I think in the web there are a lot of articles about this. Bugs 21834574,
5184776,....
None of them proposed that forcing a hard parse would help..
-----Original-Nachricht-----
Betreff: Re: shared pool waits
Datum: 2021-09-21T10:46:25+0200
Von: "Pap" <oracle.developer35@xxxxxxxxx>
An: "ahmed.fikri@xxxxxxxxxxx" <ahmed.fikri@xxxxxxxxxxx>
Thank you Ahmed. You pointed out this to be an Oracle bug. Can you please
shed some more on that regard. What the bug exactly is and the situation
when it impacts.
And yes, this database is RAC but this sql is executed from one instance
only as per the service configured. We normally see this issue when stats
gather on the underlying object which is executing from another instance.
Can this be pointing to something else?
On Tue, Sep 21, 2021 at 1:32 PM ahmed.fikri@xxxxxxxxxxx
<mailto:ahmed.fikri@xxxxxxxxxxx> <ahmed.fikri@xxxxxxxxxxx
<mailto:ahmed.fikri@xxxxxxxxxxx> > wrote:
You mentioned database 19c. I think the relevant point is whether or not
you are using RAC. From my observation, this is typical of RAC. The
problems observed do not occur every day, but when it does happen then
jobs that were only 2-3 hours in duration take more than 20 hours and
removing the SQLs that were causing the problem from the shared pool
solved the problem.
Best regards
Ahmed
-----Original-Nachricht-----
Betreff: AW: shared pool waits
Datum: 2021-09-21T09:28:25+0200
Von: "ahmed.fikri@xxxxxxxxxxx <mailto:ahmed.fikri@xxxxxxxxxxx> " <
ahmed.fikri@xxxxxxxxxxx <mailto:ahmed.fikri@xxxxxxxxxxx> >
An: "list, oracle" <oracle-l@xxxxxxxxxxxxx
<mailto:oracle-l@xxxxxxxxxxxxx> >
indeed hard parse in this case is better than waiting eternity due to
oracle bugs.
-----Original-Nachricht-----
Betreff: Re: shared pool waits
Datum: 2021-09-21T09:11:20+0200
Von: "Pap" <oracle.developer35@xxxxxxxxx
<mailto:oracle.developer35@xxxxxxxxx> >
An: "ahmed.fikri@xxxxxxxxxxx <mailto:ahmed.fikri@xxxxxxxxxxx> " <
ahmed.fikri@xxxxxxxxxxx <mailto:ahmed.fikri@xxxxxxxxxxx> >
Not getting it fully though. If we purge a sql from cursor cache, doesn't
it mean there will be hard parsing and that will cause more contention
and take more CPU to parse subsequently? Though I am also trying to
understand Lok's point, as we should have noticed the in_hard_parse flag
as 'Y' in case this issue would have been caused by hard parse.
But again , when I see the P1 and P1text while the ASH shows the event as
'library cache:mutex x' or 'cursor mutex s', they are pointing to the
same sql_text only. Does that mean even doing soft parsing only but
because of the so many number of executions/soft parses we are suffering?
But then I am wondering why do we see these issues mainly while stats
gathering is running from another session on the same underlying object?
How can this be related?
The query looks like below.
SELECT MAX (ID1) FROM TAB1 WHERE C1= :B1 AND C2 = :B2
----------------------------------------------------------------------------------------------------------------
| Id | Operation | Name | Rows | Bytes | Cost
(%CPU)| Time | Pstart| Pstop |
----------------------------------------------------------------------------------------------------------------
| 0 | SELECT STATEMENT | | | | 4
(100)| | | |
| 1 | PARTITION RANGE SINGLE | | 1 | 52 |
| | KEY | KEY |
| 2 | SORT AGGREGATE | | 1 | 52 |
| | | |
| 3 | FIRST ROW | | 1 | 52 | 4
(0)| 00:00:01 | | |
| 4 | INDEX RANGE SCAN (MIN/MAX)| IDX1 | 1 | 52 | 4 (0)|
00:00:01 | KEY | KEY |
----------------------------------------------------------------------------------------------------------------
On Tue, Sep 21, 2021 at 12:02 PM ahmed.fikri@xxxxxxxxxxx
<mailto:ahmed.fikri@xxxxxxxxxxx> <ahmed.fikri@xxxxxxxxxxx
<mailto:ahmed.fikri@xxxxxxxxxxx> > wrote:
Hi,
pragmatically I used to solve these problems (in my opinion oracle
bugs) by creating a scheduler job to detect and remove the SQLs, that
cause such problems, from the shared pool (using
dbms_shared_pool.purge).
Since one can not change the vendor's code.
Best regards
Ahmed
-----Original-Nachricht-----
Betreff: shared pool waits
Datum: 2021-09-20T21:25:36+0200
Von: "Pap" <oracle.developer35@xxxxxxxxx
<mailto:oracle.developer35@xxxxxxxxx> >
An: "Oracle L" <oracle-l@xxxxxxxxxxxxx <mailto:oracle-l@xxxxxxxxxxxxx>
>
Hi , We have a customer application in which we see high wait events
like 'cursor:mutex ' and 'library cache lock' for a select query
occasionally and thus a specific functionality impacted. This select
query(which is part of a plsql procedure) is quick query which runs ~5
million times/hr. But even though number of execution is same mostly
throughout the day, it still went through these odd wait events making
the per execution time went higher for around ~15 minutes duration
causing slowness. And during this period, the ASH shows fro this query,
the value of column in_hard_parse as 'N' but in_parse as 'Y' and 'N'
both. And we saw we were having stats gather running on that base
object during same time. We have no_invalidate set as 'FALSE" as table
stats preference, So wanted to understand from experts, can it be
really because of 'parsing' issue and we should delete this
no_invalidate preference so that it can inherit the default global
preference i.e no_invalidate=>auto? The database version is 19C.
