Skip to end of metadata
Go to start of metadata

Date

Attendees

Fritz MuellerFabrice JammesJohn GatesBrian Van KlaverenAndy SalnikovAndy HanushevskyNate PeaseVaikunth ThukralKian-Tat Lim

Discussion items

L1/AP Database

  • Andy S is sorting through rotAngle schema issues, and has opened a related RFC

Shared scans

  • John has been testing with memman enabled on the IN2P3 cluster, and has seen some odd results, and will continue to investigate: 
    • SELECT COUNT(*) FROM Object takes about 3min (should be 30sec)
    • scans were running in 1hr 15m, now done in 1hr
    • only one scheduler seems to be executing at a time
    • the shared scan subsystem doesn't seem to be grabbing all the memory if could

Large results

  • Proxy and czar now survive queries with result sets ~7G, but the mysql client dies
  • Suggested to try client APIs as well

Butler

  • RFC-184 is closing.  Nate started implementation in parallel and is largely done
  • Nate is working on refactoring the data-id class.  DM-5558 is wrapped up in this as well

DAX

  • Brian is starting work on the DAX service containerization design.
  • Will use stack base container that Fabrice has is place for qserv

VO

Docker

  • Oualid is finishing up opestack cluster spin-up scripts.  These are in plain Python, but may be switched over to ansible.
  • Fabrice has been working with Oualid, and will rebase his Swarm changes on Oualid's changes after Oualid finishes up.

XSWAP

  • Vaikunth is working an a strategy to transform logs to JSON, then load and do analysis with R
  • Needs some querid changes to make logs of start and end of queries consistent.  In the meantime, will work with data from czar only.
  • Vaikunth has designed a test matrix for machine configuration sizing tests  

Documentation

  • Andy H working on updating XRootD documentation for new features: single-session, logging plug-in, holding results

Sizing Models

  • Fritz queried about notebook style sizing models, vs. the excel spreadsheet style.  K-T informs that SKA sizing models are Python notebook based, and people seem to like them.  Some of the origin motivations for us to use Excel no longer apply now that LSST is in construction.  We willl be considering adopting notebook style models.  This might show up first with KPMs.

Misc

  • John has a few reviews up that require attention; please check
  • Andy S on vacation 6/10-6/15
  • Andy H is starting to think on disconnected query design and czar-failover
  • Andy H will be at CERN next week