we can save some memory in our OOOQ multinode jobs by specifying custom
role data -- getting rid of resources and YAQL crunching for roles that
aren't used at all in the end. Shout out to shardy for suggesting we
should look into doing this.
First observations, hopefully at least somewhat telling:
nonha-multinode-oooq job: comparing  the RSS memory usage by the 4
heat-engine processes on undercloud drops from 803 MB to 690 MB.
containers-multinode-upgrades job: comparing  heat-engine memory
usage drops from 1221 MB to 968 MB.
I expected some time savings as well but wasn't able to spot any, looks
like concurrency works well in Heat :)
Patches are here:
Have a good day,
OpenStack Development Mailing List (not for usage questions)