Weblog entry #104 for simonw
See all weblog entries made by simonw.
/etc/init.d/apache2 restart
Forcing reload of web server: Apache2(98)Address already in use: make_sock: could not bind to address 0.0.0.0:8001
no listening sockets available, shutting down
Unable to open logs
Looks like there is a race condition in the "/etc/init.d/apache2 restart". It is intermittent (very), and I suspect dependent on the state of the mod_perl2 modules inside Apache, since the logging suggests it is having trouble killing off the old tasks in a timely manner. However because logrotate directs the output of the command to /dev/null I assume this information is normally thrown away when a problem occurs.
I keenly joined the Debian apache mailing list, but time is too short, and in the end I installed apache-mpm-prefork, as we've never had this issue on any other Apache server at work, which all have to use prefork due to the need to run PHP.
Comments on this Entry
[ Parent | Reply to this comment ]
[ Send Message | View dkg's Scratchpad | View Weblogs ]
The runit-services package even has service configuration scripts for apache (maybe not apache2), but you can find apache2 and many other supervision scripts here.
As a bonus, you get clean process state and several other benefits over the old sysv scheme.
[ Parent | Reply to this comment ]
[ Send Message | View Steve's Scratchpad | View Weblogs ]
I've found people switching to runit having a lot of problems come package-upgrade time.
e.g. The last time I upgraded MySQL, the preinst script ran "mysql stop", and this didn't work. So the upgrade failed.
I had to manually stop the service, run the upgrade, then start it up again, since the Debian package knew nothing about using runsvc.
If the init scripts were supported then this would be much less pain.
[ Parent | Reply to this comment ]
[ Send Message | View dkg's Scratchpad | View Weblogs ]
If you plan on installing something like runit, it's probably wise to replace your /etc/init.d scripts with scripts which invoke /usr/bin/sv to control the relevant service directory, so that tools which want to use the invoke-rc.d interface can do so without knowing that they're controlling a runit-supervised service underneath.
[ Parent | Reply to this comment ]
Since whilst Apache is running, it effectively keeps itself going, I don't think it calls for init style restarting of the service (that is built-in), the service will run as long as root doesn't tell it to stop, or some other catastrophic system event occurs.
The more crucial question is what is going wrong, since it seems that the "/proc/nnnn" directory is removed before the port Apache is bound to is freed. I suspect a well placed "sleep 1" would also resolve it from a practical perspective. However I hadn't realised that the Debian init script authors were writing scripts quite so intimately connected to the (current) kernel in use. Also these scripts are duplicating code rather than using shared functions, since several try and check for process death in this fashion.
Perhaps I'm missing something here (Debian usually has a good reason for these sort of things), but this all seems to be storing up trouble for the future, and the init scripts seem to be overly arcane in nature.
[ Parent | Reply to this comment ]