In the Swap Rebalance upgrade strategy documentation, you can either swap one or several nodes at a time.
From what I understood, doing this swap (removal and add) in one step (it’s the rebalance that effectively trigger the two operations) avoid putting too much stress on the remaining nodes of the cluster, as opposed to the remove and rebalance upgrade, is that correct ?
I need to migrate a cluster of three nodes, and I tried hereunder to describe the pro’s and con’s of each option:
Option 1: 3 swap/rebalance operations with one node
- Rollback possible until there’s still one old node in place
- Put less stress on the cluster as three “smaller/shorter” rebalances, affect less the performances of the cluster than a big rebalance with all 6 nodes
- Can reuse one of the removed nodes, thus requiring only 4 nodes in total instead of 6 nodes
- Slower than option 2, because (I assume that) three rebalances of 4 nodes are slower than a rebalance of 6 nodes, which also requires more “manual” migration operations
Option 2: One swap/rebalance operation with three nodes
- Cluster more affected (or need more resources ? ) during the rebalance as all nodes are fully impacted by it
- No rollback possible as it’s only one operation
Option 3: one swap/rebalance of one node + another swap for all the remaining nodes
We could also swap/rebalance first one node to see if everything run smoothly, and then swap all the remaining nodes in another operation: that would be more adequate I think in a cluster with a huge number of nodes to avoid too much operations. Or do the upgrades by batches of N nodes.
Can you give me your advice on that, and if I missed something ?
Thanks in advance,