Re: autotest: Temporarily starting daemons

[Allan Clark]

On Sun, Aug 2, 2009 at 18:00, address@hidden <address@hidden> wrote:

> I'm trying to temporarily fork off a daemon (BIND's named) within a
> testsuite. Yet the testsuite hangs despite having started the daemon. Below
> is an excerpt of the circumstances. Running testsuite -v shows that the hang
> occurs not as one would assume during the fork of the daemon but in the
> subsequent check for its pid. This occurs both when making the daemon
> startup a separate AT_CHECK item as well as when making it a simple shell
> statement. However, named starts properly and, since I'm not killing it for
> now, continues to run past SIGINT of the test suite .
>
> I understand that things enclosed between AT_SETUP and AT_CLEANUP run
> within their own subshell hence my plan was to start named, do some tests
> and kill it within those two. Could it be that named is capturing
> stdout/stderr irretrievably away from autotest or is forking generally a
> problem?


In my testing of Nagios/LDAP, I created a tool similar to "nohup"; instead
of changing signal-handling, I made "timeout" to kill the child process
after a given delay, giving a return code based on what cases/reasons it
stopped.

I'm using this in MySQLfs (http://sourceforge.net/projects/mysqlfs/develop)
-- you can grab a copy from the mysqlfs/tests-autotest/timeout.c.  Check the
Makefile.am, you'll see that I cheat the dependencies to make sure "timeout"
is created before running the test:

---
check-local: atconfig atlocal $(TESTSUITE) timeout
        $(SHELL) $(TESTSUITE)
        rm -fr $(subdir)/testsuite.dir

check_PROGRAMS = timeout
timeout_SOURCES = timeout.c
---

I get pretty good results from this, but it occasionally "timeout" isn't
aggressive enough in failed tests (and it's timeout expires, and it goes on
a killing spree).  in Nagios/LDAP, I kill the timeout by pidfile mentioned
in the ldap config; in MySQLfs, I got weak, and I just killall.  I could
probably write out the pidfile from timeout's fork() for better/safer
assassination.

Doxygen'd at http://mysqlfs.sourceforge.net/html/timeout_8c-source.html .  GPL2

Allan
-- 
address@hidden  "金鱼" http://linkedin.com/in/goldfish