X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Ftests%2Ftest-framework.sh;h=d17bf3c367a0734ff3e225885bce61db8ec14c78;hp=4b8ef21526f9aa10fd5ed98b287cfd1cde502378;hb=5d61648b99997406ae897c076c0d3eb1ca84a0db;hpb=4c84d2739170c2f084b9df446f63206f6eae6d02

diff --git a/lustre/tests/test-framework.sh b/lustre/tests/test-framework.sh
index 4b8ef21..d17bf3c 100644
--- a/lustre/tests/test-framework.sh
+++ b/lustre/tests/test-framework.sh
@@ -98,11 +98,14 @@ init_test_env() {
     if ! echo $PATH | grep -q $LUSTRE/tests; then
 	export PATH=$PATH:$LUSTRE/tests
     fi
-    export MDSRATE=${MDSRATE:-"$LUSTRE/tests/mdsrate"}
+    export MDSRATE=${MDSRATE:-"$LUSTRE/tests/mpi/mdsrate"}
     [ ! -f "$MDSRATE" ] && export MDSRATE=$(which mdsrate 2> /dev/null)
     if ! echo $PATH | grep -q $LUSTRE/tests/racer; then
         export PATH=$PATH:$LUSTRE/tests/racer
     fi
+    if ! echo $PATH | grep -q $LUSTRE/tests/mpi; then
+        export PATH=$PATH:$LUSTRE/tests/mpi
+    fi
     export LCTL=${LCTL:-"$LUSTRE/utils/lctl"}
     [ ! -f "$LCTL" ] && export LCTL=$(which lctl)
     export LFS=${LFS:-"$LUSTRE/utils/lfs"}
@@ -515,8 +518,13 @@ stop() {
 }
 
 # save quota version (both administrative and operational quotas)
+# add an additional parameter if mountpoint is ever different from $MOUNT
 quota_save_version() {
     local fsname=${2:-$FSNAME}
+
+    $LFS quotaoff -ug $MOUNT # just in case
+    [ -n "$1" ] && { $LFS quotacheck -$1 $MOUNT || error "quotacheck has failed"; }
+
     do_facet mgs "lctl conf_param ${fsname}-MDT*.mdd.quota_type=$1"
     local varsvc
     local osts=$(get_facets OST)
@@ -542,9 +550,7 @@ restore_quota_type () {
    if [ ! "$old_QUOTA_TYPE" ] || [ "$quota_type" = "$old_QUOTA_TYPE" ]; then
         return
    fi
-   $LFS quotaoff $mntpt
    quota_save_version $old_QUOTA_TYPE
-   $LFS quotacheck -ug $mntpt
 }
 
 setup_quota(){
@@ -560,7 +566,6 @@ setup_quota(){
     if [ "$quota_type" != "$QUOTA_TYPE" ]; then
         export old_QUOTA_TYPE=$quota_type
         quota_save_version $QUOTA_TYPE
-        $LFS quotacheck -ug $mntpt
     fi
 
     local quota_usrs=$QUOTA_USERS
@@ -867,23 +872,46 @@ check_client_load () {
     local TESTLOAD=run_${!var}.sh
 
     ps auxww | grep -v grep | grep $client | grep -q "$TESTLOAD" || return 1
-
-    check_catastrophe $client || return 2
-
-    # see if the load is still on the client
+    
+    # bug 18914: try to connect several times not only when
+    # check ps, but  while check_catastrophe also
     local tries=3
     local RC=254
     while [ $RC = 254 -a $tries -gt 0 ]; do
         let tries=$tries-1
         # assume success
         RC=0
+        if ! check_catastrophe $client; then
+            RC=${PIPESTATUS[0]}
+            if [ $RC -eq 254 ]; then
+                # FIXME: not sure how long we shuold sleep here
+                sleep 10
+                continue
+            fi
+            echo "check catastrophe failed: RC=$RC "
+            return $RC
+        fi
+    done
+    # We can continue try to connect if RC=254
+    # Just print the warning about this
+    if [ $RC = 254 ]; then
+        echo "got a return status of $RC from do_node while checking catastrophe on $client"
+    fi
+
+    # see if the load is still on the client
+    tries=3
+    RC=254
+    while [ $RC = 254 -a $tries -gt 0 ]; do
+        let tries=$tries-1
+        # assume success
+        RC=0
         if ! do_node $client "ps auxwww | grep -v grep | grep -q $TESTLOAD"; then
             RC=${PIPESTATUS[0]}
             sleep 30
         fi
     done
     if [ $RC = 254 ]; then
-        echo "got a return status of $RC from do_node while checking (i.e. with 'ps') the client load on the remote system"
+        echo "got a return status of $RC from do_node while checking (catastrophe and 'ps') the client load on $client"
         # see if we can diagnose a bit why this is
     fi
 
@@ -964,12 +992,13 @@ wait_update () {
         local RESULT
         local WAIT=0
         local sleep=5
-        while [ $WAIT -lt $MAX ]; do
+        while [ true ]; do
             RESULT=$(do_node $node "$TEST")
             if [ "$RESULT" == "$FINAL" ]; then
                 echo "Updated after $WAIT sec: wanted '$FINAL' got '$RESULT'"
                 return 0
             fi
+            [ $WAIT -ge $MAX ] && break
             echo "Waiting $((MAX - WAIT)) secs for update"
             WAIT=$((WAIT + sleep))
             sleep $sleep
@@ -1663,6 +1692,29 @@ init_facets_vars () {
     done
 }
 
+mds_sanity_check () {
+    local timeout=$1
+    local period=0
+
+    while [ $period -lt $timeout ]; do
+        count=$(do_facet $SINGLEMDS "lctl dl | grep 'osc.*mdtlov_UUID' | grep ' IN ' 2>/dev/null | wc -l")
+        if [ $count -eq 0 ]; then
+            break
+        fi
+
+        echo "There are $count OST are inactive, wait $period seconds, and try again"
+        sleep 3
+        period=$((period+3))
+    done
+
+    [ $period -lt $timeout ] || log "$count OST are inactive after $timeout seconds, give up"
+}
+
+som_check() {
+    SOM_ENABLED=$(do_facet $SINGLEMDS "$LCTL get_param mdt.*.som" | awk -F= ' {print $2}' | head -n 1)
+    echo $SOM_ENABLED
+}
+
 init_param_vars () {
     if ! remote_ost_nodsh && ! remote_mds_nodsh; then
         export MDSVER=$(do_facet $SINGLEMDS "lctl get_param version" | cut -d. -f1,2)
@@ -1675,6 +1727,11 @@ init_param_vars () {
 
     log "Using TIMEOUT=$TIMEOUT"
 
+    mds_sanity_check $TIMEOUT
+
+    if [ x"$(som_check)" = x"enabled" ]; then
+        ENABLE_QUOTA=""
+    fi
     if [ "$ENABLE_QUOTA" ]; then
         setup_quota $MOUNT  || return 2
     fi
@@ -1692,8 +1749,7 @@ check_config () {
     mgshost=$(echo $mgshost | awk -F: '{print $1}')
 
     if [ "$mgshost" != "$myMGS_host" ]; then
-        FAIL_ON_ERROR=true \
-            error "Bad config file: lustre is mounted with mgs $mgshost, but mgs_HOST=$mgs_HOST, NETTYPE=$NETTYPE
+            log "Bad config file: lustre is mounted with mgs $mgshost, but mgs_HOST=$mgs_HOST, NETTYPE=$NETTYPE
                    Please use correct config or set mds_HOST correctly!"
     fi
 
@@ -1831,19 +1887,9 @@ get_facets () {
 ##################################
 # Adaptive Timeouts funcs
 
-at_is_valid() {
-    if [ -z "$AT_MAX_PATH" ]; then
-        AT_MAX_PATH=$(do_facet $SINGLEMDS "find /sys/ -name at_max")
-        [ -z "$AT_MAX_PATH" ] && echo "missing /sys/.../at_max " && return 1
-    fi
-    return 0
-}
-
 at_is_enabled() {
-    at_is_valid || error "invalid call"
-
     # only check mds, we assume at_max is the same on all nodes
-    local at_max=$(do_facet $SINGLEMDS "cat $AT_MAX_PATH")
+    local at_max=$(do_facet $SINGLEMDS "lctl get_param -n at_max")
     if [ $at_max -eq 0 ]; then
         return 1
     else
@@ -1854,13 +1900,11 @@ at_is_enabled() {
 at_max_get() {
     local facet=$1
 
-    at_is_valid || error "invalid call"
-
     # suppose that all ost-s has the same at_max set
     if [ $facet == "ost" ]; then
-        do_facet ost1 "cat $AT_MAX_PATH"
+	do_facet ost1 "lctl get_param -n at_max"
     else
-        do_facet $facet "cat $AT_MAX_PATH"
+	do_facet $facet "lctl get_param -n at_max"
     fi
 }
 
@@ -1868,20 +1912,19 @@ at_max_set() {
     local at_max=$1
     shift
 
-    at_is_valid || error "invalid call"
-
     local facet
     for facet in $@; do
         if [ $facet == "ost" ]; then
             for i in `seq $OSTCOUNT`; do
-                do_facet ost$i "echo $at_max > $AT_MAX_PATH"
+		do_facet ost$i "lctl set_param at_max=$at_max"
+
             done
         elif [ $facet == "mds" ]; then
             for i in `seq $MDSCOUNT`; do
-                do_facet mds$i "echo $at_max > $AT_MAX_PATH"
+		do_facet mds$i "lctl set_param at_max=$at_max"
             done
         else
-            do_facet $facet "echo $at_max > $AT_MAX_PATH"
+	    do_facet $facet "lctl set_param at_max=$at_max"
         fi
     done
 }
@@ -1962,12 +2005,7 @@ clear_failloc() {
 }
 
 set_nodes_failloc () {
-    local nodes=$1
-    local node
-
-    for node in $nodes ; do
-        do_node $node lctl set_param fail_loc=$2
-    done
+    do_nodes $(comma_list $1)  lctl set_param fail_loc=$2
 }
 
 cancel_lru_locks() {
@@ -2029,10 +2067,7 @@ error_noexit() {
     ERRLOG=$TMP/lustre_${TESTSUITE}_${TESTNAME}.$(date +%s)
     echo "Dumping lctl log to $ERRLOG"
     # We need to dump the logs on all nodes
-    local NODES=$(nodes_list)
-    for NODE in $NODES; do
-        do_node $NODE $LCTL dk $ERRLOG
-    done
+    do_nodes $(comma_list $(nodes_list)) $NODE $LCTL dk $ERRLOG
     debugrestore
     [ "$TESTSUITELOG" ] && echo "$0: ${TYPE}: $TESTNAME $@" >> $TESTSUITELOG
     TEST_FAILED=true
@@ -2169,10 +2204,7 @@ log() {
     MSG=${MSG//\>/\\\>}
     MSG=${MSG//\</\\\<}
     MSG=${MSG//\//\\\/}
-    local NODES=$(nodes_list)
-    for NODE in $NODES; do
-        do_node $NODE $LCTL mark "$MSG" 2> /dev/null || true
-    done
+    do_nodes $(comma_list $(nodes_list)) $LCTL mark "$MSG" 2> /dev/null || true
 }
 
 trace() {
@@ -2195,13 +2227,8 @@ check_mds() {
 }
 
 reset_fail_loc () {
-    local myNODES=$(nodes_list)
-    local NODE
-
     echo -n "Resetting fail_loc on all nodes..."
-    for NODE in $myNODES; do
-        do_node $NODE "lctl set_param -n fail_loc=0 2>/dev/null || true"
-    done
+    do_nodes $(comma_list $(nodes_list)) "lctl set_param -n fail_loc=0 2>/dev/null || true"
     echo done.
 }
 
@@ -2529,6 +2556,19 @@ multiop_bg_pause() {
     return 0
 }
 
+do_and_time () {
+    local cmd=$1
+    local rc
+
+    SECONDS=0
+    eval '$cmd'
+    
+    [ ${PIPESTATUS[0]} -eq 0 ] || rc=1
+
+    echo $SECONDS
+    return $rc
+}
+
 inodes_available () {
     local IFree=$($LFS df -i $MOUNT | grep ^$FSNAME | awk '{print $4}' | sort -un | head -1) || return 1
     echo $IFree
@@ -2623,12 +2663,12 @@ get_mds_dir () {
     rm -f $file
     sleep 1
     local iused=$(lfs df -i $dir | grep MDT | awk '{print $3}')
-    local oldused=($iused)
+    local -a oldused=($iused)
 
     touch $file
     sleep 1
     iused=$(lfs df -i $dir | grep MDT | awk '{print $3}')
-    local newused=($iused)
+    local -a newused=($iused)
 
     local num=0
     for ((i=0; i<${#newused[@]}; i++)); do
@@ -2674,3 +2714,87 @@ delayed_recovery_enabled () {
     do_facet $SINGLEMDS lctl get_param -n mdd.${!var}.stale_export_age > /dev/null 2>&1
 }
 
+########################
+convert_facet2name() {
+    case "$1" in
+        "ost" ) echo "OST0000" ;;
+        "ost1") echo "OST0000" ;;
+        "ost2") echo "OST0001" ;;
+        "ost3") echo "OST0002" ;;
+        "ost4") echo "OST0003" ;;
+        "ost5") echo "OST0004" ;;
+        *) error "unknown facet!" ;;
+    esac
+}
+
+get_clientosc_proc_path() {
+    local ost=$1
+
+    echo "{$1}-osc-*"
+}
+
+get_lustre_version () {
+    local node=${1:-"mds"}    
+    do_facet $node $LCTL get_param -n version |  awk '/^lustre:/ {print $2}'
+}
+
+get_mds_version_major () {
+    local version=$(get_lustre_version mds)
+    echo $version | awk -F. '{print $1}'
+}
+
+get_mds_version_minor () {
+    local version=$(get_lustre_version mds)
+    echo $version | awk -F. '{print $2}'
+}
+
+get_mdtosc_proc_path() {
+    local ost=$1
+    local major=$(get_mds_version_major)
+    local minor=$(get_mds_version_minor)
+    if [ $major -le 1 -a $minor -le 8 ] ; then
+        echo "${ost}-osc"
+    else
+        echo "${ost}-osc-MDT0000"
+    fi
+}
+
+get_osc_import_name() {
+    local node=$1
+    local ost=$2
+    local name=$(convert_facet2name $ost)
+
+    if [ "$node" == "mds" ]; then
+        get_mdtosc_proc_path $name
+        return 0
+    fi
+
+    get_clientosc_proc_path $name
+    return 0
+}
+
+wait_osc_import_state() {
+    local node=$1
+    local ost_facet=$2
+    local expected=$3
+    local ost=$(get_osc_import_name $node $ost_facet)
+    local CONN_PROC
+    local CONN_STATE
+    local i=0
+
+    CONN_PROC="osc.${FSNAME}-${ost}.ost_server_uuid"
+    CONN_STATE=$(do_facet $node lctl get_param -n $CONN_PROC | cut -f2)
+    while [ "${CONN_STATE}" != "${expected}" ]; do
+        # for disconn we can check after proc entry is removed
+        [ "x${CONN_STATE}" == "x" -a "${expected}" == "DISCONN" ] && return 0
+        # disconnect rpc should be wait not more obd_timeout
+        [ $i -ge $(($TIMEOUT * 3 / 2)) ] && \
+            error "can't put import for ${ost}(${ost_facet}) into ${expected} state" && return 1
+        sleep 1
+        CONN_STATE=$(do_facet $node lctl get_param -n $CONN_PROC | cut -f2)
+        i=$(($i + 1))
+    done
+
+    log "${ost_facet} now in ${CONN_STATE} state"
+    return 0
+}