b=20918 t-f max recovery time estimation

[fs/lustre-release.git] / lustre / tests / test-framework.sh

diff --git a/lustre/tests/test-framework.sh b/lustre/tests/test-framework.sh
index 6925483..81a4b7b 100644 (file)
--- a/lustre/tests/test-framework.sh
+++ b/lustre/tests/test-framework.sh
@@ -262,7 +262,7 @@ load_module() {
            (($# > 0)) && shift 2
 
             # Ensure we have accept=all for lnet
-            if [ $module = lnet ]; then
+            if [ $(basename $module) = lnet ]; then
                 # OK, this is a bit wordy...
                 local arg accept_all_present=false
                 for arg in "$@"; do
@@ -404,7 +404,7 @@ check_gss_daemon_nodes() {
     local list=$1
     dname=$2
 
-    do_nodes --verbose $list "num=\\\$(ps -o cmd -C $dname | grep $dname | wc -l);
+    do_nodesv $list "num=\\\$(ps -o cmd -C $dname | grep $dname | wc -l);
 if [ \\\"\\\$num\\\" -ne 1 ]; then
     echo \\\$num instance of $dname;
     exit 1;
@@ -950,6 +950,8 @@ start_client_loads () {
         testnum=$((nodenum % numloads))
         start_client_load ${clients[nodenum]} ${CLIENT_LOADS[testnum]}
     done
+    # bug 22169: wait the background threads to start
+    sleep 2
 }
 
 # only for remote client 
@@ -1535,6 +1537,10 @@ do_node() {
     return ${PIPESTATUS[0]}
 }
 
+do_nodev() {
+    do_node --verbose "$@"
+}
+
 single_local_node () {
    [ "$1" = "$HOSTNAME" ]
 }
@@ -1563,7 +1569,7 @@ do_nodes() {
 
     if single_local_node $rnodes; then
         if $verbose; then
-           do_node --verbose $rnodes "$@"
+           do_nodev $rnodes "$@"
         else
            do_node $rnodes "$@"
         fi
@@ -1597,6 +1603,10 @@ do_facet() {
     do_node $HOST "$@"
 }
 
+do_nodesv() {
+    do_nodes --verbose "$@"
+}
+
 add() {
     local facet=$1
     shift
@@ -2597,13 +2607,8 @@ run_test() {
     return $?
 }
 
-EQUALS="======================================================================"
 equals_msg() {
-    msg="$@"
-
-    local suffixlen=$((${#EQUALS} - ${#msg}))
-    [ $suffixlen -lt 5 ] && suffixlen=5
-    log `echo $(printf '===== %s %.*s\n' "$msg" $suffixlen $EQUALS)`
+    banner "$*"
 }
 
 log() {
@@ -2658,18 +2663,15 @@ reset_fail_loc () {
 # Log a message (on all nodes) padded with "=" before and after. 
 # Also appends a timestamp and prepends the testsuite name.
 # 
+
+EQUALS="===================================================================================================="
 banner() {
     msg="== ${TESTSUITE} $*"
-    # pad the message out to 70 with "="
     last=${msg: -1:1}
-    [[ $last != "=" && $last != " " ]] && msg+=" "
-    for i in $(seq $((68 - ${#msg})) ); do
-       msg+="="
-    done
+    [[ $last != "=" && $last != " " ]] && msg="$msg "
+    msg=$(printf '%s%.*s'  "$msg"  $((${#EQUALS} - ${#msg})) $EQUALS )
     # always include at least == after the message
-    msg+="=="
-
-    log "$msg $(date +"%H:%M:%S (%s)")"
+    log "$msg== $(date +"%H:%M:%S (%s)")"
 }
 
 #
@@ -2714,7 +2716,7 @@ run_one_logged() {
     rm -rf $LOGDIR/err
 
     echo
-    run_one $1 "$2" 2>&1 | tee $test_log
+    (run_one $1 "$2") 2>&1 | tee $test_log
     local RC=${PIPESTATUS[0]}
 
     [ $RC -ne 0 ] && [ ! -f $LOGDIR/err ] && \
@@ -2979,9 +2981,10 @@ mixed_mdt_devs () {
 generate_machine_file() {
     local nodes=${1//,/ }
     local machinefile=$2
-    rm -f $machinefile || error "can't rm $machinefile"
+    rm -f $machinefile
     for node in $nodes; do
-        echo $node >>$machinefile
+        echo $node >>$machinefile || \
+            { echo "can not generate machinefile $machinefile" && return 1; }
     done
 }
 
@@ -3000,7 +3003,7 @@ setstripe_nfsserver () {
 
     [ -z $nfsserver ] && echo "$dir is not nfs mounted" && return 1
 
-    do_node --verbose $nfsserver lfs setstripe "$@"
+    do_nodev $nfsserver lfs setstripe "$@"
 }
 
 check_runas_id_ret() {
@@ -3152,7 +3155,7 @@ calc_osc_kbytes () {
 # generate a stream of formatted strings (<node> <param name>=<param value>)
 save_lustre_params() {
         local s
-        do_nodes --verbose $1 "lctl get_param $2 | while read s; do echo \\\$s; done"
+        do_nodesv $1 "lctl get_param $2 | while read s; do echo \\\$s; done"
 }
 
 # restore lustre parameters from input stream, produces by save_lustre_params
@@ -3366,7 +3369,7 @@ do_rpc_nodes () {
 
     # Add paths to lustre tests for 32 and 64 bit systems.
     local RPATH="$RLUSTRE/tests:/usr/lib/lustre/tests:/usr/lib64/lustre/tests:$PATH"
-    do_nodes --verbose $list "PATH=$RPATH sh rpc.sh $@ "
+    do_nodesv $list "PATH=$RPATH sh rpc.sh $@ "
 }
 
 wait_clients_import_state () {
@@ -3527,7 +3530,7 @@ gather_logs () {
         return
     fi
 
-    do_nodes --verbose $list \
+    do_nodesv $list \
         "$LCTL dk > ${prefix}.debug_log.\\\$(hostname).${suffix};
          dmesg > ${prefix}.dmesg.\\\$(hostname).${suffix}"
     if [ ! -f $LOGDIR/shared ]; then
@@ -3569,6 +3572,24 @@ do_ls () {
     return $rc
 }
 
+# target_start_and_reset_recovery_timer()
+#        service_time = at_est2timeout(service_time);
+#        service_time += 2 * (CONNECTION_SWITCH_MAX + CONNECTION_SWITCH_INC +
+#                             INITIAL_CONNECT_TIMEOUT);
+# CONNECTION_SWITCH_MAX : min(25U, max(CONNECTION_SWITCH_MIN,obd_timeout))
+#define CONNECTION_SWITCH_INC 1
+#define INITIAL_CONNECT_TIMEOUT max(CONNECTION_SWITCH_MIN,obd_timeout/20)
+#define CONNECTION_SWITCH_MIN 5U
+
+max_recovery_time () {
+    local init_connect_timeout=$(( TIMEOUT / 20 ))
+    [[ $init_connect_timeout > 5 ]] || init_connect_timeout=5 
+
+    local service_time=$(( $(at_max_get client) + $(( 2 * $(( 25 + 1  + init_connect_timeout)) )) ))
+
+    echo $service_time 
+}
+
 get_clients_mount_count () {
     local clients=${CLIENTS:-`hostname`}
 
@@ -3590,7 +3611,7 @@ combination()
         R=0
     else
         N=$((N + 1))
-        while [ $N -le $M ]; do
+        while [ $N -lt $M ]; do
             R=$((R * N))
             N=$((N + 1))
         done
@@ -3801,6 +3822,7 @@ wait_flavor()
             return 0
         else
             echo "found $res $flavor connections of $dir, not ready ($expect)"
+            return 0
             sleep 4
         fi
     done
@@ -3914,3 +3936,17 @@ log_sub_test() {
     yml_log_sub_test $@ >> $YAML_LOG
 }
 
+run_llverdev()
+{
+        local dev=$1
+        local devname=$(basename $1)
+        local size=$(grep "$devname"$ /proc/partitions | awk '{print $3}')
+        size=$(($size / 1024 / 1024)) # Gb
+
+        local partial_arg=""
+        # Run in partial (fast) mode if the size
+        # of a partition > 10 GB
+        [ $size -gt 10 ] && partial_arg="-p"
+
+        llverdev --force $partial_arg $dev
+}