hadoop 集群 杀僵尸进程

code :

comm=$1

if [ -z $comm ]; then
    hosts=`cat /opt/sohuhadoop/conf/slaves`
    for host in $hosts
    do
        file=`echo $0|perl -F"/" -anle 'print $F[-1]' `
        scp -q $0 $host:/tmp/ ;
        echo $host mapred.Child
        ssh $host "sh /tmp/$file run"
    done ;

elif [ $comm = "run" ];then
    source /etc/profile;
    for c in `jps -ml |grep org.apache.hadoop.mapred.Child|perl -nle 'print $1 if /^(\d+).*/' `
        do
            jmap -histo $c |perl -ne '
                $sum+=$1 if /:\s+\d+\s+(\d+)/;
                
                END{
                    $mem=$sum/1024/1024;printf (" '$c' >%dm",$mem);
                }
            '
            ps -A -opid,etime|grep $c |perl -nle '
                if(/(.*)\s(\d+-\d+:\d+)/){
                    print " run=$2 ; kill";
                    system("kill -9 $1");
                }
                if(/(.*)\s(\d+:\d+)/){
                    print " run=$2 ;";
                }
                
            ' ;
        done;
    echo ;

fi;