Question

We use the following to refresh statistics for all tables in a given schema:

exec dbms_stats.gather_schema_stats(ownname => 'some_schema', estimate_percent => dbms_stats.auto_sample_size, cascade => true, method_opt => 'FOR ALL COLUMNS SIZE AUTO', degree => 12);

This however, sets row-counts for our materialized views to zero and has the unwanted side effect of causing inefficient query plans for queries against materialized views. We work around this by gathering table stats against the specific mviews after the schema stats have run.

My question is: can I change the parameters to gather_schema_stats in any way that will cause mview row-counts not to be set to zero?

Answer 1

You can’t tell GATHER_SCHEMA_STATS to exclude certain objects. You could do a GATHER STALE to gather statistics only on the objects where statistics are stale but it is entirely possible that would include your materialized views. A few ways to work around that

1) Use the LOCK_TABLE_STATS procedure to lock the statistics on your materialized views. That will prevent GATHER_SCHEMA_STATS from gathering statistics on those objects until you call the UNLOCK_TABLE_STATS procedure (presumably as part of the process that refreshes the materialized view statistics periodically).

2) Use the EXPORT_TABLE_STATS procedure to save the statistics for the materialized views before gathering schema statistics and then call RESTORE_TABLE_STATS after the GATHER_SCHEMA_STATS call completes to put the materialized view statistics back.

3) Don’t use GATHER_SCHEMA_STATS. Call GATHER_TABLE_STATS in a loop where you exclude whatever objects you want. Something like

BEGIN
  FOR x IN (SELECT *
              FROM dba_tables
             WHERE owner = 'SOME_SCHEMA'
               AND table_name NOT IN (<<list of MVs>>))
  LOOP
     dbms_stats.gather_table_stats( x.owner, x.table_name, ... );
  END LOOP;
END;