Production and PASTE Meeting, 13th October 2009, CERN
Chair: A.Tsaregorodtsev
Minutes: S.Paterson
CERN: J.Closier, Z.Mathe, S.Paterson, M.Rashid, R.Santinelli, A.C.Smith, F.Stagni
Phone: V.Fernandez, R.Graciani, R.Nandakumar, M.Sapunov, A.Tsaregorodtsev
Apologies: P.Charpentier, V.Romanovsky
0. Ongoing Action List
SVN #11124
Joel commented that the scripts for migration have to be discussed after the new structure. The Savannah action will
be updated after the issues are understood.
CREAM CE #11223
Roberto mentioned it was possible to query the CE for availability of the queues but this is not available by default.
Andrei suggested requesting this CE monitoring plugin at the GDB level. There was some discussion regarding the
treatment of output sandboxes. The CREAM CE should accept a SURL as destination for the output sandbox, however,
currently this cannot talk to SRM. Resurrecting the classic SE would be another solution, as would storing the logs on
the local file system. All these solutions will be conveyed to the developers.
1. Production Status Report
- Report from the Production meeting
Available disk space at the Tier-1 sites is becoming a concern, how to proceed with simulation productions? Should be
discussed at PPG in case old data can be cleaned up (perhaps cleaning some copies of DC06 DSTs). All the sites
should also be solicited to provide more disk space. ( ACTION: Philippe ) The first two 'Upgrade' productions have
completed successfully, nice to demonstrate productions with different configuration versions are handled correctly.
Stripping has been ongoing, attempted 5 and 10 input data files with both productions only achieving 40% of data
processed after the first full pass. The vast majority of failures are due to data access at the dCache Tier-1s (with the
exception of PIC). FEST reconstruction (full and express) has been ongoing in the past week, several iterations to fix
AppConfig options but this is now running smoothly. FEST stripping is almost in place (testing starting from RAW) just
have to confirm the last DaVinci step options.
2. GDB Report
See slides attached to agenda. Roberto is willing to accept comments via mail until the end of today.
3. DIRAC Release Update
Andrei mentioned that the JobDB has 1M jobs, an effort will be made to clean these up.
4. Hardware Configuration
See Ricardo's report attached to agenda. Scalability tests have been running for over a month on volhcb14 (8
disks). Splitting the databases onto different disks we can gain an order of magnitude in query rate. Only at 10k Hz
did we start to hit the MySQL limit. We can guarantee that the capacity of the system will be enough to run the
production system for the foreseeable future. A meeting with the IT people should be organized for the next DIRAC
week starting 2nd November. ( ACTION: Joel )
5. DIRAC Tasks
- Infrastructure monitoring (SAM,SLS,Dashboard)
A Nagios box is being set up, we can start to migrate our tests. Space token monitoring informations will start to be
aggregated after the GDB. Some lemon alarm sensors will be added for our VO-boxes to allow 24/7 coverage.
- Bookkeeping status
BK web page was moved to production, a bug fix was made to fix a problem with saving python options files.
- WMS status
The DIRAC site was tested with Florian with 20k jobs running successfully. At some point the monitoring threads
caused some problems in the development system. A default timeout was added for all RPCClients.
- DMS status
The online file monitoring page was added to the web portal. Work started on analysing the storage dumps provided
by some T1 sites.
- Documentation
Ongoing, a new twiki page is being prepared.
6. AOB
Next DIRAC week starting at 2nd November. The major issues to discuss there will be code restructuring and the final
hardware configuration of the production system.
There are minutes attached to this event.
Show them.