...

# This `batch_size` should be set to a small number (e.g. 1)
# since all other events which are part of this intg. event will ben sent
# in the same batch.
batch_size = 1

from psana import DataSource
ds = DataSource(exp='xpptut15', run=1, dir='/cds/data/psdm/prj/public01/xtc/intg_det',
        intg_det='andor',         
        batch_size=batch_size)
run = next(ds.runs())
hsd = run.Detector('hsd')
andor = run.Detector('andor')

# Test calculating sum of the hsd for each integrating event.
sum_hsd = 0
for i_evt, evt in enumerate(run.events()):
    hsd_calib = hsd.raw.calib(evt)
    andor_calib = andor.raw.calib(evt)

    # Keep summing the value of the other detector (hsd in this case)
    sum_hsd += np.sum(hsd_calib[:])/np.prod(hsd_calib.shape)
    
    # When an integrating event is found, print out and reset the sum variable
    if andor_calib is not None:
        val_andor = np.sum(andor_calib[:])/np.prod(andor_calib.shape)
        print(f'i_evt: {i_evt} andor: {val_andor} sum_hsd:{sum_hsd}')
        sum_hsd = 0

Deadtime and Integrating Detectors

Sometimes data for a particular shot can not be read out because DAQ data buffers are unavailable.  This is called "deadtime".  Deadtime can make it impossible to reliably use high-rate shots to perform operations like normalizations for lower-rate integrating detectors.  Matt Weaver is adding counters to the timing system so the DAQ can count how many shots whose data has been dropped due to deadtime so experimenters can make a decision about whether there is sufficient data for a normalization-style analysis.

AMI and Integrating Detectors

AMI is also affected by the DAQ deadtime issue described above but, in addition, events can also be thrown away going into the AMI shared memory (this feature is what allows AMI analysis to stay real-time).  This can cause additional issues with AMI integrating-detector normalization analyses that try to use high-rate shots to do normalizations for low-rate integrating detectors, since not all high-rate shots are guaranteed to be available.  Also, unlike psana, AMI doesn't currently have the idea of shots that "belong together" which will make an AMI normalization-style analysis of an integrating detector impossible.

MPI Task Structure To Support Scaling

...