ZooKeeper Cluster Configuration Details and Startup Verif...

This is article 27 in the Big Data series. Detailed explanation of each parameter in ZooKeeper cluster configuration file zoo.cfg, and demonstrate complete startup and verification process.

Complete illustrated version: CSDN Original | Juejin

Experimental Environment

Host	Spec	Role (After Startup)
h121.wzk.icu	2C4G	Follower
h122.wzk.icu	2C4G	Leader
h123.wzk.icu	2C2G	Follower

zoo.cfg Complete Configuration Analysis

# ---- Data Storage ----
# Directory for storing memory snapshots
dataDir=/opt/servers/apache-zookeeper-3.8.4-bin/data

# Directory for storing transaction logs (WAL), recommend separate from dataDir to avoid I/O contention
dataLogDir=/opt/servers/apache-zookeeper-3.8.4-bin/logs

# ---- Network ----
# Client connection port, default 2181
clientPort=2181

# ---- Time Parameters ----
# ZooKeeper base time unit (milliseconds), Session timeout, heartbeat all based on this
tickTime=2000

# Follower initial connection Leader timeout = initLimit × tickTime = 10 seconds
initLimit=5

# Follower and Leader sync heartbeat timeout = syncLimit × tickTime = 4 seconds
# Follower not responding within this time will be kicked out of cluster
syncLimit=2

# ---- Auto Cleanup ----
# Number of snapshots and logs to retain (keep at least 3)
autopurge.snapRetainCount=3

# Auto cleanup interval (hours), 0 means no cleanup, 1 means cleanup every hour
autopurge.purgeInterval=1

# ---- Cluster Nodes ----
# Format: server.<myid>=<host>:<quorumPort>:<electionPort>
# quorum port (2888): Follower syncs data with Leader
# election port (3888): Leader election communication
server.1=h121.wzk.icu:2888:3888
server.2=h122.wzk.icu:2888:3888
server.3=h123.wzk.icu:2888:3888

Parameter Explanation Points

tickTime: Base for all timeout calculations. Client Session timeout defaults to 2 × tickTime to 20 × tickTime (negotiated by client).

initLimit vs syncLimit:

initLimit: Maximum wait time for Follower to first connect to Leader and complete data sync. Increase when data volume is large.
syncLimit: Normal operation Follower and Leader heartbeat detection timeout. Follower will be removed from quorum on timeout.

Dual Port Design:

2888 (quorum port): Data sync, only Follower → Leader direction
3888 (election port): Used in Leader election phase, basically idle after election completes

myid File Configuration

myid is the unique identifier for ZooKeeper to recognize cluster member identity, must correspond one-to-one with server.<id> in zoo.cfg.

# Confirm dataDir directory exists
ls /opt/servers/apache-zookeeper-3.8.4-bin/data/

# h121 node
echo 1 > /opt/servers/apache-zookeeper-3.8.4-bin/data/myid

# h122 node
echo 2 > /opt/servers/apache-zookeeper-3.8.4-bin/data/myid

# h123 node
echo 3 > /opt/servers/apache-zookeeper-3.8.4-bin/data/myid

Note: myid file can only have one number, no trailing empty lines, otherwise startup will error.

Cluster Startup Process

Execute on each of three machines (order doesn’t affect final result, but recommend starting one by one for easier troubleshooting):

/opt/servers/apache-zookeeper-3.8.4-bin/bin/zkServer.sh start

Status changes during startup:

After first node starts, only 1 node, cannot form quorum, stays in LOOKING state
After second node starts, 2 nodes exceed half (3/2+1=2), triggers Leader election, generates Leader
After third node starts, joins existing cluster as Follower

Verify Cluster Status

# View each node role
zkServer.sh status

Expected output example:

# h121
Mode: follower

# h122
Mode: leader

# h123
Mode: follower

View startup logs:

tail -f /opt/servers/apache-zookeeper-3.8.4-bin/logs/zookeeper-root-server-h121.out

When LEADING or FOLLOWING appears in logs, election is complete.

Client Connection Test

# Connect to any node
/opt/servers/apache-zookeeper-3.8.4-bin/bin/zkCli.sh -server h121.wzk.icu:2181

# Verify root node
ls /

# Expected output
[zookeeper]

Firewall Port Configuration

Cluster nodes need to open these ports (distinguish between client ports and cluster internal ports):

Port	Purpose
2181	Client connection
2888	Follower and Leader data sync
3888	Leader election communication

Common Issues Troubleshooting

Q: Node stays in LOOKING state after startup

Check if myid file content is correct
Check if each node’s zoo.cfg is exactly the same (server.x configuration)
Check if firewall opens 2888/3888 ports

Q: Log reports Cannot open channel to X at election address

Hostname resolution failed between nodes, check /etc/hosts configuration

Summary

Core of zoo.cfg is three groups of parameters: storage paths (dataDir/dataLogDir), time parameters (tickTime/initLimit/syncLimit), cluster topology (server.x). With unique myid file, ZooKeeper can complete automatic election and provide services.