Online Recon - Expert

From hps run
Revision as of 15:39, 3 August 2021 by Baltzell (talk | contribs)
Jump to navigation Jump to search

Servers

The core HPS java software builds for online, multi-threaded reconstruction live in ~hpsrun/online_recon and were installed using the instructions on confluence. We may decide the hps-java installation there should be a symlink to a standard one elsewhere, but note it requires an extra installation step and (currently) a non-master git branch. If the tomcat servlet needs updating, it will need to be rebuilt and redpolyed.

Everything operates as user=hpsrun, and the startup for all server-side stuff (except tomcat) is configured in $EPICS/apps/iocBoot/procServ.conf. There are 4 components, each of which can be connected to in telnet sessions:

  • softioc_console dqm_et (just an ET ring, offline only)
  • softioc_console dqm_evio2et (pipes an EVIO file to the ET ring, offline only)
  • softioc_console dqm_server (receives instructions from the client below)
  • softioc_console dqm_client (instructs the server on parameters, e.g. #threads, steering file)

Configuration files for all are currently in their startup directory $EPICS/apps/iocBoot/dqm and require

A full teardown and restart of all components, including the tomcat server, is in user=hpsrun's $PATH:

  • hps-dqm-restart.sh

TODO

  • Can the webapp be modified to automatically reload the plots periodically, without the user manually refreshing the web page? This will be important to give users feedback that things are working (and no one wants to have to reload the page). 1 Hz would be great, every few seconds would be good enough.
  • The stop/start functionality in the server appears maybe unreliable, and the tomcat servlet also appears to need to be restarted if the recon server is restarted. These things will need to be done in order to zero the histograms, which will be every run and every time beam is away for a significant amount of time. Meanwhile, a script to do a full teardown and restart is available, but it takes many seconds to ensure things are started with appropriate, intermediate delays.

Clients

Both should work from any clon machine: