VDT Office Hours 01-September-2005

Attending:

• Nate Mueller (VDT)
• Tim Cartwright (VDT)
• Andrew Zahn (Chicago)
• Scott Koranda (LIGO)
• Leigh Grundhoefer (Indiana)
• Conrad Steenberg (Caltech)

Andrew called regarding a Condor admin ticket he had sent in a few days earlier. He also asked whether the VDT will work on 2.6 kernels. For the most part it will, but Scott mentioned that LIGO has had problems with Condor's standard universe and the most recent FC3 kernel update. We have a workaround can help Andrew out if he needs it.

Scott had two issues. First, he wanted to know if we could install MySQL in a way that it wouldn't interfere with one that's already running. This is easy to do for RLS (what Scott cares about) but trickier for other services since Java can not talk to UNIX sockets. Scott sent in info about how LIGO is currently handling this and we'll try to do something similar in VDT 1.3.7. Scott also asked about the possibility of using Pacman to update a single package. Now that we've switched to Pacman 3 syntax this should be possible, but we're still evaluating it.

Leigh had a long list of questions:

• $VDT_LOCATION/globus/tmp is mode 1755 instead of 1777. This should be fixed, but there might be unknown side effects.
• GT3 puts some GRAM job state files in $VDT_LOCATION/globus/tmp but we do not rotate them. We'll investigate with GT4 and see what we can do.
• Leigh brought up the possibility of running the GT4 web services under Tomcat instead of the custom Globus container. This is an attractive option, and something we've thought about, but it's too early to say when it can be done. VDT 1.3.7 will most likely use the default Globus container.
• Several questions about CEMon from the 1.3.7 pre-release cache. It looks like there are some problems, but that's not entirely unexpected -- work on CEMon is not yet done. We will investigate the issues she reported and make sure they're fixed before CEMon is released.
• OSG now wants to install a subset of the VDT-Gatekeeper packages. Specifically leaving out Condor and CEMon. CEMon is still new and unknown and OSG assumes that sites have a working batch system install before installing the gatekeeper. We removed CEMon for now and are working on a way to let OSG skip the Condor install. This will most likely require a new Pacman release.
• Michael Thomas has developed some scripts to advertise VOMS and other web-services to the jClarens discovery service. Leigh wondered if we could include those and augment them for other services like GUMS. We will talk to Michael and see how difficult this would be.

Conrad called in to discuss the state of Clarens. I (Nate) have been out of the office lately so not much has changed. Conrad had done some more debugging, found some more problems and fixed all but one of them. As we see it now, Clarens should fall into place as soon as I rebuild Python.