* Added find-run-hosts.sh: swiss-army tool to find hosts where a job

run, dump the process trees, etc.

sge/find-run-hosts.sh

@@ -0,0 +1,158 @@
+#!/bin/bash
+#
+# 20160713
+# Wirawan Purwanto
+
+shopt -s extglob
+
+function find_run_hosts()
+# Find where a job runs.
+{
+  local optJobNumber="$1"
+  qstat -f | _Find_run_hosts
+}
+
+function _Find_run_hosts()
+# Processing part of the routine above.
+# Takes qstat -f output.
+{
+  awk \
+    -v optPrintRaw="$optPrintRaw" \
+    -v optJobNumber="$optJobNumber" \
+    '####
+BEGIN {
+  STDERR = "/dev/stderr"
+}
+
+# Valid host status field
+($0 ~ /^[A-Za-z]/) && (NF == 5 || NF == 6) {
+  host_info = $0
+  queue_node = $1
+  core_usage_combo = $3
+  states = $6  # if any
+  next
+}
+
+$1 == optJobNumber {
+  print host_info
+  print $0
+}' # end awk script
+}
+
+
+function find_job_owner()
+# Arg: <jobnumber>
+{
+  qstat -j "$1" \
+    | awk '$1 == "owner:" { print $2; exit }'
+}
+
+
+function find_job_master_node()
+# Arg: <jobnumber>
+# Finds the "master" node of the job, i.e. where the batch job script
+# was first initially launched.
+# Unfortunately qstat -j doesn't return the desired info, we have to
+# utilize full qstat to get this info.
+{
+  qstat  \
+    | awk -v optJobNumber="$1"  '
+$1 == optJobNumber {
+  queueHost = $8
+  sub(/^.*@/, "", queueHost)
+  print queueHost 
+  exit
+}' # end awk script
+}
+
+
+function list_job_nodes()
+# Arg: <jobnumber>
+{
+  find_run_hosts "$1" \
+    | awk '
+($0 ~ /^[A-Za-z]/) && (NF == 5 || NF == 6) { 
+  host = $1
+  sub(/^.*@/, "", host)
+  print host
+}' # end awk script
+}
+
+function dump_process_tree1()
+# Arg: <host> <user>
+{
+  local HOST="$1" JOB_OWNER="$2"
+  ssh "$HOST" ps ux --forest -u "$JOB_OWNER" \
+    | awk -v JobOwner="$JOB_OWNER" '($1 == "USER") || ($1 == JobOwner) { print }'
+  # Note: some workaround was needed because it also printed other users
+  # notably the one running this job.
+}
+
+
+function dump_process_trees()
+{
+  local SGE_JOB_ID="$1"
+  local MASTER_HOST ALL_HOSTS HOST
+  local JOB_OWNER
+  echo "job_number: $SGE_JOB_ID"
+  MASTER_HOST=$(find_job_master_node "$SGE_JOB_ID")
+  JOB_OWNER=$(find_job_owner "$SGE_JOB_ID")
+  ALL_HOSTS=( $(list_job_nodes "$SGE_JOB_ID") )
+  echo "master_host: $MASTER_HOST"
+  dump_process_tree1 "$MASTER_HOST" "$JOB_OWNER"
+
+  for HOST in "${ALL_HOSTS[@]}"; do
+    if [ "$HOST" = "$MASTER_HOST" ]; then continue; fi
+    echo "host: $HOST"
+    dump_process_tree1 "$HOST" "$JOB_OWNER"
+  done
+}
+
+
+# Main program switchboard
+
+case "$1" in
+(+([0-9]))
+  SGE_JOB_ID="$1"
+  find_run_hosts "$SGE_JOB_ID"
+  ;;
+(--process*|process*|proc|procs|px)
+  # Prints where the head node of the job is (i.e. the master node of the job)
+  # where the job script was first executing
+  if [ -z "$2" ]; then
+    echo "Job ID required as arg 2" >&2
+    exit 2
+  fi
+  SGE_JOB_ID="$2"
+  dump_process_trees "$SGE_JOB_ID"
+  ;;
+(--head-node|--head|head|headnode|head-node|--master-node|--master|master|masternode|master-node)
+  # Prints where the head node of the job is (i.e. the master node of the job)
+  # where the job script was first executing
+  if [ -z "$2" ]; then
+    echo "Job ID required as arg 2" >&2
+    exit 2
+  fi
+  SGE_JOB_ID="$2"
+  find_job_master_node "$SGE_JOB_ID"
+  ;;
+(--list-node*|--list|list|listnode*|list-node*|node|nodes)
+  # Lists all the compute nodes being used by this job
+  if [ -z "$2" ]; then
+    echo "Job ID required as arg 2" >&2
+    exit 2
+  fi
+  SGE_JOB_ID="$2"
+  list_job_nodes "$SGE_JOB_ID"
+  ;;
+(*)
+  echo "Unknown action: $1" >&2
+  exit 2
+  ;;
+esac
+
+#SGE_JOB_ID="$1"
+#find_run_hosts "$SGE_JOB_ID"
+#find_job_owner "$SGE_JOB_ID"
+#find_job_head_node "$SGE_JOB_ID"
+