Need help with semaphore creation/syncronization strategy.

[Roy Smith]

I've been playing with System V IPC semaphores a bit, and I've come to a
point where I'm stuck.

I've got two processes (call them master and slave) which share a
semaphore. I'm using the race condition avoidance algorithm Stevens
describes in UNP Volume 2. The master creates a semaphore, initializes
it to zero (i.e. claims the lock), then does a semop() to force
sem_otime to go non-zero. The slave loops waiting for sem_otime.

The problem is, I don't know how to handle the case where a previous
master process may have died unexpectedly and left an orphaned semaphore
in the kernel. How would the slave recognize this condition? As far as
the slave can tell, the semaphore is initialized, so it assumes the
master is ready to go.

[another2]

okay, here's one way i thought of, i'm a bit of a newbie on semaphores
and i''ve only had experience with pthreads on solaris so if it's
stupid: i'm sorry

okay, i'd put a piece of code like this to figure out if the
server fails:
i used 2000 but, you may want a bigger number, the assumption is that
the server wont run 2000 times in one itteration of the client but, i
think i'm right with this code, if i'm wrong and you do get it please
post it, i like thread coding :)


int i = 2000;

//i used this to as an example if you wanna keep the thing
//running indefinitley
while(true)
{
if(sem_getvalue(&your_semaphore) == 0)
{
--i;
}
else
{
i = 2000;
sem_wait(&your_semaphore);
//your critical section
sem_post(&your_semaphore);
}

if (i < 1)
{
//server failed
}

}//end of while loop

--
Posted via [url]http://dbforums.com[/url]