Re: Upgrades to v12 + parallel-min-servers

Hey Daniel,
yes the behavior has changed and the formula is "CPU_COUNT * PARALLEL_THREADS_PER_CPU * 2" - please check Mike Dietrich's blog post about this change: https://mikedietrichde.com/tag/parameter-recommendations/

By the way an uncommon high amount of CPU_COUNT is easily caused by SMT and vCPUs (micro-partitioning) on AIX.

> All these new processes proceed to consume all the memory on the host. (unable to fork)

This sounds even more strange. Do you mean that some upgrade procedure uses these processes or do they just consume this amount of memory at startup? If the latter then you might hit the well known "USLA heap" issue on AIX. However you can use svmon to check / drill-down to the root cause.  

Best Regards
Stefan Koehler

Independent Oracle performance consultant and researcher Website: http://www.soocs.de
Twitter: _at_OracleSK
Upcoming online seminar: http://tinyurl.com/17-06-13-Shared-Pool-Internals

> "Hubler, Daniel" <daniel.hubler_at_aurora.org> hat am 8. Mai 2017 um 16:25 geschrieben:
>
> Curious to know if others are experiencing this. . . . and ideas on how to work around.
>
> AIX 7.1
> GRID/CRS 12.1.x
>
> Upgrading database from 11.2.0.4 to 12.1.0.2 using DBUA/command line/silent. . . . . . .on RAC.
>
> It appears that startup parameter: parallel_min_servers has changed with v12, and is no longer set manually.
> It appears to be under the control of some kind of formula.
>
> What I am experiencing is that in the middle of the upgrade process,
> when the database is started using the v12 software,
> this new formula kicks in a generates dozens of parallel_min_servers. . . .where there were previously few (or none).
>
> All these new processes proceed to consume all the memory on the host. (unable to fork)
> The upgrade process starts to error out on all remaining steps, and fails somewhere in the middle of the process.
>
> We are left with a database that is in some unknown state.
> GRID thinks the database is upgraded, and the SRVCTL commands for v11 no longer work.
>
> Thanks in advance for any anecdotes or suggestions.

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