From 9a569207ccf0aca58421ed982f90f9a114b7b897 Mon Sep 17 00:00:00 2001
From: Michael Gardner <michael.gardner@lexisnexisrisk.com>
Date: Tue, 21 May 2024 10:59:54 -0400
Subject: [PATCH] HPCC-31893 Fix conflict between multiple thor components on
 same node

Signed-off-by: Michael Gardner <michael.gardner@lexisnexisrisk.com>
---
 initfiles/bin/init_thorslave.in | 24 +++++++-----------------
 initfiles/sbin/hpcc_setenv.in   |  2 ++
 2 files changed, 9 insertions(+), 17 deletions(-)

diff --git a/initfiles/bin/init_thorslave.in b/initfiles/bin/init_thorslave.in
index 1affe3ba9c3..a7cc38d2553 100755
--- a/initfiles/bin/init_thorslave.in
+++ b/initfiles/bin/init_thorslave.in
@@ -44,24 +44,14 @@ slavename=thorslave_${hpcc_compname}
 stop_slaves()
 {
     local timer=15
-    local isAlive=0
-
-    log "Attempting to kill $slavename with SIGTERM"
-    killall -SIGTERM $slavename > /dev/null 2>&1
-    while [[ $isAlive -eq 0 && $timer -gt 0 ]];do
-        killall -0 $slavename > /dev/null 2>&1
-        isAlive=$?
-        [[ $isAlive -eq 0 ]] && sleep 0.5
-        ((timer--))
+    local _pidarray
+    local _pidname
+    
+    _pidarray=($( ls -l ${PID_DIR} | awk -v regpattern=${slavename}_[1-9][0-9]*\.pid '$0~regpattern {print $9;}' ))
+    for _pidname in ${_pidarray[@]}; do
+        kill_process ${PID_DIR}/$_pidname $slavename $timer
     done
-
-    if [[ $isAlive -eq 0 ]]; then
-        log "Failed to kill slaves with SIGTERM. Sending SIGKILL"
-        killall -SIGKILL $slavename > /dev/null
-    fi
-
-    # need regex here to prevent removing other Thor instance pid files
-    find ${PID_DIR} -maxdepth 1 -type f -regex ".*/${slavename}_[1-9][0-9]*\.pid" -delete > /dev/null 2>&1
+   
 }
 
 start_slaves()
diff --git a/initfiles/sbin/hpcc_setenv.in b/initfiles/sbin/hpcc_setenv.in
index 60de6860025..af35d0471d5 100755
--- a/initfiles/sbin/hpcc_setenv.in
+++ b/initfiles/sbin/hpcc_setenv.in
@@ -30,7 +30,9 @@ function kill_process () {
       SENTINEL=$4
       [[ -e $SENTINEL ]] && rm -f $SENTINEL
     fi
+    log "checking $PID"
     if [[ -e $PID ]]; then
+        log "pidwait_fn $PID $PROCESS $TIMEOUT 1"
         pidwait_fn $PID $PROCESS $TIMEOUT 1
         local RC_PIDWAIT=$?
         return $RC_PIDWAIT