perf enhancement for ddl operations. DDL operations where objects (tables, views) had large number of columns were running slow due to single row inserts into metadata COLUMNS table. This showed up during DSM repository creation which had 150 tables and 500 views. Changes done: -- added code to do rowwise rowsets -- enhanced metadata COLUMNS inserts to use rwrs upserts -- changed metadata calls to use upsert instead of insert -- fixed a cleanup bug
Preventing user CQDs from affecting metadata queries We now won't pass user CQDs to new CmpContext instance (CI) during embedded CI switch. This is to prevent metadata query plans being influenced by user CQDs unintentionally. Also, temporarily added a flag to retrieve the latest hbase snapshot info for user tables disrespect to whether we use the snapshots for scan or not. The CQD TRAF_TABLE_SNAPSHOT_SCAN will only affect the user query plan generations.
Reworked LP 1420826, Turning off nested_joins fails meta-data query compilation, by reverting the changes because when compiling the metadata queries, the default optimization level would be '3' and won't be affected by user setting anymore.
LP 1426479: Row mismatch between index/table cause init auth to fail. Fixed LP 1411355: User-level CQDs also seem to apply to metadata queries.