Network partitions and split-brain in RabbitMQ - Step-by-Step Execution

Process Flow - Network partitions and split-brain

Network Partition Occurs

↓

Cluster Nodes Split

↓

Partition A

↓

Both Think They Are Master

↓

Split-Brain

↓

Data Diverges / Conflicts

↓

Manual or Automatic Resolution

This flow shows how a network partition splits a cluster into parts, causing each to think it is the master, leading to split-brain and data conflicts until resolved.

Execution Sample

RabbitMQ

rabbitmqctl cluster_status
# Simulate network partition
# Observe node states
# Resolve partition
rabbitmqctl cluster_status

This sequence shows checking cluster status, simulating a network partition causing split-brain, then resolving it and checking status again.

Process Table

Step	Action	Node A State	Node B State	Cluster Status	Notes
1	Initial cluster check	Running, connected	Running, connected	Healthy cluster	All nodes connected and synced
2	Network partition occurs	Running, isolated	Running, isolated	Partitioned cluster	Nodes cannot communicate
3	Each node assumes master	Master	Master	Split-brain detected	Both nodes think they control cluster
4	Data changes on Node A	Master with new data	Master old data	Diverged data	Data conflict begins
5	Data changes on Node B	Master with new data	Master with different new data	Data conflict worsens	Split-brain causes inconsistency
6	Manual resolution starts	Paused	Paused	Cluster paused	Prevent further changes
7	Partition fixed, nodes reconnect	Running, synced	Running, synced	Cluster healthy	Data merged or reconciled
8	Final cluster check	Running, connected	Running, connected	Healthy cluster	Split-brain resolved

💡 Partition fixed and cluster reconciled, nodes connected and data consistent

Status Tracker

Variable	Start	After Step 2	After Step 3	After Step 5	After Step 7	Final
Node A State	Running, connected	Running, isolated	Master	Master with new data	Running, synced	Running, connected
Node B State	Running, connected	Running, isolated	Master	Master with different new data	Running, synced	Running, connected
Cluster Status	Healthy cluster	Partitioned cluster	Split-brain detected	Data conflict worsens	Cluster healthy	Healthy cluster

Key Moments - 3 Insights

Why do both nodes think they are the master during a network partition?

Because the network partition isolates nodes, each loses communication with the other and assumes control to keep operating, as shown in step 3 of the execution table.

What causes data conflicts in a split-brain scenario?

When both nodes accept changes independently during partition (steps 4 and 5), their data diverges causing conflicts.

How is split-brain resolved in RabbitMQ clusters?

By pausing nodes, fixing the network, reconnecting nodes, and reconciling data as shown in steps 6 and 7.

Visual Quiz - 3 Questions

Look at the execution table, at which step do both nodes start thinking they are masters?

A Step 2

B Step 4

C Step 3

D Step 5

💡 Hint

Check the 'Cluster Status' and 'Node State' columns in step 3 where split-brain is detected.

According to the variable tracker, what is the state of Node B after step 5?

A Master with different new data

B Running, connected

C Running, isolated

D Paused

💡 Hint

Look at the 'Node B State' row under 'After Step 5' in the variable tracker.

If the network partition is not fixed, what will happen to the cluster status?

A Cluster becomes healthy

B Split-brain persists

C Cluster pauses automatically

D Nodes shut down

💡 Hint

Refer to the execution table rows 2 to 5 where partition causes split-brain and data conflicts.

Concept Snapshot

Network partitions split cluster nodes into isolated groups.
Each group may assume master role causing split-brain.
Split-brain leads to data conflicts and inconsistency.
Resolution requires fixing network and reconciling data.
RabbitMQ admins must monitor and resolve partitions quickly.

Full Transcript

Network partitions happen when cluster nodes lose communication. This splits the cluster into isolated parts. Each part thinks it is the master, causing split-brain. Both nodes accept changes independently, leading to data conflicts. To fix this, administrators pause nodes, restore network connectivity, and reconcile data. After resolution, the cluster returns to a healthy state with consistent data.