MV Refresh

Here’s a funny little problem I came across some time ago when setting up some materialized views. I have two tables, orders and order_lines, and I’ve set up materialized view logs for them that allow a join materialized view (called orders_join) to be fast refreshable. Watch what happens if I refresh this view just before gathering stats on the order_lines table.

I have a little script that start with “set echo on”, then calls two packaged procedures, one to refresh the join view, the other to collect stats on the order_lines table; here’s the output from that script:

SQL>
SQL> begin
  2          dbms_mview.refresh(
  3                  list    => 'test_user.orders_join',
  4                  method  => 'F'
  5          );
  6  end;
  7  /

PL/SQL procedure successfully completed.

SQL>
SQL>
SQL> begin
  2          dbms_stats.gather_table_stats(
  3                  ownname          => user,
  4                  tabname          =>'ORDER_LINES',
  5                  method_opt       => 'for all columns size 1',
  6                  cascade          => true
  7          );
  8  end;
  9  /
begin
*
ERROR at line 1:
ORA-01760: illegal argument for function
ORA-06512: at "SYS.DBMS_STATS", line 33859
ORA-06512: at line 2

The stats gathering has failed (on one of the indexes, though that’s not immediately obvious) because the SQL statement generated by the procedure has gone through a query rewrite that takes it to the orders_join table – which won’t support some of the funny internal function used by Oracle to collect index stats. Here’s the SQL as generated (with a little bit of formatting):

select
	/*+
		no_parallel_index(t, "ORL_PK")  dbms_stats cursor_sharing_exact use_weak_name_resl
		dynamic_sampling(0) no_monitoring no_substrb_pad  no_expand index(t,"ORL_PK")
	*/
	count(*) as nrw,
	count(distinct sys_op_lbid(82196,'L',t.rowid)) as nlb,
	null as ndk,
	sys_op_countchg(substrb(t.rowid,1,15),1) as clf
from 	"TEST_USER"."ORDER_LINES" t
where	"ORDER_ID" is not null
or	"LINE_ID" is not null
;

Because of the particular way I happen to have defined my materialized view, this query gets rewritten if the view is “fresh” – which makes the hints inapplicable – and neither the sys_op_count_chg() nor the sys_op_lbid() functions are valid unless you’re doing an index fast full scan.

At first sight the fix to this problem would simply appear to require the addition of a no_rewrite() hint to the query – but maybe there’s some reason why this isn’t viable. Of course, you’d have to be a little unlucky to hit this bug – it takes a combination of timing and unlucky data to trigger it; but if you’re using join views with referential integrity all over the place you may be slightly more susceptible than average.

This bug is repeatable in 12c – and looks like the problem referenced in MOS note 317254.1 (and don’t ask which version it was first reported against!)

Related posts:

1. MV Refresh
2. Materialized view refresh change in 10G and sql tracing
3. refresh “fast” of materialized views optimized by Oracle as “complete”
4. Materialized view refresh change in 10G and sql tracing
5. fast refresh of outer-join-only materialized views – algorithm, part 1
6. ON COMMIT Refresh without a Primary Key
7. fast refresh of join-only MVs: _mv_refresh_use_stats and locking log stats
8. fast refresh of outer-join-only materialized views – algorithm, part 2
9. Materialized View Refresh ON COMMIT
10. Fast refresh of aggregate-only materialized views – introduction