Experimental Physics and
Industrial Control System

"Vigder, Mark" <[email protected]> · Wed, 20 Apr 2011 17:50:19 -0400

Andrew, Matt

On Wednesday 20 April 2011 Andrew Johnson wrote
>On Wednesday 20 April 2011 15:06:33 Matt Newville wrote:
>> On Wed, Apr 20, 2011 at 1:04 PM, Vigder, Mark
>> <[email protected]> wrote:
>> > The only difference is I created some arbitrary unrelated PV before
>> > starting the thread. I'm still
>> > trying to understand CA contexts, so I'm not sure those calls are
>> > necessary, but
>> > for some things I tried they seemed to be required. Other little changes,
>> > such as creating the
>> > PV object inside the thread rather than passing it in to the thread, also
>> > seem to eliminate the 30 second delay.
>> > Trying on my PC rather than linux co-hosted with IOC gives about a 10
>> > second delay.
>> >
>> > Any help or explanations are well appreciated.
>> 
>> The default timeout for pv.put() in "put-with-wait" mode is 30
>> seconds, which is probably where that time is coming from.
>> 
>> It appears that the put-with-wait is not announcing completeness
>> across "contexts" (that is you create a PV in the main thread, then
>> create a new context in the created thread and do the put-with-wait on
>> the PV from the original thread).  It's possible thatt the
>> put-with-wait is not complete enough for dealing with all thread cases
>> right now.
>
>Assuming that epics.ca.context_create() results in a call to 
>ca_context_create() then I worry that Mark's code ends up trying to use a chid 
>created in one CA context (associated with the object m) in a thread that is 
>running with a different CA context.  Internally a chid has a reference to the 
>context it was created in, and I have no idea what happens if you use one this 
>way, but I'm fairly sure you're not supposed to.
>
>I would recommend that you share a single CA context across all the threads in 
>your program unless you have a very good reason to not do that, it saves 
>resources.  To do that you'll need to find out what the PyEpics equivalents 
>are of the routines ca_current_context() and ca_attach_context(), and pass the 
>context pointer returned by calling ca_current_context() from the main thread 
>into the new thread and there passing it to ca_attach_context().
>
>HTH,
>
>- Andrew
>-- 

I did a quick test and this seems to solve it. Appropriate calls in PyEpics are
ca.current_context() and ca.attach_context().

Maybe one day I'll come up with a good reason not to use the same CA context
across all Python threads, but I certainly don't have one today. I'm just
starting to figure out what chid's and contexts are all about in CA.

Thanks
Mark

Experimental Physics and Industrial Control System

Experimental Physics and
Industrial Control System