Yong, thanks again!
*--*
*Att*
*Luis Santos*
2016-10-21 14:30 GMT-02:00 Yong Huang <yong321@xxxxxxxxx>:
Luis,
The 11gR2 documentation is a little inconsistent, because the Reference
manual says differently:
http://docs.oracle.com/cd/E11882_01/server.112/e40402/
initparams223.htm#REFRN10239
"RESUMABLE_TIMEOUT enables or disables resumable statements and specifies
resumable timeout at the system level."
Since both our tests show that simply setting the parameter alone will
enable resumable sessions (even though they suddenly become enabled only
when the session is suspended), I would say the Admin Guide is incorrect
and the Reference is correct.
My test shows that a session that explicitly enables resumable with "alter
session" command will get the "resumable state object" latch 3 times for
each SQL execution, while a session that is "silently" watched by the
system (i.e. you only set resumable_timeout with "alter system") will not
except once at logon or until it actually runs out of space. If we only
judge by this latch get, from the performance point of view, it may be
slightly better to set the parameter alone. (I didn't check any other
possible overhead.)
Yong Huang
--------------------------------------------
On Fri, 10/21/16, Luis Santos <lsantos@xxxxxxxxx> wrote:
...
And we did several tests here (in both 11.2.0.2 and 11.2.0.4) and we
conclude exactly as you: setting resumable_timeout instance wide is enough!
But the Oracle analyst said that the documentation says this.
I checked 10g, 11gR1 and 11gR2 docs. And, surprising, 10g and 11gR1 docs
says, OK, that setting resumable_timeout instance wide is enough... but
11gR2 docs don't!!!
...
