TCBs

[rerpha]

In a data streaming system, a number of systems could potentially need time-channel settings:

• Monitor hardware, if they are emitting histograms from hardware.
• kdae_diagnostics, for serving spectra plots
• AreaDetector / Live view implementations
• Filewriter, if the filewriter supports taking in event-mode data and writing out histogram-mode data. Note: it is not actually clear that we want to support this.

We need to decide whether we support settings TCBs or not. even if we choose to only support event mode for HRPD-X, we still need to provide TCBs for spectra plots.

we could serve these by borzoi/kdae_control over EPICS if needed so there is a central place to change them.

[Tom-Willemsen]

One question to ask is where the responsibilities lie in ensuring chosen TCBs * spectra * periods are a reasonable number for downstream programs that are (presumably) using these TCBs to histogram the data.

On one hand, different downstream programs might have different limits for memory they're willing to allocate (and therefore TCBs they're willing to use). On the other hand, those programs don't necessarily have a good way to feed back a descriptive error to the user and stop the run if they can't allocate a large array for a histogrammed version of the data.

I think we should consider putting some of these checks in kafka_dae_control where it doesn't let num_periods * num_spectra * num_time_channels * 4/8 bytes exceed some limit defined in a kafka_dae_control config file (probably somewhere around a few GB for most instruments, or hardware-dependent if they're also being used by FPGA hardware).