Closed lemaitre-aneo closed 1 month ago
How would partitionning be affected ? If, let's say, multiple sessions are using a given partition, and the given partition is only available on specific clusters, then the affectation of a session is determined by which partitions are present on which cluster. This also asks the question on how to handle a session with partitions spanning multiple independant clusters
A small architecture diagram would be helpful to capture the concept. Would there be a kind of healthcheck between the loadbalancer and the clusters? And what would happen in the case of a lost of communication?
A small architecture diagram would be helpful to capture the concept. Would there be a kind of healthcheck between the loadbalancer and the clusters? And what would happen in the case of a lost of communication?
Regarding the diagram, I don't think it will be particularly useful as it is really simple. Maybe I should add a paragraph for communication losses, but the idea is dead simple: if the LB cannot connect to a cluster, it errors
I would also add a new API to transfer data between partitions (and sessions) to allow complex workflow requiring different clusters to be implemented, ideally without having the data going back to the client.
As a data is tied to a session, it is also tied to a cluster. When there is an API to move a data from a session to another, it would just be a matter of implementing the logic within the LB.
All committers have signed the CLA.