Experimental Physics and
Industrial Control System

Jameson Graef Rollins <[email protected]> · Sat, 30 Dec 2017 09:44:57 -0800

On Fri, Dec 29 2017, Jameson Graef Rollins <[email protected]> wrote:
> The peculiar thing I'm worried about is that in my current application
> there are, under some circumstances, multiple record updates during a
> single driver.write().  For example, before pcaspy 0.7.0, I was doing
> something like this:
>
> class MyDriver(pcaspy.Driver):
>     ...
>     def write(self, channel, value):
>         ...
>         self.setParam(channel, value)
>         self.setParam(other_channel, other_value)
>         ...
>         self.pvDB[channel].flag = False
>         self.updatePVs()
>         return True
>
> The 'pvDB[channel].flag = False' was to prevent subscribed clients from
> seeing multiple updates for the primary record 'channel'.  In 0.7.0, it
> seems that neither that, nor the final updatePVs() call are necessary.
> However, I am now occasionally (less than 50% of the time) seeing
> multiple client updates for the primary channel.  The sporadic nature of
> the double updates makes me think there's some kind of race condition.
> Regardless, I can't figure out what's going on, and how to uniformly
> prevent redundant update in subscribed clients.

Xiaoqiang, I just sent a pull request for a patch that seems to fix this
issue for me [0]:

https://github.com/paulscherrerinstitute/pcaspy/pull/49

There is a race condition between the SimplePV.updateValue() calls in
the PV scan() method, and in the Driver.updatePV() method.  The above
patch is a work around for the issue and seems to eliminate the
occasional double updates that I was seeing in my test suite.

This does *not* fix the other issue that I mention in the other thread,
that I see spurious updates after initialization.

jamie.

Experimental Physics and Industrial Control System

Experimental Physics and
Industrial Control System