Open Stack

On Wed, 2015-09-23 at 11:11 -0700, Vilobh Meshram wrote:

> Accepted in Liberty [1] [2] :
> [1] Services information be stored in respective backend configured
> by CONF.servicegroup_driver and all the interfaces which plan to
> access service information go through servicegroup layer.
> [2] Add tooz specific drivers e.g. replace existing nova servicegroup
> zookeeper driver with a new zookeeper driver backed by Tooz zookeeper
> driver.
> 
> 
> Proposal for Mitaka [3][4] :
> [3] Services information be stored in nova.services (nova database)
> and liveliness information be managed by CONF.servicegroup_driver
> (DB/Zookeeper/Memcache)
> [4] Stick to what is accepted for #2. Just that the scope will be
> decided based on whether we go with #1 (as accepted for Liberty) or #3
> (what is proposed for Mitaka)
> 
I like Mitaka (#3) proposal more. We still have whole data in the
persistent database and SG driver informs only if a host is alive. This
would make transitions between SG drivers easier for administrators and
at last this is why you want to use ZooKeeper - to know about failure
early and don't schedule new VMs to such non-responding host.