Kafka的2种日志清理策略感受一下

Kafka的2种日志清理策略感受一下

Kafka是一个基于日志的流处理平台,一个topic可以有多个分区(partition),分区是复制的基本单元,在单节点上,一个分区的数据文件可以存储在多个磁盘目录中,配置项是:

# A comma separated list of directories under which to store log files
log.dirs=/home/storm/dev/kafka-logs

每个分区的日志文件存储的时候又会分成一个个的segment,默认日志段(segment)的大小是1GB,segment是日志清理的基本单元,当前正在使用的segment是不会被清理的。

# The maximum size of a log segment file. When this size is reached a new log segment will be created.
log.segment.bytes=1073741824

日志清理

Kafka Broker 的日志清理功能在配置 log.cleaner.enable=true 后会开启一些清理线程,执行定时清理任务。在kafka 0.9.0之后 log.cleaner.enable 默认是true。支持的清理策略(log.cleanup.policy)有2种:delete和compact,默认是delete。

compact 清理策略(log compaction)

log compaction 实现的是一个topic的一个分区中,只保留最近的某个key对应的value,如果要删除某个消息可以发送一个墓碑消息(tomestone):(key, null)。为了展示这个过程,修改 Broker 的配置:把segment的大小调小点,清理策略改为 compact。

  •  
# 25KBlog.segment.bytes=25600 log.cleanup.policy=compact
批量发送一些带有key的消息。
  •  
➜  test-0 ./bin/kafka-console-producer.sh --broker-list localhost:9092 --topic test --property "parse.key=true" --property "key.separator=:" < msg.txt
然后可以在日志目录中看到日志文件的结构。
  •  
➜  kafka-logs cd test-0➜  test-0 ls -alhtotal 160Kdrwxrwxr-x  2 storm storm 4.0K Sep 11 18:47 .drwxrwxr-x 53 storm storm 4.0K Sep 11 18:47 ..-rw-rw-r--  1 storm storm    0 Sep 11 17:27 00000000000000000000.index-rw-rw-r--  1 storm storm   78 Sep 11 17:27 00000000000000000000.log-rw-rw-r--  1 storm storm   12 Sep 11 17:27 00000000000000000000.timeindex-rw-rw-r--  1 storm storm    0 Sep 11 17:27 00000000000000000153.index-rw-rw-r--  1 storm storm  175 Sep 11 17:27 00000000000000000153.log-rw-rw-r--  1 storm storm   10 Sep 11 17:27 00000000000000000153.snapshot-rw-rw-r--  1 storm storm   12 Sep 11 17:27 00000000000000000153.timeindex-rw-rw-r--  1 storm storm    8 Sep 11 18:47 00000000000000000296.index-rw-rw-r--  1 storm storm  25K Sep 11 17:27 00000000000000000296.log-rw-rw-r--  1 storm storm   10 Sep 11 17:27 00000000000000000296.snapshot-rw-rw-r--  1 storm storm   12 Sep 11 18:47 00000000000000000296.timeindex-rw-rw-r--  1 storm storm    0 Sep 11 18:47 00000000000000000522.index-rw-rw-r--  1 storm storm  16K Sep 11 18:47 00000000000000000522.log-rw-rw-r--  1 storm storm   10 Sep 11 18:47 00000000000000000522.snapshot-rw-rw-r--  1 storm storm   12 Sep 11 18:47 00000000000000000522.timeindex-rw-rw-r--  1 storm storm    0 Sep 11 18:47 00000000000000000665.index-rw-rw-r--  1 storm storm  16K Sep 11 18:47 00000000000000000665.log-rw-rw-r--  1 storm storm   10 Sep 11 18:47 00000000000000000665.snapshot-rw-rw-r--  1 storm storm   12 Sep 11 18:47 00000000000000000665.timeindex-rw-rw-r--  1 storm storm  10M Sep 11 18:47 00000000000000000808.index-rw-rw-r--  1 storm storm  25K Sep 11 18:47 00000000000000000808.log-rw-rw-r--  1 storm storm   10 Sep 11 18:47 00000000000000000808.snapshot-rw-rw-r--  1 storm storm  10M Sep 11 18:47 00000000000000000808.timeindex-rw-rw-r--  1 storm storm    8 Sep 11 10:44 leader-epoch-checkpoint
➜ test-0 ls -alhtotal 164Kdrwxrwxr-x 2 storm storm 4.0K Sep 11 18:48 .drwxrwxr-x 53 storm storm 4.0K Sep 11 18:48 ..-rw-rw-r-- 1 storm storm 0 Sep 11 17:27 00000000000000000000.index-rw-rw-r-- 1 storm storm 0 Sep 11 17:27 00000000000000000000.index.deleted-rw-rw-r-- 1 storm storm 73 Sep 11 17:27 00000000000000000000.log-rw-rw-r-- 1 storm storm 78 Sep 11 17:27 00000000000000000000.log.deleted-rw-rw-r-- 1 storm storm 12 Sep 11 17:27 00000000000000000000.timeindex-rw-rw-r-- 1 storm storm 12 Sep 11 17:27 00000000000000000000.timeindex.deleted-rw-rw-r-- 1 storm storm 0 Sep 11 17:27 00000000000000000153.index.deleted-rw-rw-r-- 1 storm storm 175 Sep 11 17:27 00000000000000000153.log.deleted-rw-rw-r-- 1 storm storm 12 Sep 11 17:27 00000000000000000153.timeindex.deleted-rw-rw-r-- 1 storm storm 8 Sep 11 18:47 00000000000000000296.index.deleted-rw-rw-r-- 1 storm storm 25K Sep 11 17:27 00000000000000000296.log.deleted-rw-rw-r-- 1 storm storm 12 Sep 11 18:47 00000000000000000296.timeindex.deleted-rw-rw-r-- 1 storm storm 0 Sep 11 18:47 00000000000000000522.index-rw-rw-r-- 1 storm storm 0 Sep 11 18:47 00000000000000000522.index.deleted-rw-rw-r-- 1 storm storm 0 Sep 11 18:47 00000000000000000522.log-rw-rw-r-- 1 storm storm 16K Sep 11 18:47 00000000000000000522.log.deleted-rw-rw-r-- 1 storm storm 0 Sep 11 18:47 00000000000000000522.timeindex-rw-rw-r-- 1 storm storm 12 Sep 11 18:47 00000000000000000522.timeindex.deleted-rw-rw-r-- 1 storm storm 0 Sep 11 18:47 00000000000000000665.index-rw-rw-r-- 1 storm storm 0 Sep 11 18:47 00000000000000000665.index.deleted-rw-rw-r-- 1 storm storm 175 Sep 11 18:47 00000000000000000665.log-rw-rw-r-- 1 storm storm 16K Sep 11 18:47 00000000000000000665.log.deleted-rw-rw-r-- 1 storm storm 10 Sep 11 18:47 00000000000000000665.snapshot-rw-rw-r-- 1 storm storm 12 Sep 11 18:47 00000000000000000665.timeindex-rw-rw-r-- 1 storm storm 12 Sep 11 18:47 00000000000000000665.timeindex.deleted-rw-rw-r-- 1 storm storm 10M Sep 11 18:47 00000000000000000808.index-rw-rw-r-- 1 storm storm 25K Sep 11 18:47 00000000000000000808.log-rw-rw-r-- 1 storm storm 10 Sep 11 18:47 00000000000000000808.snapshot-rw-rw-r-- 1 storm storm 10M Sep 11 18:47 00000000000000000808.timeindex-rw-rw-r--  1 storm storm    8 Sep 11 10:44 leader-epoch-checkpoint

可以看到除了当前segment之外,前面的segments都已经得到了清理/压缩,

从偏移量(offset)出现缺失可到看出来。

  •  

kafka_2.11-2.0.0 ./bin/kafka-run-class.sh kafka.tools.DumpLogSegments --deep-iteration --files /home/storm/dev/kafka-logs/test-0/00000000000000000000.log

 

  •  
Dumping /home/storm/dev/kafka-logs/test-0/00000000000000000000.logStarting offset: 0offset: 521 position: 0 CreateTime: 1536658031117 isvalid: true keysize: 4 valuesize: 0 magic: 2 compresscodec: NONE producerId: -1 producerEpoch: -1 sequence: -1 isTransactional: false headerKeys: []➜  kafka_2.11-2.0.0 ./bin/kafka-run-class.sh kafka.tools.DumpLogSegments  --deep-iteration --files /home/storm/dev/kafka-logs/test-0/00000000000000000665.logDumping /home/storm/dev/kafka-logs/test-0/00000000000000000665.logStarting offset: 665offset: 807 position: 0 CreateTime: 1536662844868 isvalid: true keysize: 4 valuesize: 100 magic: 2 compresscodec: NONE producerId: -1 producerEpoch: -1 sequence: -1 isTransactional: false headerKeys: []
标记为deleted的segments会在1天后被清除。
  •  
➜  test-0 pwd/home/storm/dev/kafka-logs/test-0➜  test-0 dateWed Sep 12 09:48:45 CST 2018➜  test-0 ls -alhtotal 72Kdrwxrwxr-x  2 storm storm 4.0K Sep 11 18:48 .drwxrwxr-x 53 storm storm 4.0K Sep 11 19:51 ..-rw-rw-r--  1 storm storm    0 Sep 11 17:27 00000000000000000000.index-rw-rw-r--  1 storm storm   73 Sep 11 17:27 00000000000000000000.log-rw-rw-r--  1 storm storm   12 Sep 11 17:27 00000000000000000000.timeindex-rw-rw-r--  1 storm storm    0 Sep 11 18:47 00000000000000000522.index-rw-rw-r--  1 storm storm    0 Sep 11 18:47 00000000000000000522.log-rw-rw-r--  1 storm storm    0 Sep 11 18:47 00000000000000000522.timeindex-rw-rw-r--  1 storm storm    0 Sep 11 18:47 00000000000000000665.index-rw-rw-r--  1 storm storm  175 Sep 11 18:47 00000000000000000665.log-rw-rw-r--  1 storm storm   10 Sep 11 18:47 00000000000000000665.snapshot-rw-rw-r--  1 storm storm   12 Sep 11 18:47 00000000000000000665.timeindex-rw-rw-r--  1 storm storm  10M Sep 11 18:47 00000000000000000808.index-rw-rw-r--  1 storm storm  25K Sep 11 18:47 00000000000000000808.log-rw-rw-r--  1 storm storm   10 Sep 11 18:47 00000000000000000808.snapshot-rw-rw-r--  1 storm storm  10M Sep 11 18:47 00000000000000000808.timeindex-rw-rw-r--  1 storm storm    8 Sep 11 10:44 leader-epoch-checkpoint

delete 清理策略(默认)

再来看看 delete 清理策略,这种策略就是我们默认看到的数据保留特点,超过特定的数据量或者时间,日志就会被删除,这里涉及的 Broker 配置参数是:log.retention.bytes 和 log.retention.hours(等价于 log.retention.minutes,log.retention.ms)默认值为:

  •  
# 需要自己根据实际情况设置log.retention.bytes=-1
# 默认的保留时间是7天log.retention.hours=168

为了能看出日志删除的效果,这里把保留时间调小,设置为60分钟,然后可以看到,除了当前正在使用的segment,前面的segments都被删除了(标记为deleted,1天后会物理删除)。

  •  
# The minimum age of a log file to be eligible for deletion due to agelog.retention.minutes=60➜  kafka-logs ls -alh test-0total 220Kdrwxrwxr-x  2 storm storm 4.0K Sep 13 11:12 .drwxrwxr-x 53 storm storm 4.0K Sep 13 11:12 ..-rw-rw-r--  1 storm storm    0 Sep 11 17:27 00000000000000000000.index.deleted-rw-rw-r--  1 storm storm   73 Sep 11 17:27 00000000000000000000.log.deleted-rw-rw-r--  1 storm storm   12 Sep 11 17:27 00000000000000000000.timeindex.deleted-rw-rw-r--  1 storm storm    0 Sep 11 18:47 00000000000000000522.index.deleted-rw-rw-r--  1 storm storm    0 Sep 11 18:47 00000000000000000522.log.deleted-rw-rw-r--  1 storm storm    0 Sep 11 18:47 00000000000000000522.timeindex.deleted-rw-rw-r--  1 storm storm    0 Sep 11 18:47 00000000000000000665.index.deleted-rw-rw-r--  1 storm storm  175 Sep 11 18:47 00000000000000000665.log.deleted-rw-rw-r--  1 storm storm   12 Sep 11 18:47 00000000000000000665.timeindex.deleted-rw-rw-r--  1 storm storm    8 Sep 12 10:50 00000000000000000808.index.deleted-rw-rw-r--  1 storm storm  25K Sep 11 18:47 00000000000000000808.log.deleted-rw-rw-r--  1 storm storm   12 Sep 12 10:50 00000000000000000808.timeindex.deleted-rw-rw-r--  1 storm storm    0 Sep 12 10:50 00000000000000001034.index.deleted-rw-rw-r--  1 storm storm  16K Sep 12 10:50 00000000000000001034.log.deleted-rw-rw-r--  1 storm storm   12 Sep 12 10:50 00000000000000001034.timeindex.deleted-rw-rw-r--  1 storm storm    0 Sep 12 10:50 00000000000000001177.index.deleted-rw-rw-r--  1 storm storm  16K Sep 12 10:50 00000000000000001177.log.deleted-rw-rw-r--  1 storm storm   12 Sep 12 10:50 00000000000000001177.timeindex.deleted-rw-rw-r--  1 storm storm    8 Sep 12 10:51 00000000000000001320.index.deleted-rw-rw-r--  1 storm storm  25K Sep 12 10:50 00000000000000001320.log.deleted-rw-rw-r--  1 storm storm   12 Sep 12 10:51 00000000000000001320.timeindex.deleted-rw-rw-r--  1 storm storm    0 Sep 12 10:51 00000000000000001546.index.deleted-rw-rw-r--  1 storm storm  16K Sep 12 10:51 00000000000000001546.log.deleted-rw-rw-r--  1 storm storm   12 Sep 12 10:51 00000000000000001546.timeindex.deleted-rw-rw-r--  1 storm storm    0 Sep 12 10:51 00000000000000001689.index.deleted-rw-rw-r--  1 storm storm  16K Sep 12 10:51 00000000000000001689.log.deleted-rw-rw-r--  1 storm storm   12 Sep 12 10:51 00000000000000001689.timeindex.deleted-rw-rw-r--  1 storm storm    8 Sep 13 11:12 00000000000000001832.index.deleted-rw-rw-r--  1 storm storm  25K Sep 12 10:51 00000000000000001832.log.deleted-rw-rw-r--  1 storm storm   12 Sep 13 11:12 00000000000000001832.timeindex.deleted-rw-rw-r--  1 storm storm  10M Sep 13 11:12 00000000000000002058.index-rw-rw-r--  1 storm storm    0 Sep 13 11:12 00000000000000002058.log-rw-rw-r--  1 storm storm   10 Sep 13 11:08 00000000000000002058.snapshot-rw-rw-r--  1 storm storm  10M Sep 13 11:12 00000000000000002058.timeindex-rw-rw-r--  1 storm storm   11 Sep 13 11:12 leader-epoch-checkpoint
➜ kafka-logs dateFri Sep 14 09:19:41 CST 2018➜ kafka-logs ls -alh test-0total 16Kdrwxrwxr-x 2 storm storm 4.0K Sep 13 11:13 .drwxrwxr-x 53 storm storm 4.0K Sep 14 09:19 ..-rw-rw-r-- 1 storm storm 10M Sep 13 11:12 00000000000000002058.index-rw-rw-r-- 1 storm storm 0 Sep 13 11:12 00000000000000002058.log-rw-rw-r-- 1 storm storm 10 Sep 13 11:08 00000000000000002058.snapshot-rw-rw-r-- 1 storm storm 10M Sep 13 11:12 00000000000000002058.timeindex-rw-rw-r-- 1 storm storm 11 Sep 13 11:12 leader-epoch-checkpoint

参考

https://zhuanlan.zhihu.com/p/44520004

Kafka Architecture: Log Compaction

4.8 Log Compaction

推荐阅读:干货|kafka流量监控的原理及实现kafka消费者分组消费的再平衡策略spark改七行源码实现高效处理kafka数据积压

Kafka的2种日志清理策略感受一下

上一篇:i.MX6ULL系统移植:uboot移植2 - 移植NXP官方uboot


下一篇:2021-06-17