Tim Hall inspired me with his #ThanksOTN (or “OTN Appreciation Day“) campaign to continue my mini-series on leveraging SQL Developer Reports for DBA tasks. Today with visualizing historical System Statistics from Statspack for performance analyses.
#ThanksOTN I have three favourite Oracle features to rave about today:
- Analytic SQL
- SQL Developer (particularly the reports)!
Up to Oracle 11.2 it was possible to display archived SQL plans from Statspack using DBMS_XPLAN. I make use of this in some of my scripts and SQL Developer Reports since I first saw this in Christian Antognini’s Book “Troubleshooting Oracle Performance“.
But in 12c (here: 18.104.22.168 on Linux), there’s a piece missing now:
select * from table(dbms_xplan.display(
table_name => 'perfstat.stats$sql_plan',
statement_id => null,
format => 'ALL -predicate -note',
filter_preds => 'plan_hash_value = '|| &&phv
ERROR: an uncaught error in function display has happened; please contact Oracle support
Please provide also a DMP file of the used plan table perfstat.stats$sql_plan
ORA-00904: "TIMESTAMP": invalid identifier
So it looks like STATS$SQL_PLAN wasn’t synchronized to the changes in 12c’s PLAN_TABLE. Maybe because the timestamp wouldn’t make much sense there, anyway, maybe simply because Oracle forgot.
==> Quick and most certainly unsupported workaround:
ALTER TABLE perfstat.stats$sql_plan ADD timestamp INVISIBLE AS (cast(NULL AS DATE));
Another workaround could be to create a separate view with an additional dummy timestamp column an reference the view. I chose to stick with the invisible column solution so I won’t have to create new objects in the DB and change scripts to use these objects.
Hopefully, this will be solved in 12.2 – at least, I had filed an SR / Enhancement Request with Oracle Support.
Today here’s a shorter post about my experiments with Oracle SQL Developer’s user-defined reports: A report on all long running operations (“LongOps”) with details on session wait events, explain plans and live SQL monitoring.
“Wait a minute”, you might say, “there’s already the session report in SQL Developer’s standard reports that shows Session_LongOps”! – and you’re right, it’s been integrated in SQLDev for quite some time.
BUT: I wanted to view the LongOps from a different perspective: To find out what’s taking so long in the database generally, it would be better in my humble opinion to start with ALL LongOps and drill down from there.
Here’s how it looks on a live system: