Support for ON CLUSTER Migrations

[gempir]

When managing clusters of ClickHouse nodes, it's typical to run migrations with ON CLUSTER <name> to apply any changes.

https://clickhouse.com/docs/sql-reference/distributed-ddl

This would be useful in chkit aswell. This means chkit would only run this query on 1 node, and clickhouse will handle the propagation of those changes to the cluster.

[KeKs0r]

Thanks for opening this issue. I will do an initial investigation of the delta, what would be required to run migrations on cluster. Besides making sure the migration sql runs with ON CLUSTER, are there any additional things on top of your head that we should consider?
I am doing also own research now, but I think you may have more cluster experience.

[gempir]

AFAIK nothing else. I don't know how migrations run in obsession (like does it run against all nodes or just 1) but the ON CLUSTER query also just has to run against 1 node, not all.

[KeKs0r]

Our SharedMergeTree implementation avoids the need for clusters, since data (and metadata) is shared. So each node sees everything, with some mini edge cases.

I'll dig into the Cluster support over the weekend and try to give you an estimate for implementing it. Would you be willing to try it afterwards and give feedback?

[gempir]

Yes for sure