b=11171

i=cliffw,jitendra

Create a more portable way of remote shelling:
- set $DSH variable to either rsh or ssh (default)
The location of the results can be set with rslt_loc which defaults to /tmp.
Eliminate the "scp brutal hack" needed for non-shared filesystems.
A related work item could be to do a test to see if $rslt_loc is actually
shared between the nodes and not use remote_shell if it is indeed shared.
Specifically invoke gawk in obdfilter-survey where it uses gnu awk extensions

diff --git a/lustre-iokit/obdfilter-survey/libecho b/lustre-iokit/obdfilter-survey/libecho
index 3beeecf..4815445 100644 (file)
--- a/lustre-iokit/obdfilter-survey/libecho
+++ b/lustre-iokit/obdfilter-survey/libecho
@@ -30,6 +30,33 @@ declare -a vmstatpids
 declare -a do_unload_echo
 
 
 declare -a do_unload_echo
 
 

+DSH=${DSH:-"ssh"}
+
+dsh () {
+    local node="$1"
+    local user="$2"
+    shift 2
+    local command="$@"
+
+    local here=$(pwd)
+
+    command="cd $here; export PATH=/sbin:/usr/sbin:\$PATH; $command"
+
+    case $DSH in
+       ssh)
+           if [ -n "$user" ]; then
+               user="$user@"
+           fi
+           $DSH $user$node "$command"
+           ;;
+       rsh)
+           if [ -n "$user" ]; then
+               user="-l $user"
+           fi
+           $DSH $user $node "$command"
+           ;;
+    esac
+}

 # This function executes the command sent through parameters to host
 # parameters
 # 1. hostname
 # This function executes the command sent through parameters to host
 # parameters
 # 1. hostname


@@ -53,11 +80,17 @@ custom_remote_shell () {
 remote_shell () {
     host=$1
     shift
 remote_shell () {
     host=$1
     shift

-    cmds="$*"
+    cmds="$@"

     if [ "$host" = "localhost" -o "$host" = `uname -n` ]; then
        eval "$cmds"
     else
     if [ "$host" = "localhost" -o "$host" = `uname -n` ]; then
        eval "$cmds"
     else

-       custom_remote_shell $host "$cmds"
+       # split $host into $host and $user
+       local user=""
+       if [[ $host == *@* ]]; then
+           user=${host%@*}
+           host=${host#*@}
+       fi
+       dsh $host "$user" "$cmds"

     fi
 }
 
     fi
 }
 

diff --git a/lustre-iokit/obdfilter-survey/obdfilter-survey b/lustre-iokit/obdfilter-survey/obdfilter-survey
index 2d85b8f..b2abaeb 100755 (executable)
--- a/lustre-iokit/obdfilter-survey/obdfilter-survey
+++ b/lustre-iokit/obdfilter-survey/obdfilter-survey
@@ -99,7 +99,7 @@ destroy_objects () {
 
 get_stats () {
     local rfile=$1
 
 get_stats () {
     local rfile=$1

-    awk < $rfile \
+    gawk < $rfile \

        '/^Selected device [0-9]+$/ {n = 0; next}\
        /error/ {n = -1; exit}\
        /^[0-9]+\/[0-9]+ Total: [0-9]+\.[0-9]+\/second$/ {n++; v=strtonum($3); \
        '/^Selected device [0-9]+$/ {n = 0; next}\
        /error/ {n = -1; exit}\
        /^[0-9]+\/[0-9]+ Total: [0-9]+\.[0-9]+\/second$/ {n++; v=strtonum($3); \


@@ -175,7 +175,8 @@ fi
 
 # result file prefix (date/time + hostname makes unique)
 # NB ensure path to it exists
 
 # result file prefix (date/time + hostname makes unique)
 # NB ensure path to it exists

-rslt=${rslt:-"/tmp/obdfilter_survey_`date +%F@%R`_`uname -n`"}
+rslt_loc=${rslt_loc:-"/tmp"}
+rslt=${rslt:-"$rslt_loc/obdfilter_survey_`date +%F@%R`_`uname -n`"}

 
 # what tests to run (first must be write)
 tests_str=${tests_str:-""}
 
 # what tests to run (first must be write)
 tests_str=${tests_str:-""}


@@ -263,7 +264,7 @@ if [ $case == "network" ]; then
        echo "Specify the server NID"
        exit 1;
     fi
        echo "Specify the server NID"
        exit 1;
     fi

-    osc_names_string=`ssh root@"$server_nid" lctl dl`
+    osc_names_string=`dsh $server_nid root "lctl dl"`

     count=0;
     for name in $osc_names_str; do
         count=$((count+1))
     count=0;
     for name in $osc_names_str; do
         count=$((count+1))


@@ -275,7 +276,7 @@ if [ $case == "network" ]; then
     fi
     # Now do the server setup
     setup_srv_obd $server_nid "ost_testfs"
     fi
     # Now do the server setup
     setup_srv_obd $server_nid "ost_testfs"

-    op_string=`ssh root@"$server_nid" lctl dl`
+    op_string=`dsh $server_nid root "lctl dl"`

     
     obdecho=0
     ost=0 
     
     obdecho=0
     ost=0 


@@ -481,9 +482,7 @@ for ((rsz = $rszlo; rsz <= $rszhi; rsz*=2)); do
                 t0=`date +%s.%N`
                 pidcount=0
                 for host in ${unique_hosts[@]}; do
                 t0=`date +%s.%N`
                 pidcount=0
                 for host in ${unique_hosts[@]}; do

-                   # brutal hack to deal with a non-shared /tmp
-                   scp -q ${cmdsf}_${host} ${host}:/tmp > /dev/null
-                    remote_shell $host bash ${cmdsf}_${host} &
+                    remote_shell $host bash < ${cmdsf}_${host} &

                     pidarray[$pidcount]=$!
                     pidcount=$((pidcount+1))
                 done
                     pidarray[$pidcount]=$!
                     pidcount=$((pidcount+1))
                 done


@@ -509,8 +508,7 @@ for ((rsz = $rszlo; rsz <= $rszhi; rsz*=2)); do
                    tmpfi="${tmpf}_$idx"
                    echo "=============> $test $client_name" >> $workf
                    host="${host_names[$idx]}"
                    tmpfi="${tmpf}_$idx"
                    echo "=============> $test $client_name" >> $workf
                    host="${host_names[$idx]}"

-                   scp -q ${host}:$tmpfi $tmpfi > /dev/null
-                   cat $tmpfi >> $workf
+                   remote_shell $host cat $tmpfi >> $workf

                    get_stats $tmpfi >> $tmpf
                    rm $tmpfi
                done
                    get_stats $tmpfi >> $tmpf
                    rm $tmpfi
                done