Should Scylla manager be run per-region, or one manager instance for a multi region cluster?

Since the open source version is limited to 5 instances, could I in theory have 5 instances per region, and each region has a dedicated scylla manager?

I suspect that this is not advised, but in theory will this work or break the cluster? I could stagger their repairs so it’s never concurrent

Pining @Guy since he asked :slight_smile:

Technically it might work, but not advice.
You will need to ensure Managers tasks do not overlap and are limited per DC. By default, Manager tasks, like repair, are cluster wide.

How would you ensure that they are limited to a single region?

Or I guess that’s a silly question, since it seems like the agents call back to a manager instead, so it would just be a matter of pointing an agent to the correct manager.

And for reference, I am using a somewhat unusual use case where I want many small nodes in many regions to reduce latency for lookups, so I would be running into the limit of requiring the enterprise license to have scylla manager handle repairs automatically. All these nodes will be the minimum size since each node is not hit very frequently, but I want all data in all regions.

Would there be another recommended way to run this in so many regions without doing manual repairs?

I’m not aware of other tools for running recurrent repairs on ScyllaDB.
Note that repair works between all nodes, including nodes from different data centers.

Understood thanks! To clarify you are confirming that running a nodetool repair will repair on all nodes across the cluster in that single command? I.e. I only need to run that once for the whole cluster manually, not on each node? My understanding was that this is only per-node.

nodetool repair sync between the node you are running the command from, and its replicas.
It does not repair all the nodes.