Also, nearly all of the time difference is in cpu_sec_per_exec. The elapsed
time difference posted above is ~2.5 seconds, 2 seconds of which is CPU.
Henry
On Fri, Nov 1, 2019 at 11:41 AM Mark W. Farnham <mwf@xxxxxxxx> wrote:
Excellent suggestion.
Now I haven’t read the full thread yet, but I immediately have 3 questions
(that may be in there when I dig):
1) Parallel query?
2) RAC?
3) Force parallel query local?
A wide variety of hilarity can be seen in RAC parallel query NOT forced
local on most of 12.1 (and some of 12.2). Hmm actually it can happen forced
local on 12.1, but I think that at least was patched for 12.2. A symptom of
the BUG I reference would be either parallel forced to serial (having given
up trying to get identical FULL HASH on child plans) or a long parse time
finally getting the matching children to feed to the coordinator.
That may be a zebra whilst you’re looking for a sick horse, and there are
a lot of fine responses from the few I’ve read in the thread, so take my
comment with a salt shaker.
mwf
*From:* oracle-l-bounce@xxxxxxxxxxxxx [mailto:
oracle-l-bounce@xxxxxxxxxxxxx] *On Behalf Of *Sayan Malakshinov
*Sent:* Friday, November 01, 2019 1:23 AM
*To:* breitliw@xxxxxxxxxxxxx
*Cc:* ORACLE-L; JBECKSTROM@xxxxxxxxx; tanel@xxxxxxxxxxxxxx
*Subject:* Re: Force specific plan to be used
Hi all,
Since OP is talking about adaptive plans with multiple child cursors, I
would check *FULL_PLAN_HASH_VALUE *and* PLAN_HASH_2. *
