attempts have been made to terminate it, is
still in the middle of perl_destruct( ). So when
you shut down the server, you might see something like this:
[warn] child process 7269 still did not exit,
sending a SIGTERM
[error] child process 7269 still did not exit,
sending a SIGKILL
[notice] caught SIGTERM, shutting downFirst, the parent process sends the TERM signal to all of its children, without logging a thing. If any of the processes still doesn’t quit after a short period, it sends a second TERM, logs the PID of the process, and marks the event as a warning. Finally, if the process still hasn’t terminated, it sends the KILL signal, which unconditionaly terminates the process, aborting any operation in progress in the child. This event is logged as an error.
If the mod_perl scripts do not contain any END
blocks or DESTROY methods that need to be run
during shutdown, or if the ones they have are nonessential, this step
can be avoided by setting the PERL_DESTRUCT_LEVEL
environment variable to -1. (The
-1 value for
PERL_DESTRUCT_LEVEL is special to mod_perl.) For
example, add this setting to the httpd.conf
file:
PerlSetEnv PERL_DESTRUCT_LEVEL -1
What constitutes a significant cleanup? Any change of state outside the current process that cannot be handled by the operating system itself. Committing database transactions and removing the lock on a resource are significant operations, but closing an ordinary file is not. For example, if DBI is used for persistent database connections, Perl’s destructors should not be switched off.
In order to send a signal to a process, its PID must be known. But in the case of Apache, there are many httpd processes running. Which one should be used? The parent process is the one that must be signaled, so it is the parent’s PID that must be identified.
The easiest way to find the Apache parent PID is to read the
httpd.pid
file. To find this file, look in the
httpd.conf file. Open
httpd.conf and look for the
PidFile directive. Here is the line from our
httpd.conf file:
PidFile /home/httpd/httpd_perl/logs/httpd.pid
When Apache starts up, it writes its own process ID in httpd.pid in a human-readable format. When the server is stopped, httpd.pid should be deleted, but if Apache is killed abnormally, httpd.pid may still exist even if the process is not running any more.
Of course, the PID of the running Apache can also be found using the ps(1) and grep(1) utilities (as shown previously). Assuming that the binary is called httpd_perl, the command would be:
panic% ps auxc | grep httpd_perl
or, on System V:
panic% ps -ef | grep httpd_perl
This will produce a list of all the httpd_perl (parent and child) processes. If the server was started by the root user account, it will be easy to locate, since it will belong to root. Here is an example of the sort of output produced by one of the ps command lines given above:
root 17309 0.9 2.7 8344 7096 ? S 18:22 0:00 httpd_perl nobody 17310 0.1 2.7 8440 7164 ? S 18:22 0:00 httpd_perl nobody 17311 0.0 2.7 8440 7164 ? S 18:22 0:00 httpd_perl nobody 17312 0.0 2.7 8440 7164 ? S 18:22 0:00 httpd_perl
In this example, it can be seen that all the child processes are running as user nobody whereas the parent process runs as user root. There is only one root process, and this must be the parent process. Any kill signals should be sent to this parent process.
If the server is started under some other user account (e.g., when the user does not have root access), the processes will belong to that user. The only truly foolproof way to identify the parent process is to look for the process whose parent process ID (PPID) is 1 (use ps to find out the PPID of the process).
If you have the GNU tools installed on your system, there is a nifty utility that makes it even easier to discover the parent process. The tool is called pstree , and it is very simple to use. It lists all the processes showing the family hierarchy, so if we grep the output for the wanted process’s family, we can see the parent process right away. Running this utility and greping for httpd_perl, we get:
panic% pstree -p | grep httpd_perl |-httpd_perl(17309)-+-httpd_perl(17310) | |-httpd_perl(17311) | |-httpd_perl(17312)
And this one is even simpler:
panic% pstree -p | grep 'httpd_perl.*httpd_perl' |-httpd_perl(17309)-+-httpd_perl(17310)
In both cases, we can see that the parent process has the PID 17309.
ps’s f option, available on many Unix platforms, produces a tree-like report of the processes as well. For example, you can run ps axfwwww to get a tree of all processes.
The Apache distribution comes with a script to control the server called apachectl, installed into the same location as the httpd executable. For the sake of the examples, let’s assume that it is in /home/httpd/httpd_perl/bin/apachectl.
All the operations that can be performed by using signals can also be performed on the server by using apachectl. You don’t need to know the PID of the process, as apachectl will find this out for itself.
To start httpd_perl:
panic% /home/httpd/httpd_perl/bin/apachectl start
To stop httpd_perl:
panic% /home/httpd/httpd_perl/bin/apachectl stop
To restart httpd_perl (if it is running, send HUP; if it is not, just start it):
panic% /home/httpd/httpd_perl/bin/apachectl restart
Do a graceful restart by sending a USR1 signal, or start it if it’s not running:
panic% /home/httpd/httpd_perl/bin/apachectl graceful
To perform a configuration test:
panic% /home/httpd/httpd_perl/bin/apachectl configtest
There are other options for apachectl. Use the help option to see them all.
panic% /home/httpd/httpd_perl/bin/apachectl help
It is important to remember that apachectl uses
the PID file, which is specified by the PidFile
directive in httpd.conf. If the PID file is
deleted by hand while the server is running, or if the
PidFile directive is missing or in error,
apachectl will be unable to stop or restart the
server.
If the configuration file has syntax errors, attempting to restart the server will fail and the server will die. However, if a graceful restart is attempted using apachectl and the configuration file contains errors, the server will issue an error message and continue running with the existing configuration. This is because apachectl validates the configuration file before issuing the actual restart command when a graceful restart is requested.
Apache provides a method to check the configuration’s syntax without actually starting the server. You can run this check at any time, whether or not a server is currently running. The check has two forms, using the -t or -T options. For example:
panic% /home/httpd/httpd_perl/bin/httpd_perl -t
-t
will verify that the
DocumentRoot directory exists, whereas
-T will not.
-T
is most useful when using
a configuration file containing a large number of virtual hosts,
where verifying the existence of each DocumentRoot
directory can take a substantial amount of time.
Note that when running this test with a mod_perl server, the Perl
code will be executed just as it would be at server
startup—that is, from within the
httpd.conf
<Perl>
sections or a startup file.
If a group of developers need to be able to start and stop the server, there may be a temptation to give them the root password, which is probably not a wise thing to do. The fewer people that know the root password, the less likely you will encounter problems. Fortunately, an easy solution to this problem is available on Unix platforms. It is called a setuid executable (setuid root in this case).
Before continuing, we must stress that this technique should not be used unless it is absolutely necessary. If an improperly written setuid script is used, it may compromise the system by giving root privileges to system breakers (crackers).
To be on the safe side, do not deploy the techniques explained in this section. However, if this approach is necessary in a particular situation, this section will address the possible problems and provide solutions to reduce the risks to a minimum.
A setuid executable has the setuid permissions bit set, with the following command:
panic% chmod u+s filename
This sets the process’s effective user ID to that of the file upon execution. Most users have used setuid executables even if they have not realized it. For example, when a user changes his password he executes the passwd command, which, among other things, modifies the /etc/passwd file. In order to change this file, the passwd program needs root permissions. The passwd command has the setuid bit set, so when someone executes this utility, its effective ID becomes the root user ID.
Using setuid executables should be avoided as a general practice. The less setuid executables there are in a system, the less likely it is that someone will find a way to break in. One approach that crackers use is to find and exploit unanticipated bugs in setuid executables.
When the executable is setuid to root, it is vital to ensure that it does not extend read and write permissions to its group or to the world. Let’s take the passwd utility as an example. Its permissions are:
panic% ls -l /usr/bin/passwd -r-s--x--x 1 root root 12244 Feb 8 00:20 /usr/bin/passwd
The program is group- and world-executable but cannot be read or written by group or world. This is achieved with the following command:
panic% chmod 4511 filename
The first digit (4) stands for the setuid bit, the second digit (5) is a bitwise-OR of read (4) and executable (1) permissions for the user, and the third and fourth digits set the executable (1) permissions for group and world.
In the situation where several developers need to be able to start and stop an Apache server that is run by the root account, setuid access must be available only to this specific group of users. For the sake of this example, let’s assume that these developers belong to a group named apache. It is important that users who are not root or are not part of the apache group are unable to execute this script. Therefore, the following commands must be applied to the apachectl program:
panic% chgrp apache apachectl panic% chmod 4510 apachectl
The execution order is important. If the commands are executed in reverse order, the setuid bit is lost.
The file’s permissions now look like this:
panic% ls -l apachectl -r-s--x--- 1 root apache 32 May 13 21:52 apachectl
Everything is set. Well, almost...
When Apache is started, Apache and Perl modules are loaded, so code may be executed. Since all this happens with the root effective ID, any code is executed as if run by the root user. This means that there is a risk, even though none of the developers has the root password—all users in the apache group now have an indirect root access. For example, if Apache loads some module or executes some code that is writable by any of these users, they can plant code that will allow them to gain shell access to the root account.
Of course, if the developers are not trusted, this setuid solution is not the right approach. Although it is possible to try to check that all the files Apache loads are not writable by anyone but root, there are so many of them (especially with mod_perl, where many Perl modules are loaded at server startup) that this is a risky approach.
If the developers are trusted, this approach suits the situation. Although there are security concerns regarding Apache startup, once the parent process is loaded, the child processes are spawned as non-root processes.
This section has presented a way to allow non-root users to start and stop the server. The rest is exactly the same as if they were executing the script as root in the first place.
Example 5-1 shows a sample setuid Apache startup script.
Note the line marked WORKAROUND, which fixes an obscure error when starting a mod_perl-enabled Apache, by setting the real UID to the effective UID. Without this workaround, a mismatch between the real and the effective UIDs causes Perl to croak on the -e switch.
This script depends on using a version of Perl that recognizes and emulates the setuid bits. This script will do different things depending on whether it is named start_httpd, stop_httpd, or restart_httpd; use symbolic links to create the names in the filesystem.
Example 5-1. suid_apache_ctl
#!/usr/bin/perl -T
use strict;
# These constants will need to be adjusted.
my $PID_FILE = '/home/httpd/httpd_perl/logs/httpd.pid';
my $HTTPD = '/home/httpd/httpd_perl/bin/httpd_perl ';
$HTTPD .= '-d /home/httpd/httpd_perl';
# These prevent taint checking failures
$ENV{PATH} = '/bin:/usr/bin';
delete @ENV{qw(IFS CDPATH ENV BASH_ENV)};
# This sets the real to the effective ID, and prevents
# an obscure error when starting apache/mod_perl
$< = $>; # WORKAROUND
$( = $) = 0; # set the group to root too
# Do different things depending on our name
my $name = $0;
$name =~ m|([^/]+)$|;
if ($name eq 'start_httpd') {
system $HTTPD and die "Unable to start HTTPD";
print "HTTP started.\n";
exit 0;
}
# extract the process id and confirm that it is numeric
my $pid = `cat $PID_FILE`;
$pid =~ /^(\d+)$/ or die "PID $pid not numeric or not found";
$pid = $1;
if ($name eq 'stop_httpd') {
kill 'TERM', $pid or die "Unable to signal HTTPD";
print "HTTP stopped.\n";
exit 0;
}
if ($name eq 'restart_httpd') {
kill 'HUP', $pid or die "Unable to signal HTTPD";
print "HTTP restarted.\n";
exit 0;
}
# script is named differently
die "Script must be named start_httpd, stop_httpd, or restart_httpd.\n";When using a non-production development box, it is OK to start and stop the web server by hand when necessary. On a production system, however, it is possible that the machine on which the server is running will have to be rebooted. When the reboot is completed, who is going to remember to start the server? It is easy to forget this task, and what happens if no one is around when the machine is rebooted? (Some OSs will reboot themselves without human intervention in certain situations.)
After the server installation is complete, it is important to remember that a script to perform the server startup and shutdown should be put in a standard system location—for example, /etc/rc.d under Red Hat Linux, or /etc/init.d/apache under Debian GNU/Linux.
This book uses Red Hat-compatible Linux distributions in its examples. Let’s step aside for a brief introduction to the System V (SysV) init system that many Linux and other Unix flavors use to manage starting and stopping daemons. (A daemon is a process that normally starts at system startup and runs in the background until the system goes down.)
The SysV init system keeps all its files in the /etc/rc.d/ directory. This directory contains a number of subdirectories:
panic% find /etc/rc.d -type d /etc/rc.d /etc/rc.d/init.d /etc/rc.d/rc0.d /etc/rc.d/rc1.d /etc/rc.d/rc2.d /etc/rc.d/rc3.d /etc/rc.d/rc4.d /etc/rc.d/rc5.d /etc/rc.d/rc6.d
/etc/rc.d/init.d contains many scripts, one for each service that needs to be started at boot time or when entering a specific runlevel. Common services include networking, file sharing, mail servers, web servers, FTP servers, etc.
When the system boots, the special init script runs all scripts for the default runlevel. The default runlevel is specified in the /etc/inittab file. This file contains a line similar to this one:
id:3:initdefault:
The second column indicates that the default runlevel is 3, which is the default for most server systems. (5 is the default for desktop machines.)
Let’s now see how the scripts are run. We’ll first look at the contents of the /etc/rc.d/rc3.d directory:
panic% ls -l /etc/rc.d/rc3.d lrwxrwxrwx 1 root root 13 Jul 1 01:08 K20nfs -> ../init.d/nfs lrwxrwxrwx 1 root root 18 Jul 1 00:54 K92ipchains -> ../init.d lrwxrwxrwx 1 root root 17 Jul 1 00:51 S10network -> ../init.d/network lrwxrwxrwx 1 root root 16 Jul 1 00:51 S30syslog -> ../init.d/syslog lrwxrwxrwx 1 root root 13 Jul 1 00:52 S40atd -> ../init.d/atd lrwxrwxrwx 1 root root 15 Jul 1 00:51 S40crond -> ../init.d/crond lrwxrwxrwx 1 root root 15 Jul 1 01:13 S91httpd_docs -> ../init.d/httpd_docs lrwxrwxrwx 1 root root 15 Jul 1 01:13 S91httpd_perl -> ../init.d/httpd_perl lrwxrwxrwx 1 root root 17 Jul 1 00:51 S95kheader -> ../init.d/kheader lrwxrwxrwx 1 root root 11 Jul 1 00:51 S99local -> ../rc.local
(Only part of the output is shown here, since many services are started and stopped at runlevel 3.)
There are no real files in the directory. Instead, each file is a symbolic link to one of the scripts in the init.d directory. The links’ names start with a letter (S or K) and a two-digit number. S specifies that the script should be run when the service is started and K specifies that the script should be run when the service is stopped. The number following S or K is there for ordering purposes: init will start services in the order in which they appear.
init runs each script with an argument that is either start or stop, depending on whether the link’s name starts with S or K. Scripts can be executed from the command line; the following command line will stop the httpd server:
panic# /etc/rc.d/init.d/httpd_perl stop
Unfortunately, different Unix flavors implement different init systems. Refer to your system’s documentation.
Now that we’re familiar with how the init system works, let’s return to our discussion of apachectl scripts.
Generally, the simplest solution is to copy the apachectl script to the startup directory or, better still, create a symbolic link from the startup directory to the apachectl script. The apachectl utility is in the same directory as the Apache executable after Apache installation (e.g., /home/httpd/httpd_perl/bin). If there is more than one Apache server, there will need to be a separate script for each one, and of course they will have to have different names so that they can coexist in the same directory.
On one of our Red Hat Linux machines with two servers, we have the following setup:
/etc/rc.d/init.d/httpd_docs /etc/rc.d/init.d/httpd_perl /etc/rc.d/rc3.d/S91httpd_docs -> ../init.d/httpd_docs /etc/rc.d/rc3.d/S91httpd_perl -> ../init.d/httpd_perl /etc/rc.d/rc6.d/K16httpd_docs -> ../init.d/httpd_docs /etc/rc.d/rc6.d/K16httpd_perl -> ../init.d/httpd_perl
The scripts themselves reside in the /etc/rc.d/init.d directory. There are symbolic links to these scripts in /etc/rc.d/rc*.d directories.
When the system starts (runlevel 3), we want Apache to be started when
all the services on which it might depend are already running.
Therefore, we have used S91. If, for example,
the mod_perl-enabled Apache issues a connect_on_init(
), the SQL server should be started before Apache.
When the system shuts down (runlevel 6), Apache should be one of the first processes to be stopped—therefore, we have used K16. Again, if the server does some cleanup processing during the shutdown event and requires third-party services (e.g., a MySQL server) to be running at the time, it should be stopped before these services.
Notice that it is normal for more than one symbolic link to have the same sequence number.
Under Red Hat Linux and similar systems, when a machine is booted and its runlevel is set to 3 (multiuser plus network), Linux goes into /etc/rc.d/rc3.d/ and executes the scripts to which the symbolic links point with the start argument. When it sees S87httpd_perl, it executes:
/etc/rc.d/init.d/httpd_perl start
When the machine is shut down, the scripts are executed through links from the /etc/rc.d/rc6.d/ directory. This time the scripts are called with the stop argument, like this:
/etc/rc.d/init.d/httpd_perl stop
Most systems have GUI utilities to automate the creation of symbolic links. For example, Red Hat Linux includes the ntsysv and tksysv utilities. These can be used to create the proper symbolic links. Before it is used, the apachectl or similar scripts should be put into the init.d directory or an equivalent directory. Alternatively, a symbolic link to some other location can be created.
However, it’s been reported that sometimes these tools mess up and break things. Therefore, the robust chkconfig utility should be used instead. The following example shows how to add an httpd_perl startup script to the system using chkconfig.
The apachectl script may be kept in any directory, as long as it can be the target of a symbolic link. For example, it might be desirable to keep all Apache executables in the same directory (e.g., /home/httpd/httpd_perl/bin), in which case all that needs to be done is to provide a symbolic link to this file:
panic% ln -s /home/httpd/httpd_perl/bin/apachectl /etc/rc.d/init.d/httpd_perl
Edit the apachectl script to add the following lines after the script’s main header:
# Comments to support chkconfig on RedHat Linux # chkconfig: 2345 91 16 # description: mod_perl enabled Apache Server
Now the beginning of the script looks like:
#!/bin/sh # # Apache control script designed to allow an easy command line # interface to controlling Apache. Written by Marc Slemko, # 1997/08/23 # Comments to support chkconfig on Red Hat Linux # chkconfig: 2345 91 16 # description: mod_perl-enabled Apache Server # # The exit codes returned are: # ...
Adjust the line:
# chkconfig: 2345 91 16
to suit your situation. For example, the setting used above says the script should be started in levels 2, 3, 4, and 5, that its start priority should be 91, and that its stop priority should be 16.
Now all you need to do is ask chkconfig to configure the startup scripts. Before doing so, it is best to check what files and links are in place:
panic% find /etc/rc.d | grep httpd_perl /etc/rc.d/init.d/httpd_perl
This response means that only the startup script itself exists. Now execute:
panic% chkconfig --add httpd_perl
and repeat the find command to see what has changed:
panic% find /etc/rc.d | grep httpd_perl /etc/rc.d/init.d/httpd_perl /etc/rc.d/rc0.d/K16httpd_perl /etc/rc.d/rc1.d/K16httpd_perl /etc/rc.d/rc2.d/S91httpd_perl /etc/rc.d/rc3.d/S91httpd_perl /etc/rc.d/rc4.d/S91httpd_perl /etc/rc.d/rc5.d/S91httpd_perl /etc/rc.d/rc6.d/K16httpd_perl
The chkconfig program has created all the required symbolic links using the startup and shutdown priorities as specified in the line:
# chkconfig: 2345 91 16
If for some reason it becomes necessary to remove the service from the startup scripts, chkconfig can perform the removal of the links automatically:
panic% chkconfig --del httpd_perl
By running the find command once more, you can see that the symbolic links have been removed and only the original file remains:
panic% find /etc/rc.d | grep httpd_perl /etc/rc.d/init.d/httpd_perl
Again, execute:
panic% chkconfig --add httpd_perl
Note that when using symbolic links, the link name in /etc/rc.d/init.d is what matters, not the name of the script to which the link points.
When you’re developing code on a development server, anything goes: modifying the configuration, adding or upgrading Perl modules without checking that they are syntactically correct, not checking that Perl modules don’t collide with other modules, adding experimental new modules from CPAN, etc. If something goes wrong, configuration changes can be rolled back (assuming you’re using some form of version control), modules can be uninstalled or reinstalled, and the server can be started and stopped as many times as required to get it working.
Of course, if there is more than one developer working on a development server, things can’t be quite so carefree. Possible solutions for the problems that can arise when multiple developers share a development server will be discussed shortly.
The most difficult situation is transitioning changes to a live server. However much the changes have been tested on a development server, there is always the risk of breaking something when a change is made to the live server. Ideally, any changes should be made in a way that will go unnoticed by the users, except as new or improved functionality or better performance. No users should be exposed to even a single error message from the upgraded service—especially not the “database busy” or “database error” messages that some high-profile sites seem to consider acceptable.
Live services can be divided into two categories: servers that must be up 24 hours a day and 7 days a week, and servers that can be stopped during non-working hours. The latter generally applies to Intranets of companies with offices located more or less in the same time zone and not scattered around the world. Since the Intranet category is the easier case, let’s talk about it first.
An Intranet server generally serves the company’s internal staff by allowing them to share and distribute internal information, read internal email, and perform other similar tasks. When all the staff is located in the same time zone, or when the time difference between sites does not exceed a few hours, there is often no need for the server to be up all the time. This doesn’t necessarily mean that no one will need to access the Intranet server from home in the evenings, but it does mean that the server can probably be stopped for a few minutes when it is necessary to perform some maintenance work.
Even if the update of a live server occurs during working hours and goes wrong, the staff will generally tolerate the inconvenience unless the Intranet has become a really mission-critical tool. For servers that are mission critical, the following section will describe the least disruptive and safest upgrade approach.
If possible, any administration or upgrades of the company’s Intranet server should be undertaken during non-working hours, or, if this is not possible, during the times of least activity (e.g., lunch time). Upgrades that are carried out while users are using the service should be done with a great deal of care.
In very large organizations, upgrades are often scheduled events and employees are notified ahead of time that the service might not be available. Some organizations deem these periods “at-risk” times, when employees are expected to use the service as little as possible and then only for noncritical work. Again, these major updates are generally scheduled during the weekends and late evening hours.
The next section deals with this issue for services that need to be available all the time.
Internet servers are normally expected to be available 24 hours a day, 7 days a week. E-commerce sites, global B2B (business-to-business) sites, and any other revenue-producing sites may be critical to the companies that run them, and their unavailability could prove to be very expensive. The approach taken to ensure that servers remain in service even when they are being upgraded depends on the type of server in use. There are two categories to consider: server clusters and single servers.
When a service is very popular, a single machine probably will not be able to keep up with the number of requests the service has to handle. In this situation, the solution is to add more machines and to distribute the load amongst them. From the user’s point of view, the use of multiple servers must be completely transparent; users must still have a single access point to the service (i.e., the same single URL) even though there may be many machines with different server names actually delivering the service. The requests must also be properly distributed across the machines: not simply by giving equal numbers of requests to each machine, but rather by giving each machine a load that reflects its actual capabilities, given that not all machines are built with identical hardware. This leads to the need for some smart load-balancing techniques.
All current load-balancing techniques are based on a central machine that dispatches all incoming requests to machines that do the real processing. Think of it as the only entrance into a building with a doorkeeper directing people into different rooms, all of which have identical contents but possibly a different number of clerks. Regardless of what room they’re directed to, all people use the entrance door to enter and exit the building, and an observer located outside the building cannot tell what room people are visiting. The same thing happens with the cluster of servers—users send their browsers to URLs, and back come the pages they requested. They remain unaware of the particular machines from which their browsers collected their pages.
No matter what load-balancing technique is used, it should always be straightforward to be able to tell the central machine that a new machine is available or that some machine is not available any more.
How does this long introduction relate to the upgrade problem? Simple. When a particular machine requires upgrading, the dispatching server is told to stop sending requests to that machine. All the requests currently being executed must be left to complete, at which point whatever maintenance and upgrade work is to be done can be carried out. Once the work is complete and has been tested to ensure that everything works correctly, the central machine can be told that it can again send requests to the newly upgraded machine. At no point has there been any interruption of service or any indication to users that anything has occurred. Note that for some services, particularly ones to which users must log in, the wait for all the users to either log out or time out may be considerable. Thus, some sites stop requests to a machine at the end of the working day, in the hope that all requests will have completed or timed out by the morning.
How do we talk to the central machine? This depends on the load-balancing technology that is implemented and is beyond the scope of this book. The references section at the end of this chapter gives a list of relevant online resources.
It’s not uncommon for a popular web site to run on a single machine. It’s also common for a web site to run on multiple machines, with one machine dedicated to serving static objects (such as images and static HTML files), another serving dynamically generated responses, and perhaps even a third machine that acts as a dedicated database server.
Therefore, the situation that must be addressed is where just one machine runs the service or where the service is spread over a few machines, with each performing a unique task, such that no machine can be shut down even for a single minute, and leaving the service unavailable for more than five seconds is unacceptable. In this case, two different tasks may be required: upgrading the software on the server (including the Apache server), and upgrading the code of the service itself (i.e., custom modules and scripts).
There are many things that you might need to update on a server, ranging from a major upgrade of the operating system to just an update of a single piece of software (such as the Apache server itself).
One simple approach to performing an upgrade painlessly is to have a backup machine, of similar capacity and identical configuration, that can replace the production machine while the upgrade is happening. It is a good idea to have such a machine handy and to use it whenever major upgrades are required. The two machines must be kept synchronized, of course. (For Unix/Linux users, tools such as rsync and mirror can be used for synchronization.)
However, it may not be necessary to have a special machine on standby as a backup. Unless the service is hosted elsewhere and you can’t switch the machines easily, the development machine is probably the best choice for a backup—all the software and scripts are tested on the development machine as a matter of course, and it probably has a software setup identical to that of the production machine. The development machine might not be as powerful as the live server, but this may well be acceptable for a short period, especially if the upgrade is timed to happen when the site’s traffic is fairly quiet. It’s much better to have a slightly slower service than to close the doors completely. A web log analysis tool such as analog can be used to determine the hour of the day when the server is under the least load.
Switching between the two machines is very simple:
Shut down the network on the backup machine.
Configure the backup machine to use the same IP address and domain name as the live machine.
Shut down the network on the live machine (do not shut down the machine itself!).
Start up the network on the backup machine.
When you are certain that the backup server has successfully replaced the live server (that is, requests are being serviced, as revealed by the backup machine’s access_log), it is safe to switch off the master machine or do any necessary upgrades.
Why bother waiting to check that everything is working correctly with the backup machine? If something goes wrong, the change can immediately be rolled back by putting the known working machine back online. With the service restored, there is time to analyze and fix the problem with the replacement machine before trying it again. Without the ability to roll back, the service may be out of operation for some time before the problem is solved, and users may become frustrated.
We recommend that you practice this technique with two unused machines before using the production boxes.
After the backup machine has been put into service and the original machine has been upgraded, test the original machine. Once the original machine has been passed as ready for service, the server replacement technique described above should be repeated in reverse. If the original machine does not work correctly once returned to service, the backup machine can immediately be brought online while the problems with the original are fixed.
You cannot have two machines configured to use the same IP address, so the first machine must release the IP address by shutting down the link using this IP before the second machine can enable its own link with the same IP address. This leads to a short downtime during the switch. You can use the heartbeat utility to automate this process and thus possibly shorten the downtime period. See the references section at the end of this chapter for more information about heartbeat.
Using more than one machine to perform an update may not be convenient, or even possible. An alternative solution is to use the port-forwarding capabilities of the host’s operating system.
One approach is to configure the web server to listen on an unprivileged port, such as 8000, instead of 80. Then, using a firewalling tool such as iptables, ipchains, or ipfwadm, redirect all traffic coming for port 80 to port 8000. Keeping a rule like this enabled at all times on a production machine will not noticeably affect performance.
Once this rule is in place, it’s a matter of getting the new code in place, adjusting the web server configuration to point to the new location, and picking a new unused port, such as 8001. This way, you can start the “new” server listening on that port and not affect the current setup.
To check that everything is working, you could test the server by accessing it directly by port number. However, this might break links and redirections. Instead, add another port forwarding rule before the first one, redirecting traffic for port 80 from your test machine or network to port 8001.
Once satisfied with the new server, publishing the change is just a matter of changing the port-forwarding rules one last time. You can then stop the now old server and everything is done.
Now you have your primary server listening on port 8001, answering requests coming in through port 80, and nobody will have noticed the change.
Assuming that the testbed machine and the live server have an identical software installation, consider preparing an upgrade package with the components that must be upgraded. Test this package on the testbed machine, and when it is evident that the package gets installed flawlessly, install it on the live server. Do not build the software from scratch on the live server, because if a mistake is made, it could cause the live server to misbehave or even to fail.
For example, many Linux distributions use the Red Hat Package Manager (RPM) utility, rpm, to distribute source and binary packages. It is not necessary for a binary package to include any compiled code (for example, it can include Perl scripts, but it is still called a binary). A binary package allows the new or upgraded software to be used the moment you install it. The rpm utility is smart enough to make upgrades (i.e., remove previous installation files, preserve configuration files, and execute appropriate installation scripts).
If, for example, the mod_perl server needs to be upgraded, one approach is to prepare a package on a similarly configured machine. Once the package has been built, tested, and proved satisfactory, it can then be transferred to the live machine. The rpm utility can then be used to upgrade the mod_perl server. For example, if the package file is called mod_perl-1.26-10.i386.rpm, this command:
panic% rpm -Uvh mod_perl-1.26-10.i386.rpm
will remove the previous server (if any) and install the new one.
There’s no problem upgrading software that doesn’t break any dependencies in other packages, as in the above example. But what would happen if, for example, the Perl interpreter needs to be upgraded on the live machine?
If the mod_perl package described earlier was properly prepared, it would specify the packages on which it depends and their versions. So if Perl was upgraded using an RPM package, the rpm utility would detect that the upgrade would break a dependency, since the mod_perl package is supposed to work with the previous version of Perl. rpm will not allow the upgrade unless forced to.
This is a very important feature of RPM. Of course, it relies on the fact that the person who created the package has set all the dependencies correctly. Do not trust packages downloaded from the Web. If you have to use an RPM package prepared by someone else, get its source, read its specification file, and make doubly sure that it’s what you want.
The Perl upgrade task is in fact a very easy problem to solve. Have two packages ready on the development machine: one for Perl and the other for mod_perl, the latter built using the Perl version that is going to be installed. Upload both of them to the live server and install them together. For example:
panic% rpm -Uvh mod_perl-1.26-10.i386.rpm perl-5.6.1-5.i386.rpm
This should be done as an atomic operation—i.e., as a single execution of the rpm program. If the installation of the packages is attempted with separate commands, they will both fail, because each of them will break some dependency.
If a mistake is made and checks reveal that a faulty package has been installed, it is easy to roll back. Just make sure that the previous version of the properly packaged software is available. The packages can be downgraded by using the —force option—and voilà, the previously working system is restored. For example:
panic% rpm -Uvh --force mod_perl-1.26-9.i386.rpm perl-5.6.1-4.i386.rpm
Although this example uses the rpm utility, other similar utilities exist for various operating systems and distributions. Creating packages provides a simple way of upgrading live systems (and downgrading them if need be). The packages used for any successful upgrade should be kept, because they will become the packages to downgrade to if a subsequent upgrade with a new package fails.
When using a cluster of machines with identical setups, there is another important benefit of prepackaged upgrades. Instead of doing all the upgrades by hand, which could potentially involve dozens or even hundreds of files, preparing a package can save lots of time and will minimize the possibility of error. If the packages are properly written and have been tested thoroughly, it is perfectly possible to make updates to machines that are running live services. (Note that not all operating systems permit the upgrading of running software. For example, Windows does not permit DLLs that are in active use to be updated.)
It should be noted that the packages referred to in this discussion are ones made locally, specifically for the systems to be upgraded, not generic packages downloaded from the Internet. Making local packages provides complete control over what is installed and upgraded and makes upgrades into atomic actions that can be rolled back if necessary. We do not recommend using third-party packaged binaries, as they will almost certainly have been built for a different environment and will not have been fine-tuned for your system.
Yet another alternative is to use symbolic links for upgrades. This concept is quite simple: install a package into some directory and symlink to it. So, if some software was expected in the directory /usr/local/foo, you could simply install the first version of the software in the directory /usr/local/foo-1.0 and point to it from the expected directory:
panic# ln -sf /usr/local/foo-1.0 /usr/local/foo
If later you want to install a second version of the software, install it into the directory /usr/local/foo-2.0 and change the symbolic link to this new directory:
panic# ln -sf /usr/local/foo-2.0 /usr/local/foo
Now if something goes wrong, you can always switch back with:
panic# ln -sf /usr/local/foo-1.0 /usr/local/foo
In reality, things aren’t as simple as in this example. It works if you can place all the software components under a single directory, as with the default Apache installation. Everything is installed under a single directory, so you can have:
/usr/local/apache-1.3.17 /usr/local/apache-1.3.19
and use the symlink /usr/local/apache to switch between the two versions.
However, if you use a default installation of Perl, files are spread across multiple directories. In this case, it’s not easy to use symlinks—you need several of them, and they’re hard to keep track of. Unless you automate the symlinks with a script, it might take a while to do a switch, which might mean some downtime. Of course, you can install all the Perl components under a single root, just like the default Apache installation, which simplifies things.
Another complication with upgrading Perl is that you may need to
recompile mod_perl and other Perl third-party modules that use
XS extensions. Therefore, you probably want to
build everything on some other machine, test it, and when ready, just
untar everything at once on the production
machine and adjust the symbolic links.
Although new versions of mod_perl and Apache may not be released for months at a time and the need to upgrade them may not be pressing, the handlers and scripts being used at a site may need regular tweaks and changes, and new ones may be added quite frequently.
Of course, the safest and best option is to prepare an RPM (or equivalent) package that can be used to automatically upgrade the system, as explained in the previous section. Once an RPM specification file has been written (a task that might take some effort), future upgrades will be much less time consuming and have the advantage of being very easy to roll back.
But if the policy is to just overwrite files by hand, this section will explain how to do so as safely as possible.
All code should be thoroughly tested on a development machine before it is put on the live server, and both machines must have an identical software base (i.e., the same versions of the operating system, Apache, any software that Apache and mod_perl depend on, mod_perl itself, and all Perl modules). If the versions do not match, code that works perfectly on the development machine might not work on the live server.
For example, we have encountered a problem when the live and development servers were using different versions of the MySQL database server. The new code took advantage of new features added in the version installed on the development machine. The code was tested and shown to work correctly on the development machine, and when it was copied to the live server it seemed to work fine. Only by chance did we discover that scripts did not work correctly when the new features were used.
If the code hadn’t worked at all, the problem would have been obvious and been detected and solved immediately, but the problem was subtle. Only after a thorough analysis did we understand that the problem was that we had an older version of the MySQL server on the live machine. This example reminded us that all modifications on the development machine should be logged and the live server updated with all of the modifications, not just the new version of the Perl code for a project.
We solved this particular problem by immediately reverting to the old code, upgrading the MySQL server on the live machine, and then successfully reapplying the new code.
Now let’s discuss the techniques used to upgrade live server scripts and handlers.
The most common scenario is a live running service that needs to be upgraded with a new version of the code. The new code has been prepared and uploaded to the production server, and the server has been restarted. Unfortunately, the service does not work anymore. What could be worse than that? There is no way back, because the original code has been overwritten with the new but non-working code.
Another scenario is where a whole set of files is being transferred to the live server but some network problem has occurred in the middle, which has slowed things down or totally aborted the transfer. With some of the files old and some new, the service is most likely broken. Since some files were overwritten, you can’t roll back to the previously working version of the service.
No matter what file transfer technique is used, be it FTP, NFS, or anything else, live running code should never be directly overwritten during file transfer. Instead, files should be transferred to a temporary directory on the live machine, ready to be moved when necessary. If the transfer fails, it can then be restarted safely.
Both scenarios can be made safer with two approaches. First, do not overwrite working files. Second, use a revision control system such as CVS so that changes to working code can easily be undone if the working code is accidentally overwritten. Revision control will be covered later in this chapter.
We recommend performing all updates on the live server in the following sequence. Assume for this example that the project’s code directory is /home/httpd/perl/rel. When we’re about to update the files, we create a new directory, /home/httpd/perl/test, into which we copy the new files. Then we do some final sanity checks: check that file permissions are readable and executable for the user the server is running under, and run perl -Tcw on the new modules to make sure there are no syntax errors in them.
To save some typing, we set up some aliases for some of the apachectl commands and for tailing the error_log file:
panic% alias graceful /home/httpd/httpd_perl/bin/apachectl graceful panic% alias restart /home/httpd/httpd_perl/bin/apachectl restart panic% alias start /home/httpd/httpd_perl/bin/apachectl start panic% alias stop /home/httpd/httpd_perl/bin/apachectl stop panic% alias err tail -f /home/httpd/httpd_perl/logs/error_log
Finally, when we think we are ready, we do:
panic% cd /home/httpd/perl panic% mv rel old && mv test rel && stop && sleep 3 && restart && err
Note that all the commands are typed as a single line, joined by
&&, and only at the end should the Enter
key be pressed. The && ensures that if any
command fails, the following commands will not be executed.
The elements of this command line are:
-
mv rel old && Backs up the working directory to old, so none of the original code is deleted or overwritten
-
mv test rel && Puts the new code in place of the original
-
stop && Stops the server
-
sleep 3 && Allows the server a few seconds to shut down (it might need a longer sleep)
-
restart && Restarts the server
-
err tails the error_log file to make sure that everything is OK
If mv is overriden by a global alias mv -i, which requires confirming every action, you will need to call mv -f to override the -i option.
When updating code on a remote machine, it’s a good idea to prepend nohup to the beginning of the command line:
panic% nohup mv rel old && mv test rel && stop && sleep 3 && restart && err
This approach ensures that if the connection is suddenly dropped, the server will not stay down if the last command that executes is stop.
apachectl generates its status messages a little too early. For example, when we execute apachectl stop, a message saying that the server has been stopped is displayed, when in fact the server is still running. Similarly, when we execute apachectl start, a message is displayed saying that the server has been started, while it is possible that it hasn’t yet. In both cases, this happens because these status messages are not generated by Apache itself. Do not rely on them. Rely on the error_log file instead, where the running Apache server indicates its real status.
Also note that we use restart and not just start. This is because of Apache’s potentially long stopping times if it has to run lots of destruction and cleanup code on exit. If start is used and Apache has not yet released the port it is listening to, the start will fail and the error_log will report that the port is in use. For example:
Address already in use: make_sock: could not bind to port 8000
However, if restart is used, apachectl will wait for the server to quit and unbind the port and will then cleanly restart it.
Now, what happens if the new modules are broken and the newly restarted server reports problems or refuses to start at all?
The aliased err command executes tail -f on the error_log, so that the failed restart or any other problems will be immediately apparent. The situation can quickly and easily be rectified by returning the system to its pre-upgrade state with this command:
panic% mv rel bad && mv old rel && stop && sleep 3 && restart && err
This command line moves the new code to the directory bad, moves the original code back into the runtime directory rel, then stops and restarts the server. Once the server is back up and running, you can analyze the cause of the problem, fix it, and repeat the upgrade again. Usually everything will be fine if the code has been extensively tested on the development server. When upgrades go smoothly, the downtime should be only about 5-10 seconds, and most users will not even notice anything has happened.
The Concurrent Versions System (CVS) is an open source version-control system that allows multiple developers to work on code or configuration in a central repository while tracking any changes made. We use it because it’s the dominant open source tool, but it’s not the only possibility: commercial tools such as Perforce would also work for these purposes.
If you aren’t familiar with CVS, you can learn about it from the resources provided at the end of this chapter. CVS is too broad a topic to be covered in this book. Instead, we will concentrate on the CVS techniques that are relevant to our purpose.
Things are much simpler when using CVS for server updates, especially since it allows you to tag each production release. By tagging files, we mean having a group of files under CVS control share a common label. Like RCS and other revision-control systems, CVS gives each file its own version number, which allows us to manipulate different versions of this file. But if we want to operate on a group of many files, chances are that they will have different version numbers. Suppose we want to take snapshots of the whole project so we can refer to these snapshots some time in the future, after the files have been modified and their respective version numbers have been changed. We can do this using tags.
To tag the project whose module name is myproject, execute the following from any directory on any machine:
panic% cvs -rtag PRODUCTION_1_20 myproject
Now when the time comes to update the online version, go to the directory on the live machine that needs to be updated and execute:
panic% cvs update -dP -r PRODUCTION_1_20
The -P option to cvs prunes
empty directories and deleted files, the -d
option brings in new directories and files (like cvs
checkout does), and -r
PRODUCTION_1_20 tells CVS to update the current directory
recursively to the PRODUCTION_1_20 CVS version of
the project.
Suppose that after a while, we have more code updated and we need to
make a new release. The currently running version has the tag
PRODUCTION_1_20, and the new version has the tag
PRODUCTION_1_21. First we tag the files in the
current state with a new tag:
panic% cvs -rtag PRODUCTION_1_21 myproject
and update the live machine:
panic% cvs update -dP -r PRODUCTION_1_21
Now if there is a problem, we can go back to the previous working
version very easily. If we want to get back to version
PRODUCTION_1_20, we can run the command:
panic% cvs update -dP -r PRODUCTION_1_20
As before, the update brings in new files and directories not already present in the local directory (because of the -dP options).
Remember that when you use CVS to update the live server, you should avoid making any minor changes to the code on this server. That’s because of potential collisions that might happen during the CVS update. If you modify a single line in a single file and then do cvs update, and someone else modifies the same line at the same time and commits it just before you do, CVS will try to merge the changes. If they are different, it will see a conflict and insert both versions into the file. CVS leaves it to you to resolve the conflict. If this file is Perl code, it won’t compile and it will cause temporal troubles until the conflict is resolved. Therefore, the best approach is to think of live server files as being read-only.
Updating the live code directory should be done only if the update is atomic—i.e., if all files are updated in a very short period of time, and when no network problems can occur that might delay the completion of the file update.
The safest approach is to use CVS in conjunction with the safe code update technique presented previously, by working with CVS in a separate directory. When all files are extracted, move them to the directory the live server uses. Better yet, use symbolic links, as described earlier in this chapter: when you update the code, prepare everything in a new directory and, when you’re ready, just change the symlink to point to this new directory. This approach will prevent cases where only a partial update happens because of a network or other problem.
The use of CVS needn’t apply exclusively to code. It can be of great benefit for configuration management, too. Just as you want your mod_perl programs to be identical between the development and production servers, you probably also want to keep your httpd.conf files in sync. CVS is well suited for this task too, and the same methods apply.
Perl programs running on the mod_perl server may be dependent on resources that can become temporarily unavailable when they are being upgraded or maintained. For example, once in a while a database server (and possibly its corresponding DBD module) may need to be upgraded, rendering it unusable for a short period of time.
Using the development server as a temporary replacement is probably the best way to continue to provide service during the upgrade. But if you can’t, the service will be unavailable for a while.
Since none of the code that relies on the temporarily unavailable resource will work, users trying to access the mod_perl server will see either the ugly gray “An Error has occurred” message or a customized error message (if code has been added to trap errors and customize the error-reporting facility). In any case, it’s not a good idea to let users see these errors, as they will make your web site seem amateurish.
A friendlier approach is to confess to the users that some maintenance work is being undertaken and plead for patience, promising that the service will become fully functional in a few minutes (or however long the scheduled downtime is expected to be).
It is a good idea to be honest and report the real duration of the maintenance operation, not just “we will be back in 10 minutes.” Think of a user (or journalist) coming back 20 minutes later and still seeing the same message! Make sure that if the time of resumption of service is given, it is not the system’s local time, since users will be visiting the site from different time zones. Instead, we suggest using Greenwich Mean Time (GMT). Most users have some idea of the time difference between their location and GMT, or can find out easily enough. Although GMT is known by programmers as Universal Coordinated Time (UTC), end users may not know what UTC is, so using the older acronym is probably best.
If just a few
scripts
need to be disabled temporarily, and if they are running under the
Apache::Registry handler, a maintenance message
can be displayed without messing with the server. Prepare a little
script in /home/httpd/perl/down4maintenance.pl:
#!/usr/bin/perl -Tw
use strict;
print "Content-type: text/plain\n\n",
qq{We regret that the service is temporarily
unavailable while essential maintenance is undertaken.
It is expected to be back online from 12:20 GMT.
Please, bear with us. Thank you!};Let’s say you now want to disable the /home/httpd/perl/chat.pl script. Just do this:
panic% mv /home/httpd/perl/chat.pl /home/httpd/perl/chat.pl.orig panic% ln -s /home/httpd/perl/down4maintenance.pl /home/httpd/perl/chat.pl
Of course, the server configuration must allow symbolic links for this trick to work. Make sure that the directive:
Options FollowSymLinks
is in the <Location> or
<Directory> section of
httpd.conf.
Alternatively, you can just back up the real script and then copy the file over it:
panic% cp /home/httpd/perl/chat.pl /home/httpd/perl/chat.pl.orig panic% cp /home/httpd/perl/down4maintenance.pl /home/httpd/perl/chat.pl
Once the maintenance work has been completed, restoring the previous setup is easy. Simply overwrite the symbolic link or the file:
panic% mv /home/httpd/perl/chat.pl.orig /home/httpd/perl/chat.pl
Now make sure that the script has the current timestamp:
panic% touch /home/httpd/perl/chat.pl
Apache::Registry will automatically detect the
change and use the new script from now on.
This scenario is possible because Apache::Registry
checks the modification time of the script before each invocation. If
the script’s file is more recent than the version
already loaded in memory, Apache::Registry reloads
the script from disk.
Under
non-Apache::Registry
handlers, you need to modify the configuration. You must either point
all requests to a new location or replace the handler with one that
will serve the requests during the maintenance period.
Example 5-2 illustrates a maintenance handler.
Example 5-2. Book/Maintenance.pm
package Book::Maintenance;
use strict;
use Apache::Constants qw(:common);
sub handler {
my $r = shift;
$r->send_http_header("text/plain");
print qq{We regret that the service is temporarily
unavailable while essential maintenance is undertaken.
It is expected to be back online from 12:20 GMT.
Please be patient. Thank you!};
return OK;
}
1;In practice, the maintenance script may well read the
“back online” time from a variable
set with a PerlSetVar directive in
httpd.conf, so the script itself need never be
changed.
Edit httpd.conf and change the handler line from:
<Location /perl>
SetHandler perl-script
PerlHandler Book::Handler
...
</Location>to:
<Location /perl>
SetHandler perl-script
#PerlHandler Book::Handler
PerlHandler Book::Maintenance
...
</Location>Now restart the server. Users will be happy to read their email for 10 minutes, knowing that they will return to a much improved service.
Many sites use a more complicated setup in which a “light” Apache frontend server serves static content but proxies all requests for dynamic content to the “heavy” mod_perl backend server (see Chapter 12). Those sites can use a third solution to temporarily disable scripts.
Since the frontend machine rewrites all incoming requests to
appropriate requests for the backend machine, a change to the
RewriteRule is sufficient to take handlers out of
service. Just change the directives to rewrite all incoming requests
(or a subgroup of them) to a single URI. This URI simply tells users
that the service is not available during the maintenance period.
For example, the following
RewriteRule
rewrites all URIs starting with /perl to the
maintenance URI /control/maintain on the
mod_perl server:
RewriteRule ^/perl/(.*)$ http://localhost:8000/control/maintain [P,L]
The Book::Maintenance handler from the previous
section can be used to generate the response to the URI
/control/maintain.
Make sure that this rule is placed before all the other
RewriteRules so that none of the other rules need
to be commented out. Once the change has been made, check that the
configuration is not broken and restart the server so that the new
configuration takes effect. Now the database server can be shut down,
the upgrade can be performed, and the database server can be
restarted. The RewriteRule that has just been
added can be commented out and the Apache server stopped and
restarted. If the changes lead to any problems, the maintenance
RewriteRule can simply be uncommented while you
sort them out.
Of course, all this is error-prone, especially when the maintenance
is urgent. Therefore, it can be a good idea to prepare all the
required configurations ahead of time, by having different
configuration sections and enabling the right one with the help of
the IfDefine directive during server startup.
The following configuration will make this approach clear:
RewriteEngine On
<IfDefine maintain>
RewriteRule /perl/ http://localhost:8000/control/maintain [P,L]
</IfDefine>
<IfDefine !maintain>
RewriteRule ^/perl/(.*)$ http://localhost:8000/$1 [P,L]
# more directives
</IfDefine>Now enable the maintenance section by starting the server with:
panic% httpd -Dmaintain
Request URIs starting with /perl/ will be processed by the /control/maintain handler or script on the mod_perl side.
If the -Dmaintain option is not passed, the “normal” configuration will take effect and each URI will be remapped to the mod_perl server as usual.
Of course, if apachectl or any other script is used for server control, this new mode should be added so that it will be easy to make the correct change without making any mistakes. When you’re in a rush, the less typing you have to do, the better. Ideally, all you’d have to type is:
panic% apachectl maintain
Which will both shut down the server (if it is running) and start it with the -Dmaintain option. Alternatively, you could use:
panic% apachectl start_maintain
to start the server in maintenance mode. apachectl graceful will stop the server and restart it in normal mode.
If
maintenance
tasks can be scheduled when no one is using the server, you can write
a simple
PerlAccessHandler
that will
automatically disable the server and return a page stating that the
server is under maintenance and will be back online at a specified
time. When using this approach, you don’t need to
worry about fiddling with the server configuration when the
maintenance hour comes. However, all maintenance must be completed
within the given time frame, because once the time is up, the service
will resume.
The Apache::DayLimit module from http://www.modperl.com/ is a good example of
such a module. It provides options for specifying which day server
maintenance occurs. For example, if Sundays are used for maintenance,
the configuration for Apache::DayLimit is as
follows:
<Location /perl>
PerlSetVar ReqDay Sunday
PerlAccessHandler Apache::DayLimit
</Location>It is very easy to adapt this module to do more advanced filtering. For example, to specify both a day and a time, use a configuration similar to this:
<Location /perl>
PerlSetVar ReqDay Sunday
PerlSetVar StartHour 09:00
PerlSetVar EndHour 11:00
PerlAccessHandler Apache::DayTimeLimit
</Location>To facilitate transfer from the development server to the production server, the code should be free of any server-dependent variables. This will ensure that modules and scripts can be moved from one directory on the development machine to another directory (possibly in a different path) on the production machine without problems.
If two simple rules are followed, server dependencies can be safely isolated and, as far as the code goes, effectively ignored. First, never use the server name (since development and production machines have different names), and second, never use explicit base directory names in the code. Of course, the code will often need to refer to the server name and directory names, but we can centralize them in server-wide configuration files (as seen in a moment).
By trial and error, we have found that a three-tier (development, staging, and production) scheme works best:
- Development
The development tier might include a single machine or several machines (for example, if there are many developers and each one prefers to develop on his own machine).
- Staging
The staging tier is generally a single machine that is basically identical to the production machine and serves as a backup machine in case the production machine needs an immediate replacement (for example, during maintenance). This is the last station where the code is staged before it is uploaded to the production machine.
The staging machine does not have to be anywhere near as powerful as the production server if finances are stretched. The staging machine is generally used only for staging; it does not require much processor power or memory since only a few developers are likely to use it simultaneously. Even if several developers are using it at the same time, the load will be very low, unless of course benchmarks are being run on it along with programs that create a load similar to that on the production server (in which case the staging machine should have hardware identical to that of the production machine).
- Production
The production tier might include a single machine or a huge cluster comprising many machines.
You can also have the staging and production servers running on the same machine. This is not ideal, especially if the production server needs every megabyte of memory and every CPU cycle so that it can cope with high request rates. But when a dedicated machine just for staging purposes is prohibitively expensive, using the production server for staging purposes is better than having no staging area at all.
Another possibility is to have the staging environment on the development machine.
So how does this three-tier scheme work?
Developers write the code on their machines (development tier) and test that it works as expected. These machines should be set up with an environment as similar to the production server as possible. A manageable and simple approach is to have each developer running his own local Apache server on his own machine. If the code relies on a database, the ideal scenario is for each developer to have access to a development database account and server, possibly even on their own machines.
The pre-release manager installs the code on the staging tier machine and stages it. Whereas developers can change their own httpd.conf files on their own machines, the pre-release manager will make the necessary changes on the staging machine in accordance with the instructions provided by the developers.
The release manager installs the code on the production tier machine(s), tests it, and monitors for a while to ensure that things work as expected.
Of course, on some projects, the developers, the pre-release managers, and the release managers can actually be the same person. On larger projects, where different people perform these roles and many machines are involved, preparing upgrade packages with a packaging tool such as RPM becomes even more important, since it makes it far easier to keep every machine’s configuration and software in sync.
Now that we have described the theory behind the three-tier approach, let us see how to have all the code independent of the machine and base directory names.
Although the example shown below is simple, the real configuration may be far more complex; however, the principles apply regardless of complexity, and it is straightforward to build a simple initial configuration into a configuration that is sufficient for more complex environments.
Basically, what we need is the name of the machine, the port on which the server is running (assuming that the port number is not hidden with the help of a proxy server), the root directory of the web server-specific files, the base directories of static objects and Perl scripts, the appropriate relative and full URIs for these base directories, and a support email address. This amounts to 10 variables.
We prepare a minimum of three Local::Config
packages, one per tier, each suited to a particular
tier’s environment. As mentioned earlier, there can
be more than one machine per tier and even more than one web server
running on the same machine. In those cases, each web server will
have its own Local::Config package. The total
number of Local::Config packages will be equal to
the number of web servers.
For example, for the development tier, the configuration package might look like Example 5-3.
Example 5-3. Local/Config.pm
package Local::Config; use strict; use constant SERVER_NAME => 'dev.example.com'; use constant SERVER_PORT => 8000; use constant ROOT_DIR => '/home/userfoo/www'; use constant CGI_BASE_DIR => '/home/userfoo/www/perl'; use constant DOC_BASE_DIR => '/home/userfoo/www/docs'; use constant CGI_BASE_URI => 'http://dev.example.com:8000/perl'; use constant DOC_BASE_URI => 'http://dev.example.com:8000'; use constant CGI_RELATIVE_URI => '/perl'; use constant DOC_RELATIVE_URI => ''; use constant SUPPORT_EMAIL => 'stas@example.com'; 1;
The constants have uppercase names, in accordance with Perl convention.
The configuration shows that the name of the development machine is dev.example.com, listening to port 8000. Web server-specific files reside under the /home/userfoo/www directory. Think of this as a directory www that resides under user userfoo’s home directory, /home/userfoo. A developer whose username is userbar might use /home/userbar/www as the development root directory.
If there is another web server running on the same machine, create
another Local::Config with a different port number
and probably a different root directory.
To avoid duplication of identical parts of the configuration, the package can be rewritten as shown in Example 5-4.
Example 5-4. Local/Config.pm
package Local::Config;
use strict;
use constant DOMAIN_NAME => 'example.com';
use constant SERVER_NAME => 'dev.' . DOMAIN_NAME;
use constant SERVER_PORT => 8000;
use constant ROOT_DIR => '/home/userfoo/www';
use constant CGI_BASE_DIR => ROOT_DIR . '/perl';
use constant DOC_BASE_DIR => ROOT_DIR . '/docs';
use constant CGI_BASE_URI => 'http://' . SERVER_NAME . ':' . SERVER_PORT
. '/perl';
use constant DOC_BASE_URI => 'http://' . SERVER_NAME . ':' . SERVER_PORT;
use constant CGI_RELATIVE_URI => '/perl';
use constant DOC_RELATIVE_URI => '';
use constant SUPPORT_EMAIL => 'stas@' . DOMAIN_NAME;
1;Reusing constants that were previously defined reduces the risk of
making a mistake. In the original file, several lines need to be
edited if the server name is changed, but in this new version only
one line requires editing, eliminating the risk of your forgetting to
change a line further down the file. All the use
constant statements are executed at compile time, in the
order in which they are specified. The constant
pragma ensures that any attempt to change these variables in the code
leads to an error, so they can be relied on to be correct. (Note that
in certain contexts—e.g., when they’re used as
hash keys—Perl can misunderstand the use of constants. The
solution is to either prepend & or append
( ), so ROOT_DIR would become
either &ROOT_DIR or ROOT_DIR(
).)
Now, when the code needs to access the server’s global configuration, it needs to refer only to the variables in this module. For example, in an application’s configuration file, you can create a dynamically generated configuration, which will change from machine to machine without your needing to touch any code (see Example 5-5).
Example 5-5. App/Foo/Config.pm
package App::Foo::Config; use Local::Config ( ); use strict; use vars qw($CGI_URI $CGI_DIR); # directories and URIs of the App::Foo CGI project $CGI_URI = $Local::Config::CGI_BASE_URI . '/App/Foo'; $CGI_DIR = $Local::Config::CGI_BASE_DIR . '/App/Foo'; 1;
Notice that we used fully qualified variable names instead of
importing these global configuration variables into the
caller’s namespace. This saves a few bytes of
memory, and since Local::Config will be loaded by
many modules, these savings will quickly add up. Programmers used to
programming Perl outside the mod_perl environment might be tempted to
add Perl’s exporting mechanism to
Local::Config and thereby save themselves some
typing. We prefer not to use Exporter.pm under
mod_perl, because we want to save as much memory as possible. (Even
though the amount of memory overhead for using an exported name is
small, this must be multiplied by the number of concurrent users of
the code, which could be hundreds or even thousands on a busy site
and could turn a small memory overhead into a large one.)
For the
staging tier, a similar
Local::Config module with just a few changes (as
shown in Example 5-6) is necessary.
Example 5-6. Local/Config.pm
package Local::Config;
use strict;
use constant DOMAIN_NAME => 'example.com';
use constant SERVER_NAME => 'stage.' . DOMAIN_NAME;
use constant SERVER_PORT => 8000;
use constant ROOT_DIR => '/home';
use constant CGI_BASE_DIR => ROOT_DIR . '/perl';
use constant DOC_BASE_DIR => ROOT_DIR . '/docs';
use constant CGI_BASE_URI => 'http://' . SERVER_NAME . ':' . SERVER_PORT
. '/perl';
use constant DOC_BASE_URI => 'http://' . SERVER_NAME . ':' . SERVER_PORT;
use constant CGI_RELATIVE_URI => '/perl';
use constant DOC_RELATIVE_URI => '';
use constant SUPPORT_EMAIL => 'stage@' . DOMAIN_NAME;
1;We have named our staging tier machine stage.example.com. Its root directory is /home.
The production
tier version of
Local/Config.pm is shown in Example 5-7.
Example 5-7. Local/Config.pm
package Local::Config;
use strict;
use constant DOMAIN_NAME => 'example.com';
use constant SERVER_NAME => 'www.' . DOMAIN_NAME;
use constant SERVER_PORT => 8000;
use constant ROOT_DIR => '/home/';
use constant CGI_BASE_DIR => ROOT_DIR . '/perl';
use constant DOC_BASE_DIR => ROOT_DIR . '/docs';
use constant CGI_BASE_URI => 'http://' . SERVER_NAME . ':' . SERVER_PORT
. '/perl';
use constant DOC_BASE_URI => 'http://' . SERVER_NAME . ':' . SERVER_PORT;
use constant CGI_RELATIVE_URI => '/perl';
use constant DOC_RELATIVE_URI => '';
use constant SUPPORT_EMAIL => 'support@' . DOMAIN_NAME;You can see that the setups of the staging and production machines are almost identical. This is only in our example; in reality, they can be very different.
The most important point is that the Local::Config
module from a machine on one tier must never be
moved to a machine on another tier, since it will break the code. If
locally built packages are used, the Local::Config
file can simply be excluded—this will help to reduce the risk
of inadvertently copying it.
From now on, when modules and scripts are moved between machines, you
shouldn’t need to worry about having to change
variables to accomodate the different machines’
server names and directory layouts. All this is accounted for by the
Local::Config files.
Some developers prefer to run conversion scripts on the moved code that adjust all variables to the local machine. This approach is error-prone, since variables can be written in different ways, and it may result in incomplete adjustment and broken code. Therefore, the conversion approach is not recommended.
When just
one developer is
working on a specific server, there are fewer problems, because she
can have complete control over the server. However, often a group of
developers need to develop mod_perl scripts and modules concurrently
on the same machine. Each developer wants to have control over the
server: to be able to stop it, run it in single-server mode, restart
it, etc. They also want control over the location of log files,
configuration settings such as MaxClients, and so
on.
Each developer might have her own desktop machine, but all development and staging might be done on a single central development machine (e.g., if the developers’ personal desktop machines run a different operating system from the one running on the development and production machines).
One workaround for this problem involves having a few versions of the
httpd.conf file (each having different
Port, ErrorLog, etc.
directives) and forcing each developer’s server to
be started with:
panic% httpd_perl -f /path/to/httpd.conf
However, this means that these files must be kept synchronized when there are global changes affecting all developers. This can be quite difficult to manage. The solution we use is to have a single httpd.conf file and use the -Dparameter server startup option to enable a specific section of httpd.conf for each developer. Each developer starts the server with his or her username as an argument. As a result, a server uses both the global settings and the developer’s private settings.
For example, user stas would start his server with:
panic% httpd_perl -Dstas
In httpd.conf, we write:
# Personal development server for stas
# stas uses the server running on port 8000
<IfDefine stas>
Port 8000
PidFile /home/httpd/httpd_perl/logs/httpd.pid.stas
ErrorLog /home/httpd/httpd_perl/logs/error_log.stas
Timeout 300
KeepAlive On
MinSpareServers 2
MaxSpareServers 2
StartServers 1
MaxClients 3
MaxRequestsPerChild 15
# let developers to add their own configuration
# so they can override the defaults
Include /home/httpd/httpd_perl/conf/stas.conf
</IfDefine>
# Personal development server for eric
# eric uses the server running on port 8001
<IfDefine eric>
Port 8001
PidFile /home/httpd/httpd_perl/logs/httpd.pid.eric
ErrorLog /home/httpd/httpd_perl/logs/error_log.eric
Timeout 300
KeepAlive Off
MinSpareServers 1
MaxSpareServers 2
StartServers 1
MaxClients 5
MaxRequestsPerChild 0
Include /home/httpd/httpd_perl/conf/eric.conf
</IfDefine>With this technique, we have separate error_log files and full control over server starting and stopping, the number of child processes, and port selection for each server. This saves Eric from having to call Stas several times a day just to warn, “Stas, I’m restarting the server” (a ritual in development shops where all developers are using the same mod_perl server).
With this technique, developers will need to learn the PIDs of their parent httpd_perl processes. For user stas, this can be found in /home/httpd/httpd_perl/logs/httpd.pid.stas. To make things even easier, we change the apachectl script to do the work for us. We make a copy for each developer, called apachectl.username, and change two lines in each script:
PIDFILE=/home/httpd/httpd_perl/logs/httpd.pid.username HTTPD='/home/httpd/httpd_perl/bin/httpd_perl -Dusername'
For user stas, we prepare a startup script called apachectl.stas and change these two lines in the standard apachectl script:
PIDFILE=/home/httpd/httpd_perl/logs/httpd.pid.stas HTTPD='/home/httpd/httpd_perl/bin/httpd_perl -Dstas'
Now when user stas wants to stop the server, he executes:
panic% apachectl.stas stop
And to start the server, he executes:
panic% apachectl.stas start
And so on, for all other apachectl arguments.
It might seem that we could have used just one apachectl and have it determine for itself who executed it by checking the UID. But the setuid bit must be enabled on this script, because starting the server requires root privileges. With the setuid bit set, a single apachectl script can be used for all developers, but it will have to be modified to include code to read the UID of the user executing it and to use this value when setting developer-specific paths and variables.
The last thing you need to do is to provide developers with an option to run in single-process mode. For example:
panic% /home/httpd/httpd_perl/bin/httpd_perl -Dstas -X
In addition to making the development process easier, we decided to use relative links in all static documents, including calls to dynamically generated documents. Since each developer’s server is running on a different port, we have to make it possible for these relative links to reach the correct port number.
When typing the URI by hand, it’s easy. For example, when user stas, whose server is running on port 8000, wants to access the relative URI /test/example, he types http://www.example.com:8000/test/example to get the generated HTML page. Now if this document includes a link to the relative URI /test/example2 and stas clicks on it, the browser will automatically generate a full request (http://www.example.com:8000/test/example2) by reusing the server:port combination from the previous request.
Note that all static objects will be served from the same server as well. This may be an acceptable situation for the development environment, but if it is not, a slightly more complicated solution involving the mod_rewrite Apache module will have to be devised.
To use mod_rewrite, we have to configure our httpd_docs (light) server with —enable-module=rewrite and recompile, or use DSOs and load and enable the module in httpd.conf. In the httpd.conf file of our httpd_docs server, we have the following code:
RewriteEngine on
# stas's server
# port = 8000
RewriteCond %{REQUEST_URI} ^/perl
RewriteCond %{REMOTE_ADDR} 123.34.45.56
RewriteRule ^(.*) http://example.com:8000/$1 [P,L]
# eric's server
# port = 8001
RewriteCond %{REQUEST_URI} ^/perl
RewriteCond %{REMOTE_ADDR} 123.34.45.57
RewriteRule ^(.*) http://example.com:8001/$1 [P,L]
# all the rest
RewriteCond %{REQUEST_URI} ^/perl
RewriteRule ^(.*) http://example.com:81/$1 [P]The IP addresses are those of the developer desktop machines (i.e., where they run their web browsers). If an HTML file includes a relative URI such as /perl/test.pl or even http://www.example.com/perl/test.pl, requests for those URIs from user stas’s machine will be internally proxied to http://www.example.com:8000/perl/test.pl, and requests generated from user eric’s machine will be proxied to http://www.example.com:8001/perl/test.pl.
Another possibility is to use the REMOTE_USER
variable. This requires that all developers be authenticated when
they access the server. To do so, change the
RewriteRules to match
REMOTE_USER in the above example.
Remember, the above setup will work only with relative URIs in the HTML code. If the HTML output by the code uses full URIs including a port other than 80, the requests originating from this HTML code will bypass the light server listening on the default port 80 and go directly to the server and port of the full URI.
Once the production system is working, you may think that the job is done and the developers can switch to a new project. Unfortunately, in most cases the server will still need to be maintained to make sure that everything is working as expected, to ensure that the web server is always up, and much more. A large part of this job can be automated, which will save time. It will also increase the uptime of the server, since automated processes generally work faster than manual ones. If created properly, automated processes also will always work correctly, whereas human operators are likely to make occassional mistakes.
When you’re getting started, it usually helps to monitor the server interactively. Many different tools are available to do this. We will discuss a few of them now.
When writing automated monitoring tools, you should start by monitoring the tools themselves until they are reliable and stable enough to be left to work by themselves.
Even when everything is automated, you should check at regular intervals that everything is working OK, since a minor change in a single component can silently break the whole monitoring system. A good example is a silent failure of the mail system—if all alerts from the monitoring tools are delivered through email, having no messages from the system does not necessarily mean that everything is OK. If emails alerting about a problem cannot reach the webmaster because of a broken email system, the webmaster will not realize that a problem exists. (Of course, the mailing system should be monitored as well, but then problems must be reported by means other than email. One common solution is to send messages by both email and to a mobile phone’s short message service.)
Another very important (albeit often-forgotten) risk time is the post-upgrade period. Even after a minor upgrade, the whole service should be monitored closely for a while.
The first and simplest check is to visit a few pages from the service to make sure that things are working. Of course, this might not suffice, since different pages might use different resources—while code that does not use the database system might work properly, code that does use it might not work if the database server is down.
The second thing to check is the web server’s error_log file. If there are any problems, they will probably be reported here. However, only obvious syntactic or malfunction bugs will appear here—the subtle bugs that are a result of bad program logic will be revealed only through careful testing (which should have been completed before upgrading the live server).
Periodic system health checking can be done using the top utility, which shows free memory and swap space, the machine’s CPU load, etc.
The Apache::VMonitor
module provides even better
monitoring functionality than top. It supplies
all the relevant information that top does, plus
all the Apache-specific information provided by
Apache’s mod_status module (request processing time,
last request’s URI, number of requests served by
each child, etc.) In addition, Apache::VMonitor
emulates the reporting functions of the top,
mount, and df utilities.
Apache::VMonitor has a special mode for mod_perl
processes. It also has visual alerting capabilities and a
configurable “automatic refresh”
mode. A web interface can be used to show or hide all sections
dynamically.
The module provides two main viewing modes:
Multi-processes and overall system status
Single-process extensive reporting
To run Apache::VMonitor,
you need to have
Apache::Scoreboard
installed and configured in
httpd.conf.
Apache::Scoreboard, in turn, requires
mod_status to be installed
with ExtendedStatus enabled. In
httpd.conf, add:
ExtendedStatus On
Turning on extended mode will add a certain overhead to each request’s response time. If every millisecond counts, you may not want to use it in production.
You also need Time::HiRes and
GTop to be installed. And, of course, you need a
running mod_perl-enabled Apache server.
To enable Apache::VMonitor, add the following
configuration to httpd.conf:
<Location /system/vmonitor>
SetHandler perl-script
PerlHandler Apache::VMonitor
</Location>The monitor will be displayed when you request http://localhost/system/vmonitor/.
You probably want to protect this location from unwanted visitors. If you are accessing this location from the same IP address, you can use a simple host-based authentication:
<Location /system/vmonitor>
SetHandler perl-script
PerlHandler Apache::VMonitor
order deny,allow
deny from all
allow from 132.123.123.3
</Location>Alternatively, you may use Basic or other authentication schemes provided by Apache and its extensions.
You should load the module in httpd.conf:
PerlModule Apache::VMonitor
or from the the startup file:
use Apache::VMonitor( );
You can control the behavior of Apache::VMonitor
by configuring variables in the startup file or inside the
<Perl> section. To alter the monitor
reporting behavior, tweak the following configuration arguments from
within the startup file:
$Apache::VMonitor::Config{BLINKING} = 1;
$Apache::VMonitor::Config{REFRESH} = 0;
$Apache::VMonitor::Config{VERBOSE} = 0;To control what sections are to be displayed when the tool is first accessed, configure the following variables:
$Apache::VMonitor::Config{SYSTEM} = 1;
$Apache::VMonitor::Config{APACHE} = 1;
$Apache::VMonitor::Config{PROCS} = 1;
$Apache::VMonitor::Config{MOUNT} = 1;
$Apache::VMonitor::Config{FS_USAGE} = 1;You can control the sorting of the mod_perl processes report by
sorting them by one of the following columns: pid,
mode, elapsed,
lastreq, served,
size, share,
vsize, rss,
client, or request. For
example, to sort by the process size, use the following setting:
$Apache::VMonitor::Config{SORT_BY} = "size";As the application provides an option to monitor processes other than mod_perl processes, you can define a regular expression to match the relevant processes. For example, to match the process names that include “httpd_docs”, “mysql”, and “squid”, the following regular expression could be used:
$Apache::VMonitor::PROC_REGEX = 'httpd_docs|mysql|squid';
We will discuss all these configuration options and their influence on the application shortly.
The first mode is the one that’s used most often, since it allows you to monitor almost all important system resources from one location. For your convenience, you can turn different sections on and off on the report, to make it possible for reports to fit into one screen.
This mode comes with the following features:
- Automatic Refresh Mode
You can tell the application to refresh the report every few seconds. You can preset this value at server startup. For example, to set the refresh to 60 seconds, add the following configuration setting:
$Apache::VMonitor::Config{REFRESH} = 60;A
0(zero) value turns off automatic refresh.When the server is started, you can always adjust the refresh rate through the user interface.
- top Emulation: System Health Report
Like top, this shows the current date/time, machine uptime, average load, and all the system CPU and memory usage levels (CPU load, real memory, and swap partition usage).
The top section includes a swap space usage visual alert capability. As we will explain later in this chapter, swapping is very undesirable on production systems. This tool helps to detect abnormal swapping situations by changing the swap report row’s color according to the following rules:
swap usage report color --------------------------------------------------------- 5Mb < swap < 10 MB light red 20% < swap (swapping is bad!) red 70% < swap (almost all used!) red + blinking (if enabled)
Note that you can turn on the blinking mode with:
$Apache::VMonitor::Config{BLINKING} = 1;The module doesn’t alert when swap is being used just a little (< 5 Mb), since swapping is common on many Unix systems, even when there is plenty of free RAM.
If you don’t want the system section to be displayed, set:
$Apache::VMonitor::Config{SYSTEM} = 0;The default is to display this section.
- top Emulation: Apache/mod_perl Processes Status
Like top, this emulation gives a report of the processes, but it shows only information relevant to mod_perl processes. The report includes the status of the process (Starting, Reading, Sending, Waiting, etc.), process ID, time since the current request was started, last request processing time, size, and shared, virtual, and resident size. It shows the last client’s IP address and the first 64 characters of the request URI.
This report can be sorted by any column by clicking on the name of the column while running the application. The sorting can also be preset with the following setting:
$Apache::VMonitor::Config{SORT_BY} = "size";The valid choices are
pid,mode,elapsed,lastreq,served,size,share,vsize,rss,client, andrequest.The section is concluded with a report about the total memory being used by all mod_perl processes as reported by the kernel, plus an extra number approximating the real memory usage when memory sharing is taking place. We discuss this in more detail in Chapter 10.
If you don’t want the mod_perl processes section to be displayed, set:
$Apache::VMonitor::Config{APACHE} = 0;The default is to display this section.
- top Emulation: Any Processes
This section, just like the mod_perl processes section, displays the information as the top program would. To enable this section, set:
$Apache::VMonitor::Config{PROCS} = 1;The default is not to display this section.
You need to specify which processes are to be monitored. The regular expression that will match the desired processes is required for this section to work. For example, if you want to see all the processes whose names include any of the strings “http”, “mysql”, or “squid”, use the following regular expression:
$Apache::VMonitor::PROC_REGEX = 'httpd|mysql|squid';
Figure 5-1 visualizes the sections that have been discussed so far. As you can see, the swap memory is heavily used. Although you can’t see it here, the swap memory report is colored red.
- mount Emulation
This section provides information about mounted filesystems, as if you had called mount with no parameters.
If you want the mount section to be displayed, set:
$Apache::VMonitor::Config{MOUNT} = 1;The default is not to display this section.
- df Emulation
This section completely reproduces the df utility. For each mounted filesystem, it reports the number of total and available blocks for both superuser and user, and usage in percentages. In addition, it reports available and used file inodes in numbers and percentages.
This section can give you a visual alert when a filesystem becomes more than 90% full or when there are less than 10% of free file inodes left. The relevant filesystem row will be displayed in red and in a bold font. A mount point directory will blink if blinking is turned on. You can turn the blinking on with:
$Apache::VMonitor::Config{BLINKING} = 1;If you don’t want the df section to be displayed, set:
$Apache::VMonitor::Config{FS_USAGE} = 0;The default is to display this section.
Figure 5-2 presents an example of the report consisting of the last two sections that were discussed (df and mount emulation), plus the ever-important mod_perl processes report.
In Figure 5-2, the /mnt/cdrom and /usr filesystems are more than 90% full and therefore are colored red. This is normal for /mnt/cdrom, which is a mounted CD-ROM, but might be critical for the /usr filesystem, which should be cleaned up or enlarged.
- Abbreviations and hints
The report uses many abbreviations that might be new for you. If you enable the
VERBOSEmode with:$Apache::VMonitor::Config{VERBOSE} = 1;this section will reveal the full names of the abbreviations at the bottom of the report.
If you need to get in-depth information about a single process, just click on its PID. If the chosen process is a mod_perl process, the following information is displayed:
Process type (child or parent), status of the process (Starting, Reading, Sending, Waiting, etc.), and how long the current request has been being processed (or how long the previous request was processed for, if the process is inactive at the moment the report was made).
How many bytes have been transferred so far, and how many requests have been served per child and per slot. (When the child process quits, it is replaced by a new process running in the same slot.)
CPU times used by the process:
total,utime,stime,cutime,cstime.
For all processes (mod_perl and non-mod_perl), the following information is reported:
General process information: UID, GID, state, TTY, and command-line arguments
Memory usage: size, share, VSize, and RSS
Memory segments usage: text, shared lib, date, and stack
Memory maps: start-end, offset, device_major:device_minor, inode, perm, and library path
Sizes of loaded libraries
Just as with the multi-process mode, this mode allows you to automatically refresh the page at the desired intervals.
Figures Figure 5-3, Figure 5-4, and Figure 5-5 show an example report for one mod_perl process.
As we mentioned earlier, the more things are automated, the more stable the server will be. In general, there are three things that we want to ensure:
Apache is up and properly serving requests. Remember that it can be running but unable to serve requests (for example, if there is a stale lock and all processes are waiting to acquire it).
All the resources that mod_perl relies on are available and working. This might include database engines, SMTP services, NIS or LDAP services, etc.
The system is healthy. Make sure that there is no system resource contention, such as a small amount of free RAM, a heavily swapping system, or low disk space.
None of these categories has a higher priority than the others. A system administrator’s role includes the proper functioning of the whole system. Even if the administrator is responsible for just part of the system, she must still ensure that her part does not cause problems for the system as a whole. If any of the above categories is not monitored, the system is not safe.
A specific setup might certainly have additional concerns that are not covered here, but it is most likely that they will fall into one of the above categories.
Before we delve into details, we should mention that all automated tools can be divided into two categories: tools that know how to detect problems and notify the owner, and tools that not only detect problems but also try to solve them, notifying the owner about both the problems and the results of the attempt to solve them.
Automatic tools are generally called watchdogs. They can alert the owner when there is a problem, just as a watchdog will bark when something is wrong. They will also try to solve problems themselves when the owner is not around, just as watchdogs will bite thieves when their owners are asleep.
Although some tools can perform corrective actions when something goes wrong without human intervention (e.g., during the night or on weekends), for some problems it may be that only human intervention can resolve the situation. In such cases, the tool should not attempt to do anything at all. For example, if a hardware failure occurs, it is almost certain that a human will have to intervene.
Below are some techniques and tools that apply to each category.
One simple watchdog solution is to use a slightly modified apachectl script, which we have called apache.watchdog. Call it from cron every 30 minutes—or even every minute—to make sure that the server is always up.
The crontab entry for 30-minute intervals would read:
5,35 * * * * /path/to/the/apache.watchdog >/dev/null 2>&1
The script is shown in Example 5-8.
Example 5-8. apache.watchdog
--------------------
#!/bin/sh
# This script is a watchdog checking whether
# the server is online.
# It tries to restart the server, and if it is
# down it sends an email alert to the admin.
# admin's email
EMAIL=webmaster@example.com
# the path to the PID file
PIDFILE=/home/httpd/httpd_perl/logs/httpd.pid
# the path to the httpd binary, including any options if necessary
HTTPD=/home/httpd/httpd_perl/bin/httpd_perl
# check for pidfile
if [ -f $PIDFILE ] ; then
PID=`cat $PIDFILE`
if kill -0 $PID; then
STATUS="httpd (pid $PID) running"
RUNNING=1
else
STATUS="httpd (pid $PID?) not running"
RUNNING=0
fi
else
STATUS="httpd (no pid file) not running"
RUNNING=0
fi
if [ $RUNNING -eq 0 ]; then
echo "$0 $ARG: httpd not running, trying to start"
if $HTTPD ; then
echo "$0 $ARG: httpd started"
mail $EMAIL -s "$0 $ARG: httpd started" \
< /dev/null > /dev/null 2>&1
else
echo "$0 $ARG: httpd could not be started"
mail $EMAIL -s "$0 $ARG: httpd could not be started" \
< /dev/null > /dev/null 2>&1
fi
fiAnother approach is to use the Perl LWP module to
test the server by trying to fetch a URI served by the server. This
is more practical because although the server may be running as a
process, it may be stuck and not actually serving any
requests—for example, when there is a stale lock that all the
processes are waiting to acquire. Failing to get the document will
trigger a restart, and the problem will probably go away.
We set a cron job to call this LWP script every few minutes to fetch a document generated by a very light script. The best thing, of course, is to call it every minute (the finest resolution cron provides). Why so often? If the server gets confused and starts to fill the disk with lots of error messages written to the error_log, the system could run out of free disk space in just a few minutes, which in turn might bring the whole system to its knees. In these circumstances, it is unlikely that any other child will be able to serve requests, since the system will be too busy writing to the error_log file. Think big—if running a heavy service, adding one more request every minute will have no appreciable impact on the server’s load.
So we end up with a crontab entry like this:
* * * * * /path/to/the/watchdog.pl > /dev/null
The watchdog itself is shown in Example 5-9.
Example 5-9. watchdog.pl
#!/usr/bin/perl -Tw
# These prevent taint checking failures
$ENV{PATH} = '/bin:/usr/bin';
delete @ENV{qw(IFS CDPATH ENV BASH_ENV)};
use strict;
use diagnostics;
use vars qw($VERSION $ua);
$VERSION = '0.01';
require LWP::UserAgent;
###### Config ########
my $test_script_url = 'http://www.example.com:81/perl/test.pl';
my $monitor_email = 'root@localhost';
my $restart_command = '/home/httpd/httpd_perl/bin/apachectl restart';
my $mail_program = '/usr/lib/sendmail -t -n';
######################
$ua = LWP::UserAgent->new;
$ua->agent("$0/watchdog " . $ua->agent);
# Uncomment the following two lines if running behind a firewall
# my $proxy = "http://www-proxy";
# $ua->proxy('http', $proxy) if $proxy;
# If it returns '1' it means that the service is alive, no need to
# continue
exit if checkurl($test_script_url);
# Houston, we have a problem.
# The server seems to be down, try to restart it.
my $status = system $restart_command;
my $message = ($status = = 0)
? "Server was down and successfully restarted!"
: "Server is down. Can't restart.";
my $subject = ($status = = 0)
? "Attention! Webserver restarted"
: "Attention! Webserver is down. can't restart";
# email the monitoring person
my $to = $monitor_email;
my $from = $monitor_email;
send_mail($from, $to, $subject, $message);
# input: URL to check
# output: 1 for success, 0 for failure
#######################
sub checkurl {
my($url) = @_;
# Fetch document
my $res = $ua->request(HTTP::Request->new(GET => $url));
# Check the result status
return 1 if $res->is_success;
# failed
return 0;
}
# send email about the problem
#######################
sub send_mail {
my($from, $to, $subject, $messagebody) = @_;
open MAIL, "|$mail_program"
or die "Can't open a pipe to a $mail_program :$!\n";
print MAIL <<_ _END_OF_MAIL_ _;
To: $to
From: $from
Subject: $subject
$messagebody
--
Your faithful watchdog
_ _END_OF_MAIL_ _
close MAIL or die "failed to close |$mail_program: $!";
}Of course, you may want to replace a call to
sendmail with Mail::Send,
Net::SMTP code, or some other preferred
email-sending
technique.
It is not enough to have your server and service up and running. The server must be maintained and monitored even when everything seems to be fine. This includes security auditing as well as keeping an eye on the amount of remaining unused disk space, available RAM, the system’s load, etc.
If these chores are forgotten, sooner or later the system will crash, either because it has run out of free disk space, all available RAM has been used and the system has started to swap heavily, or it has been broken into. The last issue is much too broad for this book’s scope, but the others are quite easily addressed if you follow our advice.
Particular systems might require maintenance chores that are not covered here, but this section highlights some of the most important general tasks.
Apache generally logs all the web server access events in the access_log file, whereas errors and warnings go into the error_log file. The access_log file can later be analyzed to report server usage statistics, such as the number of requests made in different time spans, who issued these requests, and much more. The error_log file is used to monitor the server for errors and warnings and to prompt actions based on those reports. Some systems do additional logging, such as storing the referrers of incoming requests to find out how users have learned about the site.
The simplest logging technique is to dump the logs into a file opened for appending. With Apache, this is as simple as specifying the logging format and the file to which to log. For example, to log all accesses, use the default directive supplied in httpd.conf:
LogFormat "%h %l %u %t \"%r\" %>s %b" common CustomLog /home/httpd/httpd_perl/logs/access_log common
This setting will log all server accesses to a file named
/home/httpd/httpd_perl/logs/access_log using the
format specified by the LogFormat
directive—in this case, common. Please refer
to the Apache documentation for a complete explanation of the various
tokens that you can use when specifying log formats. If
you’re tempted to change the format of the log file,
bear in mind that some log analysis tools may expect that only the
default or one of a small subset of logging formats is used.
The only risk with log files is their size. It is important to keep log files trimmed. If they are needed for later analysis, they should be rotated and the rotation files should be moved somewhere else so they do not consume disk space. You can usually compress them for storage offline.
The most important thing is to monitor log files for possible sudden explosive growth rates. For example, if a developer makes a mistake in his code running on the mod_perl server and the child processes executing the code start to log thousands of error messages a second, all disk space can quickly be consumed, and the server will cease to function.
The first issue is solved by having a process that rotates the logs run by cron at certain times (usually off-peak hours, if this term is still valid in the 24-hour global Internet era). Usually, log rotation includes renaming the current log file, restarting the server (which creates a fresh new log file), and compressing and/or moving the rotated log file to a different disk.
For example, if we want to rotate the access_log file, we could do:
panic% mv access_log access_log.renamed panic% apachectl graceful panic% sleep 5 panic% mv access_log.renamed /some/directory/on/another/disk
The sleep delay is added to make sure that all children complete requests and logging. It’s possible that a longer delay is needed. Once the restart is completed, it is safe to use access_log.renamed.
There are several popular utilities, such as rotatelogs and cronolog, that can perform the rotation, although it is also easy to create a basic rotation script. Example 5-10 shows a script that we run from cron to rotate our log files.
Example 5-10. logrotate
#!/usr/local/bin/perl -Tw
# This script does log rotation. Called from crontab.
use strict;
$ENV{PATH}='/bin:/usr/bin';
delete @ENV{qw(IFS CDPATH ENV BASH_ENV)};
### configuration
my @logfiles = qw(access_log error_log);
umask 0;
my $server = "httpd_perl";
my $logs_dir = "/home/httpd/$server/logs";
my $restart_command = "/home/httpd/$server/bin/apachectl restart";
my $gzip_exec = "/usr/bin/gzip -9"; # -9 is maximum compression
my ($sec, $min, $hour, $mday, $mon, $year) = localtime(time);
my $time = sprintf "%0.4d.%0.2d.%0.2d-%0.2d.%0.2d.%0.2d",
$year+1900, ++$mon, $mday, $hour, $min, $sec;
chdir $logs_dir;
# rename log files
foreach my $file (@logfiles) {
rename $file, "$file.$time";
}
# now restart the server so the logs will be restarted
system $restart_command;
# allow all children to complete requests and logging
sleep 5;
# compress log files
foreach my $file (@logfiles) {
system "$gzip_exec $file.$time";
}As can be seen from the code, the rotated files will include the date and time in their filenames.
As we mentioned earlier, there are times when the web server goes wild and starts to rapidly log lots of messages to the error_log file. If no one monitors this, it is possible that in a few minutes all free disk space will be consumed and no process will be able to work normally. When this happens, the faulty server process may cause so much I/O that its sibling processes cannot serve requests.
Although this rarely happens, you should try to reduce the risk of it occurring on your server. Run a monitoring program that checks the log file size and, if it detects that the file has grown too large, attempts to restart the server and trim the log file.
Back when we were using quite an old version of mod_perl, we sometimes had bursts of “Callback called exit” errors showing up in our error_log. The file could grow to 300 MB in a few minutes.
Example 5-11 shows a script that should be executed from crontab to handle situations like this. This is an emergency solution, not to be used for routine log rotation. The cron job should run every few minutes or even every minute, because if the site experiences this problem, the log files will grow very rapidly. The example script will rotate when error_log grows over 100K. Note that this script is still useful when the normal scheduled log-rotation facility is working.
Example 5-11. emergency_rotate.sh
#!/bin/sh
S=`perl -e 'print -s "/home/httpd/httpd_perl/logs/error_log"'`;
if [ "$S" -gt 100000 ] ; then
mv /home/httpd/httpd_perl/logs/error_log \
/home/httpd/httpd_perl/logs/error_log.old
/etc/rc.d/init.d/httpd restart
date | /bin/mail -s "error_log $S kB" admin@example.com
fiOf course, a more advanced script could be written using timestamps and other bells and whistles. This example is just a start, to illustrate a basic solution to the problem in question.
Another solution is to use ready-made tools that are written for this
purpose. The daemontools package includes a
utility called multilog that saves the
STDIN stream to one or more log files. It
optionally timestamps each line and, for each log, includes or
excludes lines matching specified patterns. It automatically rotates
logs to limit the amount of disk space used. If the disk fills up, it
pauses and tries again, without losing any data.
The obvious caveat is that it does not restart the server, so while it tries to solve the log file-handling issue, it does not deal with the problem’s real cause. However, because of the heavy I/O induced by the log writing, other server processes will work very slowly if at all. A normal watchdog is still needed to detect this situation and restart the Apache server.
If you are running more than one server on the same machine, Apache offers the choice of either having a separate set of log files for each server, or using a central set of log files for all servers. If you are running servers on more than one machine, having them share a single log file is harder to achieve, but it is possible, provided that a filesharing system is used (logging into a database, or a special purpose application like syslog).
There are a few file-sharing systems that are widely used:
- Network File System (NFS)
NFS is a network file-sharing system. It’s a very useful system, when it works. Unfortunately, it breaks too often, which makes it unreliable to use on production systems. NFS is available on most Unix flavors.
- Andrew File System (AFS)
AFS is a distributed filesystem that enables cooperating hosts (clients and servers) to efficiently share filesystem resources across both local area and wide area networks. This filesystem is reliable, but it costs money and is available only on the HP, Next, DEC, IBM, SUN, and SGI operating systems. For more information, see http://www.transarc.com/ and http://www.angelfire.com/hi/plutonic/afs-faq.html.
- Coda
Coda is a distributed filesystem with its origin in AFS2. It has many features that are very desirable for network filesystems. Coda is platform-independent: you can mix and match servers and clients on any supported platform. As of this writing, it’s not clear how stable the system is; some people have reported success using it, but others have had some problems with it. For more information, see http://www.coda.cs.cmu.edu/.
Apache permits the location of the file used for logging purposes to
be specified, but it also allows you to specify a program to which
all logs should be piped. To log to a program, modify the log handler
directive (for example, CustomLog) to use the
logging program instead of specifying an explicit filename:
LogFormat "%h %l %u %t \"%r\" %>s %b" common CustomLog "| /home/httpd/httpd_perl/bin/sqllogger.pl" common
Logging into a database is a common solution, because you can do insertions from different machines into a single database. Unless the logger is programmed to send logs to a few databases at once, this solution is not reliable, since a single database constitutes a single failure point. If the database goes down, the logs will be lost. Sending information to one target is called unicast (see Figure 5-6), and sending to more than one target is called multicast (see Figure 5-7). In the latter case, if one database goes down, the others will still collect the data.
Another solution is to use a centralized logger program based on syslog(3) to send all logs to a central location on a master host. syslog(3) is not a very scalable solution, because it’s slow. It’s also unreliable—it uses UDP to send the data, which doesn’t ensure that the data will reach its destination. This solution is also unicast: if the master host goes down, the logs will be lost.
One advanced system that provides consolidated logging is
mod_log_spread. Based on the group communications
toolkit Spread, using IP multicast, mod_log_spread
provides reliable, scalable centralized logging whith minimal
performance impact on the web servers. For more
information,
see
http://www.backhand.org/mod_log_spread/.
Before we delve into swapping process details, let’s look briefly at memory components and memory management.
Computer memory is called RAM (Random Access Memory). Reading and writing to RAM is faster than doing the same operations on a hard disk, by around five orders of magnitude (and growing). RAM uses electronic memory cells (transistors) with no moving parts, while hard disks use a rotating magnetic medium. It takes about one tenth of a microsecond to write to RAM but something like ten thousand microseconds to write to hard disk. It is possible to write just one byte (well, maybe one word) to RAM, whereas the minimum that can be written to a disk is often four thousand or eight thousand bytes (a single block). We often refer to RAM as physical memory.
A program may take up many thousands of bytes on disk. However, when it is executed normally, only the parts of the code actually needed at the time are loaded into memory. We call these parts segments.
On most operating systems, swap memory is used as an extension for RAM and not as a duplication of it. Assuming the operating system you use is one of those, if there is 128 MB of RAM and a 256 MB swap partition, there is a total of 384 MB of memory available. However, the extra (swap) memory should never be taken into consideration when deciding on the maximum number of processes to be run (we will show you why in a moment). The swap partition is also known as swap space or virtual memory .
The swapping memory can be built from a number of hard disk partitions and swap files formatted to be used as swap memory. When more swap memory is required, as long as there is some free disk space, it can always be extended on demand. (For more information, see the mkswap and swapon manpages.)
System memory is quantified in units called memory pages . Usually the size of a memory page is between 1 KB and 8 KB. So if there is 256 MB of RAM installed on the machine, and the page size is 4 KB, the system has 64,000 main memory pages to work with, and these pages are fast. If there is a 256-MB swap partition, the system can use yet another 64,000 memory pages, but they will be much slower.
When the system is started, all memory pages are available for use by the programs (processes). Unless a program is really small (in which case at any one time the entire program will be in memory), the process running this program uses only a few segments of the program, each segment mapped onto its own memory page. Therefore, only a few memory pages are needed—generally fewer than the program’s size might imply.
When a process needs an additional program segment to be loaded into memory, it asks the system whether the page containing this segment is already loaded. If the page is not found, an event known as a “page fault” occurs. This requires the system to allocate a free memory page, go to the disk, and finally read and load the requested segment into the allocated memory page.
If a process needs to bring a new page into physical memory and there are no free physical pages available, the operating system must make room for this page by discarding another page from physical memory.
If the page to be discarded from physical memory came from a binary image or data file and has not been modified, the page does not need to be saved. Instead, it can be discarded, and if the process needs that page again it can be brought back into memory from the image or data file.
However, if the page has been modified, the operating system must preserve the contents of that page so that it can be accessed at a later time. This type of page is known as a dirty page, and when it is removed from memory it is saved in a special sort of file called the swap file. This process is referred to as swapping out.
Accesses to the swap file are very slow compared with the speed of the processor and physical memory, and the operating system must juggle the need to write pages to disk with the need to retain them in memory to be used again.
To try to reduce the probability that a page will be needed just after it has been swapped out, the system may use the LRU (least recently used) algorithm or some similar algorithm.
To summarize the two swapping scenarios, discarding read-only pages incurs little overhead compared with discarding data pages that have been modified, since in the latter case the pages have to be written to a swap partition located on the (very slow) disk. Thus, the fewer memory pages there are that can become dirty, the better will be the machine’s overall performance.
But in Perl, both the program code and the program data are seen as data pages by the OS. Both are mapped to the same memory pages. Therefore, a big chunk of Perl code can become dirty when its variables are modified, and when those pages need to be discarded they have to be written to the swap partition.
This leads us to two important conclusions about swapping and Perl:
Running the system when there is no free physical memory available hinders performance, because processes’ memory pages will be discarded and then reread from disk again and again.
Since the majority of the running code is Perl code, in addition to the overhead of reading in the previously discarded pages, there is the additional overhead of saving the dirty pages to the swap partition.
When the system has to swap memory pages in and out, it slows down. This can lead to an accumulation of processes waiting for their turn to run, which further increases processing demands, which in turn slows down the system even more as more memory is required. Unless the resource demand drops and allows the processes to catch up with their tasks and go back to normal memory usage, this ever-worsening spiral can cause the machine to thrash the disk and ultimately to halt.
In addition, it is important to be aware that for better performance, many programs (particularly programs written in Perl) do not return memory pages to the operating system even when they are no longer needed. If some of the memory is freed, it is reused when needed by the process itself, without creating the additional overhead of asking the system to allocate new memory pages. That is why Perl programs tend to grow in size as they run and almost never shrink.
When the process quits, it returns all the memory pages it used to the pool of available pages for other processes to use.
It should now be obvious that a system that runs a web server should never swap. Of course, it is quite normal for a desktop machine to swap, and this is often apparent because everything slows down and sometimes the system starts freezing for short periods. On a personal machine, the solution to swapping is simple: do not start up any new programs for a minute, and try to close down any that are running unnecessarily. This will allow the system to catch up with the load and go back to using just RAM. Unfortunately, this solution cannot be applied to a web server.
In the case of a web server, we have much less control, since it is
the remote users who load the machine by issuing requests to the
server. Therefore, the server should be configured such that the
maximum number of possible processes will be small enough for the
system to handle. This is achieved with the
MaxClients directive, discussed in Chapter 11. This will ensure that at peak times, the
system will not swap. Remember that for a web server, swap space is
an emergency pool, not a resource to be used routinely. If the system
is low on memory, either buy more memory or reduce the number of
processes to prevent swapping, as discussed in Chapter 14.
However, due to faulty code, sometimes a process might start running
in an infinite loop, consuming all the available RAM and using lots
of swap memory. In such a situation, it helps if there is a big
emergency pool (i.e., lots of swap memory). But the problem must
still be resolved as soon as possible, since the pool will not last
for long. One solution is to use the
Apache::Resource module, described in the next
section.
There are times when we need to prevent processes from excessive consumption of system resources. This includes limiting CPU or memory usage, the number of files that can be opened, and more.
The
Apache::Resource
module uses the
BSD::Resource module, which in turn uses the C
function setrlimit(
)
to set limits on system resources.
A resource limit is specified in terms of a soft limit and a hard
limit. When a soft limit (for example, CPU time or file size) is
exceeded, the process may receive a signal, but it will be allowed to
continue execution until it reaches the hard limit (or modifies its
resource limit). The rlimit structure is used to
specify the hard and soft limits on a resource. (See the
setrlimit manpage for OS-specific information.)
If the value of variable in rlimit is of the form
S:H, S is treated as the soft
limit, and H is the hard limit. If the value is a
single number, it is used for both soft and hard limits. So if the
value is 10:20, the soft limit is 10 and the hard
limit is 20, whereas if the value is just 20, both
the soft and the hard limits are set to 20.
The most common use of this module is to limit
CPU usage. The environment variable
PERL_RLIMIT_CPU defines the maximum amount of CPU
time the process can use. If it attempts to run longer than this
amount, it is killed, no matter what it is doing at the time, be it
processing a request or just waiting. This is very useful when there
is a bug in the code and a process starts to spin in an infinite
loop, using a lot of CPU resources and never completing the request.
The value is measured in seconds. The following example sets the soft limit for CPU usage to 120 seconds (the default is 360):
PerlModule Apache::Resource PerlSetEnv PERL_RLIMIT_CPU 120
Although 120 seconds does not sound like a long time, it represents a great deal of work for a modern processor capable of millions of instructions per second. Furthermore, because the child process shares the CPU with other processes, it may be quite some time before it uses all its allotted CPU time, and in all probability it will die from other causes (for example, it may have served all the requests it is permitted to serve before this hard limit is reached).
Of course, we should tell mod_perl to use this module, which is done by adding the following directive to httpd.conf:
PerlChildInitHandler Apache::Resource
There are other resources that we might want to limit. For example,
we can limit the data and bstack memory segment sizes
(PERL_RLIMIT_DATA and
PERL_RLIMIT_STACK), the maximum process file size
(PERL_RLIMIT_FSIZE), the core file size
(PERL_RLIMIT_CORE), the address space (virtual
memory) limit (PERL_RLIMIT_AS), etc. Refer to the
setrlimit manpage for other possible resources.
Remember to prepend PERL_ to the resource types
that are listed in the manpage.
If Apache::Status is configured, it can display
the resources set in this way. Remember that
Apache::Status must be loaded before
Apache::Resource, in order to enable the resources
display menu.
To turn on debug mode, set the
$Apache::Resource::Debug variable before loading
the module. This can be done using a Perl section in
httpd.conf.
<Perl>
$Apache::Resource::Debug = 1;
require Apache::Resource;
</Perl>
PerlChildInitHandler Apache::ResourceNow view the error_log file using tail -f and watch the debug messages show up when requests are served.
Under certain Linux setups, malloc( ) uses
mmap( ) instead of brk( ). This
is done to conserve virtual memory—that is, when a program
malloc( )s a large block of memory, the block is
not actually returned to the program until it is initialized. The
old-style brk( ) system call obeyed resource
limits on data segment sizes as set in setrlimit(
). mmap( ) does not.
Apache::Resource’s defaults put
limits on data size and stack size. Linux’s current
memory-allocation scheme does not honor these limits, so if we just
do:
PerlSetEnv PERL_RLIMIT_DEFAULTS On PerlModule Apache::Resource PerlChildInitHandler Apache::Resource
our Apache processes are still free to use as much memory as they like.
However, BSD::Resource also has a limit called
RLIMIT_AS (Address Space), which limits the total
number of bytes of virtual memory assigned to a process. Fortunately,
Linux’s memory manager does
honor this limit.
Therefore, we can limit memory usage under Linux
with Apache::Resource. Simply add a line to
httpd.conf:
PerlSetEnv PERL_RLIMIT_AS 67108864
This example sets hard and soft limits of 64 MB of total address space.
Refer to the Apache::Resource and
setrlimit(2) manpages for more
information.
Generally, limits
should be
imposed on mod_perl processes to prevent mayhem if something goes
wrong. There is no need to limit processes if the code does not have
any bugs, or at least if there is sufficient confidence that the
program will never overconsume resources. When there is a risk that a
process might hang or start consuming a lot of memory, CPU, or other
resources, it is wise to use the Apache::Resource
module.
But what happens if a process is stuck waiting for some event to
occur? Consider a process trying to acquire a lock on a file that can
never be satisfied because there is a deadlock. The process just
hangs waiting, which means that neither extra CPU nor extra memory is
used. We cannot detect and terminate this process using the
resource-limiting techniques we just discussed. If there is such a
process, it is likely that very soon there will be many more
processes stuck waiting for the same or a different event to occur.
Within a short time, all processes will be stuck and no new processes
will be spawned because the maximum number, as specified by the
MaxClients directive, has been reached. The
service enters a state where it is up but not serving clients.
If a watchdog is run that does
not just check that the process is up, but actually issues requests
to make sure that the service responds, then there is some protection
against a complete service outage. This is because the watchdog will
restart the server if the testing request it issues times out. This
is a last-resort solution; the ideal is to be able to detect and
terminate hanging processes that do not consume many resources (and
therefore cannot be detected by the
Apache::Resource module) as soon as possible, not
when the service stops responding to requests, since by that point
the quality of service to the users will have been severely degraded.
This is where the
Apache::Watchdog::RunAway
module comes in handy. This module
samples all live child processes every
$Apache::Watchdog::RunAway::POLLTIME seconds. If a
process has been serving the same request for more than
$Apache::Watchdog::RunAway::TIMEOUT seconds, it is
killed.
To perform accounting, the
Apache::Watchdog::RunAway module uses the
Apache::Scoreboard module, which in turn delivers
various items of information about live child processes. Therefore,
the following
configuration must be added to
httpd.conf:
<Location /scoreboard>
SetHandler perl-script
PerlHandler Apache::Scoreboard::send
order deny,allow
deny from all
allow from localhost
</Location>Make sure to adapt the access permission to the local environment. The above configuration allows access to this handler only from the localhost server. This setting can be tested by issuing a request for http://localhost/scoreboard. However, the returned data cannot be read directly, since it uses a binary format.
We are now ready to configure
Apache::Watchdog::RunAway. The module should be
able to retrieve the information provided by
Apache::Scoreboard, so we will tell it the URL to
use:
$Apache::Watchdog::RunAway::SCOREBOARD_URL = "http://localhost/scoreboard";
We must decide how many seconds the process is allowed to be busy serving the same request before it is considered a runaway. Consider the slowest clients. Scripts that do file uploading and downloading might take a significantly longer time than normal mod_perl code.
$Apache::Watchdog::RunAway::TIMEOUT = 180; # 3 minutes
Setting the timeout to 0 will disable the
Apache::Watchdog::RunAway module entirely.
The rate at which the module polls the server should be chosen carefully. Because of the overhead of fetching the scoreboard data, this is not a module that should be executed too frequently. If the timeout is set to a few minutes, sampling every one or two minutes is a good choice. The following directive specifies the polling interval:
$Apache::Watchdog::RunAway::POLLTIME = 60; # 1 minute
Just like the timeout value, polling time is measured in seconds.
To see what the module does, enable debug mode:
$Apache::Watchdog::RunAway::DEBUG = 1;
and watch its log file using the tail command.
The following statement allows us to specify the log file’s location:
$Apache::Watchdog::RunAway::LOG_FILE = "/tmp/safehang.log";
This log file is also used for logging information about killed
processes, regardless of the value of the $DEBUG
variable.
The module uses a lock file in order to prevent starting more than
one instance of itself. The default location of this file may be
changed using the $LOCK_FILE variable.
$Apache::Watchdog::RunAway::LOCK_FILE = "/tmp/safehang.lock";
There are two ways to invoke this process: using the Perl functions, or using the bundled utility called amprapmon (mnemonic: ApacheModPerlRunAwayProcessMonitor).
The following functions are available:
In order for mod_perl to invoke this process, all that is needed is
the start_detached_monitor( ) function. Add the
following code to startup.pl:
use Apache::Watchdog::RunAway( ); Apache::Watchdog::RunAway::start_detached_monitor( );
Another approach is to use the amprapmon utility. This can be started from the startup.pl file:
system "amprapmon start";
This will fork a new process. If the process is already running, it will just continue to run.
The amprapmon utility could instead be started from cron or from the command line.
No matter which approach is used, the process will fork itself and run as a daemon process. To stop the daemon, use the following command:
panic% amprapmon stop
If we want to test this module but have no code that makes processes
hang (or we do, but the behavior is not reproducible on demand), the
following code can be used to make the process hang in an infinite
loop when executed as a script or handler. The code writes
"\0" characters to the browser every second, so
the request will never time out. The code is shown in Example 5-12.
Example 5-12. hangnow.pl
my $r = shift;
$r->send_http_header('text/plain');
print "PID = $$\n";
$r->rflush;
while(1) {
$r->print("\0");
$r->rflush;
sleep 1;
}The code prints the PID of the process running it before it goes into
an infinite loop, so that we know which process hangs and whether it
gets killed by the Apache::Watchdog::RunAway
daemon as it should.
Of course, the watchdog is used only for prevention. If you have a serious problem with hanging processes, you have to debug your code, find the reason for the problem, and resolve it, as discussed in Chapter 21.
To limit the
number of Apache
children that can simultaneously serve a specific resource, take a
look at the Apache mod_throttle_access module.
Throttling access is useful, for example, when a handler uses a
resource that places a limitation on concurrent access or that is
very CPU-intensive. mod_throttle_access limits the
number of concurrent requests to a given URI.
Consider a service providing the following three URIs:
/perl/news/ /perl/webmail/ /perl/morphing/
The response times of the first two URIs are critical, since people want to read the news and their email interactively. The third URI is a very CPU- and RAM-intensive image-morphing service, provided as a bonus to the users. Since we do not want users to abuse this service, we have to set some limit on the number of concurrent requests for this resource. If we do not, the other two critical resources may have their performance degraded.
When compiled or loaded into Apache and enabled,
mod_throttle_access makes the
MaxConcurrentReqs directive available. For
example, the following setting:
<Location "/perl/morphing">
<Limit PUT GET POST>
MaxConcurrentReqs 10
</Limit>
</Location>will allow only 10 concurrent PUT,
GET, HEAD (as implied by
GET), or POST requests for the
URI /perl/morphing to be processed at any given
time. The other two URIs in our example remain unlimited.
Web services generally welcome search engine robots, also called spiders. Search engine robots are programs that query the site and index its documents for a search engine.
Most indexing robots are polite and pause between requests. However, some search engine robots behave very badly, issuing too many requests too often, thus slowing down the service for human users. While everybody wants their sites to be indexed by search engines, it is really annoying when an initially welcomed spider gives the server a hard time, eventually becoming an unwanted spider.
A common remedy for keeping impolite robots off a site is based on an
AccessHandler that checks the name of the robot
and disallows access to the server if it is listed in the robot
blacklist. For an example of such an
AccessHandler, see the
Apache::BlockAgent module, available from
http://www.modperl.com/.
Unfortunately, some robots have learned to work around this blocking technique, masquerading as human users by using user agent strings identifying them as conventional browsers. This prevents us from blocking just by looking at the robot’s name—we have to be more sophisticated and beat the robots by turning their own behavior against them. Robots work much faster than humans, so we can gather statistics over a period of time, and when we detect too many requests issued too fast from a specific IP, this IP can be blocked.
The Apache::SpeedLimit module, also available from
http://www.modperl.com/, provides
this advanced filtering technique.
There might be a problem with proxy servers, however, where many
users browse the Web via a single proxy. These users are seen from
the outside world (and from our sites) as coming from the
proxy’s single IP address or from one of a small set
of IP addresses. In this case, Apache::SpeedLimit
cannot be used, since it might block legitimate users and not just
robots. However, we could modify the module to ignore specific IP
addresses that we designate as acceptable.
“Stopping and Restarting Apache,” from the Apache documentation: http://httpd.apache.org/docs/stopping.html.
The Red Hat Package Manager web site: http://www.rpm.org/.
Maximum RPM, by Ed Bailey (Red Hat Press).
“RPM-HOWTO,” by Donnie Barnes: http://www.rpm.org/support/RPM-HOWTO.html.
CVS (Concurrent Versions System) resources:
http://www.cvshome.org/ is the home of the CVS project and includes a plethora of documentation. Of special interest is the Cederqvist, the official CVS manual, available at http://www.cvshome.org/docs/manual/.
Open Source Development with CVS, by Karl Fogel (Coriolis, Inc.). Most of the book is available online at http://cvsbook.red-bean.com/.
CVS Quick Reference Card: http://www.refcards.com/about/cvs.html.
daemontools, a collection of tools for managing Unix services: http://cr.yp.to/daemontools.html.
Log collecting and processing tools: http://www.apache-tools.com/search.jsp?keys=log.
cronolog, a log file-rotation program for the Apache web server: http://www.cronolog.org/.
mod_log_spread, which provides reliable distributed logging for Apache http://www.backhand.org/mod_log_spread/.
Spread, a wide area group communication system: http://www.spread.org/.
Recall, an open source library for writing distributed, fault-tolerant, replicated storage servers. A Recall-based server will allow you to save and access data even in the presence of machine failures. See http://www.fault-tolerant.org/recall/.
Chapters 2, 4, 9, 11, and 28 in UNIX System Administration Handbook, by Evi Nemeth, Garth Snyder, Scott Seebass, and Trent H. Hein (Prentice Hall).
Chapters 4 and 5 in Optimizing UNIX for Performance, by Amir H. Majidimehr (Prentice Hall).
To learn more about memory management, refer to a book that deals with operating system theory, and especially with the operating systems used on web server machines.
A good starting point is one of the classic textbooks used in operating system courses. For example:
Operating System Concepts, by Abraham Silberschatz and Peter Baer Galvin (John Wiley & Sons, Inc.).
Applied Operating System Concepts, by Abraham Silberschatz, Peter Baer Galvin, and Greg Gagne (John Wiley & Sons, Inc.).
Design of the Unix Operating System, by Maurice Bach (Prentice Hall).
The Memory Management Reference at http://www.xanalys.com/software_tools/mm/ is also very helpful.
mod_throttle_access: http://www.fremen.org/apache/mod_throttle_access.html.
mod_backhand, which provides load balancing for Apache: http://www.backhand.org/mod_backhand/.
The High-Availability Linux Project, the definitive guide to load-balancing techniques: http://www.linux-ha.org/.
The Heartbeat project is a part of the HA Linux project.
lbnamed, a load-balancing name server written in Perl: http://www.stanford.edu/~riepel/lbnamed/ or http://www.stanford.edu/~schemers/docs/lbnamed/lbnamed.html.
“Network Address Translation and Networks: Virtual Servers (Load Balancing)”: http://www.suse.de/~mha/linux-ip-nat/diplom/node4.html#SECTION00043100000000000000.
Linux Virtual Server Project: http://www.linuxvirtualserver.org/.
Linux and port forwarding: http://www.netfilter.org/ipchains/ or http://www.netfilter.org/.
“Efficient Support for P-HTTP in Cluster-Based Web Servers,” by Mohit Aron and Willy Zwaenepoel, in Proceedings of the USENIX 1999 Annual Technical Conference, Monterey, CA, June 1999: http://www.cs.rice.edu/~druschel/usenix99lard.ps.gz or http://www.usenix.org/publications/library/proceedings/usenix99/full_papers/aron/aron_html/.
IP filter: http://coombs.anu.edu.au/~avalon/. The latest IP filter includes some simple load-balancing code that allows a round-robin distribution onto several machines via ipnat.
Perl modules available from http://www.modperl.com/book/source (not on CPAN):
Apache::BlockAgent, which allows you to block impolite web agents.Apache::SpeedLimit, which allows you to limit indexing robots’ speed.
Get Practical mod_perl now with the O’Reilly learning platform.
O’Reilly members experience books, live events, courses curated by job role, and more from O’Reilly and nearly 200 top publishers.
