logkafka - Collect logs and send lines to Apache Kafka 0.8+

Introduction 中文文档

logkafka sends log file contents to kafka 0.8 line by line. It treats one line of file as one kafka message.

See FAQ if you wanna deploy it in production environment.

Features

• log collecting configuration management with zookeeper
• [log path with timeformat (collect files chronologically)](/mirrors/logkafka/tree/master/docs/Features.md#Log Path Pattern)
• log file rotating
• batching messages
• compression (none, gzip, snappy)
• [message regex filter](/mirrors/logkafka/tree/master/docs/Features.md#Regex Filter)
• [user-defined line delimiter](/mirrors/logkafka/tree/master/docs/Features.md#Line Delimiter)
• user-defined monitor

Differences with other log aggregation and monitoring tools

The main differences with flume, fluentd, logstash are

• Management of log collecting configs and state:

  flume, fluentd, logstash keep log file configs and state locally: start local server for management of log file configs and state.

  logkafka keep log file configs and state in zookeeper: watch the node in zookeeper for configs changing, record file position in local position file, and push position info to zookeeper periodically.

• Order of log collecting

  flume, fluentd, logstash all have INPUT type ‘tail’, they collecting all files simultaneously, without considering chronological order of log files.

  logkafka will collect files chronologically.

Users of logkafka

• Qihoo360 - Collecting openstack operation logs
• ICBC - Collecting ssdb operation log
• Didi Dache
• Meituan
• Let me know if you are using logkafka.

details…

Supported operating systems

• GNU/Linux: recommended
• MacOS: >=10.10

Requirements

• librdkafka

• libzookeeper_mt

• libuv

• libpcre2

• PHP 5.3 and above (with zookeeper extension)

Build

Two methods, choose accordingly.

1. Install librdkafka(>0.8.6), libzookeeper_mt, libuv(>v1.6.0), libpcre2(>10.20) manually, then

   cmake -H. -B_build -DCMAKE_INSTALL_PREFIX=_install
   cd _build
   make -j4
   make install

2. Just let cmake handle the dependencies ( cmake version >= 3.0.2 ).

   cmake -H. -B_build -DCMAKE_INSTALL_PREFIX=_install \
                      -DINSTALL_LIBRDKAFKA=ON \
                      -DINSTALL_LIBZOOKEEPER_MT=ON \
                      -DINSTALL_LIBUV=ON \
                      -DINSTALL_LIBPCRE2=ON
   cd _build
   make -j4
   make install

   If any of the libs installation fail, please manually install it, and set the corresponding config -DINSTALL_LIBXXX=OFF.

Usage

Note: If you already have kafka and zookeeper installed, you can start from step 2 and replace zk connection string with your own in the following steps, default is 127.0.0.1:2181.

1. Deploy Kafka and Zookeeper in local host

   tools/grid bootstrap

2. Start logkafka

   • local conf

   Customizing _install/conf/logkafka.conf to your needs

    zookeeper.connect = 127.0.0.1:2181
    pos.path       = ../data/pos.myClusterName
    line.max.bytes = 1048576
    ...

   • run

   Run in the foreground

   _install/bin/logkafka -f _install/conf/logkafka.conf -e _install/conf/easylogging.conf

   Or as a daemon

   _install/bin/logkafka --daemon -f _install/conf/logkafka.conf -e _install/conf/easylogging.conf

3. Configs Management

Use UI or command line tools.

3.1 UI (with kafka-manager)

We add logkafka as one kafka-manager extension. You need to install and start kafka-manager, add cluster with logkafka enabled, then you can manage logkafka with the ‘Logkafka’ menu.

• How to add cluster with logkafka enabled

• How to create new config

• How to delete configs

• How to list configs and monitor sending progress

3.2 Command line tools

We use php script (tools/log_config.php) to create/delete/list collecting configurations in zookeeper nodes.

If you do not know how to install php zookeeper module, check this.

• How to create configs

  Example:

  Collect apache access log on host “test.qihoo.net” to kafka brokers with zk connection string “127.0.0.1:2181”. The topic is “apache_access_log”.

  php tools/log_config.php --create \
                           --zookeeper_connect=127.0.0.1:2181 \
                           --logkafka_id=test.qihoo.net \
                           --log_path=/usr/local/apache2/logs/access_log.%Y%m%d \
                           --topic=apache_access_log

  Note:

  • [hosname, log_path] is the key of one config.

• How to delete configs

  php tools/log_config.php --delete \
                           --zookeeper_connect=127.0.0.1:2181 \
                           --logkafka_id=test.qihoo.net \
                           --log_path=/usr/local/apache2/logs/access_log.%Y%m%d

• How to list configs and monitor sending progress

  php tools/log_config.php --list --zookeeper_connect=127.0.0.1:2181

  shows

  logkafka_id: test.qihoo.net
   log_path: /usr/local/apache2/logs/access_log.%Y%m%d
   Array
   (
       [conf] => Array
           (
               [logkafka_id] => test.qihoo.net
               [log_path] => /usr/local/apache2/logs/access_log.%Y%m%d
               [topic] => apache_access_log
               [partition] => -1
               [key] =>
               [required_acks] => 1
               [compression_codec] => none
               [batchsize] => 1000
               [message_timeout_ms] => 0
               [follow_last] => 1
               [valid] => 1
           )
  
   )

More details about configuration management, see `php tools/log_config.php --help`.

Benchmark

We test with 2 brokers, 2 partitions

Name	Description
rtt min/avg/max/mdev	0.478/0.665/1.004/0.139 ms
message average size	1000 bytes
batchsize	1000
required_acks	1
compression_codec	none
message_timeout_ms	0
peak rates	20.5 Mb/s

Third Party

The most significant third party packages are:

• confuse

• easylogging

• tclap

• rapidjson

Thanks to the creators of these packages.

Developers

1. Make sure you have lcov installed, check this

compile with unittest and debug type

cmake -H. -B_build -DCMAKE_INSTALL_PREFIX=_install \
                   -Dtest=ON \
                   -DCMAKE_BUILD_TYPE=Debug
cd _build
make
make logkafka_coverage  # run unittest

2. Google C++ Style Guide

The code that not conform to this rule should be fixed before committing, you can use cpplint to check the modified files.

TODO

1. Multi-line mode