Attendees: Serge, Douglas, Bill, Jacek

300 node test
  - "select from obj join source using objectId where objectId = <xx>"
     results in full scan, objectId index not recognized, probably
     subchunks materialized too. Code needs to be fixed
  - still some issues with getting cluster correctly configured
    for smaller # nodes
  - need to run few low-volume queries (eg "select from source
    in a very small region")
  - run near neighbor for 10x10 deg
  - returning the nodes ~Sept 24

bugs
  - worker still hosed under some conditions, writing/reading
    results is synchronized incorrectly: results are read
    before they are populated
  - some bugs Bill found are handled through fault tolerance,
    code not yet merged to master


logging
  - Bill wrote some code for logging, thread safe,
    outputs buffers one full line at a time, with thread
    id and time stamp
  - using boost filtered strings, not using any package
    like logcxx
  - it will be easy to migrate once we settle on the package
    we should use long term


what is next (not - till the construction starts)
  - high priority: finalize documentation or FDR
  - then: code refactoring


Qserv code refactoring between now and start of
the Construction - two options:
  a) bottom up (start from the existing prototype
     and cleanup/fix/tweak/improve)
       or
  b) top down (come up with a fresh clean design based
     on all lessons learned, reuse pieces from the existing
     prototype)
  - overall sentiment based on very preliminary discussion
    (without Daniel!): "b"
  - main reason: existing prototype has too many layers,
    very hard to debug
  - other points made:
   - would be good to implement geometry support
     in c++, not from python
   - would be good to get rid of subchunks
   - would be good to push more code from python to C++
  - also... want to switch to the MariaDb distribution

The topic of code refactoring will be discussed
in details in the next few weeks...

Jacek

########################################################################
Use REPLY-ALL to reply to list

To unsubscribe from the QSERV-L list, click the following link:
https://listserv.slac.stanford.edu/cgi-bin/wa?SUBED1=QSERV-L&A=1