Skip to content
The dokimi vaquita, wearing a headset, between two Kubernetes nodes, with test traffic passing from one pod to the other, above the name dokimi and the words performance testing for Kubernetes

Performance tests for your Kubernetes cluster, against criteria you can read. Today, dokimi measures the network between two of its nodes.

dokimi is in active development

The first release, 0.1, is coming soon. Until then, a run parses its test definition, resolves its two nodes against your cluster, claims the cluster for itself, starts its own Prometheus, checks that the cluster can run the test, runs the iperf2 test through Robot Framework, reads the measurement, evaluates your pass criteria against it, and exits with its verdict. Robot's report comes back to your machine. What it does not do yet is write the CSV and JSONL result files. Everything marked works today on this site really does. Everything else is marked as planned.

What dokimi does today

  • A test definition in YAML or Robot


    Write a test as a YAML document or as a Robot Framework suite. Both formats describe the same things, and both are strict: anything dokimi does not understand is refused, with the line and the reason.

    works today Test definitions

  • Pass criteria you can read


    Throughput Should Be At Least 950 megabits per second. Criteria are part of the test definition, with the unit written out. Every run evaluates them against what each pair measured and lists each one with its value. A pair that couldn't be measured is an error, never a pass.

    works today Pass criteria

  • Runs in the cluster, as a one-shot Job


    The CLI submits a one-shot Kubernetes Job and streams its log. Close your laptop and the run carries on; reattach later from any machine. No controller stays behind between runs. What stays is each run's own Prometheus, kept so the run's history outlives it, until you remove it.

    works today How it works

  • Robot Framework runs the test


    Whichever format you wrote, the run renders it to a Robot suite and runs it. The suite deploys the client and server pods, waits until both are ready, and runs iperf2. Robot's log.html and report.html come back to your machine, next to the exact suite that ran.

    works today How it works

  • Look before it creates anything


    --dry-run renders every manifest a run would create and creates none of them. Given a kubeconfig, it also checks your node names, read-only. A missing, cordoned or tainted node is reported with the fix.

    works today Using the CLI

  • Touches only what it created


    Every object a run creates carries the run's labels, and every lookup and delete is scoped by them. Anything without those labels is never modified. A deadline stops an orphaned run on its own. A run that didn't pass keeps its test pods for you to look at, and dokimi cleanup removes them. And a cluster runs one run at a time: a second run is refused while the first holds it.

    works today Runs and exit codes

  • Exit codes made for CI


    The exit code is the verdict: 0 passed, 1 a criterion didn't hold, 2 a test couldn't be validly measured. 3 to 5 mean there is no verdict, so a run that stops early can never be mistaken for one that passed.

    works today Exit codes

What the first release adds

planned for 0.1

The first release completes one path end to end. It already runs one iperf2 UDP test between two nodes you name, next to a managed Prometheus, checks the cluster first, reads the measurement, evaluates the pass criteria against it, records where everything ran, cleans up after itself and exits with its verdict. What it adds is the rest of that path: dokimi writes the results as CSV and JSONL, and an end-to-end acceptance run checks the whole of it.

See the roadmap