Collaboration meeting



# ARGUS Collaboration Meeting - 9/10/2015

## General News

* Selection process restarted for the hired person that did not come
* The other person (Marco) has started and will work partly on ARGUS

* UMD-4 still being build: small delay
  * Internal testbed will use Puppet
  * BDII already verified
  * First announcement postponed

## EL7 and Java8 Support

Update of dependencies (Jetty, VOMS, Canl) completed for all components
* No stress tests done yet
* To build the real packages, need to push the last version of dependencies to the repos
* Packages available for beta testing
  * CERN ready to do it in some production-like environment
  * Put these packages in UMD-4 repos as soon as we are confident there is no known flaw: need to provide an automatic test suite
* No YAIM config provided with the new packages: need to rely on Puppet, Quattor or something else
  * Configuration files not changed since last version
  * Need to be stated clearly for sites who may want to migrate
* No functional change: only a few minor fixes

Client/server compatibility matrix using the ARGUS test suite
* Still fixing some of the tests that seem impossible to succeed
* Run through a Jenkins instance
* The new person at CNAF also spotted a few changes required: would be good to coordinate

## Open Issues - New needs

CERN issue
* Hypothesis last month was that CMS VOBOX was calling ARGUS with its own cert (no VOMS extension) rather than the user cert, confusing/overloading ARGUS... Almost clear that this is not the only trigger.
  * Cause for the CMS problem remains to be understood... still tuned
* Recent analysis showed that an additional problem was that when the ARGUS server was unresponsive, CREAM would remember it during a significant period (up to several hours if unlucky).
  * when CREAM was complaining about impossibility of getting a pepd answer, using the pep client would work on the same machine...
  * Problem in the pep Java client? connection reused?
  * Some tuning on one CE was done: need to wait next incident to see the exact impact
* ARGUS overload seen recently due to many WNs contacting the server at the same time (glexec): may be connected to some application workflow, to be analyzed
  * NFS used for the shared gridmapdir may be the bottleneck in this kind of situation: in this case, ARGUS is more a victim than a culprit, may require improving ARGUS to use a database rather than the legacy gridmapdir. In fact planned in INDIGO.
  * NFS/gridmapdir is not involved in the CMS VOBOX problem

Andrea : the ARGUS Java client library may rely on a http client library that may be old and not used by any other component. Should look at it
* First guess: the problem seen on CREAM at CERN may be related to a connection pooling problem

Extending the policy engine to support authz in IOTA context (see Misha's email after last meeting): decision based on VO+CA
* Andrea: should be no major problem to add the support for this. Need to agree on what we want to do exactly.

## AOB

Add Marco to GGUS support unit and to ARGUS future discussion list.

Next meeting: November 6, 2 pm CET
* Agreement to continue with this monthly meeting with the generally the first Friday of the month as the probable date

There are minutes attached to this event. Show them.
    • 2:00 PM 2:10 PM
      General news 10m
      INDIGO recruitment
    • 2:10 PM 2:35 PM
      EL7 and Java8 support 25m
      Status of dependency updates Server/client compatibility matrix
    • 2:35 PM 2:50 PM
      Open issues and new features 15m
      CERN issue status authZ decisions based on the combined VO+CA (M. Salle)
    • 2:50 PM 3:00 PM
      AOB 10m