Core Minutes 11/02/2004
Present: Joanne Bogart, Toby Burnett, Pol D'Avezac,
James Chiang, Seth Digel, Richard Dubois, Riccardo Giannitrapani, Navid
Golpayegani, Heather Kelly, Michael Kuss, Matt Langston, Chuck
Patterson, James Peachey, Dirk Petry, Robert Schaefer, Tom Stephens
-
Installer: (Navid) He has made a new
version which fixes the design flaw described in this
message to the Core Software List.
-
Memory leak plugged: (Navid) He found and
fixed a leak in Onboard Filter, perhaps the leak Julie had seen.
(Richard) Would be good to verify by running the particular System Test for
which the leak was easily detectable. Julie would know the details, but is
currently travelling.
- Pipeline:
(Richard for Dan) Pipeline was broken last week. The problem was traced
to a field too small to accommodate long names I&T was using. Dan
increased the field size in the database, but originally neglected to make
necessary changes to all scripts referencing the field. After this fix,
it seemed that 20-30% of Pipeline jobs were failing. In fact the jobs
were not failing, but the Pipeline couldn't always detect the true outcome.
Background: LSF (SLAC batch system) has a blocking mode
under which submitting process is supposed to be blocked until the job is
done. But "done" doesn't always quite mean, well, done, and the
lag between "LSF done" and done was often long enough that Dan's ad
hoc attempts to wait it out weren't always successful.
Discussions are now underway with SCS to get this fixed at their end;
meanwhile Dan is trying out alternate work-arounds . (Richard) In addition to
I&T use of Pipeline, would like to try out all-gamma and DC1-style runs.
- Pipeline configuration interface:
(Matt)
Read about about the XML files used to configure a pipeline instance. The
write-up includes an example file and the XML schema it must conform to. If
you're within the SLAC firewall (won't be necessary ultimately. We have done
our part; still waiting for SCS to fix this.) you can
see the page which allows you to upload such an XML file.
- New RM status: (Joanne for Alex,
Navid) The new RM, which has considerably more flexibility and extensibility
than the existing facility and will ultimately handle Windows as well as Linux
builds, is necessarily a complicated beast with many components. It needs to
be implemented and tested incrementally. Almost all the bits are there to
exercise it with a representative part of Linux-only builds, in parallel with
production RM. Pieces include a two-part database (one generic part to
supervise and initiate stages of activity and one RM-specific part), Perl
scripts to do the actual checkouts, builds, etc., a piece to handle LSF job
submission, and so forth. Before this test can take place the Perl scripts
which do RM application functions on Linux have to be modified to communicate
with the RM-specific part of the database. After the Linux-only design has
been validated with this test, incorporating the full functionality of the
current RM on Linux should be relatively straightforward, though far from
instantaneous.
The Windows implementation has additional wrinkles since there is no batch
system and builds are normally done interactively. Riccardo and Marco have
already provided a solution for the second problem in MRcmt, a script which
may be invoked from the command line to do the kinds of things (recursive
checkout, configure, build) needed by RM. (Richard) In fact LSF is advertised
to run under Windows. SCS hasn't yet figured out how to configure it properly
for SLAC, but this is in the works. If it succeeds the new RM could use it.
[Later addition from Alex: SCS tried to get LSF to work was tried some time
ago and failed; seems to have been languishing ever since. Probably the
project doesn't have much priority. Still later addition from Richard: it's
not entirely moribund. Randy Melen of SCS reports there has been some progress
recently.]
- ScienceTools checkout complete: (Seth)
Issues to be addressed include prompting, error messages, and conformance of FITS
files to standards. (James) It's recently been demonstrated that discrepancies
between FITS files created and standards are small to non-existent. Seth will
enlarge on
this summary at a later date.
- GSSC software testing: (Dirk) See
this
page for plans and status. (Richard) The Missions
Ground Systems manager (breaking a tie) has decided that
FASTCopy will be
used to move files from MOC to SLAC and from SLAC to SSC.
- DC2 Coordinating Committee: (Richard)
Julie will chair it. Proposal is for next steering meeting to take place
following next week's Core meeting, which should preferable be kept to an hour
or less.
- InstallArea fully disabled: (Toby)
Earlier changes didn't quite eliminate all of it [in particular, not lib
subdirectory. ed.] but the latest tag of GlastPolicy does.
∫ Flight(w, ..) dw
- EM status:
(Heather)
New tag
v3r0406p0,
made yesterday, includes, most notably, the new CalibGenCal. Will make another
tag shortly to pick up new EM2. She would like to get System Tests going, but
Julie is on the road. We need a sub!
- ROOT upgrade: (Heather) ..is perhaps
lagging somewhat relative to Xerces work; if so, Xerces changes should go in
first. Richard suggested contacting James discuss possible impact of upgrade
on ScienceTools, which Heather cheerfully agreed to do. Only tenuously related
to this were a couple remarks about PyROOT: Toby says it now can be used on
Windows; Richard reports that Berrie is turning into a fan (using it on Linux,
of course).
- Xerces upgrade: (Joanne) ..is much
less sure that Xerces work will be ready before ROOT, but progress has been
made since last week. Jim will take care of ScienceTools-only packages needing
conversion, which he says should be very quick. Joanne hopes to convert the
remaining 2 or 3 packages in a small number of days. However she has run into
a possible snag providing a suitable binary distribution of Xerces for
Windows. The Xerces project only provides Release binaries, built with
VS 6.0. Sample programs which are part of this distribution run properly
on a XP 2003 .Net system; she hasn't yet tried linking our code against the
Xerces library. But even if it seems ok, since it's Release rather than Debug
we would be running a risk: mixing Release and Debug can cause problems with
heap memory. So far her attempts to build Debug binaries from the source
distribution have failed, but this is likely to be operator error.
- LDF file validation: (Richard) We need
it! At Heather's request, Tony Waite has retrieved a program which does more
or less the right set of things for balloon data. The hope is that someone
from ISOC could use this as a guide to write something appropriate for LDF
files, though certain kinds of validation will have to be done by SAS.
- CAL: (Zach) New CalibGenCal has been
released and draft documentation (Doxygen
and
comprehensive guide) is available.
- TKR: (Leon) Of restructuring tasks, pattern recognition still
needs to be addressed. He, Tracy and Bill had a planning meeting and
have their marching orders. Work is underway.
Coming Soon
- Monthly Status Review: ..takes place
tomorrow. We're towards the end of the
agenda.
- Richard remote: He'll be working at
home for approximately two weeks, starting Thursday.
J. Bogart Last Modified:
01-Jun-2010 15:45:57 -0700