Handling ElastiCache Maintenance ~Rainy Season 2018~

This is a memo from around June 2018 when a large number of AWS ElastiCache maintenance notifications came in all at once.

Links I referred to during maintenance

Amazon ElastiCache for Redis

Basically, replacing the node is required.

Below are the steps for doing it manually.

When node = 1

For a standalone ElastiCache:

1. Take a backup
2. Add a read replica
3. Promote the replica to become the primary
4. Delete the former node that has become the replica

This completes the node replacement.

node = N (>=2)

When there are multiple nodes:

1. Take a backup
2. Perform a failover

According to AWS Support:

The failover API simulates a failure, so the node is also replaced after the failover.

So that’s how it works.

How to confirm that maintenance is complete

I think there is no way to check the status accurately other than asking Support.

The events keep an operation log, but no log indicating that maintenance was actually performed is left behind at all.

Also, the alert does not disappear immediately. I’m not exactly sure, but it had disappeared after a few hours had passed.

What happens if you leave it alone without running it manually

The maintenance was performed during the time window specified in the maintenance window.

By the way, the event log when leaving it alone looked like the following.

n = 1

n = 2

You can see that for n = 2, the failover is executed automatically.

The elapsed time of the maintenance itself depends on the amount of data, so it cannot be stated uniformly. Therefore, I think it’s a good idea to do a rehearsal once before production to get a rough estimate.

That’s all.
I hope this is helpful.