From 6477502ca963a78f87727f0215716b7e90a69eda Mon Sep 17 00:00:00 2001
From: Charles Antonelli <cja@umich.edu>
Date: Mon, 2 Mar 2020 15:22:50 -0500
Subject: [PATCH] make tskserver port selection dynamic

If multiple tskservers are started concurrently on the same host,
only one of them will be able to connect to the (hardcoded) tskserver
port.  This patch allows dynamic port selection by searching for
a monotonically increasing port number for each tskserver instance.
This is highly desirable for installations that do not have whole
node allocation policies.
---
 paramrun | 19 +++++++++++--------
 1 file changed, 11 insertions(+), 8 deletions(-)

diff --git a/paramrun b/paramrun
index 6f1c61a..d623b30 100755
--- a/paramrun
+++ b/paramrun
@@ -174,13 +174,14 @@ if [[ "$LAUNCHER_SCHED" == "dynamic" ]]; then
   RETRY=0
   while [[ "$RUNNING" == "false" ]]
   do
+    let TSKPORT=9471+RETRY
     if [[ "$windowsP" == "true" ]]; then
-        $LAUNCHER_DIR/tskserver $LAUNCHER_NJOBS localhost 9471 2>/dev/null &
+        $LAUNCHER_DIR/tskserver $LAUNCHER_NJOBS localhost $TSKPORT 2>/dev/null &
         LAUNCHER_DYN_PID=$!
         disown $LAUNCHER_DYN_PID
         sleep 1s
     else
-        $LAUNCHER_DIR/tskserver $LAUNCHER_NJOBS $HOSTNAME 9471 2>/dev/null &
+        $LAUNCHER_DIR/tskserver $LAUNCHER_NJOBS $HOSTNAME $TSKPORT 2>/dev/null &
         LAUNCHER_DYN_PID=$!
         disown $LAUNCHER_DYN_PID
         sleep 1s
@@ -190,7 +191,7 @@ if [[ "$LAUNCHER_SCHED" == "dynamic" ]]; then
     then
       if [[ $RETRY -ne 10 ]]
       then
-        lwarn WARNING "Unable to start dynamic task service. Retrying..."
+        lwarn WARNING "Unable to start dynamic task service on port $TSKPORT. Retrying..."
         RETRY=`expr $RETRY + 1`
         sleep 10s
       else
@@ -198,6 +199,7 @@ if [[ "$LAUNCHER_SCHED" == "dynamic" ]]; then
         lexit
       fi
     else
+      lwarn NOTE "Started dynamic task service on port $TSKPORT."
       RUNNING="true"
     fi
   done
@@ -207,23 +209,24 @@ if [[ "$LAUNCHER_SCHED" == "dynamic" ]]; then
   else
       export LAUNCHER_DYN_COUNT="$HOSTNAME"
   fi
-  export LAUNCHER_DYN_COUNT_PORT=9471
+  export LAUNCHER_DYN_COUNT_PORT=$TSKPORT
 
   if [[ $LAUNCHER_USE_PHI -ne "0" ]]; then
+    let TSKPORT=TSKPORT+1
     if [[ $windowsP==true ]]; then
 	#Start another tskserver for the Intel Xeon Phi cards
-        $LAUNCHER_DIR/tskserver `wc -l $PHI_WORKDIR/$PHI_CONTROL_FILE` localhost 9472 2>/dev/null &
+        $LAUNCHER_DIR/tskserver `wc -l $PHI_WORKDIR/$PHI_CONTROL_FILE` localhost $TSKPORT 2>/dev/null &
         LAUNCHER_PHI_DYN_PID=$1
         disown $LAUNCHER_PHI_DYN_PID
         export LAUNCHER_PHI_DYN_COUNT=localhost
-        export LAUNCHER_PHI_DYN_COUNT_PORT=9472
+        export LAUNCHER_PHI_DYN_COUNT_PORT=$TSKPORT
     else
 	#Start another tskserver for the Intel Xeon Phi cards
-        $LAUNCHER_DIR/tskserver `wc -l $PHI_WORKDIR/$PHI_CONTROL_FILE` $HOSTNAME 9472 2>/dev/null &
+        $LAUNCHER_DIR/tskserver `wc -l $PHI_WORKDIR/$PHI_CONTROL_FILE` $HOSTNAME $TSKPORT 2>/dev/null &
         LAUNCHER_PHI_DYN_PID=$1
         disown $LAUNCHER_PHI_DYN_PID
         export LAUNCHER_PHI_DYN_COUNT="$HOSTNAME"
-        export LAUNCHER_PHI_DYN_COUNT_PORT=9472
+        export LAUNCHER_PHI_DYN_COUNT_PORT=$TSKPORT
     fi
   fi
 fi