For quite some time I’ve been having a problem with my httpd.pid file going missing every now and then. Whenever this happens, I can’t use the init script to start and stop the server anymore, which is pretty annoying. For some reason I had learned to live with this – I would just ‘killall httpd’, start the server again and the httpd.pid would be back for some time, until the problem would reappear.

Finally I decided to do something about it. I added this small script to my hourly cron jobs:

if [ ! -e "/usr/local/apache2/logs/httpd.pid" ]; then
        echo "httpd.pid missing"
fi

A few weeks later this trap was finally triggered – less than an hour after my daily logrotate, which produced this output:

Starting httpd: [  OK  ]
Stopping httpd: [  OK  ]
(repeated a number of times...)
Starting httpd: [  OK  ]
Stopping httpd: [FAILED]
Starting httpd: (98)Address already in use: AH00072: make_sock: could not bind to address [::]:80
(98)Address already in use: AH00072: make_sock: could not bind to address 0.0.0.0:80
no listening sockets available, shutting down
AH00015: Unable to open logs
[FAILED]
error: error running postrotate script for /usr/local/apache2/logs/*_log 

This was not the first time I had seen this, but it was the first time I realized that the logrotate problem was causing the httpd.pid problem, and not the other way around.

So I checked my /etc/logrotate.d/apache config file:

/usr/local/apache2/logs/*_log {
    rotate 1024
    size 4M
    notifempty
    postrotate
        /etc/init.d/httpd restart
    endscript
}

After a few minutes of reading the logrotate man page, I realized what was wrong (had a hunch, though): The restart between the postrotate/endscript directives was performed for each logfile, i.e. multiple times, instead of just a single time after the last one was rotated. This in itself is pretty bad, but it would also happen asynchronously, thus create a mess. This problem was easily solved using the sharedscript directive. Also, a simple “reload” instead of “restart” is sufficient to make httpd reopen the logfiles, thus create new filehandles.

After fixing these two problems, the file ended up like this:

/usr/local/apache2/logs/*_log {
    rotate 1024
    size 4M
    notifempty
    sharedscripts
    postrotate
        /etc/init.d/httpd reload
    endscript
}

I’m fully expecting this to solve the problem once and for all.

A few weeks ago at work, I needed to rename a webhost. To avoid breaking a lot of links to the old hostname, I set up permanent redirection, but ran into an infinite loop. I tried both a simple Redirect statement like this:

Redirect permanent / http://newhost/


And the same thing using mod_rewrite. I have done this many times before at home, but what was special about this case is that the web server was set up as default host. So I just added a virtual host with the old hostname. Eventually I got it to work with mod_rewrite by using a RewriteCond statement to break the loop:

<VirtualHost *:80>
ServerName oldhost

<IfModule rewrite_module>
RewriteEngine on
RewriteCond %{HTTP_HOST} oldhost
RewriteRule ^/(.*)$ http://newhost/$1 [R=permanent,L]
</IfModule>
</VirtualHost>


However, I’m still not sure why this is neccesarry, since the first rewritten URL should end at the default host which doesn’t rewrite anything.

Today I wanted to upgrade Courier-Authlib from 0.63.0 and read this in the ChangeLog:

2010-03-06 Sam Varshavchik

* Remove the bundled libtdl library. Require the system-installed
libltdl library.

As expected, this gave me some problems with my old CentOS 5.9 release:

/bin/sh ./libtool –tag=CC –mode=link gcc -g -O2 -Wall -I.. -I./.. -export-dynamic -dlopen libauthuserdb.la -dlopen libauthpam.la -dlopen libauthshadow.la -dlopen libauthcustom.la -dlopen libauthpipe.la -o authdaemondprog authdaemond.o libltdl/libltdlc.la libcourierauthcommon.la liblock/liblock.la libhmac/libhmac.la md5/libmd5.la sha1/libsha1.la rfc822/libencode.la numlib/libnumlib.la -ldl
libtool: link: cannot find the library `libltdl/libltdlc.la’ or unhandled argument `libltdl/libltdlc.la’
make[2]: *** [authdaemondprog] Error 1
make[2]: Leaving directory `/usr/local/src/courier-authlib-0.65.0′
make[1]: *** [all-recursive] Error 1
make[1]: Leaving directory `/usr/local/src/courier-authlib-0.65.0′
make: *** [all] Error 2

I fixed this by adding this line to my build script (after configure):
sed -i -e 's/^LIBLTDL = ${top_build_prefix}libltdl\/libltdlc.la/LIBLTDL = -lltdl/' Makefile


Update, February 15th 2013: Today I found out that in a freshly installed VMware machine with CentOS 5.9, ltdl was missing. So I needed to add the following in my pre-build script as well:
rpm -q libtool-ltdl >/dev/null
if [ $? != 0 ]; then yum -q -y install libtool-ltdl; fi
rpm -q libtool-ltdl-devel >/dev/null
if [ $? != 0 ]; then yum -q -y install libtool-ltdl-devel; fi


Update, December 7th 2015: I wanted to build 0.66.4 today, and the problem reappeared. Luckily, I found this posting which saved me some time. Updated ‘sed’ command:
sed -i -e 's/^LIBLTDL = $(top_build_prefix)libltdl\/libltdlc.la/LIBLTDL = -lltdl/' Makefile