Re: Race condition with AckDaemonProcess

[Adam Jocksch <ajocksch at redhat dot com>]

Andrew Cagney wrote:
> Adam Jocksch wrote:
> > I've been using AckDaemonProcess to write the test case for the stack 
> > trace work I've been doing, and I've recently stumbled upon an 
> > interesting race condition.
> >
> > I'm telling AckDaemonProcess to launch a program called looper2 (that 
> > I've checked the source for into frysk-core/rt/tests/) that makes a 
> > few function calls to establish some sort of meaningful context, then 
> > signals Frysk to continue and busywaits while frysk attaches to it. 
> > When I build this file outside of the build system using just -g the 
> > daemon process returns right away and the test continues as normal. 
> > However since moving looper2 into the build system I find that the 
> > AckDaemonProcess constructor times out whenever the test case is run 
> > without logging. Running the test case with logging (i.e. with -c 
> > FINE) allows the test case to run normally again, which is what leads 
> > me to believe that there is a race condition somewhere.
> Something must be causing it to miss or drop the ack signal :-/  Does 
> it work when just that test is run from within the test framework?  
> What does ftrace show?
>
> Andrew
>
> > I'll be looking into this, but if anyone has any insights they'd like 
> > to share they would be very welcome :)
> >
> > Adam
I think I've resolved this issue. The test program I was using was 
reaching the kill call before frysk was ready to recieve it. I've 
changed the test program and it seems to work now, although I'll be 
watching closely for it's recurrance.

Adam