Not to mention that, by default, they bill by the whole calendar month for each infrastructure and APM host. Scale your Kubernetes cluster up and then back down? That'll be an extra $18 + $36 per additional node (not $15 + $31 – that's the contract pricing, not the on-demand pricing), even if they were only online for a few days – even if they were only online for thirty seconds. Swap out a node? By default they bill by unique instances, not by number of instances, so they'll bill you for that, too.

If you ask them about it, they'll “happily” put you onto hourly on-demand billing (which seems to fix the unique vs. count thing, too), which is more expensive if you let something run on-demand for a whole month... but isn't the point of an on-demand service that you're not running it for a whole billing period?

Also, their agent logs fairly noisily, and of course its logs count toward your quota! I upgraded a cluster without also upgrading the agent, and didn't notice for about a week that each agent was happily spamming away about some long-deprecated Kubernetes API no longer being available[0]. At $2.55/million log lines and fewer than a million lines logged, this was not a costly mistake, but it's the principle of the thing. Why should an incompatibility in their agent (which their dashboard could specifically alert about, but doesn't!) cost me money?

[0] https://github.com/DataDog/helm-charts/issues/620#issuecomme...

Their billing doesn't match up with AWS's (AWS is by the second, Datadog by the hour, or at least when we used it) even though it's by the cloud instance (doesn't roll over). So we ended up paying more for the monitoring than the actual resources being monitored. When we asked for a break, they agreed to give us a 50% break IF we signed up for additional services.