b=17229

i=Adilger
remote multiop fix

diff --git a/lustre/tests/replay-vbr.sh b/lustre/tests/replay-vbr.sh
index 6410436..119555b 100644 (file)
--- a/lustre/tests/replay-vbr.sh
+++ b/lustre/tests/replay-vbr.sh
@@ -447,17 +447,41 @@ run_test 7a "fail MDS, delayed recovery, fail MDS"
 rmultiop_start() {
     local client=$1
     local file=$2
-    do_node $client LUSTRE="" sh runmultiop_bg_pause $file O_tSc
-    eval export ${client}_pid=$(do_node $client cat /tmp/multiop_bg.pid)
+
+    # We need to run do_node in bg, because pdsh does not exit
+    # if child process of run script exists.
+    # I.e. pdsh does not exit when runmultiop_bg_pause exited,
+    # because of multiop_bg_pause -> $MULTIOP_PROG &
+    # By the same reason we need sleep a bit after do_nodes starts 
+    # to let runmultiop_bg_pause start muliop and
+    # update /tmp/multiop_bg.pid ;
+    # The rm /tmp/multiop_bg.pid guarantees here that 
+    # we have the updated by runmultiop_bg_pause
+    # /tmp/multiop_bg.pid file
+
+    local pid_file=$TMP/multiop_bg.pid.$$
+    do_node $client "rm -f $pid_file && MULTIOP_PID_FILE=$pid_file LUSTRE= runmultiop_bg_pause $file O_tSc" & 
+    local pid=$!
+    sleep 3
+    local multiop_pid
+    multiop_pid=$(do_node $client cat $pid_file)
+    [ -n "$multiop_pid" ] || error "$client : Can not get multiop_pid from $pid_file "
+    eval export ${client}_multiop_pid=$multiop_pid
+    eval export ${client}_do_node_pid=$pid
+    local var=${client}_multiop_pid
+    echo client $client multiop_bg started multiop_pid=${!var}
     return $?
 }
 
 rmultiop_stop() {
     local client=$1
-    local pid=${client}_pid
-    echo "Stopping pid=${!pid}"
-    # how to wait for pid in that case?
-    do_node $client "kill -USR1 ${!pid}; sleep 3"
+    local multiop_pid=${client}_multiop_pid
+    local do_node_pid=${client}_do_node_pid
+
+    echo "Stopping multiop_pid=${!multiop_pid} (kill ${!multiop_pid} on $client)"
+    do_node $client kill -USR1 ${!multiop_pid}
+
+    wait ${!do_node_pid} || true
 }
 
 test_8a() {