• H
    Correctly compute distribution policy in phase 1 of gpexpand (#555) · b896d839
    Hans Zeller 提交于
    Consider a hashed distributed partitioned table foo. After first stage
    of gpexpand, the root table will have distribution polcy as hashed but
    the leaf partitions will have random distribution. In such a case,
    ORCA sets m_convert_hash_to_random to true in the table
    descriptor. When the flag is true we should treat the distribution
    policy of table as random during planning.
    
    Note that this code will need to change when we plan to handle
    gpexpand phase 1 in GPDB 6 and later. At that time, we won't be able
    to treat these tables as random-partitioned anymore.
    
    Co-authored-by: Sambitesh Dash sdash@pivotal.io
    Co-authored-by: Hans Zeller hzeller@pivotal.io
    b896d839
CMakeLists.txt 4.8 KB