sneak/ppss
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

Improved distributed ppss

Browse Source
This commit is contained in:
Louwrentius 2009-09-20 09:04:23 +00:00
parent 7ca08b87cf
commit 1da1de4b18
1 changed files with 26 additions and 12 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

38
ppss.sh
View File

@ -38,7 +38,7 @@ trap 'kill_process; ' INT
# Setting some vars. Do not change. # Setting some vars. Do not change.
SCRIPT_NAME="Distributed Parallel Processing Shell Script" SCRIPT_NAME="Distributed Parallel Processing Shell Script"
SCRIPT_VERSION="2.20" SCRIPT_VERSION="2.22"
# The first argument to this script is always the 'mode'. # The first argument to this script is always the 'mode'.
MODE="$1" MODE="$1"
@ -48,7 +48,7 @@ shift
# export PPSS_DIR=/path/to/workingdir # export PPSS_DIR=/path/to/workingdir
if [ -z "$PPSS_DIR" ] if [ -z "$PPSS_DIR" ]
then then
PPSS_DIR="./ppss" PPSS_DIR="ppss"
fi fi
if [ ! -e "$PPSS_DIR" ] if [ ! -e "$PPSS_DIR" ]
@ -159,7 +159,7 @@ showusage () {
echo echo
echo -e "The following options are used for distributed execution of PPSS." echo -e "The following options are used for distributed execution of PPSS."
echo echo
echo -e "--server | -s Specifies the SSH server that is used for communication between nodes." echo -e "--master | -m Specifies the SSH server that is used for communication between nodes."
echo -e " Using SSH, file locks are created, informing other nodes that an item " echo -e " Using SSH, file locks are created, informing other nodes that an item "
echo -e " is locked. Also, often items, such as files, reside on this host. SCP " echo -e " is locked. Also, often items, such as files, reside on this host. SCP "
echo -e " is used to transfer files from this host to nodes for local procesing." echo -e " is used to transfer files from this host to nodes for local procesing."
@ -167,7 +167,7 @@ showusage () {
echo -e "--node | -n File containig a list of nodes that act as PPSS clients. One IP / DNS " echo -e "--node | -n File containig a list of nodes that act as PPSS clients. One IP / DNS "
echo -e " name per line." echo -e " name per line."
echo echo
echo -e "--key | -k The SSH key that a node uses to connect to the server." echo -e "--key | -k The SSH key that a node uses to connect to the master."
echo echo
echo -e "--known-hosts | -K The file that contains the server public key. Can often be found on " echo -e "--known-hosts | -K The file that contains the server public key. Can often be found on "
echo -e " hosts that already once connected to the server. See the file " echo -e " hosts that already once connected to the server. See the file "
@ -243,11 +243,13 @@ exec_cmd () {
CMD="$1" CMD="$1"
if [ ! -z "$SSH_SERVER" ] && [ "$SECURE_COPY" == "1" ] if [ ! -z "$SSH_SERVER" ]
then then
ssh $SSH_OPTS $SSH_KEY $USER@$SSH_SERVER $CMD ssh $SSH_OPTS $SSH_KEY $USER@$SSH_SERVER $CMD
return $?
else else
eval "$CMD" eval "$CMD"
return $?
fi fi
} }
@ -482,7 +484,7 @@ do
shift 2 shift 2
fi fi
;; ;;
--server|-s ) --master|-m )
SSH_SERVER="$2" SSH_SERVER="$2"
add_var_to_config SSH_SERVER "$SSH_SERVER" add_var_to_config SSH_SERVER "$SSH_SERVER"
shift 2 shift 2
@ -533,7 +535,7 @@ check_for_running_instances () {
JOBS=`ps axu | grep -v grep | grep ${USER} | grep -v -i screen | grep ppss.sh | wc -l` JOBS=`ps axu | grep -v grep | grep ${USER} | grep -v -i screen | grep ppss.sh | wc -l`
#echo "$(date) : ${JOBS}" #echo "$(date) : ${JOBS}"
get_min_jobs get_min_jobs
log INFO "Minjobs is $MIN_JOBS" log DEBUG "Minjobs is $MIN_JOBS"
if [ "$JOBS" -gt "$MIN_JOBS" ] if [ "$JOBS" -gt "$MIN_JOBS" ]
then then
@ -750,9 +752,15 @@ erase_ppss () {
then then
for NODE in `cat $NODES_FILE` for NODE in `cat $NODES_FILE`
do do
log INFO "Erasing PPSS homedir $PPSS_HOME_DIR from node $NODE." does_file_exist "ppss"
ssh -q $SSH_KEY $SSH_OPTS $USER@$NODE "./$PPSS_HOME_DIR/$0 kill" if [ "$?" == "0" ]
ssh -q $SSH_KEY $SSH_OPTS $USER@$NODE "rm -rf $PPSS_HOME_DIR" then
log INFO "Erasing PPSS homedir $PPSS_HOME_DIR from node $NODE."
ssh -q $SSH_KEY $SSH_OPTS $USER@$NODE "./$PPSS_HOME_DIR/$0 kill"
ssh -q $SSH_KEY $SSH_OPTS $USER@$NODE "rm -rf $PPSS_HOME_DIR"
else
log INFO "PPSS was not present on node $NODE."
fi
done done
else else
log INFO "Aborting.." log INFO "Aborting.."
@ -1464,7 +1472,7 @@ start_all_workers () {
get_status_of_node () { get_status_of_node () {
NODE="$1" NODE="$1"
STATUS=`ssh -o ConnectTimeout=10 $SSH_KEY $USER@$NODE cat "$PPSS_HOME_DIR/$NODE_STATUS"` STATUS=`ssh -o ConnectTimeout=10 $SSH_KEY $USER@$NODE cat "$PPSS_DIR/$NODE_STATUS"`
ERROR="$?" ERROR="$?"
if [ ! "$ERROR" == "0" ] if [ ! "$ERROR" == "0" ]
then then
@ -1508,7 +1516,13 @@ show_status () {
for x in `cat $NODES_FILE` for x in `cat $NODES_FILE`
do do
NODE=`get_status_of_node "$x" | awk '{ print $1 }'` NODE=`get_status_of_node "$x" | awk '{ print $1 }'`
RES=`exec_cmd "grep $NODE ~/$JOB_LOG_DIR/* | wc -l"` RES=`exec_cmd "grep $NODE ~/$JOB_LOG_DIR/* >> /dev/null 2>&1"`
if [ ! "$ERROR" == "0" ]
then
RES=0
else
RES=`exec_cmd "grep $NODE ~/$JOB_LOG_DIR/* | wc -l"`
fi
let PROCESSED=$PROCESSED+$RES let PROCESSED=$PROCESSED+$RES
STATUS=`get_status_of_node "$x" | awk '{ print $2 }'` STATUS=`get_status_of_node "$x" | awk '{ print $2 }'`
LINE=`echo "$x $NODE $RES $STATUS" | awk '{ printf ("%-16s %-18s % 10s %10s\n",$1,$2,$3,$4) }'` LINE=`echo "$x $NODE $RES $STATUS" | awk '{ printf ("%-16s %-18s % 10s %10s\n",$1,$2,$3,$4) }'`