Quickstart
From signup to a running Flink job reading Kafka.
This is the whole path, end to end. It assumes nothing except a credit card and a Flink job you want to run.
1. Create an environment
An environment is a dedicated Kubernetes cluster that belongs only to you. Nothing is shared with another customer — not the control plane, not the nodes, not the storage.
From the dashboard, choose New environment, pick a name and a size, and create it.
Provisioning takes 10–20 minutes. That is AWS building a real cluster, not a queue we control, and we would rather tell you the honest number than show a fast spinner. The page shows each stage as it completes; you can close it and come back.
2. Enable Kafka (optional)
If your job reads or writes Kafka, turn it on from the environment's Kafka tab.
Enabling allocates a dedicated 3-broker cluster on its own machines. It costs 6 vNodes from your environment's capacity, and the page shows you that number and how much you have free before you confirm. Brokers take a few minutes to form a cluster.
You do not configure anything else. Replication, retention defaults, and broker settings are fixed — that is the deal: you get a Kafka that works, and we get one shape to support.
3. Create a topic
Also on the Kafka tab. You choose the name, the partition count (up to 12) and the retention (up to 7 days). Durability comes from S3, not a replication factor — brokers are stateless, so there is no replica count to choose.
Why retention is capped at 7 days
Kafka retention lives on the brokers' disks, which are part of what you are paying for. For longer-term storage, write from Flink to S3 — that is cheaper per GB by a wide margin and it is what object storage is for. See Checkpoints & storage.
4. Deploy a job
Jobs → New job. Pick the Read from Kafka preset if you are consuming from your built-in Kafka: it wires the bootstrap address and credentials in for you and turns on exactly-once checkpointing, which is what stops a restart re-processing or skipping messages.
Point it at your JAR, set the parallelism, and deploy.
Before it writes anything, we check the things that would otherwise fail after a successful-looking deploy — parallelism that exceeds your task slots, checkpointing with nowhere to write, a Kafka job on an environment without Kafka. If something is wrong you get a sentence, not a CrashLoopBackOff.
5. Watch it run
- Status updates as the job moves through DEPLOYING → RUNNING.
- Logs has a live tail and a searchable history (kept as long as you configured).
- Grafana has throughput, checkpoint duration and backpressure. The link is on the environment page.
- Flink UI is proxied through the dashboard — no port-forwarding, no kubeconfig.