Flink中关于Save point的详解

1- Savepoint概述

• Flink流式计算，提供Checkpoint机制，程序自动将State进行快速Snapshot，然后进行Checkpoint保存。此外，还支持用户可以手动进行Snapshot，保存State数据，称为：SavePoint保存点。
  
• Savepoint：保存点，类似于以前玩游戏的时候，遇到难关/遇到boss，赶紧手动存个档，然
  后接着玩，如果失败了，赶紧从上次的存档中恢复，然后接着玩。
  
• Checkpoint 与Savepoint区别：
  

2- Savepoint演示

案例演示说明：运行前面Checkpoint程序（配置自动重启策略），将其运行在YARN集群上，采用WEB UI界面方式部署运行（flink on yarn Session会话模式）。

# 启动yarn session
/export/server/flink/bin/yarn-session.sh -jm 1024 -tm 1024 -s 2 -d

# 运行job-会自动执行Checkpoint
/export/server/flink/bin/flink run -d \
--class cn.itcast.flink.checkpoint.StreamCheckpointDemo \
/root/ckpt.jar hdfs://node1.itcast.cn:8020/flink-checkpoint/checkpoint

# 手动创建savepoint--相当于手动做了一次Checkpoint
/export/server/flink/bin/flink savepoint 461e4d99b207e435022a6bd2f7b2e5bd \
hdfs://node1.itcast.cn:8020/flink-checkpoint/savepoint/

# 停止job
/export/server/flink/bin/flink cancel 461e4d99b207e435022a6bd2f7b2e5bd 

# 重新启动job,手动加载savepoint数据
/export/server/flink/bin/flink run -d \ -s hdfs://node1.itcast.cn:8020/flink-checkpoint/savepoint/savepoint-461e4d-fb937e8062f7 \
--class cn.itcast.flink.checkpoint.StreamCheckpointDemo \
/root/ckpt.jar hdfs://node1.itcast.cn:8020/flink-checkpoint/checkpoint

# 停止yarn session
yarn application -kill application_1598434063229_0001