Core Minutes 9 April 2013

Core Minutes 4/9/2013

Present: Joanne Bogart, Jim Chiang, Warren Focke, Tom Glanzman, Heather Kelly, David Landriu, Leon Rochester, Tom Stephens, Tracy Usher

ScienceTools: (Jim) There are two new items on the Science Tools Development Notes page: one concerning an enhanced Pipeline interface for some apps; another, described in detail in JIRA LK-85, concerning default contents of output source maps.

Jeremy was here last week. He has signed up for various tasks, including adding doc strings in pyLikelihood.
Reprocessing (Tom G.) FITS file generation is still on hold, awaiting new interstellar emission model and IRFs.

There have been rumors about ACD calibrations which, if true, would imply that we have to redo all reprocessing since last September.
Pass 8 (Tracy) GR 20-08-12 exists! We need Bill to be able to do tuning work and there is not time to get him set up on fermi-ts with SCons builds, so it should be installed on glast-ts as CMT build. (Heather) This could take some work but she'll try.. She will install the SCons build on fermi-ts as well.
Core talk:
- Proposed new externals containers (Joanne) There are various ways we can handle the sym links in the CVS repository linking packages to our application containers (the ones for GR and ST). The simplest — just delete the ones we no longer want in the ST and GR containers — is not going to fly because then we would no longer be able to check out old tags. The best option seems to be to create new ST and GR containers with only the sym links we need (excludig those packages which will be in the new externals container or containers) — unless this would cause trouble elsewhere, e.g. for RM. (Tom S.) doesn't see any serious problems with this approach for RM. (Heather) The Installers would have to know about the new names.
  
  (Joanne) will see about building and using the new external or externals in a local sandbox. To start she will put irfs and utility packages in just a single external. There don't seem to be any strong arguments for separating into two. Containers which need the proposed set of utility packages also need irfs.
- Documentation (Heather) is about to make new prod Workbook release, including changes for SCons/GoGui etc. (Joanne) has updated documentation for SCons, GoGui and the extra recipe page for Windows users, but these things are spread around in several places (Workbook, Confluence, stand-alone page for GoGui) and not very easy for a new user to navigate. Still to be written: a bit of fermi-ts-specific documentation. It is probably not feasible to unify everything in the Workbook (without Chuck, alas).
  
  (Tom G.) Enhanced documentation on batch usage is needed as well. There was a melt-down of u55 last weekend and the machine had to be rebooted. Before that happened Ganglia logged lots of "nfs bad calls" messages. The cause ultimately was a full partition. Someone had submitted a bunch (not that large a bunch, either — around 150) of batch jobs. There was not enough space on their user disk for the output, causing jobs to crash and create core dumps, further exacerbating the problem. Recommended practices to be documented:
  - Check disk space intended for output before you submit
  - Consider setting max core dump size to 0 (in which case no file will be created at all.
- Windows SCons support (Leon) Joanne has been very responsive; we're nearly there. (Joanne) made a new release of GoGui, 1.4.0, in which the Visual Studio button behaves more nicely. If the top-level GR container (base or supersede) is selected, VS will start with all.sln. If Gleam package is selected it will use allGleam.sln.
- Windows, RM and Jenkins (Tom S.) When the builds try to run normally in Jenkins (which runs as a Windows system service) they fail. However if Tony logs in and runs the Jenkins script by hand it works fine. He's trying to track down why that is.
  
  Then, apart from Jenkins, there is the problem that several Windows test programs fail and put up error assertion dialogs. Since, typically, there is no one logged on to acknowledge them, they sit there for 30 minutes before timing out, adding hours to the build time. He's found a way to avoid putting up the assertion dialog but it would involve modifying the code. To the extent that the failing programs are derived from a single lower-level class, it could be done all at once. All the errors have to do with vector iterators with messages like "vector iterator not dereferenceable" or "vector subscript out of range". (Jim) Why don't we see problems on other platforms? He's willing to investigate, but has no access to a Windows platform. [That is now being addressed.]
- Overlay (Leon) The xml file that tells Gleam where to get the overlay.root files is in xroot, but Gleam can't read xroot. This leads to confusion, because all the xml files have the same name (McIlwain_L.xml).
  
  With the pass8 overlay package, we have the option to rename the xml file to be more informative about where it came from, like McIlwain_L_4.xml to go with OverlayMaker4 (aka "files for Pass8"), which takes care of the problem there.
  
  (Heather) just made a new L1 GR (17-35-24-lp41, then 17-35-24-lp42) incorporating all the latest Overlay code, so we should be in good shape there as well.

minutes index

J. Bogart, Last Modified: