flinkit

flinkit / docs / getting started

Getting started

Install flinkit, point it at a Flink cluster, and get your first diagnosis in a couple of minutes.

Install

flinkit is in private beta. If you have access, install it from the Homebrew tap. You'll need a GitHub token that can read the beta repositories:

shell
$ export HOMEBREW_GITHUB_API_TOKEN=<your token>
$ brew install smanjakuppam-at-s3i/tap/flinkit

Verify the binary is on your PATH:

shell
$ flinkit --version

Don't have access yet? flinkit ships with a 30-day trial. Request access at hello@flinkit.org and we'll get you a token and install instructions.

Point flinkit at a cluster

flinkit talks to a job over its JobManager REST API. You tell it where the cluster is with a config file at ~/.flinkit/config.yaml that defines one or more named profiles. A minimal open-source Flink setup looks like:

~/.flinkit/config.yaml
default: local-oss

profiles:
  local-oss:
    source: oss
    platform: oss
    endpoint: http://localhost:8081

With that in place, commands connect to the default profile automatically. See Configuration & connecting for the full file format, pointing at an alternate file with FLINKIT_CONFIG, and per-platform setup (Kubernetes, Confluent).

Your first run

The normal flow is fleet → one job → root cause.

1. List the jobs — scan

shell
$ flinkit scan
JOB                 STATE        PARALLELISM   RESTARTS   CHECKPOINT
orders-rollup       RESTARTING   2             21         failing
sessionize-clicks   RUNNING      8             0          ok

Every command accepts a job by its name or its id — so you can just use orders-rollup.

2. Triage one job — inspect

shell
$ flinkit inspect orders-rollup --watch

This is the main event: a live dashboard with the diagnosis headline, checkpoint and restart stats, per-operator I/O and backpressure, ranked findings, and the bottleneck. It refreshes in place with --watch. See inspect for the panels and keybindings.

3. Pinpoint the fix — bottleneck

shell
$ flinkit bottleneck orders-rollup

When a job is throughput-bound, bottleneck walks the operator graph to the true choke point, classifies the cause, and suggests a fix.

Next steps