drill-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Jason Altekruse <altekruseja...@gmail.com>
Subject Meeting Notes 10/15/13
Date Tue, 15 Oct 2013 18:30:40 GMT
Hello Drillers!

Here are the notes from the Hangout this week. Michael, can you paste this
into the google doc?

Date 10/15/13
Jason Altekruse, Michael Hasenblas, Steven Phillips, Tim Chen, Jacques
Nadeau, Ben Becker

Important note, we have finally ironed out the hangout link expiration.
Until it breaks again this is the link:


    - talks have been going well
    - lots of interested users anticipating GA
    - can the current release run in distributed mode?
        - response: steven
            - need to run physical plans
            - need to manually insert exchanges
            - need to add exchanges to optimizer
        - setting up cluster is easy, just connect drillbits to zooKeeper
        - right now we are focused on validating performance
        - so not as much focus on sql query submission
    - if someone would write 3 or 4 steps to
        - launch 3 drillbits
        - connect to zookeeper
        - run a distributed query
            - even if its not use SQL yet
        - Michael can expand on it
        - Tim has patch for Apache Whirr, still waiting for merge
            - launches drillbits and connects to zookeeper
        - would be nice if we could run REST client at start of cluster
        - Jacques: Stateful client that sits in front of drillclient

    - empty batch issue
    - update the clear contract for what record batch implementation should
be used
    - limit looks really close

    - merging receiver operator
        - lot of progress, no real blocks

    - more or less done with ordered range partitioner
        - checking in soon
    - problems with hash exchange
        - for larger batches its duplicating rows
    - next task spooling to disk
        - beginning of a fragment
        - doesn't solve the situation where a blocking operator has too
          much to handle
        - HDFS writes?
            - want to use the same disks, not just the OS disk
            - drill directories set up like MapReduce directories
            - config for each node
                - could be /tmp, or a directory, or directory on data node

    - patched Whirr
    - no feedback on review patch tool
        - does it below in source control
        - should we have a tools directory for this and like IDE settings?
        - apache doesn't really have a concept of multiple repos for a

    - working on reader/writer
    - BitWeaving integration grad students at UW Madison

Patch coming from Mehant more people from MapR will be helping with Drill

    - drill release notes
    - cannot edit wiki, need to talk to Apache infrastructure people

  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message