Core Minutes 2/21/2012

• ScienceTools: (Jim) made a new ST tag, 09-27-01, and all appears to be well. This tag uses the new version of cfitsio. See Science Tools Development Notes for details.

• glastlnx07 (Heather) There were more problems with this machine, but things have been better since the last outage. (Tom G.) It's been ok since the memory transplant, which also involved doubling the amount of memory.

• rhel4 lingering life (Heather) We've purchased an extended license for one year for rhel4 on glastlnx14. She believes everyone who needs access now has it. If not, let her know. And those who have been given access should make sure they can log on.

• MySQL migration (Heather) This move, which has been in the works for the better part of a year, is finally upon us. Originally we were going to upgrade from MySQL 4.1 to 5.0; now 5.5 is the target. The machines involved are glastlnx01 (aka glastDB, aka glastCalibDB) and glastlnx02. See a list of affected databases in Confluence. The ones which are predominantly our responsibility are those used for calibrations and those belonging to CMT RM (SCons RM databases are already on mysql-node03 which is running MySQL Server version 5.0). (Joanne) It is possible to re-direct jobs needing access to the calibration database to a different node by means of job options. Once the database has has been copied, I'll run a test program, directing it to the new database. Then Tom can do the same with a production job or two. (Heather, Michael) The CMT RM is a more complicated case since processes normally write to it much more frequently than is the case with the calibration database. Some care must be taken to be sure no writes take place after the databases have been copied but before we have switched over to using them. We can easily avoid HEAD and release builds during critical periods since they are initiated manually. LATEST is more of a problem. It would be best to eliminate any activity connected with LATEST by temporarily commenting out relevant trscrontab settings. We don't have the ability to do this ourselves. (Tom G.) But we can ask unix-admin to do it for us.

  (Heather) has received a list from Arash of databases and accounts. He would like to eliminate any that aren't in active use. Michael has been trying to determine if all the CMT RM databases and their tables are actually needed, but in MySQL 4.1 no general-purpose tools are provided to determine, e.g., when a table was last written. (Joanne) Depending on when we're expecting to drop CMT (3 months? 6 months? longer?) it might not be worthwhile to expend much effort on finding the odd unused table.

• Disk — post-meeting addendum (Tom G.) The new Dell storage arrived at SLAC on Friday. It will likely take a couple of weeks for it to be put into production.

• Pass7 reprocessing First sample (about a month's worth) is done as is AGN skim, which Art is looking at.

  State of truncation software (making old-style data from new) is the same as last week: works on Windows, crashes on Linux.

• Pass8 (Tracy) Last week Heather made a new new GR tag, incorporating patches for the most pressing problems detected in the previous new GR tag, in particular a correction for track energy and a bug fix from Philippe in profile fit code. Stefan redid MC datasets within a day; we're set for the Pisa meeting and expect to be in a steady state for some weeks.

  (Tracy) was getting a bus error. The problem was fixed by properly updating an old job options file. This might have some relevance for the crash Leon is seeing on Linux.

• Truncation runs (Leon) There will be a CCB meeting on Thursday at which a truncation scheme will be proposed. Then some time around March 1 we'll do a very short run (10-20 minutes) to make sure the change of configuration goes smoothly. If so, at a later time we'll do a series of runs with normal configuration, include some limb pointing and then at a still later time (when position and orientation are similar to the time of the first series) do another set with the new configuration.

• Core talk:

  • rhel5 migration and testing (Heather) would like to make a systest comparison between CMT and SCons builds of the new P8 tag, GR 19-04-01-gr13, as was already done for an L1proc GR tag. Next up would be to compare rhel4 and rhel5 SCons builds.

  • Observer pattern tamed (Heather) Randoms and source selection are now working properly for new gaudi builds. There is still a problem with event display, however. Both Fred and WIRED fail to come up, with slightly different symptoms (Fred gets a little further). We're using a slightly newer version of OmniOrb with the SCons builds: 4.1.4 rather than 4.1.2. That might be a factor. However, even without a working event display we can validate batch behavior and give M-E a tag she can build her stuff against.

  • u30 disk (Heather) Michael has cleaned it up. It's now down to 72% full, less than it's been for some time.

  • SCons RM (Tom S.) It's doing well. There were problems last week with Windows builds: CVS check-outs were failing, even though the same commands worked from an interactive session. glast-win04 was rebooted and that apparently fixed the problem.

  • GR vc90 stumbling block (Joanne) learned a little more about last week's problem (test_Gleam crash). The crash occcurs in TkrRecon initialization, specifically on this line:

    TkrCovMatrix(int p, int q): CLHEP::HepMatrix(p,q) {}

    where p and q are both 4. Since there is very little going on in C++ I tried stepping through the assembler. My best guess is that there is disagreement about the layout of one of the structures so that a field which is expected to be a pointer doesn't have a pointer-like value. If that's correct, it could have something to do with the vc90 CLHEP external. There are many warnings at link time about a missing pdb file. Whether or not this is relevant, it would be good to figure out how to eliminate those warnings.

  • CMT RM (Michael) Assuming unix-admin will update trscrontab for us, we still need a way to test RM against the dbs in its new location. To start, he needs to find all places where the host is specified.

previous

minutes index

next

J. Bogart, Last Modified: