Re: Weird CAS hangup on IOC

[Torsten Bögershausen via Core-talk <core-talk at aps.anl.gov>]

Hej again,
I think there is a risk for race conditions .
for me it seams as if bind() succeeded,
but then listen() fails.

From my (limited) understanding, 2 processes are allowed to bind() to 
the same
port, for whatever reasons.
process 1 may use connect() on the port (specifying local port to be 
something
defined.
   This is not usually used, but technically possible and allowed.)
process 2 may use listen() on the same port, listening for incomming 
connections.
No both processes don’t interfere by using the same local port.

From that point of view, I would suspect that bind() can be done on the 
same port
from 2 different processes without problems.

The “trouble” starts, when we want to listen() from 2 differnent 
processes on
the same port,
that is not a valid use-case.

In this case, our EPICS code could handle this situation more gently,
by closing the bind() port, bind() to the next one and triy to listen.

Does this make any sense ?

/Torsten



On 06/02/20 13:44, Ralph Lange via Core-talk wrote:
> Hej Torsten,
>
> On Thu, 6 Feb 2020 at 13:35, Torsten Bögershausen 
> <torsten.bogershausen at ess.eu <mailto:torsten.bogershausen at ess.eu>> wrote:
>           if ( listen ( IOC_sock, 20 ) < 0 ) {
>               char sockErrBuf[64];
>               epicsSocketConvertErrnoToString (
>                   sockErrBuf, sizeof ( sockErrBuf ) );
>               errlogPrintf ( "CAS: Listen error: %s\n",
>                   sockErrBuf );
>               epicsSocketDestroy (IOC_sock);
>               epicsThreadSuspendSelf ();
>           }
>
>     We should be able to print out errno,
>     to see what is going on and why listen fails.
>     (could be EAGAIN ??) We need to find out...
>
>
> Well, we have the error string, right? "Address already in use" is not 
> EAGAIN.
> Also, the hanging IOC reports (casr 4) to be running on the 5064 port, 
> which matches the missing "dynamically assigned TCP port" warning - 
> sorry, I should have mentioned that in my original mail.
> Cheers,
> ~Ralph