Experimental Physics and
Industrial Control System

Michael Davidsaver <[email protected]> · Mon, 18 Dec 2017 17:46:32 -0600

I haven't worked through the whole situation, but this might give you some hints.

> The CP monitor finds the record busy (PACT=1) and arranges for re-processing when the record is done

'slow_aSub' will never be found with PACT=1 as it's not asynchronous,
and there will never be a DB_LINK processing loop to break.
Rather, a concurrent dbPutField() on this record will block in dbScanLock() until the sleep ends.

Because seqRecord is _always_ asynchronous, the lockset is unlocked here with seq.PACT=1 but slow_aSub.PACT=0
At which point a concurrent dbPutField() could re-lock and process 'slow_aSub' again (with mode==0).
Of course this is racing with the callback task for the seqRecord (in which case 'slow_aSub' is processed with mode==1).

Needless to say, my first advice is to be careful about using seqRecord in a loop!

On 12/15/2017 10:02 AM, Dirk Zimoch wrote:
> Hi Folks,
> 
> A colleague puzzled me with a db that shows surprising behavior.
> 
> An aSub record is triggered periodically though a CP input link. It looks up if it needs re-initialization (the user can press a button to request that). If yes it does an additional calculation. After processing it resets the record that requested the re-initialization.
> 
> See the attached .db file.
> 
> Usually when the user presses the button to re-initialize, the aSub does the additional calculation once and then goes back to normal processing.
> 
> But the calculation may may take a bit too long -- longer than the rate of the CP input link monitors. In that case, the record never leaves initialization mode!
> 
> I have simulated the slow calculation with the attached subroutine code which simply waits. (I know that is is a bad idea to wait in a synchronous subroutine. The original code does not wait but calculates.)
> 
> I think that this is an effect of lock sets because the "mode" record is as well input being processed by a caput as well as output of the aSub record. But is this the intended behavior?
> 
> The CP monitor finds the record busy (PACT=1) and arranges for re-processing when the record is done. But should not the FLNK be processed and the mode record go to 1 before the aSub runs again?
> 
> Any idea how to fix this? Is it necessary to make the subroutine asynchonous? Making the links from and to "mode" CA links does not fix it. Scanning the aSub itself with ".1 second" does not show the strange behavior, it simply delays the scan thread.
> 
> Puzzle for the weekend...
> Dirk

Experimental Physics and Industrial Control System

Experimental Physics and
Industrial Control System