Let's see. I tried:
1. v$sql_monitor on both nodes and also gv$sql_monitor. My sql_id did not
appear
2. ran repot_sql_monitor from both nodes. Nothing
3. ran the alter system set events 'trace[sql_mon.*] prior to executing my
sql. Didn't see any trc file in diag which looked useful. I also did a grep
there for keswx which diid not show up in any recent traces.
4. I did not yet force parallel
5. David, this is 12.1.0.2, not 19c, but I will give sqlt a look. I
remember checking it out a while back, but don't remember much.
Henry
brought to you by: Larry as Lucy and rsm as the football.
On Wed, May 12, 2021 at 8:50 AM David Barbour <david.barbour1@xxxxxxxxx>
wrote:
Is the query using a global temp or private temp (19c) table? Since these
are populated at runtime, the optimizer can (and has in my experience)
create a totally off-the-wall plan.
Something else you might consider for analysis is sqlt. I found if OEM
and the native tables/research functionality aren't providing answers, sqlt
can really help in pinpointing issues. It's easy to install and you'd be
amazed at the detail.
On Tue, May 11, 2021 at 4:53 PM Henry Poras <henry.poras@xxxxxxxxx> wrote:
Hi Sayan,
Thanks for checking back. I've been busy with some other stuff, but hope
to have some more information on this either tonight or tomorrow.
Henry
On Mon, May 10, 2021 at 6:17 PM Sayan Malakshinov <xt.and.r@xxxxxxxxx>
wrote:
Hi Henry,
Did you get an RTSM trace? It should contain keswx* functions
On Mon, May 10, 2021 at 11:36 PM Jonathan Lewis <jlewisoracle@xxxxxxxxx>
wrote:
Two possibilities, though neither seems likely.
a) gv$sql_plan_monitor (and gv$sql_monitor) may be one of those (rare)
views that DOESN'T run cross-instance, so if the query is running on
instance but the session you're using to look at the monitor report is
connect to another instance you wouldn't see it.
b) there is another hidden parameter that limits the number of plans
that can be monitored at any one time. This is _sqlmon_max_plan, default
80, and is the number of plans PER CPU. Maybe something has jammed the
counter of v$sql_monitor at the maximum so your query can't be captured.
Regards
Jonathan Lewis
On Fri, 7 May 2021 at 17:50, Henry Poras <henry.poras@xxxxxxxxx> wrote:
I think I'm missing something silly here, but I don't know what.
I'm having an issue with getting a sql_monitor report and I'm running
(well, ran) out of ideas on what to try next. I have a long running query
(~2 hours on 12.1.0.2) for which I am trying to get a report using
dbms_sqltune.report_sql_monitor. The result returned is always empty. So I
tried:
- setting _sqlmon_max_planlines to 1000 (execution plan is ~600 lines
using display_cursor('adaptive'))
- checked statistics_level which is correct. It is TYPICAL
- control_management_pack_access is DIAGNOSTIC+TUNING (also correct)
I thought it might be aging out of memory since it ran long, but
running the report and looking in gv$sql_monitor after 2, and 10-15
minutes
still showed nothing. Adding a MONITOR hint to the CTE and body of the sql
didn't help. Neither did running an ALTER SYSTEM SET EVENTS 'sql_monitor
[sql:...] force=true'.
I can't figure out why I am getting nothing. Anybody have any ideas?
Thanks.
Henry
--
Best regards,
Sayan Malakshinov
Oracle performance tuning engineer
Oracle ACE Associate
http://orasql.org
