writeNotify fails to notify for Acquire PV

[mfurseman]

I have come across a strange problem while using caproto alongside areaDetector. This seems to be present on at least ADAndor and ADSim detectors.

When attempting to change the $(P)$(R)Acquire PV to the value of 1 with caprotos's PV.write method, which initiates a CA writeNotify, the callback notification is never sent from the areaDetector IOC.

I have been unable to trackdown the root cause of this problem but it seems that the callback to send the Notify event is created as Extra Labour for the CAS_Event thread, but in this particular case the event mutux is never granted to the thread and it hangs.

[ulrikpedersen]

Hi @mfurseman good to see you around here 😄

When setting Acquire to 1 you start an acquisition but the callback shouldn't return until the acquisition is complete...

Docs about $(P)$(R)Acquire:

Start (1) or stop (0) image acquisition. This record forward links to $(P)$(R)AcquireBusy which is an EPICS busy record that does not process its forward link until acquisition is complete. Clients should write 1 to the Acquire record to start acquisition, and wait for AcquireBusy to go to 0 to know that acquisition is complete. This can be done automatically with ca_put_callback.

So does your camera/ADSim complete the acquisition and set Acquire back down to 0? Does it work plain old caput -c?

[MarkRivers]

If PV.write effectively does ca_put_callback then the callback will not occur until Acquire goes back to 0. If you did a Continuous acquisition then it will never occur until you manually stop the camera.

You will see the same behavior with

caput -c

but in that case there is a default 1 second timeout, so you will see a timeout error. If you set a 10 second acquisition and do

caput -c -w 15

then you should get a callback after 10 seconds, rather than a timeout.

[mfurseman]

Hi @ulrikpedersen,

Nice to hear from you. We're implementing some EPICS solutions for diagnostics on the MAST-U Experiment.

Thank you for your insights, I suppose I've been looking for a complicated solution to a simple 'problem'! Indeed the callback is processed when acquisition stops.