--- title: "Run an active/active pair" description: "Start the two-region active/active sandbox, where both regions take writes and mirror to each other, then lose one region and watch the other carry on." url: https://monedula.dev/flock/docs/kafka-simulator/how-to/run-active-active/ --- ## Steps ### 1. Start the sandbox In the left rail of the [playground](/kafka-simulator/playground/), set **cluster type** to **Active / active**. On a narrow screen, pick it from the **scenario** menu in the top bar, under **✦ Free play**. The sandbox starts with: - two regions, `west` with `broker-1` and `broker-2`, and `east` with `broker-3` and `broker-4`, each region with its own three-voter KRaft quorum; - a topic owned by each region, `west.orders` and `east.orders`, and a mirror link each way that copies it to the other region under the same name; - a producer per region: `pw` writes to `west.orders` and `pe` to `east.orders`; - a consumer group per region, `cg1` in `west` and `cg2` in `east`, each reading both topics in its own region. ### 2. Read the two mirrors Select either arrow between the regions. The Inspector opens on that mirror link, for example **Mirror link · mm-west-east**, with its kind, direction, destination topic and per-partition **lag**. Neither link has a failover section, because neither region is a standby. The **Why** tab ends with a **DR posture** card. Here it reads "Active-active · bidirectional mirror", with the RPO as the mirror lag in each direction and the RTO as "≈ 0". ### 3. Lose one region Select the `west` region band and **kill · hard west**. The producer `pw` can no longer write, while `pe` keeps writing to `east.orders` and `cg2` keeps reading in `east`: select `pe` and it keeps getting acknowledgements, while `pw` reports an error. The records `west` took but had not mirrored are the loss window, the RPO the **DR posture** card reported, and the card still reads "≈ 0" for the RTO. Losing a region is half of a failover. Active-active has no standby to promote, because `east` already takes its own writes, so the other half is moving `west`'s clients to `east`. Select the **Failure** tab: on this sandbox it has a **DR drill (active-active)** section with a row per region. Select **lose west**, which takes down the same brokers and KRaft voters as the kill above, then **move west clients to east**: - `pw` reconnects to `east` as a new producer session, with a new producer ID. `west.orders` is a read-only mirror in `east`, so `pw` now writes `east.orders`, next to `pe`. - `cg1` moves to `east` and reads both topics there. On the mirror of `west.orders` it resumes at its committed offset translated by MirrorMaker 2, capped at what the mirror holds, so the records `west` never mirrored are skipped. On `east.orders` it keeps the same offset. The `west` row is tagged **clients on east**, and in the **Failover (DR)** section the **west → east** row names the records lost with `west` and the offset the moved group resumes at. The **DR posture** card says the same. With an **offset checkpoint (ms)** set in the timing settings, `cg1` resumes at MirrorMaker 2's last checkpoint of its offset instead, which can trail its last commit, and the row counts the records it reads again. ### 4. Bring it back Select the `west` region band again and **restart west**. Its brokers and controllers come back, and the mirror from `east` copies the records `west` missed while it was down. `west` comes back with its disks, so the mirror into `east` also copies the records it had not mirrored before the loss: they reach `east` late rather than never. A group that read them in `west` before the loss and now reads the `east` mirror reads them a second time. If `west` had not come back, those records would be gone. To finish the drill, select **move west clients back** in the **DR drill (active-active)** section. `pw` reconnects to `west` as a new session and writes `west.orders` again, and `cg1` moves home, carrying its positions back. Wait for the mirror into `west` to catch up first: a position is capped at what `west`'s copy holds, so moving back early makes `cg1` read the gap again. `pe` and `cg2` never moved. The **Failure** tab's **Inter-region link** section cuts the link between the regions with **cut inter-region link** and restores it with **heal**, and its replication speed buttons throttle both mirrors. It has no **fail over** button for this sandbox, for the same reason the mirror links have no failover section: neither region is a standby to promote. In the terminal, each broker answers for its own region's cluster: `kafka-topics --list` on `broker-1` prints `east.orders` and `west.orders`, both held in `west`, and `kafka-consumer-groups --list` after `ssh broker-3` prints only `g2`, the canvas's `cg2`. ## Related - [Run an active/passive pair and fail over](/flock/docs/kafka-simulator/how-to/run-active-passive/) is the standby model, with a failover and a failback. - [Features by release](/flock/docs/kafka-simulator/reference/features/) lists the free-play topologies this release offers. - The [Active-active multi-cluster](/kafka-simulator/scenarios/active-active/) scenario page walks through the same design as a scripted scenario. - The [Active-active DR drill](/kafka-simulator/scenarios/active-active-dr-drill/) scenario page walks through steps 3 and 4 as a scripted scenario: lose `west`, move its clients to `east`, restart `west` and move them back.