Databricks Serverless: Before You Commit
Critical Questions to Answer
Know before you move it, not after you're billed.
Before a workload moves to serverless, you need real answers on its duration, data volume, concurrency, and custom requirements, plus what the move will actually cost. Databricks native tooling answers some of that on its own. The rest is on you, unless you know exactly where to look.
The same five questions apply to every workload you run, not just one. Native tooling answers some of this manually, one workload at a time, if you go looking. A few of these questions, it can't answer at all. Here's what that actually looks like for one real workload, then the breakdown for each question below.
Nightly ETL job
What to ask, why, and how to answers it.
Each question below carries the threshold it's checking, how far Databricks native tooling gets you, and what only Unravel can tell you.
Does this job run under or over 30 minutes?
Serverless bills per second and starts almost instantly, which rewards short workloads that spin up and finish. Classic clusters have real startup and idle overhead, but once a cluster's running, flat hourly pricing doesn't care how long the job takes. 30 minutes is roughly where those two economics cross: below it, serverless's speed and per-second billing win; above it, classic's flat rate starts winning instead.
