Re: libezca and EPICS-3.14

[Till Straumann <strauman@SLAC.Stanford.EDU>]

Jeff.

Thanks for your reply.

Oops - I had forgotten the attachment, sorry

-- Till

Jeff Hill wrote:
> 1) Be careful about using R3.14's new feature allowing preemptive
> callback with EZCA. Being a legacy single threaded code, the EZCA
> library probably isn't properly protected against CA callbacks occurring
> preemptively from another thread, at any time, even when it is not
> executing in a CA function.

I guess it's not safe.

>
> 2) If the CA client library does not see the server's beacon then it
> sends an "are you there" message to the server over TCP. It is supposed
> to work (in preemptive callback mode or otherwise) in a way that if you
> do not call ca_pend_event() very often and a UDP beacon is lost, then it
> will still send the "are you there" message and will not disconnect
> until ca_pend_event() is called again and the "are you there" has not
> been replied to within a reasonable amount of time.
>
> 3) This disconnect message that you are seeing, assuming that it is not
> the false message occurring only when the process calls ca_task_exit()

I believe that not being the case. ezca never calls ca_task_exit().
BTW: I could distill a trivial test program (attached) to reproduce the
problem. It essentially does:

	 ezcaGet();
          sleep(120);
          ezcaGet();

After about 1min, the message:

   Unable to contact CA repeater after 50 tries
   Silence this message by starting a CA repeater daemon

appears (note that I _have_ a repeater running) when the
2min expire and the second ezcaGet() attempt is made, I
finally get the message (2nd ezcaGet() fails):

   CA.Client.Exception...............................................
       Warning: "Virtual circuit disconnect"
       Context: "134.79.35.35:5064"
       Source File: ../cac.cpp line 1809
       Current Time: Thu Dec 05 2002 12:29:24.852541000
   ..................................................................
   get_callback() : Virtual circuit disconnect

Neither of these messages appears when I either enable the preemptive
callback or when I use ezcaDelay() instead of sleep() (I suppose because
the ezcaDelay() does ca_pend()).

-- Till

> and channels are still connected (that has been fixed after R3.14 beta
> 2), could be a bug related to proper implementation of (2) above.
> However you are using R3.14 beta one which was a very preliminary (and
> certainly buggy) release of R3.14. Given that we are about to release
> R3.13.1, then it would be very interesting (at least for me) to see if
> the same problem exists with R3.14 beta 2 and ezca.
>
> Jeff
>
>
>
>>-----Original Message-----
>>From: Till Straumann [mailto:strauman@SLAC.Stanford.EDU]
>>Sent: Wednesday, December 04, 2002 9:24 PM
>>To: Jeff Hill
>>Cc: 'Tech-talk'
>>Subject: Re: libezca and EPICS-3.14
>>
>>Having Stephanie next door pays off :-)
>>
>>My application behaves like this:
>>
>>    ezcaGet();
>>
>>    /* user works; this can take minutes, hours, days */
>>
>>    ezcaGet();
>>    /* second ezcaGet() fails if more than ~80sec expire
>>     * between the ezca calls
>>     */
>>
>>Steph pointed out that the CA library still needs to handle
>>beacons even when idle and recommended to initialize the
>>application's CA context with ca_enable_preemptive_callback.
>>As a (undocumented) side effect, this would result in beacons
>>being handled by one of the CA background threads.
>>(At least, this is what I recall from our conversation - if
>>not completely accurate, I'm to blame for misquoting)
>>
>>Anyways, this helped, the disconnects are gone and I'm  happy
>>because I no longer need 3.13 around :-)
>>
>>-- Till
>>
>>PS: the app now does
>>
>>     ca_context_create(ca_enable_preemptive_callback);
>>
>>     ...
>>     ezcaGet();
>>
>>     /* user works for hours */
>>
>>     ezcaGet();
>>     /* SUCCESS */
>>
>>
>>Jeff Hill wrote:
>>
>>>>CA.Client.Exception..........................................
>>>
>>..
>>
>>>>...
>>>>    Warning: "Virtual circuit disconnect"
>>>>    Context: "134.79.35.35:5064"
>>>>    Source File: ../cac.cpp line 1809
>>>>    Current Time: Wed Dec 04 2002 13:48:18.219267000
>>>>.............................................................
>>>
>>..
>>
>>>
>>>Do you know for certain that the "read command" is failing or
>>
>>are you
>>
>>>encountering the following situation?
>>>
>>>In R3.14 beta 1 and beta 2 the ca client library will print
>>
>>the above
>>
>>>message if a connected channel still exists when you call
>>>ca_task_exit(). To remain strictly backwards compatibility
>>
>>this message
>>
>>>is suppressed, only for this specific situation, in the
>>
>>latest version
>>
>>>of R3.14.
>>>
>>>Does ezca block in ca_pend_io() when it is waiting for the
>>
>>"read
>>
>>>command" to complete? If so, then how long is the timeout
>>
>>which is
>>
>>>provided to ca_pend_io()?
>>>
>>>Jeff
>>>
>>>
>>>
>>>>-----Original Message-----
>>>>From: Till Straumann [mailto:strauman@SLAC.Stanford.EDU]
>>>>Sent: Wednesday, December 04, 2002 3:12 PM
>>>>To: Tech-talk
>>>>Subject: libezca and EPICS-3.14
>>>>
>>>>Hi.
>>>>
>>>>I discussed this briefly with Marty at JLAB. He suggested
>>>>I should post to the list. I was not sure whether the
>>>>EZCA library is supported under 3.14.
>>>>
>>>>Here's my problem:
>>>>
>>>>I have put together a CA interface for SCILAB (a nice open
>>>>source tool a la matlab) who builds on top of libezca. Works
>>>>fine with 3.13.
>>>>Using 3.14_beta1, I can access (e.g. read using ezcaget) a PV
>>>>into SCILAB without problem. An attempt to access the same PV
>>>>a couple of minutes later fails, however, giving the
>>>
>>following
>>
>>>>error:
>>>>
>>>>CA.Client.Exception..........................................
>>>
>>..
>>
>>>>...
>>>>    Warning: "Virtual circuit disconnect"
>>>>    Context: "134.79.35.35:5064"
>>>>    Source File: ../cac.cpp line 1809
>>>>    Current Time: Wed Dec 04 2002 13:48:18.219267000
>>>>.............................................................
>>>
>>..
>>
>>>>.....
>>>>
>>>>Strange enough, if I repeat the read command for a couple
>>>>of times, it works again. Let some time expire and it again
>>>>fails.
>>>>And so on.
>>>>
>>>>Does somebody have an explanation or, yet better, a fix?
>>>>
>>>>-- Till
>>>
>>>
>
>



#include <cadef.h>
#include "ezca.h"
#include <assert.h>

#include <stdio.h>

static void
usage(char *n)
{
	fprintf(stderr,"usage: %s PV\n",n);
}

int
main(int argc, char **argv)
{
int n=1,i;
double db;
	if ( argc<2 ) {
		usage(argv[0]);
		exit(1);
	}
//	assert(ECA_NORMAL == ca_context_create(ca_enable_preemptive_callback));
	for ( i=1; i>=0; i-- ) {

		if ( ezcaGet(argv[1], ezcaDouble ,n,&db) ) {
			fprintf(stderr,"ezca error");
			exit(1);
		}
		printf("%s: %g\n",argv[1],db);

		if ( i )
#if 0
			ezcaDelay(120.);
#else
			sleep(120);
#endif
	}
}