Handling node replacements

[stratacast]

I am reaching a point where I am planning to reinstall the OS on my nodes for upgrades. The problem is, I am hitting an issue of adding a new node back to the cluster after the OS is reinstalled in my testing. Even simply removing my radata contents and starting afresh I run into "preventing overlapping partitions" errors and ra will not start. This is alleviated if I stop the entire cluster and start it. Not ideal. So what is the best way to essentially replace a node to prevent partition overlaps?

[michaelklishin]

@stratacast overlapping partitions is a kernel setting and has little to do with ra. You can be upgrading Erlang from a version where overlapping partitions were not enabled by default to the latest version or something like that.

There isn't much specific that can be recommended from a few line long description.

Grow-then-Shrink upgrades is one approach usually recommended against but for replacing a Ra node it can work.

ra:remove_member/3 is the function that removes cluster members assuming there's an online majority of Ra servers.

[stratacast]

I think I was able to resolve how to get upgrades going. This will be a rolling upgrade. I have to take the node offline, reinstall the OS, reinstall the software, then add the node back to the cluster. My process that avoids overlapping partitions involves doing a ra:remove_member/2, then ra:add_member/2 then I have to run a manual ra:start_server/5 on the new node. This joins the upgraded node nicely. This has avoided the overlapping partitions issue I mentioned. The ra:start_server/5 has been necessary due to a mistake in my code