Triggering of xspress3 at BMM takes 2.5s for 0.5s exposure

[mrakitin]

This part of the code seems to be a bottleneck (see the attached picture with the ts_msg_hook-generated output):

nslsii/nslsii/detectors/xspress3.py

Lines 623 to 638 in 18c0632

	def trigger(self):
	if self._staged != Staged.yes:
	raise RuntimeError("not staged")
	self._status = DeviceStatus(self)
	self.settings.erase.put(1)
	self._acquisition_signal.put(1, wait=False)
	trigger_time = ttime.time()
	for sn in self.read_attrs:
	if sn.startswith('channel') and '.' not in sn:
	ch = getattr(self, sn)
	self.dispatch(ch.name, trigger_time)
	self._abs_trigger_count += 1
	return self._status

[mrakitin]

Did a few tests today with @bruceravel, all repeatable, even with the disabled HDF5 plugin.

@tacaswell, what would be the best way to debug it? Which kinds of prints can help here? I guess this dispatching is doing something very slowly, but not sure how to best approach it.

[jklynch]

https://github.com/joerick/pyinstrument

[tacaswell]

I think we may need to throw a whole lot of camonitors on the AD signals to see where it is getting stuck. If we are at the wait message, than that means we are through trigger and are waiting now waiting for the code in

nslsii/nslsii/detectors/xspress3.py

Lines 615 to 621 in 18c0632

	def _acquire_changed(self, value=None, old_value=None, **kwargs):
	"This is called when the 'acquire' signal changes."
	if self._status is None:
	return
	if (old_value == 1) and (value == 0):
	# Negative-going edge means an acquisition just finished.
	self._status._finished()

to fire.