GlueX Software Meeting, January 7, 2020
GlueX Software Meeting
Tuesday, January 7, 2020
2:00 pm EST
JLab: CEBAF Center A110
BlueJeans: 968 592 007
- 1 Agenda
- 2 Minutes
- 2.1 Announcements
- 2.2 Review of Minutes from the Last Software Meeting
- 2.3 Review of Minutes from the Last HDGeant4 Meeting
- 2.4 Rebooting the Work Disk Server
- 2.5 Report from the Last SciComp Meeting
- 2.6 Fix for Recent Raw Data Runs
- 2.7 New Farm Priority Scheme
- 2.8 Review of recent issues and pull requests
- 2.9 Recent Discussion on the GlueX Software Help List
- 2.10 New Meeting Time?
- 2.11 Action Item Review
- Review of Minutes from the Last Software Meeting (all)
- Review of Minutes from the Last HDGeant4 Meeting (all)
- Rebooting the work disk server (Mark)
- Report from the Last SciComp Meeting (Mark)
- Fix for recent raw data runs (from Sean)
- New Farm Priority Scheme (Mark)
- Review of recent issues and pull requests:
- Review of recent discussion on the GlueX Software Help List (all)
- New meeting time? (all)
- Action Item Review (all)
- CMU: : Naomi Jarvis
- JLab: Alex Austregesilo, Mark Dalton, Mark Ito (chair), Igal Jaegle, David Lawrence, Keigo Mizutani, Justin Stevens, Simon Taylor, Beni Zihlmann
There is a recording of his meeting on the BlueJeans site. Use your JLab credentials to access it.
- New version set: version_4.12.0.xml. This version set is suitable for analyzing data from the Fall 2019 run.
- new version sets, new simulation (halld_sim-4.11.0), old reconstruction. These versions sets use the same version of halld_sim as in version_4.12.0.xml, but with old version of halld_recon, versions used in previous analysis launches. New branches of those old versions were needed to accommodate the new particle_type.h used in the latest halld_sim version.
Review of Minutes from the Last Software Meeting
We went over the minutes from December 10.
- Mark has made some progress on a path forward for conversion to Python 3. More correctly, he found there is an easy way to use Python-2-compatible SCons on systems with Python 3 as the default and vice-versa. It turns out that at least for RedHat-like systems, modern distributions ship SCons in both flavors.
- There has been no news on the release of CCDB 2.0.
- The RCDB errors reported at the meeting were solved.
- The problem reported with using CCDB from hallddb-farm still appear to be with us. Jobs hang forever at start-up when accessing calibration constants.
- The problem that Mark Dalton reported with genBH was due to a bug in HDDM that Richard has since fixed.
Review of Minutes from the Last HDGeant4 Meeting
We went over the minutes from the meeting on December 17. Alex has done work exploring the effect of widening the timing cuts on charged hadrons which hit the BCAL. See the discussion below in the section on halld_recon pull requests.
Rebooting the Work Disk Server
The reboot of the work disk server, done on December 17, seems to have fixed the file locking problems we have been experiencing for months now. Brad Sawatzky from Hall C also reports success with his builds, builds that would reliably fail in the recent past.
Getting all of the Halls to agree to bring the disk down took some convincing given that any benefit was speculative. Kurt Strosahl of SciComp did see some anomalies while doing some pre-reboot diagnostics. There is no proof that the fix is permanent; we will have to continue to monitor the server.
Report from the Last SciComp Meeting
Mark led us through his notes from the meeting on December 19.
- We are on track for a 4 PB expansion of our Lustre capacity for Experiment Nuclear Physics (i.e, all Halls).
- There will be work this summer to transition to a single tape library. We have two at present. With two tape bandwidth can get siloed (see sense 3).
Fix for Recent Raw Data Runs
Mark I. highlighted Sean Dobb's recent fix for anomalies found in the raw data from last Fall. Naomi pointed out that there are actually three such anomalies related to the the new firmware used in the Flash-250s. David addressed one of them in halld_recon Pull Request #247
Mark D. has compiled a collection of information on this and related issues.
[Added in press: Mark D. reports that currently all FADCs are running with the latest version of the firmware (Version C13).]
New Farm Priority Scheme
In December, several folks reported anomalies in how gxproj accounts were being scheduled on the JLab farm. Starting yesterday, a new Slurm configuration for farm job priorities was rolled out. Priorities are assigned in a hierarchy such that resources at each level are apportioned according to "shares," and groups that are inactive at any given time have their shares apportioned among the active groups. The following figure shows the hierarchy and the current share assignments.
N.B.: Production accounts must use the the "gluex-pro" project to the the benefit if increased priority.
Beni reported seeing extremely long retrieval times for tape files in December.
Review of recent issues and pull requests
- halld_recon Issue #257: "Broken beam bunch selection with wide timing cuts." Alex reported on recent work in studying the effect of the width of timing cuts in the BCAL on the efficiency calculated in Monte Carlo for ρ events. There has been a long-standing problem in the efficiency comparison for this topology between Geant3 and Geant4 when plotted as a function of beam photon energy. Using a very wide cut (±5.0 ns versus the standard ±1.0 ns) he saw the two efficiencies come together, but the overall level drop. He traced this to a bug in the analysis library where the RF bunch assignment, in the case of multiple acceptable RF bunches, was based on the assignment with the highest χ2 rather than the lowest. After correcting the code the overall efficiency level rose to a level higher than that with the narrow timing cut, as expected.
- halld_recon Issue #256: "ReactionFilter plugin is not working with an (atomic) electron target." Igal reported a problem where no events were selected by the ReactionFilter when run on Compton events from his recently installed generator. The simulation was done with no magnetic field and the resulting straight tracks from electrons were all being assigned a positive charge (the default when the charge cannot be determined) and so events appeared to be electron-free. There was an extended discussion of how to reconcile this problem; the discussion will continue offline.
- halld_sim Pull Request #102 "Ijaegle primex evt" and halld_sim Pull Request #104: "Increase gen. config. length name." These pull requests from Igal add an η generator and a double Compton generator respectively to the repository.
- halld_recon Pull Request #253: "Add trd to fit." Simon has been working to add TRD data from last year's DIRC commissioning run to the track fit. This is work in progress.
Recent Discussion on the GlueX Software Help List
We looked over the list without significant discussion.
New Meeting Time?
Teaching schedules have changed. We will move this meeting and the HDGeant4 meeting time to 3:30 pm. They will remain on Tuesdays.
Action Item Review
- Figure out how to do build on a system with Python 3 as the default.
- Look at getting CCDB constants from hallddb-farm.
- Develop a plan for calculate efficiencies when timing cuts on charged tracks hitting the BCAL depend on distributions with long, not-well-understood tails