April/May priorities: (Richard) See them all in Confluence.
GR releases (Heather) Note that improvement in the v13r11 series have already been propagated to v14rX. We apparently need some features from the latest astro but not all. (Toby) has already made a suitable branch tag. It includes code to intercept ill-formed quaternions and throw an exception, but disables the Swig module.
(Heather) has made a Confluence page to keep track of requests to upgrade packages for one or another of the GR branches. If you have such a request, add a comment to the page mentioning packages to be upgraded, tags for the packages, the branch (e.g. L1, Big Run, main development branch), and a reason for the request. Some day it would be nice to have something spiffier which will scale better, but this should reduce confusion considerably.
Pass 6 reprocessing (Richard) Tom pushed the button and.. (Tom) out of 1000 jobs, 340 crashed. This appears to be due to the nature of these particular jobs and in particular their impact on xrootd. Each one needs to read 100 merit files, 100 digi and 100 recon. (Richard) These jobs were reading the data directly from xrootd. One tactic used in the past which could be tried here would be to write them to local disk.
(Tom) The list of file paths is passed to the program as the value of an environment variable. Somewhere around 100 files this gets to be too long. Could change the implementation to write the list to a file rather than setting an environment variable.
Science Tools : (Eric W.) mentioned the following which concern releases he makes of Science Tools for distribution by GSSC:
Seth then went through the Science Tools Update for May 6th.
Data handling: (Dan) On Friday we reviewed how the newest version of the Pipeline II software was performing in DEV and TEST and decided that it was stable and performant enough to warrant moving it to PROD. Version 1.1.1 of the Pipeline software was installed into PROD Friday afternoon. The release included major performance enhancements in the stored procedures which handle process state-transitions and dependency evaluation. I have since been monitoring Oracle and accepting some of its recommendations to alter query execution plans to be more efficient. [thanks to Dan for this report]
Documentation: (Chuck) The next newsletter is underway as well as continuing work on How-to-fix. There are recent additions in the critical Applications section for FASTCopy incoming and FASTCopy outgoing. Tony Johnson is making a list.. (Richard) The goal is to monitor systems and detect incipient problems before melt-down. For example, we'll start pinging collaborator sites so we have early warning of network failures.
Skimmer (David C.) is investigating a problem Tom reported. It may require some consultation with ROOT experts.
New RM, SCons (Emmanuel) has been working on Mac SCons builds of Science Tools. He ran into one problem with an undefined environment variable which he was able to work around, but now there is a problem loading a library. (Eric W.) has successfully built Science Tools for GSSC for Mac, bot Intel and PowerPC hardware, and remembers encountering (and overcoming) problems similar to what Emmanuel describes. They will will be in touch.
(Navid) The new RM builds on Windows are now working. There are still some unit test failures but they also show up on Linux. He has managed to avoid hangs due to a dialog window popping up in batch jobs but the new behavior is far from ideal: the job fails with no output in the log file, hence no way to know what went wrong. His next item of business is to bring up the new version of SCons which has some desirable features.
gcc feature (Toby) A use of the string "bitor" as the name of an iterator caused the gcc compile to go haywire. Apparently "bitor" (and several other strings, like "or", "and", "bitand", etc.) are defined in the file iso646.h. [According to the Wikipedia entry it isn't even necessary to include the file; these strings are already keywords in ISO C++. ed.]
Studio oddity (Richard) was disconcerted during a debugging session when the file displayed did not match the code he was stepping through. The actual class was FileSpectrum, but the file displayed was for another class, Filespectrum [or do I have that backwards?]. Presumably Studio is using standard services, insensitive to case, to find the file and is finding the wrong file first. (Heather) Could we disambiguate with namespaces? (Jim) Other constraints prevent this. However, one of these classes is obsolete; we could just eliminate it altogether.
previous | minutes index | next |