您的位置：首页 > 其它

Kafka集群安装

2014-08-29 09:51 399 查看

安装Kafka集群

假设我们有集群中，需要配置4个broker，形成下面图表的Kafka集群。

2.1 配置文件

配置所有的Kafka的Producer文件，其中，brokerpid的值是独一无二的数字值。几个核心属性如下：

<span style="font-size:18px;"># The id of the broker.This must be set to a unique integer for each broker.
broker.id=11
# The port the socketserver listens on
port=9092
# Hostname the broker willbind to. If not set, the server will bind to all interfaces
host.name=hadoop-master
#zookeeper 集群
zookeeper.connect=machine-1:2222,machine-2:2222,machine-0:2222</span>

由于Kafka依赖于Zookeeper集群，所以，必须先启动Zookeeper集群。这里不作具体介绍。

2.2 broker属性配置

在机器machine-0和Hadoop-master，配置好对应broker配置文件，两个机器上的配置属性相同。

Hadoop-master 上，Kafka的server.properties配置:

<span style="font-size:18px;"># The id of the broker.This must be set to a unique integer for each broker.
broker.id=11

#############################Socket Server Settings #############################

# The port the socketserver listens on
port=9092

# Hostname the broker willbind to. If not set, the server will bind to all interfaces
host.name=hadoop-master

# Hostname the broker willadvertise to producers and consumers. If not set, it uses the
# value for"host.name" if configured. Otherwise, it will use the value returned from
#java.net.InetAddress.getCanonicalHostName().
#advertised.host.name=<hostnameroutable by clients>

# The port to publish toZooKeeper for clients to use. If this is not set,
# it will publish the sameport that the broker binds to.
#advertised.port=<portaccessible by clients>

# The number of threadshandling network requests
num.network.threads=3

# The number of threadsdoing disk I/O
num.io.threads=8

# The send buffer(SO_SNDBUF) used by the socket server
socket.send.buffer.bytes=1048576

# The receive buffer(SO_RCVBUF) used by the socket server
socket.receive.buffer.bytes=1048576

# The maximum size of arequest that the socket server will accept (protection against OOM)
socket.request.max.bytes=104857600

#############################Log Basics #############################

# A comma seperated list ofdirectories under which to store log files
log.dirs=/opt/kafka/logs

# The default number of logpartitions per topic. More partitions allow greater
# parallelism forconsumption, but this will also result in more files across
# the brokers.
num.partitions=3

#############################Log Flush Policy #############################

# Messages are immediatelywritten to the filesystem but by default we only fsync() to sync
# the OS cache lazily. Thefollowing configurations control the flush of data to disk.
# There are a few importanttrade-offs here:
# 1. Durability: Unflushed data may be lostif you are not using replication.
# 2. Latency: Very large flush intervals maylead to latency spikes when the flush does occur as there will be a lot of datato flush.
# 3. Throughput: The flush is generally themost expensive operation, and a small flush interval may lead to exceessiveseeks.
# The settings below allowone to configure the flush policy to flush data after a period of time or
# every N messages (orboth). This can be done globally and overridden on a per-topic basis.

# The number of messages toaccept before forcing a flush of data to disk
#log.flush.interval.messages=10000

# The maximum amount oftime a message can sit in a log before we force a flush
#log.flush.interval.ms=1000

#############################Log Retention Policy #############################

# The followingconfigurations control the disposal of log segments. The policy can
# be set to delete segmentsafter a period of time, or after a given size has accumulated.
# A segment will be deletedwhenever *either* of these criteria are met. Deletion always happens
# from the end of the log.

# The minimum age of a logfile to be eligible for deletion
log.retention.hours=168

# A size-based retentionpolicy for logs. Segments are pruned from the log as long as the remaining
# segments don't drop belowlog.retention.bytes.
#log.retention.bytes=1073741824

# The maximum size of a logsegment file. When this size is reached a new log segment will be created.
log.segment.bytes=536870912

# The interval at which logsegments are checked to see if they can be deleted according
# to the retention policies
log.retention.check.interval.ms=60000

# By default the logcleaner is disabled and the log retention policy will default to just deletesegments after their retention expires.
# Iflog.cleaner.enable=true is set the cleaner will be enabled and individual logscan then be marked for log compaction.
log.cleaner.enable=false

#############################Zookeeper #############################

# Zookeeper connectionstring (see zookeeper docs for details).
# This is a comma separatedhost:port pairs, each corresponding to a zk
# server. e.g."127.0.0.1:3000,127.0.0.1:3001,127.0.0.1:3002".
# You can also append anoptional chroot string to the urls to specify the
# root directory for allkafka znodes.
zookeeper.connect=machine-1:2222,machine-2:2222,machine-0:2222
#server.1=machine-0:2888:3888
#server.2=machine-1:2888:3888
#server.3=machine-2:2888:3888

# Timeout in ms forconnecting to zookeeper
zookeeper.connection.timeout.ms=1000000
</span>

Hadoop-master 上，Kafka的server-1.properties配置:

<span style="font-size:18px;"># The id of the broker.This must be set to a unique integer for each broker.
broker.id=12