X-Git-Url: https://git.whamcloud.com/?a=blobdiff_plain;f=lustre%2Ftests%2Fha.sh;h=d30424a33ae1347529999bb6e2d115e0197c639f;hb=80e1c8ac5d509a286427b72a85c0d6888d7a58b3;hp=0c47ed67e7ef73448aff15a6e35a43ae596e23ce;hpb=1a7c352e97d8b9c41ed8b9c881c813ee87b3789c;p=fs%2Flustre-release.git

diff --git a/lustre/tests/ha.sh b/lustre/tests/ha.sh
index 0c47ed6..d30424a 100755
--- a/lustre/tests/ha.sh
+++ b/lustre/tests/ha.sh
@@ -82,6 +82,9 @@
 #   The number of clients run MPI loads is configured by parameter
 #   ha_mpi_instances. Only one client runs MPI workloads by default.
 #
+#   MPI workloads can be run from several users. The list of users to use is
+#   configured by parameter ha_mpi_users, default is "mpiuser".
+#
 # PROCESS STRUCTURE AND IPC
 #
 #   On the node where this script is run, the processes look like this:
@@ -212,19 +215,27 @@ declare     ha_mpi_instances=${ha_mpi_instances:-1}
 
 declare     ha_mpi_loads=${ha_mpi_loads="ior simul mdtest"}
 declare -a  ha_mpi_load_tags=($ha_mpi_loads)
+declare -a  ha_mpiusers=(${ha_mpi_users="mpiuser"})
 
 declare     ha_ior_params=${IORP:-'" -b $ior_blockSize -t 2m -w -W -T 1"'}
 declare     ha_simul_params=${SIMULP:-'" -n 10"'}
 declare     ha_mdtest_params=${MDTESTP:-'" -i 1 -n 1000"'}
 declare     ha_mpirun_options=${MPIRUN_OPTIONS:-""}
+declare     ha_clients_stripe=${CLIENTSSTRIPE:-'"$STRIPEPARAMS"'}
+declare     ha_nclientsset=${NCLIENTSSET:-1}
+declare     ha_ninstmustfail=${NINSTMUSTFAIL:-0}
+
+declare     ha_racer_params=${RACERP:-"MDSCOUNT=1"}
 
 eval ha_params_ior=($ha_ior_params)
 eval ha_params_simul=($ha_simul_params)
 eval ha_params_mdtest=($ha_mdtest_params)
+eval ha_stripe_clients=($ha_clients_stripe)
 
 declare ha_nparams_ior=${#ha_params_ior[@]}
 declare ha_nparams_simul=${#ha_params_simul[@]}
 declare ha_nparams_mdtest=${#ha_params_mdtest[@]}
+declare ha_nstripe_clients=${#ha_stripe_clients[@]}
 
 declare -A  ha_mpi_load_cmds=(
 	[ior]="$IOR -o {}/f.ior {params}"
@@ -232,12 +243,15 @@ declare -A  ha_mpi_load_cmds=(
 	[mdtest]="$MDTEST {params} -d {}"
 )
 
+declare racer=${RACER:-"$(dirname $0)/racer/racer.sh"}
+
 declare     ha_nonmpi_loads=${ha_nonmpi_loads="dd tar iozone"}
 declare -a  ha_nonmpi_load_tags=($ha_nonmpi_loads)
-declare -a  ha_nonmpi_load_cmds=(
-	"dd if=/dev/zero of={}/f.dd bs=1M count=256"
-	"tar cf - /etc | tar xf - -C {}"
-	"iozone -a -e -+d -s $iozone_SIZE {}/f.iozone"
+declare -A  ha_nonmpi_load_cmds=(
+	[dd]="dd if=/dev/zero of={}/f.dd bs=1M count=256"
+	[tar]="tar cf - /etc | tar xf - -C {}"
+	[iozone]="iozone -a -e -+d -s $iozone_SIZE {}/f.iozone"
+	[racer]="$ha_racer_params $racer {}"
 )
 
 ha_usage()
@@ -404,6 +418,10 @@ ha_repeat_mpi_load()
 	local load=$2
 	local status=$3
 	local parameter=$4
+	local machines=$5
+	local stripeparams=$6
+	local mpiuser=$7
+	local mustpass=$8
 	local tag=${ha_mpi_load_tags[$load]}
 	local cmd=${ha_mpi_load_cmds[$tag]}
 	local dir=$ha_test_dir/$client-$tag
@@ -416,10 +434,12 @@ ha_repeat_mpi_load()
 	cmd=${cmd//"{}"/$dir}
 	cmd=${cmd//"{params}"/$parameter}
 
+	[[ -n "$ha_postcmd" ]] && ha_postcmd=${ha_postcmd//"{}"/$dir}
 	ha_info "Starting $tag"
 
-	local machines="-machinefile $ha_machine_file"
+	machines="-machinefile $machines"
 	while [ ! -e "$ha_stop_file" ] && ((rc == 0)); do
+		ha_info "$client Starts: $cmd" 2>&1 |  tee -a $log
 		{
 		local mdt_index
 		if $ha_mdt_index_random && [ $ha_mdt_index -ne 0 ]; then
@@ -429,21 +449,37 @@ ha_repeat_mpi_load()
 		fi
 		ha_on $client $LFS mkdir -i$mdt_index -c$ha_dir_stripe_count "$dir" &&
 		ha_on $client $LFS getdirstripe "$dir" &&
+		ha_on $client $LFS setstripe $stripeparams $dir &&
+		ha_on $client $LFS getstripe $dir &&
 		ha_on $client chmod a+xwr $dir &&
-		ha_on $client "su mpiuser sh -c \" $mpirun $ha_mpirun_options \
+		ha_on $client "su $mpiuser sh -c \" $mpirun $ha_mpirun_options \
 			-np $((${#ha_clients[@]} * mpi_threads_per_client )) \
-			$machines $cmd \" " &&
+			$machines $cmd \" " || rc=$?
+		[[ -n "$ha_postcmd" ]] && ha_info "$ha_postcmd" &&
+			ha_on $client $ha_postcmd >>"$log" 2>&1
+		(( ((rc == 0)) && (( mustpass != 0 )) )) ||
+		(( ((rc != 0)) && (( mustpass == 0 )) )) &&
 			ha_on $client rm -rf "$dir";
 		} >>"$log" 2>&1 || rc=$?
 
-		ha_info rc=$rc
-
-		if ((rc != 0)); then
+		ha_info rc=$rc mustpass=$mustpass
+
+		# mustpass=0 means that failure is expected
+		if (( rc !=0 )); then
+			if (( mustpass != 0 )); then
+				touch "$ha_fail_file"
+				touch "$ha_stop_file"
+				ha_dump_logs "${ha_clients[*]} ${ha_servers[*]}"
+			else
+				# Ok to fail
+				rc=0
+			fi
+		elif (( mustpass == 0 )); then
 			touch "$ha_fail_file"
 			touch "$ha_stop_file"
 			ha_dump_logs "${ha_clients[*]} ${ha_servers[*]}"
 		fi
-		echo $rc >"$status"
+		echo rc=$rc mustpass=$mustpass >"$status"
 
 		nr_loops=$((nr_loops + 1))
 	done
@@ -451,7 +487,8 @@ ha_repeat_mpi_load()
 	[ $nr_loops -ne 0 ] &&
 		avg_loop_time=$((($(date +%s) - start_time) / nr_loops))
 
-	ha_info "$tag stopped: rc $rc avg loop time $avg_loop_time"
+	ha_info "$tag stopped: rc=$rc mustpass=$mustpass \
+		avg loop time $avg_loop_time"
 }
 
 ha_start_mpi_loads()
@@ -462,24 +499,44 @@ ha_start_mpi_loads()
 	local status
 	local n
 	local nparam
+	local machines
+	local m
+	local -a mach
+	local mpiuser
 
-	for client in ${ha_clients[@]}; do
-		ha_info ha_machine_file=$ha_machine_file
-		echo $client >> $ha_machine_file
+	# ha_mpi_instances defines the number of
+	# clients start mpi loads; should be <= ${#ha_clients[@]}
+	# do nothing if
+	#    ha_mpi_instances = 0
+	# or
+	#    ${#ha_mpi_load_tags[@]} =0
+	local inst=$ha_mpi_instances
+	(( inst == 0 )) || (( ${#ha_mpi_load_tags[@]} == 0 )) &&
+		ha_info "no mpi load to start" &&
+		return 0
+
+	(( inst <= ${#ha_clients[@]} )) || inst=${#ha_clients[@]}
+
+	# Define names for machinefiles for each client set
+	for (( n=0; n < $ha_nclientsset; n++ )); do
+		mach[$n]=$ha_machine_file$n
+	done
+
+	for ((n = 0; n < ${#ha_clients[@]}; n++)); do
+		m=$(( n % ha_nclientsset))
+		machines=${mach[m]}
+		ha_info machine_file=$machines
+		echo ${ha_clients[n]} >> $machines
 	done
 	local dirname=$(dirname $ha_machine_file)
 	for client in ${ha_clients[@]}; do
 		ha_on $client mkdir -p $dirname
-		scp $ha_machine_file $client:$ha_machine_file
+		scp $ha_machine_file* $client:$dirname
 	done
 
-	# ha_mpi_instances defines the number of
-	# clients start mpi loads; should be <= ${#ha_clients[@]}
-	local inst=$ha_mpi_instances
-	(( inst <= ${#ha_clients[@]} )) || inst=${#ha_clients[@]}
-
 	for ((n = 0; n < $inst; n++)); do
 		client=${ha_clients[n]}
+		mpiuser=${ha_mpiusers[$((n % ${#ha_mpiusers[@]}))]}
 		for ((load = 0; load < ${#ha_mpi_load_tags[@]}; load++)); do
 			tag=${ha_mpi_load_tags[$load]}
 			status=$ha_status_file_prefix-$tag-$client
@@ -489,7 +546,16 @@ ha_start_mpi_loads()
 			nparam=$((n % num))
 			local aref=ha_params_$tag[nparam]
 			local parameter=${!aref}
-			ha_repeat_mpi_load $client $load $status "$parameter" &
+			local nstripe=$((n % ha_nstripe_clients))
+			aref=ha_stripe_clients[nstripe]
+			local stripe=${!aref}
+			local m=$(( n % ha_nclientsset))
+			machines=${mach[m]}
+			local mustpass=1
+			[[ $ha_ninstmustfail == 0 ]] ||
+				mustpass=$(( n % ha_ninstmustfail ))
+			ha_repeat_mpi_load $client $load $status "$parameter" \
+				$machines "$stripe" "$mpiuser" "$mustpass" &
 				ha_status_files+=("$status")
 		done
 	done
@@ -501,7 +567,7 @@ ha_repeat_nonmpi_load()
 	local load=$2
 	local status=$3
 	local tag=${ha_nonmpi_load_tags[$load]}
-	local cmd=${ha_nonmpi_load_cmds[$load]}
+	local cmd=${ha_nonmpi_load_cmds[$tag]}
 	local dir=$ha_test_dir/$client-$tag
 	local log=$ha_tmp_dir/$client-$tag
 	local rc=0
@@ -509,11 +575,12 @@ ha_repeat_nonmpi_load()
 	local avg_loop_time=0
 	local start_time=$(date +%s)
 
-    cmd=${cmd//"{}"/$dir}
+	cmd=${cmd//"{}"/$dir}
 
-    ha_info "Starting $tag on $client"
+	ha_info "Starting $tag on $client"
 
 	while [ ! -e "$ha_stop_file" ] && ((rc == 0)); do
+		ha_info "$client Starts: $cmd" 2>&1 |  tee -a $log
 		ha_on $client "mkdir -p $dir &&                              \
 			$cmd &&                                              \
 			rm -rf $dir" >>"$log" 2>&1 || rc=$?
@@ -664,17 +731,20 @@ ha_lfsck_repaired()
 
 ha_start_loads()
 {
-    trap ha_trap_stop_signals $ha_stop_signals
-    ha_start_nonmpi_loads
-    ha_start_mpi_loads
+	$ha_lfsck_bg && ha_lfsck_bg
+	trap ha_trap_stop_signals $ha_stop_signals
+	ha_start_nonmpi_loads
+	ha_start_mpi_loads
 }
 
 ha_stop_loads()
 {
-    touch $ha_stop_file
-    trap - $ha_stop_signals
-    ha_info "Waiting for workloads to stop"
-    wait
+	touch $ha_stop_file
+	# true because of lfsck_bg could be stopped already
+	$ha_lfsck_bg && wait $LFSCK_BG_PID || true
+	trap - $ha_stop_signals
+	ha_info "Waiting for workloads to stop"
+	wait
 }
 
 ha_wait_loads()
@@ -878,8 +948,6 @@ ha_main()
 	ha_on ${ha_clients[0]} " \
 		$LFS setstripe $ha_stripe_params $ha_test_dir"
 
-	$ha_lfsck_bg && ha_lfsck_bg
-
 	ha_start_loads
 	ha_wait_loads
 
@@ -896,9 +964,6 @@ ha_main()
 
 	$ha_lfsck_fail_on_repaired && ha_lfsck_repaired
 
-	# true because of lfsck_bg could be stopped already
-	$ha_lfsck_bg && wait $LFSCK_BG_PID || true
-
 	if [ -e "$ha_fail_file" ]; then
 		exit 1
 	else