Experimental Physics and
Industrial Control System

Ru Igarashi via Tech-talk <tech-talk at aps.anl.gov> · Sat, 17 Apr 2021 06:22:12 +0000

From: Michael Davidsaver <mdavidsaver at gmail.com>, Thursday, April 15, 2021 14:47
>On 4/13/21 11:29 PM, Ru Igarashi via Tech-talk wrote:
>> [normally, I'd inline my response but I'm stuck with Web Outlook.  My apologies.]
>>
> Hello Andrew,
>>
>> re: camonitor: the errant sequence is the same with and without camonitor
>> running in parallel.  My thinking was that the reliance on the Channel
>> Access monitors meant that update occasionally necessarily would
>> eventually be out of sync. ...

> I don't think anyone else addressed this point specifically,
> and I think it is an important point to make.  CA (and PVA)
> subscriptions should always, eventually, be "in-sync". This is an
> essential property for a robust distributed control system to have!

Subscriptions and even caget(), so far, are not impacted with the
observed problem, and, up until I ran my tests on one very specific
configuration, asynchronous caput() weren't affected (at least to
nowhere near this extent).  Then I ran on this one very specific
configuration and observed a huge uptick in failed updates in
asynchronous caputs.  I think I've simply run across an odd
combination of processor, OS and EPICS versions.  I'm just trying
to determine what I need to change to make it work.

> If a PV produces a burst of updates which exceeds some internal
> buffers, the most recent update will never be discarded.  So all
> clients will eventually receive the most recent update produced.

I'm seeing that even when the asynchronous caput fail to update
properly: immediate caget after put are updating properly.  I have
yet to run my tests on camonitors.  That is, EPICS is not failing in
that regard, either.

ru

Experimental Physics and Industrial Control System

Experimental Physics and
Industrial Control System