Re: setting the parameter parallel_force_local to false leads to data loss

Were you setting the parameter with ALTER SESSION or ALTER SYSTEM?

It has always struck me as a bit odd that a parameter related to multi-session processing can be changed on a single session with ALTER SESSION.  I understand that the session being changed is the QC and that is the only session that matters in this case, but still I wouldn't fully trust anything unless running ALTER SYSTEM for something involving parallelism.

Just a shot in the dark...

On 4/26/2021 2:45 AM, ahmed.fikri_at_t-online.de wrote:
>
> Hi all,
>
> does someone has experienced a problem with the parallel_force_local
> parameter?
>
> In a new created envirement, the database is EE 12.1.0.2.0 64 bit
> (Redhat), Iobserved following: In a PL/SQL Block several inserts +
> commit after each insert are performed. Sometime not all data are
> commited:
>
> Expected inserted rows is 1000. I repeated the test 5 times (each time
> the table is truncated) and I got this:
>
> 1. only 127 rows are inserted (the sum of sql%rowcount = 1000)
>
> 2. 920 rows are inserted   (the sum of sql%rowcount = 1000)
>
> 3. 1000 rows are inserted (the sum of sql%rowcount = 1000)
>
> 4. 1000 rows are inserted (the sum of sql%rowcount = 1000)
>
> 5. 1000 rows are inserted (the sum of sql%rowcount = 1000)
>
> In the past I observed the same problem in our production environement
> (after a migration from AIX to Linux) our regression test had same
> problem, after each call a different result. My gut feeling I
> suspected the parallel_force_local parameter that was set to false.
> After setting the parameter to true everything works as expected (reg
> test is consistent).
>
> Best regards
>
> Ahmed
>
> 

--
http://www.freelists.org/webpage/oracle-l