On 07/06/2019 15:03, Yi Yang (杨燚)-云服务集团 wrote:
> Robert, let us use inventory as an example, assume I started 127 ODL nodes,
> how many leader shards can we have for it? I think it is one, so for this
> case, only one node, i.e. inventory leader shard node is handling all the
> inventory read and write (including serving for remote read and write), isn't
> it?

Right now you can have as many shards as there are models which define
root-level nodes. This is the default, and suffers exactly from the
'inventory leader' syndrome, where you just cannot partition the data to
fit the access pattern. I am sorry, but this is how shards were contributed.

> Please let me know how we can split inventory into many shards (I don't mean
> replica shards) in order that we can have any leader shards per requirement,
> for example, we have 100 openflowplugin clusters, every has its own inventory
> leader shard, that is what I want, can you tell me how I can do this in
> current ODL clustering implementation? You told me prefix-based shard, but it
> isn't ready for this, right?

So this is really about data partitioning, which is what shards are
supposed to be. The use case is that each OVS switch has some
master-related state and therefore, OFP needs to be able to tell CDS to
move the master to co-locate that state with where the OFP master lives.

This functionality is available with prefix-based shards, which I think
does provide compatibility APIs DOMStore, but I'd have to check.

Any testing feedback is appreciated.

> BTW, you Pantheon sales Martin Varga contacted me, he told me he would ask 
> you 
> to have a meeting with me about our requirement, I would like to have more 
> discussion with you, I'm not sure if you catch our pain point.

Let's follow up separately.

Regards,
Robert

Attachment: signature.asc
Description: OpenPGP digital signature

_______________________________________________
controller-dev mailing list
controller-dev@lists.opendaylight.org
https://lists.opendaylight.org/mailman/listinfo/controller-dev

Reply via email to