fluentd 推送 mariadb audit log

说明:

mariadb audit log是 mariadb 的审计日志

目的是把日志拆分成 tab 键分隔的字段

直接附上 fluentd 配置文件

<system>
  log_level error
</system>

<source>
  @type tail
  path /data/mysql_audit/*
  limit_recently_modified 86400
  open_on_every_update true
  tag mysql_audit
  read_from_head true
  pos_file /tmp/fluentd.pos
  <parse>
    @type multiline
    format_firstline /^d{8}/
    format1 /^(?<Date>d{8}) (?<Hour>d{2}):(?<Min>d{2}):(?<Sec>d{2}),(?<host>[^,]+),(?<user>[^,]+),(?<ip>[^,]+),(?<connid>[^,]+),(?<queryid>[^,]+),(?<action>[^,]+),(?<db>[^,]+),(?<message>.*),(?<retcode>d+)$/
  </parse>
</source>

<filter mysql_audit>
  @type grep
  <regexp>
    key action
    pattern QUERY
  </regexp>
  <exclude>
    key user
    pattern lagou_status
  </exclude>
  <exclude>
    key db
    pattern information_schema
  </exclude>
</filter>

<filter mysql_audit>
  @type record_transformer
  enable_ruby
  <record>
    message ${record["message"].gsub(/s/, ' ')}
    message ${record["message"].gsub(/s+/, ' ')}
  </record>
</filter>

<match mysql_audit>
  #@type stdout
  @type webhdfs
  host oss-hadoop-namenode-bjc-002
  path /mysql_audit/${Date}/${host}_${Hour}
  append true
  compress gzip
  <format>
    @type csv
    fields Date,Hour,Min,Sec,host,user,ip,action,db,message,retcode
    delimiter '    '
  </format>
  <buffer host,Date,Hour>
    @type memory
    flush_interval 20s
  </buffer>
</match>

fluentd 比 logstash 内存占用大大下降

分析同样的日志 logstash 占用700M, fluentd 占用35M

不过 cpu 占用相当,对于日志量大的机器 cpu 到100%

看来对日志做正则过滤很损耗 cpu

如果不加 open_on_every_update true  那么 td-agent 会一直保持打开过的文件描述符