Core Minutes 8 February 2011

Core Minutes 2/8/2011

Present: Joanne Bogart, Anders Borgland, Jim Chiang, Elizabeth Hays, Heather Kelly, Michael Kuss, Chuck Patterson, Leon Rochester, Tom Stephens, Tracy Usher, Eric Winter

ScienceTools: (Jim) There are two items since last week:
- The new interface mentioned previously has been tagged though so far is only in LATEST.
- Minor bug fix to gtexpcube2 which will not affect most people.
See entries dated February 4th in the Science Tools Development Notes for details
FSSC: (Tom S.) The Data Server is now out of quarantine! However other dynamic scripts still have to go through the process.

They've set up and internal web server for testing.
Documentation: (Chuck) sends the following:

I created a new tag last week, and Berrie has updated the mirror site. The Workbook is now at 08-00-02, and I'm continuing to update the Science Tools.
Pass8: (Tracy) Everything is fine! Various datasets — overlay, fake overlay, mega all-gamma... — are being churned out with GR v18r8p6. It's all in good shape for the Face-to-face in Pisa next week.

(Leon) The meeting will have a format similar to the last one: status talks in the morning followed by working sessions in the afternoon. By the end of the meeting we aim to develop a plan for future work, as was done at the last face-to-face.
WIRED: (Heather) She met with Leon, Tracy, Tony J and the new guy, Dmitri. (Leon) Dmitri will refactor WIRED in order to make it less spaghetti-like. Once this is done (no time estimate yet) he expects that implementing all the features people want will go very quickly.
Core talk:
- Chiefly CMT RM (Heather) Over the past week, we have continued to have troubles with the ReleaseManagers, and in particular with CMT RM Here are this week's highlights:
  
  Again lost the ssh files stored on the Z: drive - this prevented windows builds from going through for either SCons or CMT. This has been fixed.
  
  Filled up the V:\Glast_Software\Toaster area which contains externals, and windows builds for both SCons and CMT. When we delete builds, the database entries are cleaned up, but the actual build may or may not be cleaned up as well. This is true of the automated cleanups that happen for LATEST builds, as well as when we run the deleteBuild script by hand. There were a number of old builds, particularly in the CMT build area, that have since been deleted. Heather is talking to Charlotte about adding this area to nagios monitoring.
  
  Although the new glast-win04 box is set up for SCons builds and has been added to the glastwinq queue we cannot yet run CMT builds on it. Kim is working with windows-admin (particularly Rodney) to get the permissions necessary to install the software need to set up CMT RM. In the meantime, we must prevent the CMT RM from submitting jobs to glast-win04. Kim inserted the necessary -R option, but later we realized there is already a -R option being specified in the same bsub command. The result is that the submission fails. However, CMT RM is set up to resubmit jobs when there is a failure. (This was done to handle the case where a host goes down, and we desire to rerun the same build on another running host.) In this case, the submission always failed, and the RM continued to resubmit the build. This in turn filled up the /tmp and /var areas of glastlnx20 and glastlnx14, due to the repeated submissions...and the load on those boxes was abnormally high. That brought CMT RM to its knees — no builds of any kind could proceed. We have since killed the offending jobs, and Kim cleaned up the jobs table to prevent the jobs from being resubmitted. /tmp is automatically cleaned up and that seems ok now. Kim is reworking the bsub command for windows and in the meantime, windows CMT RM builds are offline until further notice.
- SCons RM (Tom S.) All clear on this front. Today he plans to properly define the environment variable LATCalibRoot (probably in glastrm's .cshrc or .bashrc) for the benefit of certain GR test programs which are failing due to lack of a definition. This will take care of Linux builds; something else needs to be done for Windows.
- Snow Leopard (Tom S.) He still has to work around the problem of lost AFS tokens by logging in periodically. The token loss is correlated with the clean-up of cron jobs.
- TMine external (Joanne) All is well for Linux and Windows, but Eric C. was unable to build it on Mac. There are at least two issues, one minor and one more critical.
  - For SCons builds on most platforms, our code will compose a suitable default variant name, but it doesn't do the right thing for snow leopard. This shouldn't be hard to fix. Meanwhile one can work around this by specifying the variant explicitly with the --variant option, as SCons RM does.
  - Most links are failing, perhaps because certain of the TMine libraries are being created with no extension at all, which is not what libraries depending on them expect. The lack of the proper extension (.so) could be because some set-up specific to macs has traditionally been associated with the python external, and TMine doesn't use this external. That set-up can be done elsewhere for TMine. [Update: Eric added the extra set-up, but that didn't fix the lack-of-extension problem.]
- GoGui (Joanne) Builds of GoGui 1.0.0 for Windows and Linux are now available for download. On central SLAC linux machines with standard set-up the command GoGui will resolve to this version. The changes between this version and the preceding are not great. The most notable is the ability to specify arbitrary SCons options from within GoGui. Documentation has been brought up to date.
- SCons meeting Wednesday at 10:30 as usual.

minutes index

J. Bogart, Last Modified: