Anatomy baseline
Anatomy
RF=3 across 3 racks. Read ISR, leader, follower, LEO, and HW on a healthy cluster.
Visualize Apache Kafka in the browser
Kafka Simulator: break a cluster and see why it reacts.
Change a setting or kill a broker, then step through what producers, brokers, replicas, consumers and the metrics that watch them do.
Produce committed. Healthy: three replicas of every partition are in sync, so the producer's acks=all write commits.
broker-2 hard kill. broker-2 is offline. ISR shrinks on every partition it replicated to.
Leader re-elected. broker-2 led partition-1, so the controller moved leadership to broker-1. Two replicas are still in sync, which meets min.insync.replicas=2, so acks=all writes still commit.
broker-3 hard kill. Below min.insync.replicas - ISR is down to 1. Any acks=all write would now be refused with NotEnoughReplicas until a replica returns.
Recovery (broker-2 restart). broker-2 rejoined; ISR is back to 2 on every partition. broker-3 is still down.
Open this run in the playground- Models Apache Kafka 4.4. Every run is seeded, so it replays the same way.
- 136 scenarios, plus free play (single-DC to stretched 3-DC).
- Free, with no signup and no backend: everything runs in your browser.
Watch acks=all cross the durability boundary.
RF=3, min.insync.replicas=2, acks=all. Kill one broker; the cluster degrades but writes still commit. Kill a second; durability cannot be satisfied, and produce fails with NotEnoughReplicas. The figure above is visualized from a real simulator run: it is the simulator's own canvas at each frame of scenario 03.2.1, drawn when the site is built. Below is what the second kill leaves in the log and in the metrics.
[t+14.802] produce_request → orders/p1
[t+14.803] leader: broker-1 (epoch 3)
[t+14.803] isr: [broker-1]
[ERROR] NotEnoughReplicas - acks=all needs 2 in-sync replicas,
but only 1 is currently in ISR.| Metric | Δ | Was | Now |
|---|---|---|---|
| UnderReplicatedPartitions | +3 | 0 | 3 |
| IsrShrinksPerSec | +1.2 | 0.0 | 1.2 |
| FailedProduceRequestsPerSec | +4.2 | 0.0 | 4.2 |
| LeaderElectionRateAndTimeMs | - | 0 | 3, 38ms |
136 scenarios, each checked against a recorded run.
The curriculum, 16 modules
- Anatomy 8 scenarios
- Producers & EOS 13 scenarios
- Consumer groups 13 scenarios
- Replication 12 scenarios
- Transactions 10 scenarios
- Storage 12 scenarios
- KRaft 4 scenarios
- Operations 6 scenarios
- Chaos 6 scenarios
- Disaster recovery 22 scenarios
- Tiered storage 2 scenarios
- Schema registry 3 scenarios
- Authorization (ACLs & RBAC) 2 scenarios
- Share groups 9 scenarios
- Capstones 5 scenarios
- Terminal ops 9 scenarios
These 136 scenarios cover every pack shipped through v2.0 - Governance, queues & capstones - the complete 136-scenario library. The changelog lists what each pack added.
Canonical failure
Replication
Kill two brokers under acks=all, min.ISR=2: the run shown above.
acks trade-off
Producer
The same workload at acks=0, 1 and all, compared on throughput, latency and durability.
Unclean leader
Replication
Why unclean.leader.election.enable=true can silently lose data.
Idempotence retry
Producer
Retry under failure - with and without idempotence. Watch a duplicate appear, then disappear.
min.ISR boundary
Replication
Move min.insync.replicas from 1 to 3 on a running cluster with one slider.
Slow follower
Replication
A slow follower drops out of ISR. How replica.lag.time.max.ms decides.
Compaction
Anatomy
Repeated keys to a compact topic. Then a tombstone, and the key is gone.
Scenarios and free play share one playground.
Brokers, partitions, RF, min.ISR, racks, regions, timing and a deterministic seed are all yours to set, then share by URL.
- Zoom - level of detail and pinch-to-zoom on a phone keep a growing cluster readable.
- Frames - step every tick with a per-event inspector; entities link straight to the parts they touch.
- WHY tab - live health checks explain why the cluster sits where it does.
- Steps tab - a click-to-jump walkthrough of every frame, each captioned with what just happened.
Visualize your own Kafka cluster in the playground.
monedula-sim-link takes a read-only snapshot of a real Kafka cluster and prints a playground link that opens the simulator shaped like it. Its MCP server lets an AI agent build the same links.
monedula-sim-link connect --brokers localhost:9092 --topics orders,paymentsWhat it models, and what it does not.
Modeled
- Producer → leader → follower fetch → HW/LEO advance, across multiple topics
- Producer tuning - acks, idempotence &
max.in.flightordering, compression (zstd/gzip/snappy),buffer.memorybackpressure, byte-rate quotas, rack-aware partitioning (partitioner.rack.aware, KIP-1123) - Transactions & EOS - PIDs, sequences, LSO, zombie fencing
- ISR membership, leader epoch, unclean leader election, throttled reassignment (
replication.throttled.rate) - Retention - time- & size-based (
retention.bytes), compaction & tombstones, tiered storage withlocal.retention.ms(KIP-405) - Consumer groups (eager & cooperative rebalances), static membership (KIP-345), share groups (KIP-932)
- Control plane - KRaft quorum, controller epoch & failover, plus the pre-4.0 ZooKeeper control plane for comparison (Kafka 4.x has no ZooKeeper mode)
- Multi-DC DR - stretched 2.5-DC / MRC, observer promote & demote, active-passive failover & failback, active-active; MirrorMaker 2 offset translation & checkpoints; Cluster Linking (offset-preserving mirror, promote, bidirectional switchover)
- Schema Registry - subjects, versions, compatibility modes
- Authorization - ACLs & RBAC (principal / resource / operation, role bindings, deny-by-default)
- Hard kill, graceful shutdown, restart, slow follower, network partition
- Real Kafka metric names, directionally correct
Not modeled
- GC pauses, OS page cache, network buffers, byte-exact sizes
- SASL, mTLS, encrypted transport - the auth handshake & wire crypto (ACL/RBAC authorization itself is modeled)
- Kafka Streams & Connect - core broker/client protocol only
- New in Kafka 4.4 - share-group dead-letter queues (KIP-1191) and delayed tiered-storage upload (KIP-1241)
- Magnitudes of metric movement - calibrated direction, illustrative scale
Simplified. 28 behaviors differ from Kafka 4.4 on purpose, such as
batch.size counting records and timeouts shortened so a scenario plays in seconds, and the
scenarios say so where it matters. See the fidelity page.
Modeled baseline: Apache Kafka 4.4. Not for production-as-is.
Part of Monedula Flock
The same people build tools for real clusters.
The simulator is where we explain Kafka. These are what we reach for when the cluster is not a simulation.
- monedula-gitops GitOps for Apache Kafka: topics, access, quotas, schemas, SCRAM users and Confluent RBAC, from a CLI or a Kubernetes operator.
- monedula-acl-rbac-converter A CLI that converts Apache Kafka ACLs into Confluent RBAC role bindings through a checksummed plan and verifies effective access before any ACL is deleted.
- monedula-metrics-reporter A Kafka MetricsReporter plugin that ships broker, JVM and KIP-714 client metrics over OTLP; a collector outage never blocks Kafka.
- monedula-sim-link A read-only Go CLI that turns a real Kafka cluster into a shareable Kafka Simulator link, with an MCP server for AI agents.
Questions
Does anything leave my browser?
Nothing about your scenarios or actions is sent anywhere. The site counts page views with cookieless analytics (GoatCounter), as every page on monedula.dev does.
Which Kafka version does it model?
Apache Kafka 4.4.
How faithful is it?
It models how Kafka behaves, approximates some things by their effect, and leaves others out, such as GC pauses and the page cache. The fidelity page lists what is modeled, approximated and left out.
Can an application connect to it like a Kafka emulator?
No. It models how a cluster behaves; it does not speak the Kafka protocol. For a local broker, run Apache Kafka in Docker.
Can I see my own cluster?
Yes, with monedula-sim-link.
Does it work on a phone?
Yes. On a phone the cluster is drawn as a vertical stack you can pinch to zoom, the details open in a bottom sheet that always shows the current step, and the secondary actions fold into a menu in the top bar.
How often do new packs ship?
The simulator grows in numbered releases, each a themed pack of scenarios and free-play features. The roadmap lists every version and its theme.
Built by Monedula. Apache Kafka and Kafka are trademarks of The Apache Software Foundation. Monedula Kafka Simulator is an independent educational tool and is not endorsed by or affiliated with The Apache Software Foundation.