RE: Problem with huge waveforms in EPICS 7

[Mark Rivers via Core-talk <core-talk at aps.anl.gov>]

I tested Michael's patch and it fixes the problem when I run the ADSimDetector test.

 

(base) J:\epics\devel\asyn\iocBoot\ioctestAsynPortDriver>python monitor_test_ad.py

2022-06-27 15:50:43.877772: Adding BIGARRAY callback function

2022-06-27 15:50:43.878761: Sleeping for 5 seconds

2022-06-27 15:50:45.787905: pv_callback counter=1, pvName=13SIM1:image1:ArrayData, size=20000000, value=[0]=99

2022-06-27 15:50:48.880187: Starting acquisition

2022-06-27 15:50:48.901216: Waiting for acquisition to complete

2022-06-27 15:50:50.290341: pv_callback counter=2, pvName=13SIM1:image1:ArrayData, size=20000000, value=[0]=100

2022-06-27 15:50:51.699526: pv_callback counter=3, pvName=13SIM1:image1:ArrayData, size=20000000, value=[0]=112

2022-06-27 15:50:52.709903: pv_callback counter=4, pvName=13SIM1:image1:ArrayData, size=20000000, value=[0]=126

2022-06-27 15:50:53.127619: pv_callback counter=5, pvName=13SIM1:image1:ArrayData, size=20000000, value=[0]=136

2022-06-27 15:50:53.545305: pv_callback counter=6, pvName=13SIM1:image1:ArrayData, size=20000000, value=[0]=140

2022-06-27 15:50:53.956352: pv_callback counter=7, pvName=13SIM1:image1:ArrayData, size=20000000, value=[0]=144

2022-06-27 15:50:54.371361: pv_callback counter=8, pvName=13SIM1:image1:ArrayData, size=20000000, value=[0]=148

2022-06-27 15:50:55.201308: pv_callback counter=9, pvName=13SIM1:image1:ArrayData, size=20000000, value=[0]=152

2022-06-27 15:50:55.626239: pv_callback counter=10, pvName=13SIM1:image1:ArrayData, size=20000000, value=[0]=161

2022-06-27 15:50:56.240083: pv_callback counter=11, pvName=13SIM1:image1:ArrayData, size=20000000, value=[0]=165

2022-06-27 15:50:56.654982: pv_callback counter=12, pvName=13SIM1:image1:ArrayData, size=20000000, value=[0]=171

2022-06-27 15:50:57.071780: pv_callback counter=13, pvName=13SIM1:image1:ArrayData, size=20000000, value=[0]=175

2022-06-27 15:50:57.495975: pv_callback counter=14, pvName=13SIM1:image1:ArrayData, size=20000000, value=[0]=179

2022-06-27 15:50:57.909298: pv_callback counter=15, pvName=13SIM1:image1:ArrayData, size=20000000, value=[0]=183

2022-06-27 15:50:58.536483: pv_callback counter=16, pvName=13SIM1:image1:ArrayData, size=20000000, value=[0]=188

2022-06-27 15:50:59.158556: pv_callback counter=17, pvName=13SIM1:image1:ArrayData, size=20000000, value=[0]=194

2022-06-27 15:50:59.574255: pv_callback counter=18, pvName=13SIM1:image1:ArrayData, size=20000000, value=[0]=199

2022-06-27 15:50:59.576321: Acquisition complete

2022-06-27 15:50:59.576321: Sleeping for 60 seconds

2022-06-27 15:51:59.579580: Exiting

 

There are no callbacks after acquisition is complete.

 

Mark

 

 

-----Original Message-----
From: Core-talk <core-talk-bounces at aps.anl.gov> On Behalf Of Michael Davidsaver via Core-talk
Sent: Monday, June 27, 2022 9:51 AM
To: Zimoch Dirk (PSI) <dirk.zimoch at psi.ch>
Cc: core-talk at aps.anl.gov
Subject: Re: Problem with huge waveforms in EPICS 7

 

On 6/27/22 06:56, Zimoch Dirk (PSI) via Core-talk wrote:

> On Mon, 2022-06-27 at 15:12 +0200, Zimoch Dirk wrote:

>> The problem appears between commit 4ab98081 and 56f05d72. (The two

>> commits in between do not produce stable code.)

>> 

>

> Ben did some code refactoring of db_field_log which changed array handling.

>

> In particular in dbEvent.c the following snipped disappeared from

> db_queue_event_log()

>

>      /*

>       * if we have an event on the queue and both the last

>       * event on the queue and the current event are emtpy

>       * (i.e. of type dbfl_type_rec), simply ignore duplicate

>       * events (saving empty events serves no purpose)

>       */

>      if (pevent->npend > 0u &&

>          (*pevent->pLastLog)->type == dbfl_type_rec &&

>          pLog->type == dbfl_type_rec) {

>          db_delete_field_log(pLog);

>          UNLOCKEVQUE (ev_que);

>          return;

>      }

>

> I have added a debug printf in this code and see that it triggers

> whenever a frame is dropped. That is about 3 out of 4 frames in my test case.

>

> It is not trivial to put back the code because dbfl_type_rec disappeared from enum dbfl_type too.

 

I think the (untested) equivalent would now be:

 

> diff --git a/modules/database/src/ioc/db/dbEvent.c

> b/modules/database/src/ioc/db/dbEvent.c

> index 0164f55b9..480784d1e 100644

> --- a/modules/database/src/ioc/db/dbEvent.c

> +++ b/modules/database/src/ioc/db/dbEvent.c

> @@ -789,6 +789,18 @@ static void db_queue_event_log (evSubscrip

> *pevent, db_field_log *pLog)

> 

>      LOCKEVQUE (ev_que);

> 

> +    /* if we have an event on the queue and both the last

> +     * event on the queue and the current event are empty,

> +     * simply ignore duplicate events.

> +     */

> +    if (pevent->npend > 0u

> +            && !dbfl_has_copy(*pevent->pLastLog)

> +            && !dbfl_has_copy(pLog)) {

> +        db_delete_field_log(pLog);

> +        UNLOCKEVQUE (ev_que);

> +        return;

> +    }

> +

>      /*

>       * add to task local event que

>       */