My favorite moment as a customer dealing with all of the random times _they_ would take us down (even with triple redundancy) was their suggestion of operating a duplicate 3 datacenter cluster in production as a hot spare.

They must think their customers are cash machines.

I mean if they were truely best of breed they wouldn’t have needed a license change now would they?

Indeed, although I'd say for a majority of situations, Amazon's hosted Elasticsearch service is a complete non-starter.

Why is it a non-starter? I have a few ElasticSearch clusters that have been running for years with zero effort/downtime/hassle.

Personal experience is that AWS elasticsearch has often been missing some really useful stuff, index rebalancing, some of the utility endpoints that avoid me having to spend forever rebuilding indexes, etc. I'd be a little spooked to run something really huge and customer-facing (maybe it's possible and I'm a n00b, but for the money we pay, it should be n00b friendly).

It's always great (really, it was quite easy to get started and usually works) until it's not (a couple times have had indexes break, or had to reindex to a fresh cluster to fix balancing problems).

I'm sure LOTS of people use aws elasticsearch for big, user-facing stuff, but I often feel you'd be better off managing it yourself if it were truly critical.

This was my experience too: the AWS ES hosted offering was ok, but we had an outage and some associated needs and we basically lost all of our data.

However, we found that running our own cluster on EC2 with Cloudformation was relatively painless.

What the other guy said.

Also my saying that is extremely colored by experiences with pre-ES6 versions, where AWS's offering didn't have many of the configuration knobs available that you really _need_ to operate a decent cluster.

It's pretty bad when you reach the 40-50 node scale with 10's or 100's of TB of data. I've had about half dozen calls with their service team about this over the last year.