[Openstack] Cinder block device mapping concurrency
I have a small 5 server openstack lab with RDO
1 controller + 4 compute
The controller also runs the storage on LVM backed with some decent disk -> +- 40.000 iops
Glance on file on the same server
When I request 10 instances of centos to be spun up, I see every instance go to block device mapping
And then see 1 by 1 the glance image being cloned to the volume, then mapped, then nova spins up the instance.
However this all works serial, and so my last 2 instances go in error, as they say they tried 61 times.
Is there a way to make cinder do more concurrent clones, as my storage system, the cpu and mem are not taxed at all, and not an HW performance issue.
-------------- next part --------------
An HTML attachment was scrubbed...