Re: Unpredictable behaviour with PyEpics

[Matt Newville via Tech-talk <tech-talk at aps.anl.gov>]

Hi Simon, 

On Wed, May 27, 2020 at 1:33 AM Simon Rose <Simon.Rose at ess.eu> wrote:

Hi all –

 

I am not using a python interpreter from within the IOC; I am using python outside the IOC (but on the same VM) to test the IOC.

OK, I think that simplifies the problem.

 

Something that I have noted: the default delay of 2 seconds can be shortened quite a bit (about 0.05 seconds seems as far as I can push it).

Yes, there does seem to be a real and possibly unavoidable time between creating a PV and actually being fully connected.  20 to 50 ms is about normal, I think.  That time can be shared over multiple connections so that one can create hundreds of PVs and be able to use them quickly.

I think that I was wrong earlier in the expectation that  `assert pv.connect()` should just work.   In fact, `pv.connet()` is now more like an extra-explicit request to connect.  It will not set `PV.connected` itself but merely mark "connection has started".  So the right way to verify connection after creating a PV should really be

    mypy = PV(pvname)

    mypv.wait_for_connection()

But this should not really be necessary - connection will happen and getting/putting data for the PV will also wait for connection.  

 

However, if I run wait_for_connect() and debug what path it takes through that function in order to connect, I see the same consistent behavior as initially described: about 1 in 5 times it fails to connect on the call to connect(), but then succeeds when it hits the call to ca.poll() (although it may take a few poll()s before it succeeds).

 

Note that this does not depend (!) on the length of the delay: a delay of 10 seconds yields the same pattern as a delay of 0.05 seconds.

You should use 'wait_for_connection()` as the method that will wait to return until connection has completed.  If that is not working, can you post the actual code you're having trouble with?

 

(Regarding Jaroslav’s earlier question, although it may be moot: the PYEPICS_LIBCA in both cases is the same)

Setting PYEPICS_LIBCA should not be necessary for your case. 

Hope that helps, 

--Matt