By using multiple compute flavors can we make the overcloud deployment faster ?

asked 2015-11-30 05:29:11 -0600

Vinu gravatar image

I'm trying to deploy an openstack overcloud via undercloud. I've total of 63 baremetal servers which am planning to deploy as part of my environment. I'm planning to have the overcloud controllers in HA, so 3 nodes are reserved for controllers. Remaining 60 hosts I was to deploy as compute nodes. After all introspection of nodes, I created a flavor by name control and compute and assigned it to respective nodes capability:profile. 'nova flavor-list' also shows up those 2 flavors. Next I went ahead and executed 'openstack overcloud deplpoy --templates --control-scale 3 --compute-scale <in increments="" of="" 5=""> --control-flavor control --compute-flavor compute -e <> '

in batches of 5 nodes per run. By the time I scaled up to 35 nodes, the overall execution time increased to 3 hours 30 mins. I've read that the default deployment timeouts set is 4 hours has part of heat stack. Along with that keystone token expiry time is also 4 hours. At this point if I go ahead and start the next batch of 5 nodes for deploy, its going to time out for sure. Could you please suggest how to reduce the over all deployment time ?

  • Will tagging the nodes with different different flavors and then deploying them separately, will it help ? If yes then could you please provide the deployment command in that case to be used ?
  • Will clearing all old and unused data of MySQL database can fasten the queries and there by speed up the deployment ? If yes, then will keystone-manage toke_flush command help here ?

Or is there any other good way to optimize the Openstack overcloud scale deployments ? Any help is highly appreciated ....

edit retag flag offensive close merge delete