Collaboration Meeting Minutes (14-March-2012)

XRootD Collaboration Meeting held at CERN 14-March-2012

Attendees: * Doug Benjamin (Duke) * Dirk Duellmann (CERN) * Gerardo Ganis (CERN) * Andrew Hanushevsky (SLAC) * Lukasz Janyst (CERN) * Danila Oleynik (JINR) * Andreas Peters (CERN) * Artem Petrosyan (JINR)

Discussion Topics (No particular Order)

New Client status

Mostly done but some functions are not yet implemented, more work on proper timeout handling is needed and the client relies on the use of root’s TTree Cache.
We discussed some pressing requirements desired by all of the experiments, especially in a federated environment. We decided that the priory would be:
- Timeouts for single source reads with a switch to another server (about 3 months) and
- Multi-source random reads with single source timeouts driving it (about 6 months).
There was a lot of discussion on how to transition to the new client. Specifically, just give people a deadline to move, support the old client indefinitely (only serious bugs fixed), provide an old-style client wrapper to the new client.
There was some agreement but no consensus. It’s still not clear how difficult it would be to devise an old wrapper for the new client to ease transition. With no wrapper, we would set a 2 year timeline to switch to the new client when it is released. We would likely support the old client for two years after that with no upgrades and only fault conditions fixed. Everyone thought this was reasonable but yet fully flushed out to commit to this scenario.

xrdcp

As part of the move to the new client, the front-end to xrdcp is completely revised in a way that can be used by the old client and the new one. This provides the identical user interface regardless of which client is being used (deemed a requirement).
Creating a new interface that uses common interface standards proved to be challenging as the old interface used some esoteric conventions. The solution was to support the esoteric as legacy options but otherwise just document the “standard” option list.
One conflict arose that was not clear how to fix and that was use of the -force option to allow the client to try the copy for about a week before giving up. We decided it wasn’t worth keeping but we should check.
Action Item: AH will e-mail Costin about whether the -force is still needed.

Versioning

Once we switched to using shared libraries, corresponding to EPEL guideline, people not using RPMS have had problems mixing and matching executables with the correct library versions.
We decided that the real problem was that we didn’t have a fool-proof mechanism to determine when a shared library evolves in a binary incompatible way.
Action Item: LJ will bump the so name for every master release. No binary incompatible changes will be accepted to patch releases.

Git hub

There was a long discussion of whether to move the project repository from SLAC to GitHub. It would raise visibility but that brings its own set of issues. The new client is already there. We decided to continue internal discussions at each of our respective sites to determine the ups and downs of such a move.

Monitoring

Discussion of how work would be split between JINR and UCSD. The approach is that, by large, JINR will focus on summary monitoring with some detail traffic included (e.g. file copy records). UCSD would concentrate on detailed monitoring.
There is some overlap and it’s not clear how we handle that. Specifically, we need plug-ins for rending agents (i.e., ganglia, Mona Lisa, etc) and plug-ins for monitoring record storage (i.e., Cassandra, root, etc). What was recognized is that for this to work smoothly, a common record format must be developed for each set of plug-ins.
Action Item: AH will ask Ian Bird for standard monitoring record format document circulating in the WLCG. As we are moving in the direction discussed at the Lyon Federation Workshop, it would be good to use the format if possible.

Third Party Copy Status

The server-side of 3rd party copy works and we need xrdcp to provide the correct cgi information to trigger the copy. This is tied up with the new xrdcp front-end (see above). It should be done within a month.
The interface is http friendly so it can be carried forward to http servers as well.
While there was a lot on interest from Atlas and CMS, that has waned with only Alice and CMS/Castor expressing great interest. So, we move forward.

Licensing

We decided that we would move our license to an LGPL model on the next release, subject to the action items below.
Action Item: LJ will provide the license text (done).
Action Item: AH will consult with SLAC lawyers to get this moving to completion.

Logging

This issue surfaced once again, triggered by RAL’s interest in deploying xrootd for a large project. We decided to address the concerns by:
Action Item: AH will put in support for the logrotate command for sites wishing to use this mechanism as opposed to the native mechanism.
Action Item: AH will factor in the desire to have machine parse able logs vs. human readable logs. This is always a problem with tensions between large sites and small sites. It may be possible to include key-value information as an option. This is clearly tied into the larger project of redoing log messaging.

RPM’s in Epel & OSG

We’ve always had historical tensions on where and how distributes RPM’s. Some rely on the EPEL repository while others on the OSG repository which is identical to the official RPM’s are hosted at SLAC.
To make things go smoother, LJ got the appropriate accounts to create fedora builds and make RPM’s compatible with EPEL. This should be done in 2-3 weeks. This will likely remove OSG out of the loop. However, OSG does the DSI plug-in packaging. So, this may be a stumbling block.
Action Item: LJ should become EPEL package co-owner to speed package deployment.
Action Item: AP will provide the DSI plug-in for RH6 (OSG does not) and as a separate RPM. This will provide a smooth transition.

Python vs Perl

Recently, a user got an unsupported Python binding to the xrootd API and implemented site critical services with it (never mind that was bad form). In the latest release, that binding no longer works because we moved some API’s to different libraries to solve other problems. Now, we are getting requests to resolve this problem. We have a Perl binding and feel that we cannot continue to support both. The group preferred the Python bindings.
Action Item: AH will mail to xrootd-l asking which one we should support. Ask who uses what and indicate we can only support one.

IPV6

While we see the world moving to IPV6 we see that movement so slow that this appears to not be a critical issue at this time. However, we decided to start putting in the basic blocks to make a smooth transition.
Action Item: AH will create a new IPV4/6 XrdNetAddress class and integrate it in the current code (it’s a significant effort). This will provide a firm foundation for LJ to do the development. Our time-line for full IPV6 support is 3Q 2013.

Simple Example Configurations

Many people have requested simplified templates for configuration files dealing with a simple site.
Action Item: DB will head this up.

Xrootd Protocol Plugin

GG requested that an so version of the xrootd protocol be created so that he could use it as a plug-in. We agreed that we would build an so version of the protocol as this is pretty trivial to do.

Miscellaneous

Action Item: LJ will create an automatic monthly telecom reminder e-mail. We will have a telecom only if there something we need to discuss as a group.

Next Meeting

Next collaboration meeting will be held at CERN during the week of June 11th (in conjunction with Atlas software week). LJ will get a phone bridge going so other can remotely attend.