flink基于savepoint点的重启脚本

flink版本：1.12.4

#!/bin/bash
job_name=
start_log=
stop_log=
savepointpath=

#如果之前已经启动,从start_log里面拿到jobid
#根据job_name从yarn拿到application id
#根据jobid和application id停止任务，注意缺少application id可能停止不了任务
#start_log应该可以拿到jobid和application id，要看日志打印全不全
#hdfs://emr-cluster/flink/savepoints是一个目录,savepoint生成时会有一个随机数子目录生成
#stop过程可能要耗时，所以睡眠10s
if [ -f $start_log ] ; then
        jobid=$(grep 'Job has been submitted with JobID' $start_log | awk -F ' ' '{print $NF}')
        yid=$(yarn application -list | grep  ${job_name} | awk -F ' ' '{print $1}')
        ${FLINKHOME}/bin/flink stop -p hdfs://emr-cluster/flink/savepoints -D rest.port=8083 -d -yid  $yid  $jobid > $stop_log  2>&1
        savepointpath=$(grep 'Savepoint completed' $stop_log | awk -F ' ' '{print $NF}')
        sleep 10s
        yid=$(yarn  application -list -appStates RUNNING | awk -F ' ' '{print $1}' | grep $yid)
        if [ "$yid" != "" ] ; then
            yarn application -kill $yid
        fi
fi

#从stop_log中拿到savepoint路径，如果是首次启动savepoint是空
if [ -f $stop_log ] ; then
    savepointpath=$(grep 'Savepoint completed' $stop_log | awk -F ' ' '{print $NF}')
    if [ "$savepointpath" != ""  ] ; then
        savepointpath=" -s $savepointpath"
    fi
fi

${FLINKHOME}/bin/flink run $savepointpath -m yarn-cluster -ynm ${job_name} -yD rest.port=8083 -c com.TestJob -d service-1.0.jar $start_log 2>&1

 savepointpath：

  hdfs://emr-cluster/flink/savepoints/savepoint-a68877-43ad4f697e11

 验证：

更好的思路：

  把savepoint路径和任务信息同时保存到mysql，防止丢失

奇怪的坑：

如果不加  rest.port=8083  参数，我这边会报奇怪错误，实际不一定要加，看情况。