大数据监控脚本

#!/bin/sh

hdfs dfsadmin -report |awk -F":" 'BEGIN{print "Node_Name Status Capacity DFS_Used Non_DFS DFS_Remain DFS_Used% Contact_Time ";print "--------------- ---------- ---------- ---------- ---------- ---------- ---------- --------------- ";Flag=0}{ if($1=="Name") Flag=1; if(Flag==1 && $1=="Name") printf ("%-16s",$2); if(Flag==1 && $1=="Decommission Status ") printf ("%-11s",$2); if(Flag==1 && $1=="Configured Capacity") printf ("%-11s",substr($2,index($2,"(")+1,index($2,")")-1-index($2,"("))); if(Flag==1 && $1=="DFS Used") printf ("%-11s",substr($2,index($2,"(")+1,index($2,")")-1-index($2,"("))); if(Flag==1 && $1=="Non DFS Used") printf ("%-11s",substr($2,index($2,"(")+1,index($2,")")-1-index($2,"("))); if(Flag==1 && $1=="DFS Remaining") printf ("%-11s",substr($2,index($2,"(")+1,index($2,")")-1-index($2,"("))); if(Flag==1 && $1=="DFS Used%") printf ("%-11s",$2); if(Flag==1 && $1=="Last contact") printf ("%s:%s\n",substr($2,6,15),$3); if($1==null) Flag=0; if(substr($1,1,14)=="Live datanodes") { Total_nodes=substr($1,index($1,"(")+1,index($1,")")-index($1,"(")-1); } if(substr($1,1,14)=="Dead datanodes") {Dead_nodes=substr($1,index($1,"(")+1,index($1,")")-index($1,"(")-1);} if($1=="Configured Capacity" && Flag==0) Total_Config_Capacity=substr($2,index($2,"(")+1,index($2,")")-1-index($2,"(")); if($1=="Present Capacity" && Flag==0) Total_Present_Capacity=substr($2,index($2,"(")+1,index($2,")")-1-index($2,"(")); if($1=="DFS Remaining" && Flag==0) Total_DFS_Remain=substr($2,index($2,"(")+1,index($2,")")-1-index($2,"(")); if($1=="DFS Used" && Flag==0) Total_Used=substr($2,index($2,"(")+1,index($2,")")-1-index($2,"(")); if($1=="DFS Used%" && Flag==0) Total_Used_per=$2; }

END{

print "\n DFS Total Config_Capacity: ",Total_Config_Capacity ;

print " DFS Total Present_Capacity: ",Total_Present_Capacity ;

print " DFS Total Used ( Used% ): ",Total_Used,"(",Total_Used_per,")\n" ;

print " DFS Total Datanodes: ",Total_nodes ;

print " DFS Dead Datanodes: ",Dead_nodes,"\n";

#get every zookeeper

echo "----------------------------------------------"

echo "checking : Zookeeper status"

#ssh admin1 'source /bigdata/.bash_profile;hostname;/bigdata/zookeeper/bin/zkServer.sh status'

#ssh admin2 'source /bigdata/.bash_profile;hostname;/bigdata/zookeeper/bin/zkServer.sh status'

#ssh admin3 'source /bigdata/.bash_profile;hostname;/bigdata/zookeeper/bin/zkServer.sh status'

sh /bigdata/work/zk/check_zk.sh

echo "----------------------------------------------"

echo "checking : Namenode status"

n=0

n=`ssh admin1 'source /bigdata/.bash_profile;/bigdata/hadoop/bin/hdfs haadmin -getServiceState nn1' | grep -i active | wc -l`

m=0

m=`ssh admin2 'source /bigdata/.bash_profile;/bigdata/hadoop/bin/hdfs haadmin -getServiceState nn2' | grep -i active | wc -l`

if [ $n -eq 1 ] ;then

echo "admin1 is Namenode active"

else

echo "admin1 is Namenode standby"

if [ $m -eq 1 ];then

echo "admin2 is Namenode active"

else

echo "admin2 is Namenode standby"

echo "------------------hdfs zkfc----------------------------"

echo "checking : zkfc status"

n=0

n=`ssh admin1 'ps -ef|grep DFSZKFailoverController'|grep -v grep |wc -l`

if [ $n -eq 1 ] ;then

echo "admin1 zkfc status OK"

else

echo "admin1 zkfc status ERROR"

m=0

m=`ssh admin2 'ps -ef|grep DFSZKFailoverController'|grep -v grep|wc -l`

if [ $m -eq 1 ] ;then

echo "admin2 zkfc status OK"

else

echo "admin2 zkfc status ERROR"

echo "------------------hdfs JournalNode----------------------------"

echo "checking : JournalNode status"

n=0

n=`ssh admin1 'ps -ef|grep JournalNode'|grep -v grep|wc -l`

if [ $n -eq 1 ] ;then

echo "admin1 JournalNode status OK"

else

echo "admin1 JournalNode status ERROR"

m=0

m=`ssh admin2 'ps -ef|grep JournalNode'|grep -v grep|wc -l`

if [ $m -eq 1 ] ;then

echo "admin2 JournalNode status OK"

else

echo "admin2 JournalNode status ERROR"

k=0

k=`ssh admin3 'ps -ef|grep JournalNode'|grep -v grep|wc -l`

if [ $k -eq 1 ] ;then

echo "admin3 JournalNode status OK"

else

echo "admin3 JournalNode status ERROR"

#echo "----------------------------------------------"

#echo "checking : Hmaster status"

#ssh admin1 'source .bash_profile;get-active-master.rb'

#echo "checking : hadpmn04 status"

#ssh admin2 'source .bash_profile;get-active-master.rb'

echo "----------------------------------------------"

echo "checking : yarn node status"

大数据监控 脚本

大数据监控脚本