b=19737

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

diff --git a/lustre/tests/test-framework.sh b/lustre/tests/test-framework.sh
index 11fae0e..9507a56 100644 (file)
--- a/lustre/tests/test-framework.sh
+++ b/lustre/tests/test-framework.sh
@@ -666,7 +666,7 @@ sanity_mount_check_nodes () {
 
     local rc=0
     for mnt in $mnts ; do
-        do_nodes $nodes "set -x; running=\\\$(grep -c $mnt' ' /proc/mounts);
+        do_nodes $nodes "running=\\\$(grep -c $mnt' ' /proc/mounts);
 mpts=\\\$(mount | grep -w -c $mnt);
 if [ \\\$running -ne \\\$mpts ]; then
     echo \\\$(hostname) env are INSANE!;
@@ -731,7 +731,7 @@ zconf_mount_clients() {
 
     echo "Starting client $clients: $OPTIONS $device $mnt"
 
-    do_nodes $clients "set -x;
+    do_nodes $clients "
 running=\\\$(mount | grep -c $mnt' ');
 rc=0;
 if [ \\\$running -eq 0 ] ; then
@@ -759,7 +759,7 @@ zconf_umount_clients() {
     [ "$3" ] && force=-f
 
     echo "Stopping clients: $clients $mnt (opts:$force)"
-    do_nodes $clients "set -x; running=\\\$(grep -c $mnt' ' /proc/mounts);
+    do_nodes $clients "running=\\\$(grep -c $mnt' ' /proc/mounts);
 if [ \\\$running -ne 0 ] ; then
 echo Stopping client \\\$(hostname) client $mnt opts:$force;
 lsof -t $mnt || need_kill=no;
@@ -834,10 +834,14 @@ check_progs_installed () {
     shift
     local progs=$@
 
-    do_nodes $clients "set -x ; PATH=:$PATH status=true; for prog in $progs; do
-        which \\\$prog || { echo \\\$prog missing on \\\$(hostname) && status=false; }
-        done;
-        eval \\\$status"
+    do_nodes $clients "PATH=:$PATH; status=true;
+for prog in $progs; do
+    if ! [ \\\"\\\$(which \\\$prog)\\\"  -o  \\\"\\\${!prog}\\\" ]; then
+       echo \\\$prog missing on \\\$(hostname);
+       status=false;
+    fi
+done;
+eval \\\$status"
 }
 
 client_var_name() {
@@ -1157,9 +1161,6 @@ facet_failover() {
     shutdown_facet $facet
     [ -n "$sleep_time" ] && sleep $sleep_time
     reboot_facet $facet
-    client_df &
-    DFPID=$!
-    echo "df pid is $DFPID"
     change_active $facet
     local TO=`facet_active_host $facet`
     echo "Failover $facet to $TO"
@@ -1215,7 +1216,7 @@ ost_evict_client() {
 
 fail() {
     facet_failover $* || error "failover: $?"
-    df $MOUNT || error "post-failover df: $?"
+    client_df || error "post-failover df: $?"
 }
 
 fail_nodf() {
@@ -1228,9 +1229,9 @@ fail_abort() {
     stop $facet
     change_active $facet
     mount_facet $facet -o abort_recovery
-    df $MOUNT || echo "first df failed: $?"
+    client_df || echo "first df failed: $?"
     sleep 1
-    df $MOUNT || error "post-failover df: $?"
+    client_df || error "post-failover df: $?"
 }
 
 do_lmc() {
@@ -1470,6 +1471,8 @@ stopall() {
 }
 
 cleanupall() {
+    nfs_client_mode && return
+
     stopall $*
     unload_modules
     cleanup_gss
@@ -1584,6 +1587,8 @@ writeconf_all () {
 }
 
 setupall() {
+    nfs_client_mode && return
+
     sanity_mount_check ||
         error "environments are insane!"
 
@@ -1746,7 +1751,24 @@ init_param_vars () {
     fi
 }
 
+nfs_client_mode () {
+    if [ "$NFSCLIENT" ]; then
+        echo "NFSCLIENT mode: setup, cleanup, check config skipped"
+        local clients=$CLIENTS
+        [ -z $clients ] && clients=$(hostname)
+
+        # FIXME: remove hostname when 19215 fixed
+        do_nodes $clients "echo \\\$(hostname); grep ' '$MOUNT' ' /proc/mounts"
+        declare -a nfsexport=(`grep ' '$MOUNT' ' /proc/mounts | awk '{print $1}' | awk -F: '{print $1 " "  $2}'`)
+        do_nodes ${nfsexport[0]} "echo \\\$(hostname); df -T  ${nfsexport[1]}"
+        return
+    fi
+    return 1
+}
+
 check_config () {
+    nfs_client_mode && return
+
     local mntpt=$1
 
     local mounted=$(mount | grep " $mntpt ")
@@ -1800,6 +1822,8 @@ check_timeout () {
 }
 
 check_and_setup_lustre() {
+    nfs_client_mode && return
+
     local MOUNTED=$(mounted_lustre_filesystems)
     if [ -z "$MOUNTED" ] || ! $(echo $MOUNTED | grep -w -q $MOUNT); then
         [ "$REFORMAT" ] && formatall
@@ -1811,6 +1835,11 @@ check_and_setup_lustre() {
         check_config $MOUNT
         init_facets_vars
         init_param_vars
+
+        do_nodes $(comma_list $(nodes_list)) "lctl set_param debug=\\\"$PTLDEBUG\\\";
+            lctl set_param subsystem_debug=\\\"${SUBSYSTEM# }\\\";
+            lctl set_param debug_mb=${DEBUG_SIZE};
+            sync"
     fi
     if [ "$ONLY" == "setup" ]; then
         exit 0
@@ -2686,7 +2715,7 @@ get_stripe_info() {
 
        stripe_size=`awk '$1 ~ /size/ {print $2}' $tmp_file`
        stripe_count=`awk '$1 ~ /count/ {print $2}' $tmp_file`
-       stripe_index=`awk '/obdidx/ {start = 1; getline; print $1; exit}' $tmp_file`
+       stripe_index=`awk '$1 ~ /stripe_offset/ {print $2}' $tmp_file`
        rm -f $tmp_file
 }
 
@@ -2716,30 +2745,6 @@ get_mds_dir () {
     error "mdt-s : inodes count OLD ${oldused[@]} NEW ${newused[@]}"
 }
 
-mpi_run () {
-    local mpirun="$MPIRUN $MPIRUN_OPTIONS"
-    local command="$mpirun $@"
-    local mpilog=$TMP/mpi.log
-    local rc
-
-    if [ "$MPI_USER" != root -a $mpirun ]; then
-        echo "+ chmod 0777 $MOUNT"
-        chmod 0777 $MOUNT
-        command="su $MPI_USER sh -c \"$command \""
-    fi
-
-    ls -ald $MOUNT
-    echo "+ $command"
-    eval $command 2>&1 > $mpilog || true
-
-    rc=${PIPESTATUS[0]}
-    if [ $rc -eq 0 ] && grep -q "p4_error: : [^0]" $mpilog ; then
-       rc=1
-    fi
-    cat $mpilog
-    return $rc
-}
-
 mdsrate_cleanup () {
     mpi_run -np $1 -machinefile $2 ${MDSRATE} --unlink --nfiles $3 --dir $4 --filefmt $5 $6
 }
@@ -2820,8 +2825,13 @@ wait_osc_import_state() {
     CONN_PROC="osc.${FSNAME}-${ost}.ost_server_uuid"
     CONN_STATE=$(do_facet $node lctl get_param -n $CONN_PROC 2>/dev/null | cut -f2)
     while [ "${CONN_STATE}" != "${expected}" ]; do
-        # for disconn we can check after proc entry is removed
-        [ "x${CONN_STATE}" == "x" -a "${expected}" == "DISCONN" ] && return 0
+        if [ "${expected}" == "DISCONN" ]; then 
+            # for disconn we can check after proc entry is removed
+            [ "x${CONN_STATE}" == "x" ] && return 0
+            #  with AT we can have connect request timeout ~ reconnect timeout
+            # and test can't see real disconnect
+            [ "${CONN_STATE}" == "CONNECTING" ] && return 0
+        fi
         # 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