[squid-users] issue with start / stop scripts

[Amos Jeffries]

On 29/11/2015 11:32 a.m., Eliezer Croitoru wrote:
> A check on what?
> Basically to verify if squid is still running you need to verify that
> there are is not one squid instance running.
> The PID is kind of a hack to make sure squid is still there or not.
> In most cases you can cancel the timeout and check only for the PID.
> Also notice that there is a "rm -rf" there which was inherited from an
> old script that I got as a "gift" since my own script got lost in a
> server migration.
> 
> You can run three checks in parallel:
> - the pid exists or not
> - the process exists or not(using "ps aux|grep squid")
> - check if the port in netstat is still in listening mode.
> 


Also, be aware the shutdown signal for squid is SIGHUP. If you need to
abort the internal timeout use a second SIGHUP.

So the shutdown process as seen/done by external scripts should be:

* scan squid.conf for unusual pidfile_path and shutdown_lifetime values
* if squid.pid exists; send SIGHUP to process indicated inside
* else locate running squid process and send SIGHUP
* wait desired timeout + a few seconds

* if process still exists; repeat SIGHUP
* wait a few seconds more

* if process still exists; send SIGKILL; repeat until closed
* if squid.pid still exists; rm squid.pid


Notice the two points where waiting "a few seconds more" is needed for
the signals to have effects. That is the time Squid *actually* takes to
shutdown.

The shutdown_lifetime is a grace period where normal proxying operations
are still taking place to try and finish client transactions off. It is
essentially the *minimum* time needed for Squid to shutdown. If you just
abruptly abort/SIGKILL right at the same point where squid should only
be starting to seriousy wind up its own internal actions it can lead
cache and FD problems.

HTH
Amos