Number of nodes that can be removed at a time

Operations
1

Hi,

We have a 18-node cluster (v3.14 EE).

How many nodes we can remove at a time during maintenance?

Please let me know what else information is needed.

Thank you.

2

If using rack-aware, you can remove a rack at a time. But be careful in regards to capacity (especially in a version without migrate-fill-delay), the remaining racks will still want to replicate replication-factor copies.

Otherwise, you can take down replication-factor - 1 nodes at a time (this is typically one node since replication-factor is often 2).

3

You can also quiesce multiple nodes at once, and allow data to move off several nodes… For example, in the past I’ve quiesced 10-20 nodes at once in a 100 node cluster. Have to ensure you wont lose too much data and the remaining nodes will be able to deal though, and make sure you wait for master objects and transactions to fall off the quiesced nodes…

4

What do you mean by “ensure you wont lose too much data”?

I didn’t mention quiesce since 3.14 doesn’t support it.

5

Oh right forgot! Oh I just mean that each node would have a larger amount of records, and may or may not be okay :slight_smile:

6

@mvineza what is the maintenance anyway, if you don’t mind me asking

7

Thanks @kporter for clarifying. :slight_smile:

8

Hi @Albot Thanks also for responding on my question. We will be doing kernel upgrade as maintenance.