export GSS_KRB5=false
export GSS_PIPEFS=false
export IDENTITY_UPCALL=default
+export QUOTA_AUTO=1
#export PDSH="pdsh -S -Rssh -w"
-# eg, assert_env LUSTRE MDSNODES OSTNODES CLIENTS
-assert_env() {
- local failed=""
- for name in $@; do
- if [ -z "${!name}" ]; then
- echo "$0: $name must be set"
- failed=1
- fi
- done
- [ $failed ] && exit 1 || true
-}
+# function used by scripts run on remote nodes
+LUSTRE=${LUSTRE:-$(cd $(dirname $0)/..; echo $PWD)}
+. $LUSTRE/tests/functions.sh
+
+LUSTRE_TESTS_CFG_DIR=${LUSTRE_TESTS_CFG_DIR:-${LUSTRE}/tests/cfg}
+
+EXCEPT_LIST_FILE=${EXCEPT_LIST_FILE:-${LUSTRE_TESTS_CFG_DIR}/tests-to-skip.sh}
+
+if [ -f "$EXCEPT_LIST_FILE" ]; then
+ echo "Reading test skip list from $EXCEPT_LIST_FILE"
+ cat $EXCEPT_LIST_FILE
+ . $EXCEPT_LIST_FILE
+fi
assert_DIR () {
local failed=""
}
print_summary () {
+ trap 0
[ "$TESTSUITE" == "lfscktest" ] && return 0
[ -n "$ONLY" ] && echo "WARNING: ONLY is set to ${ONLY}."
local form="%-13s %-17s %s\n"
done
for O in $TESTSUITE_LIST; do
- [ "${!O}" = "no" ] && \
- printf "$form" "Skipped" "$O" ""
+ if [ "${!O}" = "no" ]; then
+ # FIXME.
+ # only for those tests suits which are run directly from acc-sm script:
+ # bonnie, iozone, etc.
+ if [ -f "$TESTSUITELOG" ] && grep FAIL $TESTSUITELOG | grep -q ' '$O ; then
+ printf "$form" "UNFINISHED" "$O" ""
+ else
+ printf "$form" "Skipped" "$O" ""
+ fi
+ fi
done
for O in $TESTSUITE_LIST; do
export LUSTRE=`absolute_path $LUSTRE`
export TESTSUITE=`basename $0 .sh`
export TEST_FAILED=false
+ export FAIL_ON_SKIP_ENV=${FAIL_ON_SKIP_ENV:-false}
export MKE2FS=${MKE2FS:-mke2fs}
export DEBUGFS=${DEBUGFS:-debugfs}
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/test/racer; then
+ 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:-"$LUSTRE/utils/lfs"}
export TUNEFS=${TUNEFS:-"$LUSTRE/utils/tunefs.lustre"}
[ ! -f "$TUNEFS" ] && export TUNEFS=$(which tunefs.lustre)
export CHECKSTAT="${CHECKSTAT:-"checkstat -v"} "
+ export LUSTRE_RMMOD=${LUSTRE_RMMOD:-$LUSTRE/scripts/lustre_rmmod}
+ [ ! -f "$LUSTRE_RMMOD" ] && export LUSTRE_RMMOD=$(which lustre_rmmod 2> /dev/null)
export FSTYPE=${FSTYPE:-"ldiskfs"}
export NAME=${NAME:-local}
export LGSSD=${LGSSD:-"$LUSTRE/utils/gss/lgssd"}
[ "$GSS_PIPEFS" = "true" ] && [ ! -f "$LGSSD" ] && \
export LGSSD=$(which lgssd)
export LSVCGSSD=${LSVCGSSD:-"$LUSTRE/utils/gss/lsvcgssd"}
- [ ! -f "$LSVCGSSD" ] && export LSVCGSSD=$(which lsvcgssd)
+ [ ! -f "$LSVCGSSD" ] && export LSVCGSSD=$(which lsvcgssd 2> /dev/null)
export KRB5DIR=${KRB5DIR:-"/usr/kerberos"}
export DIR2
export SAVE_PWD=${SAVE_PWD:-$LUSTRE/tests}
IDENTITY_UPCALL=false
;;
esac
+ export LOAD_MODULES_REMOTE=${LOAD_MODULES_REMOTE:-false}
# Paths on remote nodes, if different
export RLUSTRE=${RLUSTRE:-$LUSTRE}
export RPWD=${RPWD:-$PWD}
export I_MOUNTED=${I_MOUNTED:-"no"}
+ if [ ! -f /lib/modules/$(uname -r)/kernel/fs/lustre/mds.ko -a \
+ ! -f `dirname $0`/../mds/mds.ko ]; then
+ export CLIENTMODSONLY=yes
+ fi
# command line
[ "$TESTSUITELOG" ] && rm -f $TESTSUITELOG || true
rm -f $TMP/*active
-
}
case `uname -r` in
*) EXT=".ko"; USE_QUOTA=yes;;
esac
+
+module_loaded () {
+ /sbin/lsmod | grep -q $1
+}
+
load_module() {
EXT=".ko"
module=$1
shift
BASE=`basename $module $EXT`
- lsmod | grep -q ${BASE} || \
- if [ -f ${LUSTRE}/${module}${EXT} ]; then
+
+ module_loaded ${BASE} && return
+
+ if [ -f ${LUSTRE}/${module}${EXT} ]; then
insmod ${LUSTRE}/${module}${EXT} $@
else
# must be testing a "make install" or "rpm" installation
fi
}
-load_modules() {
+load_modules_local() {
if [ -n "$MODPROBE" ]; then
# use modprobe
return 0
echo Loading modules from $LUSTRE
load_module ../libcfs/libcfs/libcfs
- [ "$PTLDEBUG" ] && lctl set_param debug=$PTLDEBUG
- [ "$SUBSYSTEM" ] && lctl set_param subsystem_debug=${SUBSYSTEM# }
+ [ "$PTLDEBUG" ] && lctl set_param debug="$PTLDEBUG"
+ [ "$SUBSYSTEM" ] && lctl set_param subsystem_debug="${SUBSYSTEM# }"
local MODPROBECONF=
[ -f /etc/modprobe.conf ] && MODPROBECONF=/etc/modprobe.conf
[ ! "$MODPROBECONF" -a -d /etc/modprobe.d ] && MODPROBECONF=/etc/modprobe.d/Lustre
load_module osc/osc
load_module lov/lov
load_module mgc/mgc
- if [ -z "$CLIENTONLY" ] && [ -z "$CLIENTMODSONLY" ]; then
+ if ! client_only; then
grep -q crc16 /proc/kallsyms || { modprobe crc16 2>/dev/null || true; }
+ grep -q jbd /proc/kallsyms || { modprobe jbd 2>/dev/null || true; }
[ "$FSTYPE" = "ldiskfs" ] && load_module ../ldiskfs/ldiskfs/ldiskfs
load_module mgs/mgs
load_module mds/mds
[ -f $LUSTRE/utils/mount.lustre ] && cp $LUSTRE/utils/mount.lustre /sbin/. || true
}
-RMMOD=rmmod
-if [ `uname -r | cut -c 3` -eq 4 ]; then
- RMMOD="modprobe -r"
-fi
-
-wait_for_lnet() {
- local UNLOADED=0
- local WAIT=0
- local MAX=60
- MODULES=$($LCTL modules | awk '{ print $2 }')
- while [ -n "$MODULES" ]; do
- sleep 5
- $RMMOD $MODULES > /dev/null 2>&1 || true
- MODULES=$($LCTL modules | awk '{ print $2 }')
- if [ -z "$MODULES" ]; then
- return 0
- else
- WAIT=$((WAIT + 5))
- echo "waiting, $((MAX - WAIT)) secs left"
- fi
- if [ $WAIT -eq $MAX ]; then
- echo "LNET modules $MODULES will not unload"
- lsmod
- return 3
- fi
- done
-}
-
-unload_dep_module() {
- #lsmod output
- #libcfs 107852 17 llite_lloop,lustre,obdfilter,ost,...
- local MODULE=$1
- local DEPS=$(lsmod | awk '($1 == "'$MODULE'") { print $4 }' | tr ',' ' ')
- for SUBMOD in $DEPS; do
- unload_dep_module $SUBMOD
- done
- [ "$MODULE" = "libcfs" ] && $LCTL dk $TMP/debug || true
- $RMMOD $MODULE || true
+load_modules () {
+ load_modules_local
+ # bug 19124
+ # load modules on remote nodes optionally
+ # lustre-tests have to be installed on these nodes
+ if $LOAD_MODULES_REMOTE ; then
+ local list=$(comma_list $(remote_nodes_list))
+ echo loading modules on $list
+ do_rpc_nodes $list load_modules
+ fi
}
check_mem_leak () {
unload_modules() {
wait_exit_ST client # bug 12845
- lsmod | grep libcfs > /dev/null && $LCTL dl
- unload_dep_module $FSTYPE
- unload_dep_module libcfs
-
- local MODULES=$($LCTL modules | awk '{ print $2 }')
- if [ -n "$MODULES" ]; then
- echo "Modules still loaded: "
- echo $MODULES
- if [ "$(lctl dl)" ]; then
- echo "Lustre still loaded"
- lctl dl || true
- lsmod
- return 2
- else
- echo "Lustre stopped but LNET is still loaded, waiting..."
- wait_for_lnet || return 3
- fi
+ $LUSTRE_RMMOD $FSTYPE || return 2
+ if $LOAD_MODULES_REMOTE ; then
+ local list=$(comma_list $(remote_nodes_list))
+ echo unloading modules on $list
+ do_rpc_nodes $list $LUSTRE_RMMOD $FSTYPE
+ do_rpc_nodes $list check_mem_leak
fi
+
HAVE_MODULES=false
check_mem_leak || return 254
mount_facet() {
local facet=$1
shift
- local dev=${facet}_dev
+ local dev=$(facet_active $facet)_dev
local opt=${facet}_opt
echo "Starting ${facet}: ${!opt} $@ ${!dev} ${MOUNT%/*}/${facet}"
do_facet ${facet} mount -t lustre ${!opt} $@ ${!dev} ${MOUNT%/*}/${facet}
echo "mount -t lustre $@ ${!dev} ${MOUNT%/*}/${facet}"
echo "Start of ${!dev} on ${facet} failed ${RC}"
else
- do_facet ${facet} "lctl set_param debug=$PTLDEBUG; \
- lctl set_param subsystem_debug=${SUBSYSTEM# }; \
+ do_facet ${facet} "lctl set_param debug=\\\"$PTLDEBUG\\\"; \
+ lctl set_param subsystem_debug=\\\"${SUBSYSTEM# }\\\"; \
lctl set_param debug_mb=${DEBUG_SIZE}; \
sync"
shift
eval export ${facet}_dev=${device}
eval export ${facet}_opt=\"$@\"
+
+ local varname=${facet}failover_dev
+ if [ -n "${!varname}" ] ; then
+ eval export ${facet}failover_dev=${!varname}
+ else
+ eval export ${facet}failover_dev=$device
+ fi
+
do_facet ${facet} mkdir -p ${MOUNT%/*}/${facet}
mount_facet ${facet}
RC=$?
wait_exit_ST ${facet}
}
+# 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}
+ local spec=$1
+ local ver=$(tr -c -d "123" <<< $spec)
+ local type=$(tr -c -d "ug" <<< $spec)
+
+ [ -n "$ver" -a "$ver" != "3" ] && error "wrong quota version specifier"
+
+ [ -n "$type" ] && { $LFS quotacheck -$type $MOUNT || error "quotacheck has failed"; }
+
+ do_facet mgs "lctl conf_param ${fsname}-MDT*.mdd.quota_type=$spec"
+ local varsvc
+ local osts=$(get_facets OST)
+ for ost in ${osts//,/ }; do
+ varsvc=${ost}_svc
+ do_facet mgs "lctl conf_param ${!varsvc}.ost.quota_type=$spec"
+ done
+}
+
+# client could mount several lustre
+quota_type () {
+ local fsname=${1:-$FSNAME}
+ local rc=0
+ do_facet mgs lctl get_param mdd.${fsname}-MDT*.quota_type || rc=$?
+ do_nodes $(comma_list $(osts_nodes)) \
+ lctl get_param obdfilter.${fsname}-OST*.quota_type || rc=$?
+ return $rc
+}
+
+restore_quota_type () {
+ local mntpt=${1:-$MOUNT}
+ local quota_type=$(quota_type $FSNAME | grep MDT | cut -d "=" -f2)
+ if [ ! "$old_QUOTA_TYPE" ] || [ "$quota_type" = "$old_QUOTA_TYPE" ]; then
+ return
+ fi
+ quota_save_version $old_QUOTA_TYPE
+}
+
+setup_quota(){
+ local mntpt=$1
+
+ # We need save the original quota_type params, and restore them after testing
+
+ # Suppose that quota type the same on mds and ost
+ local quota_type=$(quota_type | grep MDT | cut -d "=" -f2)
+ [ ${PIPESTATUS[0]} -eq 0 ] || error "quota_type failed!"
+ echo "[HOST:$HOSTNAME] [old_quota_type:$quota_type] [new_quota_type:$QUOTA_TYPE]"
+ if [ "$quota_type" != "$QUOTA_TYPE" ]; then
+ export old_QUOTA_TYPE=$quota_type
+ quota_save_version $QUOTA_TYPE
+ else
+ qtype=$(tr -c -d "ug" <<< $QUOTA_TYPE)
+ $LFS quotacheck -$qtype $mntpt || error "quotacheck has failed for $type"
+ fi
+
+ local quota_usrs=$QUOTA_USERS
+
+ # get_filesystem_size
+ local disksz=$(lfs df $mntpt | grep "filesystem summary:" | awk '{print $3}')
+ local blk_soft=$((disksz + 1024))
+ local blk_hard=$((blk_soft + blk_soft / 20)) # Go 5% over
+
+ local Inodes=$(lfs df -i $mntpt | grep "filesystem summary:" | awk '{print $3}')
+ local i_soft=$Inodes
+ local i_hard=$((i_soft + i_soft / 20))
+
+ echo "Total disk size: $disksz block-softlimit: $blk_soft block-hardlimit:
+ $blk_hard inode-softlimit: $i_soft inode-hardlimit: $i_hard"
+
+ local cmd
+ for usr in $quota_usrs; do
+ echo "Setting up quota on $HOSTNAME:$mntpt for $usr..."
+ for type in u g; do
+ cmd="$LFS setquota -$type $usr -b $blk_soft -B $blk_hard -i $i_soft -I $i_hard $mntpt"
+ echo "+ $cmd"
+ eval $cmd || error "$cmd FAILED!"
+ done
+ # display the quota status
+ echo "Quota settings for $usr : "
+ $LFS quota -v -u $usr $mntpt || true
+ done
+}
+
zconf_mount() {
local OPTIONS
local client=$1
do_node $client mkdir -p $mnt
do_node $client mount -t lustre $OPTIONS $device $mnt || return 1
- do_node $client "lctl set_param debug=$PTLDEBUG;
- lctl set_param subsystem_debug=${SUBSYSTEM# };
+ do_node $client "lctl set_param debug=\\\"$PTLDEBUG\\\";
+ lctl set_param subsystem_debug=\\\"${SUBSYSTEM# }\\\";
lctl set_param debug_mb=${DEBUG_SIZE}"
return 0
fi
}
+# nodes is comma list
+sanity_mount_check_nodes () {
+ local nodes=$1
+ shift
+ local mnts="$@"
+ local mnt
+
+ # FIXME: assume that all cluster nodes run the same os
+ [ "$(uname)" = Linux ] || return 0
+
+ local rc=0
+ for mnt in $mnts ; do
+ 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!;
+ exit 1;
+fi"
+ [ $? -eq 0 ] || rc=1
+ done
+ return $rc
+}
+
+sanity_mount_check_servers () {
+ [ "$CLIENTONLY" ] &&
+ { echo "CLIENTONLY mode, skip mount_check_servers"; return 0; } || true
+ echo Checking servers environments
+
+ # FIXME: modify get_facets to display all facets wo params
+ local facets="$(get_facets OST),$(get_facets MDS),mgs"
+ local node
+ local mnt
+ local facet
+ for facet in ${facets//,/ }; do
+ node=$(facet_host ${facet})
+ mnt=${MOUNT%/*}/${facet}
+ sanity_mount_check_nodes $node $mnt ||
+ { error "server $node environments are insane!"; return 1; }
+ done
+}
+
+sanity_mount_check_clients () {
+ local clients=${1:-$CLIENTS}
+ local mntpt=${2:-$MOUNT}
+ local mntpt2=${3:-$MOUNT2}
+
+ [ -z $clients ] && clients=$(hostname)
+ echo Checking clients $clients environments
+
+ sanity_mount_check_nodes $clients $mntpt $mntpt2 ||
+ error "clients environments are insane!"
+}
+
+sanity_mount_check () {
+ sanity_mount_check_servers || return 1
+ sanity_mount_check_clients || return 2
+}
+
# mount clients if not mouted
zconf_mount_clients() {
- local OPTIONS
local clients=$1
local mnt=$2
-
+ local OPTIONS=${3:-$MOUNTOPT}
# Only supply -o to mount if we have options
- if [ -n "$MOUNTOPT" ]; then
- OPTIONS="-o $MOUNTOPT"
+ if [ "$OPTIONS" ]; then
+ OPTIONS="-o $OPTIONS"
fi
local device=$MGSNID:/$FSNAME
if [ -z "$mnt" -o -z "$FSNAME" ]; then
fi
echo "Starting client $clients: $OPTIONS $device $mnt"
- do_nodes $clients "mount | grep $mnt || { mkdir -p $mnt && mount -t lustre $OPTIONS $device $mnt || false; }"
+
+ do_nodes $clients "
+running=\\\$(mount | grep -c $mnt' ');
+rc=0;
+if [ \\\$running -eq 0 ] ; then
+ mkdir -p $mnt;
+ mount -t lustre $OPTIONS $device $mnt;
+ rc=\\\$?;
+fi;
+exit \\\$rc" || return ${PIPESTATUS[0]}
echo "Started clients $clients: "
- do_nodes $clients "mount | grep $mnt"
+ do_nodes $clients "mount | grep -w $mnt"
- do_nodes $clients "lctl set_param debug=$PTLDEBUG;
- lctl set_param subsystem_debug=${SUBSYSTEM# };
+ do_nodes $clients "lctl set_param debug=\\\"$PTLDEBUG\\\";
+ lctl set_param subsystem_debug=\\\"${SUBSYSTEM# }\\\";
lctl set_param debug_mb=${DEBUG_SIZE};"
return 0
[ "$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
+echo Stopping client \\\$(hostname) $mnt opts:$force;
+lsof -t $mnt || need_kill=no;
if [ "x$force" != "x" -a "x\\\$need_kill" != "xno" ]; then
pids=\\\$(lsof -t $mnt | sort -u);
- if [ -n \\\$pids ]; then
- kill -9 \\\$pids
+ if [ -n \\\"\\\$pids\\\" ]; then
+ kill -9 \\\$pids;
fi
-fi
-busy=\\\$(umount $force $mnt 2>&1 | grep -c "busy")
+fi;
+busy=\\\$(umount $force $mnt 2>&1 | grep -c "busy");
if [ \\\$busy -ne 0 ] ; then
- echo "$mnt is still busy, wait one second" && sleep 1
- umount $force $mnt
+ echo "$mnt is still busy, wait one second" && sleep 1;
+ umount $force $mnt;
fi
fi"
}
+shudown_node_hard () {
+ local host=$1
+ local attempts=3
+
+ for i in $(seq $attempts) ; do
+ $POWER_DOWN $host
+ sleep 1
+ ping -w 3 -c 1 $host > /dev/null 2>&1 || return 0
+ echo "waiting for $host to fail attempts=$attempts"
+ [ $i -lt $attempts ] || \
+ { echo "$host still pingable after power down! attempts=$attempts" && return 1; }
+ done
+}
+
+shutdown_client() {
+ local client=$1
+ local mnt=${2:-$MOUNT}
+ local attempts=3
+
+ if [ "$FAILURE_MODE" = HARD ]; then
+ shudown_node_hard $client
+ else
+ zconf_umount_clients $client $mnt -f
+ fi
+}
+
shutdown_facet() {
local facet=$1
if [ "$FAILURE_MODE" = HARD ]; then
- $POWER_DOWN `facet_active_host $facet`
- sleep 2
+ shudown_node_hard $(facet_active_host $facet)
elif [ "$FAILURE_MODE" = SOFT ]; then
stop $facet
fi
local node=$1
if [ "$FAILURE_MODE" = HARD ]; then
$POWER_UP $node
+ wait_for_host $node
fi
}
shift
local progs=$@
- 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 $progs; do
+ if ! [ \\\"\\\$(which \\\$prog)\\\" -o \\\"\\\${!prog}\\\" ]; then
+ echo \\\$prog missing on \\\$(hostname);
+ status=false;
+ fi
+done;
+eval \\\$status"
}
-start_client_load() {
- local list=(${1//,/ })
- local nodenum=$2
+client_var_name() {
+ echo __$(echo $1 | tr '-' 'X')
+}
- local numloads=${#CLIENT_LOADS[@]}
- local testnum=$((nodenum % numloads))
+start_client_load() {
+ local client=$1
+ local load=$2
+ local var=$(client_var_name $client)_load
+ eval export ${var}=$load
- do_node ${list[nodenum]} "PATH=$PATH MOUNT=$MOUNT ERRORS_OK=$ERRORS_OK \
+ do_node $client "PATH=$PATH MOUNT=$MOUNT ERRORS_OK=$ERRORS_OK \
BREAK_ON_ERROR=$BREAK_ON_ERROR \
END_RUN_FILE=$END_RUN_FILE \
LOAD_PID_FILE=$LOAD_PID_FILE \
TESTSUITELOG=$TESTSUITELOG \
- run_${CLIENT_LOADS[testnum]}.sh" &
+ run_${load}.sh" &
CLIENT_LOAD_PIDS="$CLIENT_LOAD_PIDS $!"
- log "Started client load: ${CLIENT_LOADS[testnum]} on ${list[nodenum]}"
+ log "Started client load: ${load} on $client"
- eval export ${list[nodenum]}_load=${CLIENT_LOADS[testnum]}
return 0
}
start_client_loads () {
- local clients=(${1//,/ })
+ local -a clients=(${1//,/ })
+ local numloads=${#CLIENT_LOADS[@]}
+ local testnum
- for ((num=0; num < ${#clients[@]}; num++ )); do
- start_client_load $1 $num
+ for ((nodenum=0; nodenum < ${#clients[@]}; nodenum++ )); do
+ testnum=$((nodenum % numloads))
+ start_client_load ${clients[nodenum]} ${CLIENT_LOADS[testnum]}
done
}
# only for remote client
check_client_load () {
local client=$1
- local var=${client}_load
-
+ local var=$(client_var_name $client)_load
local TESTLOAD=run_${!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
for client in $clients; do
check_client_load $client
- rc=$?
+ rc=${PIPESTATUS[0]}
if [ "$rc" != 0 ]; then
log "Client load failed on node $client, rc=$rc"
return $rc
fi
done
}
+
+restart_client_loads () {
+ local clients=${1//,/ }
+ local expectedfail=${2:-""}
+ local client=
+ local rc=0
+
+ for client in $clients; do
+ check_client_load $client
+ rc=${PIPESTATUS[0]}
+ if [ "$rc" != 0 -a "$expectedfail" ]; then
+ local var=$(client_var_name $client)_load
+ start_client_load $client ${!var}
+ echo "Restarted client load ${!var}: on $client. Checking ..."
+ check_client_load $client
+ rc=${PIPESTATUS[0]}
+ if [ "$rc" != 0 ]; then
+ log "Client load failed to restart on node $client, rc=$rc"
+ # failure one client load means test fail
+ # we do not need to check other
+ return $rc
+ fi
+ else
+ return $rc
+ fi
+ done
+}
# End recovery-scale functions
# verify that lustre actually cleaned up properly
[ "`lctl dl 2> /dev/null | wc -l`" -gt 0 ] && lctl dl && \
echo "$0: lustre didn't clean up..." 1>&2 && return 202 || true
- if [ "`/sbin/lsmod 2>&1 | egrep 'lnet|libcfs'`" ]; then
+ if module_loaded lnet || module_loaded libcfs; then
echo "$0: modules still loaded..." 1>&2
/sbin/lsmod 1>&2
return 203
return 0
}
+wait_update () {
+ local node=$1
+ local TEST=$2
+ local FINAL=$3
+ local MAX=${4:-90}
+
+ local RESULT
+ local WAIT=0
+ local sleep=5
+ 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
+ done
+ echo "Update not seen after $MAX sec: wanted '$FINAL' got '$RESULT'"
+ return 3
+}
+
+wait_update_facet () {
+ local facet=$1
+ wait_update $(facet_active_host $facet) "$@"
+}
+
wait_delete_completed () {
local TOTALPREV=`lctl get_param -n osc.*.kbytesavail | \
awk 'BEGIN{total=0}; {total+=$1}; END{print total}'`
sleep 1
TOTAL=`lctl get_param -n osc.*.kbytesavail | \
awk 'BEGIN{total=0}; {total+=$1}; END{print total}'`
- [ "$TOTAL" -eq "$TOTALPREV" ] && break
+ [ "$TOTAL" -eq "$TOTALPREV" ] && return 0
echo "Waiting delete completed ... prev: $TOTALPREV current: $TOTAL "
TOTALPREV=$TOTAL
WAIT=$(( WAIT + 1))
done
- echo "Delete completed."
+ echo "Delete is not completed in $MAX_WAIT sec"
+ return 1
}
wait_for_host() {
- HOST=$1
- check_network "$HOST" 900
- while ! do_node $HOST "ls -d $LUSTRE " > /dev/null; do sleep 5; done
+ local host=$1
+ check_network "$host" 900
+ while ! do_node $host hostname > /dev/null; do sleep 5; done
}
wait_for() {
- facet=$1
- HOST=`facet_active_host $facet`
- wait_for_host $HOST
+ local facet=$1
+ local host=`facet_active_host $facet`
+ wait_for_host $host
}
-wait_mds_recovery_done () {
- local timeout=`do_facet $SINGLEMDS lctl get_param -n timeout`
-#define OBD_RECOVERY_TIMEOUT (obd_timeout * 5 / 2)
-# as we are in process of changing obd_timeout in different ways
-# let's set MAX longer than that
- MAX=$(( timeout * 4 ))
- WAIT=0
+wait_recovery_complete () {
+ local facet=$1
+
+ # Use default policy if $2 is not passed by caller.
+ #define OBD_RECOVERY_TIMEOUT (obd_timeout * 5 / 2)
+ # as we are in process of changing obd_timeout in different ways
+ # let's set MAX longer than that
+ local MAX=${2:-$(( TIMEOUT * 4 ))}
+
+ local var_svc=${facet}_svc
+ local procfile="*.${!var_svc}.recovery_status"
+ local WAIT=0
+ local STATUS=
+
while [ $WAIT -lt $MAX ]; do
- STATUS=`do_facet $SINGLEMDS "lctl get_param -n mdt.*-MDT0000.recovery_status | grep status"`
- echo $STATUS | grep COMPLETE && return 0
+ STATUS=$(do_facet $facet lctl get_param -n $procfile | grep status)
+ [[ $STATUS = "status: COMPLETE" ]] && return 0
sleep 5
WAIT=$((WAIT + 5))
- echo "Waiting $(($MAX - $WAIT)) secs for MDS recovery done"
+ echo "Waiting $((MAX - WAIT)) secs for $facet recovery done. $STATUS"
done
- echo "MDS recovery not done in $MAX sec"
+ echo "$facet recovery not done in $MAX sec. $STATUS"
+ return 1
+}
+
+wait_mds_ost_sync () {
+ # just because recovery is done doesn't mean we've finished
+ # orphan cleanup. Wait for llogs to get synchronized.
+ echo "Waiting for orphan cleanup..."
+ # MAX value includes time needed for MDS-OST reconnection
+ local MAX=$(( TIMEOUT * 2 ))
+ local WAIT=0
+ while [ $WAIT -lt $MAX ]; do
+ local -a sync=($(do_nodes $(comma_list $(osts_nodes)) \
+ "$LCTL get_param -n obdfilter.*.mds_sync"))
+ local con=1
+ for ((i=0; i<${#sync[@]}; i++)); do
+ [ ${sync[$i]} -eq 0 ] && continue
+ # there is a not finished MDS-OST synchronization
+ con=0
+ break;
+ done
+ sleep 2 # increase waiting time and cover statfs cache
+ [ ${con} -eq 1 ] && return 0
+ echo "Waiting $WAIT secs for $facet mds-ost sync done."
+ WAIT=$((WAIT + 2))
+ done
+ echo "$facet recovery not done in $MAX sec. $STATUS"
+ return 1
+}
+
+wait_destroy_complete () {
+ echo "Waiting for destroy to be done..."
+ # MAX value shouldn't be big as this mean server responsiveness
+ # never increase this just to make test pass but investigate
+ # why it takes so long time
+ local MAX=5
+ local WAIT=0
+ while [ $WAIT -lt $MAX ]; do
+ local -a RPCs=($($LCTL get_param -n osc.*.destroys_in_flight))
+ local con=1
+ for ((i=0; i<${#RPCs[@]}; i++)); do
+ [ ${RPCs[$i]} -eq 0 ] && continue
+ # there are still some destroy RPCs in flight
+ con=0
+ break;
+ done
+ sleep 1
+ [ ${con} -eq 1 ] && return 0 # done waiting
+ echo "Waiting $WAIT secs for destroys to be done."
+ WAIT=$((WAIT + 1))
+ done
+ echo "Destroys weren't done in $MAX sec."
return 1
}
local pids=$(ps uax | grep "$PDSH.*$prog.*$MOUNT" | grep -v grep | awk '{print $2}')
[ -z "$pids" ] && return 0
echo "$PDSH processes still exists after $WAIT seconds. Still running: $pids"
+ # FIXME: not portable
for pid in $pids; do
cat /proc/${pid}/status || true
cat /proc/${pid}/wchan || true
return $rc
}
-client_df() {
+clients_up() {
# not every config has many clients
+ sleep 1
if [ ! -z "$CLIENTS" ]; then
- $PDSH $CLIENTS "df $MOUNT" > /dev/null
+ $PDSH $CLIENTS "stat -f $MOUNT" > /dev/null
+ else
+ stat -f $MOUNT > /dev/null
+ fi
+}
+
+client_up() {
+ local client=$1
+ # usually checked on particular client or locally
+ sleep 1
+ if [ ! -z "$client" ]; then
+ $PDSH $client "stat -f $MOUNT" > /dev/null
else
- df $MOUNT > /dev/null
+ stat -f $MOUNT > /dev/null
fi
}
+client_evicted() {
+ ! client_up $1
+}
+
client_reconnect() {
uname -n >> $MOUNT/recon
if [ -z "$CLIENTS" ]; then
}
facet_failover() {
- facet=$1
- sleep_time=$2
+ local facet=$1
+ local sleep_time=$2
echo "Failing $facet on node `facet_active_host $facet`"
shutdown_facet $facet
[ -n "$sleep_time" ] && sleep $sleep_time
reboot_facet $facet
- client_df &
- DFPID=$!
- echo "df pid is $DFPID"
change_active $facet
- TO=`facet_active_host $facet`
+ local TO=`facet_active_host $facet`
echo "Failover $facet to $TO"
wait_for $facet
mount_facet $facet || error "Restart of $facet failed"
do_facet $facet sync
df $MOUNT
local svc=${facet}_svc
- do_facet $facet $LCTL --device %${!svc} readonly
do_facet $facet $LCTL --device %${!svc} notransno
+ do_facet $facet $LCTL --device %${!svc} readonly
do_facet $facet $LCTL mark "$facet REPLAY BARRIER on ${!svc}"
$LCTL mark "local REPLAY BARRIER on ${!svc}"
}
do_facet $facet sync
local svc=${facet}_svc
echo Replay barrier on ${!svc}
- do_facet $facet $LCTL --device %${!svc} readonly
do_facet $facet $LCTL --device %${!svc} notransno
+ do_facet $facet $LCTL --device %${!svc} readonly
do_facet $facet $LCTL mark "$facet REPLAY BARRIER on ${!svc}"
$LCTL mark "local REPLAY BARRIER on ${!svc}"
}
local facet=$1 echo running=${running}
local svc=${facet}_svc
echo Replay barrier on ${!svc}
- do_facet $facet $LCTL --device %${!svc} readonly
do_facet $facet $LCTL --device %${!svc} notransno
+ do_facet $facet $LCTL --device %${!svc} readonly
do_facet $facet $LCTL mark "$facet REPLAY BARRIER on ${!svc}"
$LCTL mark "local REPLAY BARRIER on ${!svc}"
}
fail() {
facet_failover $* || error "failover: $?"
- df $MOUNT || error "post-failover df: $?"
+ clients_up || error "post-failover df: $?"
}
fail_nodf() {
stop $facet
change_active $facet
mount_facet $facet -o abort_recovery
- df $MOUNT || echo "first df failed: $?"
- sleep 1
- df $MOUNT || error "post-failover df: $?"
+ clients_up || echo "first df failed: $?"
+ clients_up || error "post-failover df: $?"
}
do_lmc() {
change_active() {
local facet=$1
- failover=${facet}failover
+ local failover=${facet}failover
host=`facet_host $failover`
[ -z "$host" ] && return
- curactive=`facet_active $facet`
+ local curactive=`facet_active $facet`
if [ -z "${curactive}" -o "$curactive" == "$failover" ] ; then
eval export ${facet}active=$facet
else
eval export ${facet}active=$failover
fi
# save the active host for this facet
- activevar=${facet}active
+ local activevar=${facet}active
echo "$activevar=${!activevar}" > $TMP/$activevar
}
do_node() {
- HOST=$1
+ local verbose=false
+ # do not stripe off hostname if verbose, bug 19215
+ if [ x$1 = x--verbose ]; then
+ shift
+ verbose=true
+ fi
+
+ local HOST=$1
shift
local myPDSH=$PDSH
if [ "$HOST" = "$HOSTNAME" ]; then
[ -n "$($myPDSH $HOST cat $command_status)" ] && return 1 || true
return 0
fi
- $myPDSH $HOST "(PATH=\$PATH:$RLUSTRE/utils:$RLUSTRE/tests:/sbin:/usr/sbin; cd $RPWD; sh -c \"$@\")" | sed "s/^${HOST}: //"
+
+ if $verbose ; then
+ # print HOSTNAME for myPDSH="no_dsh"
+ if [[ $myPDSH = no_dsh ]]; then
+ $myPDSH $HOST "(PATH=\$PATH:$RLUSTRE/utils:$RLUSTRE/tests:/sbin:/usr/sbin; cd $RPWD; sh -c \"$@\")" | sed -e "s/^/${HOSTNAME}: /"
+ else
+ $myPDSH $HOST "(PATH=\$PATH:$RLUSTRE/utils:$RLUSTRE/tests:/sbin:/usr/sbin; cd $RPWD; sh -c \"$@\")"
+ fi
+ else
+ $myPDSH $HOST "(PATH=\$PATH:$RLUSTRE/utils:$RLUSTRE/tests:/sbin:/usr/sbin; cd $RPWD; sh -c \"$@\")" | sed "s/^${HOST}: //"
+ fi
return ${PIPESTATUS[0]}
}
}
do_nodes() {
+ local verbose=false
+ # do not stripe off hostname if verbose, bug 19215
+ if [ x$1 = x--verbose ]; then
+ shift
+ verbose=true
+ fi
+
local rnodes=$1
shift
if $(single_local_node $rnodes); then
- do_node $rnodes $@
+ if $verbose; then
+ do_node --verbose $rnodes $@
+ else
+ do_node $rnodes $@
+ fi
return $?
fi
$myPDSH $rnodes $LCTL mark "$@" > /dev/null 2>&1 || :
fi
- $myPDSH $rnodes "(PATH=\$PATH:$RLUSTRE/utils:$RLUSTRE/tests:/sbin:/usr/sbin; cd $RPWD; sh -c \"$@\")" | sed -re "s/\w+:\s//g"
+ if $verbose ; then
+ $myPDSH $rnodes "(PATH=\$PATH:$RLUSTRE/utils:$RLUSTRE/tests:/sbin:/usr/sbin; cd $RPWD; sh -c \"$@\")"
+ else
+ $myPDSH $rnodes "(PATH=\$PATH:$RLUSTRE/utils:$RLUSTRE/tests:/sbin:/usr/sbin; cd $RPWD; sh -c \"$@\")" | sed -re "s/\w+:\s//g"
+ fi
return ${PIPESTATUS[0]}
}
do_facet() {
- facet=$1
+ local facet=$1
shift
- HOST=`facet_active_host $facet`
+ local HOST=`facet_active_host $facet`
[ -z $HOST ] && echo No host defined for facet ${facet} && exit 1
do_node $HOST "$@"
}
fail mds1
fi
- # assume client mount is local
- grep " $MOUNT " /proc/mounts && zconf_umount $HOSTNAME $MOUNT $*
- grep " $MOUNT2 " /proc/mounts && zconf_umount $HOSTNAME $MOUNT2 $*
+ local clients=$CLIENTS
+ [ -z $clients ] && clients=$(hostname)
- if [ -n "$CLIENTS" ]; then
- zconf_umount_clients $CLIENTS $MOUNT "$*" || true
- [ -n "$MOUNT2" ] && zconf_umount_clients $CLIENTS $MOUNT2 "$*" || true
- fi
+ zconf_umount_clients $clients $MOUNT "$*" || true
+ [ -n "$MOUNT2" ] && zconf_umount_clients $clients $MOUNT2 "$*" || true
[ "$CLIENTONLY" ] && return
# The add fn does rm ${facet}active file, this would be enough
}
cleanupall() {
+ nfs_client_mode && return
+
stopall $*
unload_modules
cleanup_gss
# We need ldiskfs here, may as well load them all
load_modules
[ "$CLIENTONLY" ] && return
- echo "Formatting mdts, osts"
+ echo Formatting mgs, mds, osts
+ if [[ $MDSDEV1 != $MGSDEV ]] || [[ $mds1_HOST != $mgs_HOST ]]; then
+ add mgs $mgs_MKFS_OPTS $FSTYPE_OPT --reformat $MGSDEV || exit 10
+ fi
+
for num in `seq $MDSCOUNT`; do
echo "Format mds$num: $(mdsdevname $num)"
if $VERBOSE; then
zconf_mount `hostname` $1 || error "mount failed"
}
-set_obd_timeout() {
- local facet=$1
- local timeout=$2
-
- do_facet $facet lsmod | grep -q obdclass || \
- do_facet $facet "modprobe obdclass"
-
- do_facet $facet "lctl set_param timeout=$timeout"
-}
-
writeconf_facet () {
local facet=$1
local dev=$2
}
setupall() {
+ nfs_client_mode && return
+
+ sanity_mount_check ||
+ error "environments are insane!"
+
load_modules
init_gss
if [ -z "$CLIENTONLY" ]; then
- echo "Setup mdts, osts"
+ echo Setup mgs, mdt, osts
echo $WRITECONF | grep -q "writeconf" && \
writeconf_all
+ if [[ $mds1_HOST != $mgs_HOST ]] || [[ $MDSDEV1 != $MGSDEV ]]; then
+ start mgs $MGSDEV $mgs_MOUNT_OPTS
+ fi
+
for num in `seq $MDSCOUNT`; do
DEVNAME=$(mdsdevname $num)
- set_obd_timeout mds$num $TIMEOUT
start mds$num $DEVNAME $MDS_MOUNT_OPTS
# We started mds, now we should set failover variables properly.
done
for num in `seq $OSTCOUNT`; do
DEVNAME=$(ostdevname $num)
- set_obd_timeout ost$num $TIMEOUT
start ost$num $DEVNAME $OST_MOUNT_OPTS
# We started ost$num, now we should set ost${num}failover variable properly.
[ -n "$CLIENTS" ] && zconf_mount_clients $CLIENTS $MOUNT2
fi
- init_versions_vars
+ init_param_vars
# by remounting mdt before ost, initial connect from mdt to ost might
# timeout because ost is not ready yet. wait some time to its fully
}
init_facet_vars () {
+ [ "$CLIENTONLY" ] && return 0
local facet=$1
shift
local device=$1
if [ -z "${!varname}" ]; then
eval $varname=$(facet_host $facet)
fi
+
+ # ${facet}failover_dev is set in cfg file
+ varname=${facet}failover_dev
+ if [ -n "${!varname}" ] ; then
+ eval export ${facet}failover_dev=${!varname}
+ else
+ eval export ${facet}failover_dev=$device
+ fi
}
init_facets_vars () {
local DEVNAME
- for num in `seq $MDSCOUNT`; do
- DEVNAME=`mdsdevname $num`
- init_facet_vars mds$num $DEVNAME $MDS_MOUNT_OPTS
- done
+ if ! remote_mds_nodsh; then
+ for num in `seq $MDSCOUNT`; do
+ DEVNAME=`mdsdevname $num`
+ init_facet_vars mds$num $DEVNAME $MDS_MOUNT_OPTS
+ done
+ fi
+
+ remote_ost_nodsh && return
for num in `seq $OSTCOUNT`; do
DEVNAME=`ostdevname $num`
done
}
-init_versions_vars () {
- export MDSVER=$(do_facet $SINGLEMDS "lctl get_param version" | cut -d. -f1,2)
- export OSTVER=$(do_facet ost1 "lctl get_param version" | cut -d. -f1,2)
- export CLIVER=$(lctl get_param version | cut -d. -f 1,2)
+osc_ensure_active () {
+ local facet=$1
+ local type=$2
+ local timeout=$3
+ local period=0
+
+ while [ $period -lt $timeout ]; do
+ count=$(do_facet $facet "lctl dl | grep '${FSNAME}-OST.*-osc-${type}' | 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 $SINGLEMDS "lctl get_param version" | cut -d. -f1,2)
+ export OSTVER=$(do_facet ost1 "lctl get_param version" | cut -d. -f1,2)
+ export CLIVER=$(lctl get_param version | cut -d. -f 1,2)
+ fi
+
+ remote_mds_nodsh ||
+ TIMEOUT=$(do_facet $SINGLEMDS "lctl get_param -n timeout")
+
+ log "Using TIMEOUT=$TIMEOUT"
+
+ osc_ensure_active $SINGLEMDS M $TIMEOUT
+ osc_ensure_active client c $TIMEOUT
+
+ if [ x"$(som_check)" = x"enabled" ]; then
+ ENABLE_QUOTA=""
+ echo "disable quota temporary when SOM enabled"
+ fi
+ if [ $QUOTA_AUTO -ne 0 ]; then
+ if [ "$ENABLE_QUOTA" ]; then
+ echo "enable quota as required"
+ setup_quota $MOUNT || return 2
+ else
+ echo "disable quota as required"
+ $LFS quotaoff -ug $MOUNT > /dev/null 2>&1
+ fi
+ fi
+
+ return 0
+}
+
+nfs_client_mode () {
+ if [ "$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 () {
+check_config_client () {
local mntpt=$1
+
+ local mounted=$(mount | grep " $mntpt ")
+ if [ "$CLIENTONLY" ]; then
+ # bug 18021
+ # CLIENTONLY should not depend on *_HOST settings
+ local mgc=$($LCTL device_list | awk '/MGC/ {print $4}')
+ # in theory someone could create a new,
+ # client-only config file that assumed lustre was already
+ # configured and didn't set the MGSNID. If MGSNID is not set,
+ # then we should use the mgs nid currently being used
+ # as the default value. bug 18021
+ [[ x$MGSNID = x ]] &&
+ MGSNID=${mgc//MGC/}
+
+ if [[ x$mgc != xMGC$MGSNID ]]; then
+ if [ "$mgs_HOST" ]; then
+ local mgc_ip=$(ping -q -c1 -w1 $mgs_HOST | grep PING | awk '{print $3}' | sed -e "s/(//g" -e "s/)//g")
+# [[ x$mgc = xMGC$mgc_ip@$NETTYPE ]] ||
+# error_exit "MGSNID=$MGSNID, mounted: $mounted, MGC : $mgc"
+ fi
+ fi
+ return 0
+ fi
+
local myMGS_host=$mgs_HOST
if [ "$NETTYPE" = "ptl" ]; then
myMGS_host=$(h2ptl $mgs_HOST | sed -e s/@ptl//)
mgshost=$(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=$NETTYPE
Please use correct config or set mds_HOST correctly!"
fi
+
+}
+
+check_config_clients () {
+ local clients=${CLIENTS:-$HOSTNAME}
+ local mntpt=$1
+
+ nfs_client_mode && return
+
+ do_rpc_nodes $clients check_config_client $mntpt
+
+ sanity_mount_check ||
+ error "environments are insane!"
+}
+
+check_timeout () {
+ local mdstimeout=$(do_facet $SINGLEMDS "lctl get_param -n timeout")
+ local cltimeout=$(lctl get_param -n timeout)
+ if [ $mdstimeout -ne $TIMEOUT ] || [ $mdstimeout -ne $cltimeout ]; then
+ error "timeouts are wrong! mds: $mdstimeout, client: $cltimeout, TIMEOUT=$TIMEOUT"
+ return 1
+ fi
+}
+
+is_mounted () {
+ local mntpt=$1
+ local mounted=$(mounted_lustre_filesystems)
+
+ echo $mounted' ' | grep -w -q $mntpt' '
}
check_and_setup_lustre() {
+ nfs_client_mode && return
+
local MOUNTED=$(mounted_lustre_filesystems)
- if [ -z "$MOUNTED" ] || ! $(echo $MOUNTED | grep -w -q $MOUNT); then
+
+ local do_check=true
+ # 1.
+ # both MOUNT and MOUNT2 are not mounted
+ if ! is_mounted $MOUNT && ! is_mounted $MOUNT2; then
[ "$REFORMAT" ] && formatall
+ # setupall mounts both MOUNT and MOUNT2 (if MOUNT_2 is set)
setupall
- MOUNTED=$(mounted_lustre_filesystems | head -1)
- [ -z "$MOUNTED" ] && error "NAME=$NAME not mounted"
+ is_mounted $MOUNT || error "NAME=$NAME not mounted"
export I_MOUNTED=yes
- else
- check_config $MOUNT
+ do_check=false
+ # 2.
+ # MOUNT2 is mounted
+ elif is_mounted $MOUNT2; then
+ # 3.
+ # MOUNT2 is mounted, while MOUNT_2 is not set
+ if ! [ "$MOUNT_2" ]; then
+ cleanup_mount $MOUNT2
+ export I_UMOUNTED2=yes
+
+ # 4.
+ # MOUNT2 is mounted, MOUNT_2 is set
+ else
+ # FIXME: what to do if check_config failed?
+ # i.e. if:
+ # 1) remote client has mounted other Lustre fs ?
+ # 2) it has insane env ?
+ # let's try umount MOUNT2 on all clients and mount it again:
+ if ! check_config_clients $MOUNT2; then
+ cleanup_mount $MOUNT2
+ restore_mount $MOUNT2
+ export I_MOUNTED2=yes
+ fi
+ fi
+
+ # 5.
+ # MOUNT is mounted MOUNT2 is not mounted
+ elif [ "$MOUNT_2" ]; then
+ restore_mount $MOUNT2
+ export I_MOUNTED2=yes
+ fi
+
+ if $do_check; then
+ # FIXME: what to do if check_config failed?
+ # i.e. if:
+ # 1) remote client has mounted other Lustre fs?
+ # 2) lustre is mounted on remote_clients atall ?
+ check_config_clients $MOUNT
init_facets_vars
- init_versions_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 [ "$ONLY" == "setup" ]; then
exit 0
fi
}
+restore_mount () {
+ local clients=${CLIENTS:-$HOSTNAME}
+ local mntpt=$1
+
+ zconf_mount_clients $clients $mntpt
+}
+
+cleanup_mount () {
+ local clients=${CLIENTS:-$HOSTNAME}
+ local mntpt=$1
+
+ zconf_umount_clients $clients $mntpt
+}
+
cleanup_and_setup_lustre() {
if [ "$ONLY" == "cleanup" -o "`mount | grep $MOUNT`" ]; then
lctl set_param debug=0 || true
}
check_and_cleanup_lustre() {
- if [ "`mount | grep $MOUNT`" ]; then
+ if is_mounted $MOUNT; then
[ -n "$DIR" ] && rm -rf $DIR/[Rdfs][0-9]*
+ [ "$ENABLE_QUOTA" ] && restore_quota_type || true
+ fi
+
+ if [ "$I_UMOUNTED2" = "yes" ]; then
+ restore_mount $MOUNT2 || error "restore $MOUNT2 failed"
+ fi
+
+ if [ "$I_MOUNTED2" = "yes" ]; then
+ cleanup_mount $MOUNT2
fi
+
if [ "$I_MOUNTED" = "yes" ]; then
cleanupall -f || error "cleanup failed"
+ unset I_MOUNTED
fi
- unset I_MOUNTED
}
#######
local WAIT=0
local MAX=$2
while [ $NETWORK -eq 0 ]; do
- ping -c 1 -w 3 $1 > /dev/null
- if [ $? -eq 0 ]; then
+ if ping -c 1 -w 3 $1 > /dev/null; then
NETWORK=1
else
WAIT=$((WAIT + 5))
echo "$*" | tr -s " " "\n" | sort -b -u | tr "\n" " " | sed 's/ \([^$]\)/,\1/g'
}
-# list is comma separated list
-exclude_item_from_list () {
+# list, excluded are the comma separated lists
+exclude_items_from_list () {
local list=$1
local excluded=$2
+ local item
list=${list//,/ }
- list=$(echo " $list " | sed -re "s/\s+$excluded\s+/ /g")
+ for item in ${excluded//,/ }; do
+ list=$(echo " $list " | sed -re "s/\s+$item\s+/ /g")
+ done
echo $(comma_list $list)
}
+# list, expand are the comma separated lists
+expand_list () {
+ local list=${1//,/ }
+ local expand=${2//,/ }
+ local expanded=
+
+ expanded=$(for i in $list $expand; do echo $i; done | sort -u)
+ echo $(comma_list $expanded)
+}
+
+testslist_filter () {
+ local script=$LUSTRE/tests/${TESTSUITE}.sh
+
+ [ -f $script ] || return 0
+
+ local start_at=$START_AT
+ local stop_at=$STOP_AT
+
+ local var=${TESTSUITE//-/_}_START_AT
+ [ x"${!var}" != x ] && start_at=${!var}
+ var=${TESTSUITE//-/_}_STOP_AT
+ [ x"${!var}" != x ] && stop_at=${!var}
+
+ sed -n 's/^test_\([^ (]*\).*/\1/p' $script | \
+ awk ' BEGIN { if ("'${start_at:-0}'" != 0) flag = 1 }
+ /^'${start_at}'$/ {flag = 0}
+ {if (flag == 1) print $0}
+ /^'${stop_at}'$/ { flag = 1 }'
+}
+
absolute_path() {
(cd `dirname $1`; echo $PWD/`basename $1`)
}
-##################################
-# Adaptive Timeouts funcs
+get_facets () {
+ local name=$(echo $1 | tr "[:upper:]" "[:lower:]")
+ local type=$(echo $1 | tr "[:lower:]" "[:upper:]")
-at_is_valid() {
- if [ -z "$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
+ local list=""
+ local count=${type}COUNT
+ for ((i=1; i<=${!count}; i++)) do
+ list="$list ${name}$i"
+ done
+ echo $(comma_list $list)
}
-at_is_enabled() {
- at_is_valid || error "invalid call"
+##################################
+# Adaptive Timeouts funcs
+at_is_enabled() {
# only check mds, we assume at_max is the same on all nodes
- local at_max=$(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
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
}
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
}
}
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() {
return 1
fi
done
+ if [[ $MDSDEV1 != $MGSDEV ]]; then
+ stop mgs
+ fi
+
return 0
}
}
debugrestore() {
- [ -n "$DEBUGSAVE" ] && lctl set_param debug="${DEBUGSAVE}"
+ [ -n "$DEBUGSAVE" ] && \
+ do_nodes $(comma_list $(nodes_list)) "$LCTL set_param debug=\\\"${DEBUGSAVE}\\\";"
DEBUGSAVE=""
}
+debug_size_save() {
+ DEBUG_SIZE_SAVED="$(lctl get_param -n debug_mb)"
+}
+
+debug_size_restore() {
+ [ -n "$DEBUG_SIZE_SAVED" ] && \
+ do_nodes $(comma_list $(nodes_list)) "$LCTL set_param debug_mb=$DEBUG_SIZE_SAVED"
+ DEBUG_SIZE_SAVED=""
+}
+
+start_full_debug_logging() {
+ debugsave
+ debug_size_save
+
+ local FULLDEBUG=-1
+ local DEBUG_SIZE=150
+
+ do_nodes $(comma_list $(nodes_list)) "$LCTL set_param debug_mb=$DEBUG_SIZE"
+ do_nodes $(comma_list $(nodes_list)) "$LCTL set_param debug=$FULLDEBUG;"
+}
+
+stop_full_debug_logging() {
+ debug_size_restore
+ debugrestore
+}
+
##################################
# Test interface
##################################
error_noexit() {
local TYPE=${TYPE:-"FAIL"}
local ERRLOG
- lctl set_param fail_loc=0 2>/dev/null || true
+
+ local dump=true
+ # do not dump logs if $1=false
+ if [ "x$1" = "xfalse" ]; then
+ shift
+ dump=false
+ fi
+
log " ${TESTSUITE} ${TESTNAME}: @@@@@@ ${TYPE}: $@ "
- ERRLOG=$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
+
+ if $dump; then
+ ERRLOG=$TMP/lustre_${TESTSUITE}_${TESTNAME}.$(date +%s)
+ echo "Dumping lctl log to $ERRLOG"
+ # We need to dump the logs on all nodes
+ do_nodes $(comma_list $(nodes_list)) $NODE $LCTL dk $ERRLOG
+ fi
debugrestore
[ "$TESTSUITELOG" ] && echo "$0: ${TYPE}: $TESTNAME $@" >> $TESTSUITELOG
TEST_FAILED=true
error() {
error_noexit "$@"
- $FAIL_ON_ERROR && exit 1 || true
+ if $FAIL_ON_ERROR; then
+ reset_fail_loc
+ exit 1
+ fi
}
error_exit() {
error_noexit "$@"
}
+skip_env () {
+ $FAIL_ON_SKIP_ENV && error false $@ || skip $@
+}
+
skip () {
- log " SKIP: ${TESTSUITE} ${TESTNAME} $@"
- [ "$TESTSUITELOG" ] && \
- echo "${TESTSUITE}: SKIP: $TESTNAME $@" >> $TESTSUITELOG || true
+ echo
+ log " SKIP: ${TESTSUITE} ${TESTNAME} $@"
+ [ "$TESTSUITELOG" ] && \
+ echo "${TESTSUITE}: SKIP: $TESTNAME $@" >> $TESTSUITELOG || true
}
build_test_filter() {
+ EXCEPT="$EXCEPT $(testslist_filter)"
+
[ "$ONLY" ] && log "only running test `echo $ONLY`"
for O in $ONLY; do
eval ONLY_${O}=true
done
}
-_basetest() {
- echo $*
-}
-
basetest() {
- IFS=abcdefghijklmnopqrstuvwxyz _basetest $1
+ if [[ $1 = [a-z]* ]]; then
+ echo $1
+ else
+ echo ${1%%[a-z]*}
+ fi
}
# print a newline if the last test was skipped
log() {
echo "$*"
- lsmod | grep lnet > /dev/null || load_modules
+ module_loaded lnet || load_modules
local MSG="$*"
- # Get rif of '
+ # Get rid of '
MSG=${MSG//\'/\\\'}
MSG=${MSG//\(/\\\(}
MSG=${MSG//\)/\\\)}
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() {
}
reset_fail_loc () {
- local myNODES=$(nodes_list)
- local NODE
-
- for NODE in $myNODES; do
- do_node $NODE "lctl set_param fail_loc=0 2>/dev/null || true"
- done
+ echo -n "Resetting fail_loc on all nodes..."
+ do_nodes $(comma_list $(nodes_list)) "lctl set_param -n fail_loc=0 2>/dev/null || true"
+ echo done.
}
run_one() {
local SAVE_UMASK=`umask`
umask 0022
- BEFORE=`date +%s`
- log "== test $testnum: $message ============ `date +%H:%M:%S` ($BEFORE)"
+ local BEFORE=`date +%s`
+ echo
+ log "== test $testnum: $message == `date +%H:%M:%S` ($BEFORE)"
#check_mds
export TESTNAME=test_$testnum
TEST_FAILED=false
remote_mds_nodsh()
{
+ [ "$CLIENTONLY" ] && return 0 || true
remote_mds && [ "$PDSH" = "no_dsh" -o -z "$PDSH" -o -z "$mds_HOST" ]
}
remote_ost_nodsh()
{
+ [ "$CLIENTONLY" ] && return 0 || true
remote_ost && [ "$PDSH" = "no_dsh" -o -z "$PDSH" -o -z "$ost_HOST" ]
}
+remote_mgs_nodsh()
+{
+ local MGS
+ MGS=$(facet_host mgs)
+ remote_node $MGS && [ "$PDSH" = "no_dsh" -o -z "$PDSH" -o -z "$ost_HOST" ]
+}
+
mdts_nodes () {
local MDSNODES
local NODES_sort
rnodes=${rnodes//,/ }
- local nodes=($rnodes)
+ local -a nodes=($rnodes)
local num=${#nodes[@]}
- local i=$((RANDOM * num / 65536))
+ local i=$((RANDOM * num * 2 / 65536))
echo ${nodes[i]}
}
+client_only () {
+ [ "$CLIENTONLY" ] || [ "$CLIENTMODSONLY" = yes ]
+}
+
is_patchless ()
{
lctl get_param version | grep -q patchless
add user $myRUNAS_UID:$myRUNAS_GID on these nodes."
}
+# obtain the UID/GID for MPI_USER
+get_mpiuser_id() {
+ local mpi_user=$1
+
+ MPI_USER_UID=$(do_facet client "getent passwd $mpi_user | cut -d: -f3;
+exit \\\${PIPESTATUS[0]}") || error_exit "failed to get the UID for $mpi_user"
+
+ MPI_USER_GID=$(do_facet client "getent passwd $mpi_user | cut -d: -f4;
+exit \\\${PIPESTATUS[0]}") || error_exit "failed to get the GID for $mpi_user"
+}
+
+# obtain and cache Kerberos ticket-granting ticket
+refresh_krb5_tgt() {
+ local myRUNAS_UID=$1
+ local myRUNAS_GID=$2
+ shift 2
+ local myRUNAS=$@
+ if [ -z "$myRUNAS" ]; then
+ error_exit "myRUNAS command must be specified for refresh_krb5_tgt"
+ fi
+
+ CLIENTS=${CLIENTS:-$HOSTNAME}
+ do_nodes $CLIENTS "set -x
+if ! $myRUNAS krb5_login.sh; then
+ echo "Failed to refresh Krb5 TGT for UID/GID $myRUNAS_UID/$myRUNAS_GID."
+ exit 1
+fi"
+}
+
# Run multiop in the background, but wait for it to print
# "PAUSING" to its stdout before returning from this function.
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
}
calc_osc_kbytes () {
+ df $MOUNT > /dev/null
$LCTL get_param -n osc.*[oO][sS][cC][-_][0-9a-f]*.$1 | calc_sum
}
# generate a stream of formatted strings (<node> <param name>=<param value>)
save_lustre_params() {
local s
- do_node $1 "lctl get_param $2" | while read s; do echo "$1 $s"; done
+ do_nodes --verbose $1 "lctl get_param $2 | while read s; do echo \\\$s; done"
}
# restore lustre parameters from input stream, produces by save_lustre_params
local name
local val
while IFS=" =" read node name val; do
- do_node $node "lctl set_param -n $name $val"
+ do_node ${node//:/} "lctl set_param -n $name $val"
done
}
-check_catastrophe () {
+check_catastrophe() {
local rnodes=${1:-$(comma_list $(remote_nodes_list))}
+ local C=$CATASTROPHE
+ [ -f $C ] && [ $(cat $C) -ne 0 ] && return 1
- [ -f $CATASTROPHE ] && [ $(cat $CATASTROPHE) -ne 0 ] && return 1
if [ $rnodes ]; then
- do_nodes $rnodes "set -x; [ -f $CATASTROPHE ] && { [ \`cat $CATASTROPHE\` -eq 0 ] || false; } || true"
+ do_nodes $rnodes "rc=\\\$([ -f $C ] && echo \\\$(< $C) || echo 0);
+if [ \\\$rc -ne 0 ]; then echo \\\$(hostname): \\\$rc; fi
+exit \\\$rc;"
fi
}
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
}
# CMD: determine mds index where directory inode presents
get_mds_dir () {
local dir=$1
- local file=$dir/$tfile
+ local file=$dir/f0.get_mds_dir_tmpfile
+ mkdir -p $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
+ openfile -f O_CREAT:O_LOV_DELAY_CREATE -m 0644 $file > /dev/null
sleep 1
iused=$(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
if [ ${oldused[$i]} -lt ${newused[$i]} ]; then
echo $(( i + 1 ))
- rm -f $dir/$tfile
+ rm -f $file
return 0
fi
done
error "mdt-s : inodes count OLD ${oldused[@]} NEW ${newused[@]}"
}
-mpi_run () {
- local mpirun="$MPIRUN $MPIRUN_OPTIONS"
- local command="$mpirun $@"
+mdsrate_cleanup () {
+ if [ -d $4 ]; then
+ mpi_run -np $1 -machinefile $2 ${MDSRATE} --unlink --nfiles $3 --dir $4 --filefmt $5 $6
+ rmdir $4
+ fi
+}
+
+delayed_recovery_enabled () {
+ local var=${SINGLEMDS}_svc
+ do_facet $SINGLEMDS lctl get_param -n mdd.${!var}.stale_export_age > /dev/null 2>&1
+}
+
+########################
+
+convert_facet2label() {
+ local facet=$1
+
+ if [ x$facet = xost ]; then
+ facet=ost1
+ fi
+
+ local varsvc=${facet}_svc
- if [ "$MPI_USER" != root -a $mpirun ]; then
- echo "+ chmod 0777 $MOUNT"
- chmod 0777 $MOUNT
- command="su $MPI_USER sh -c \"$command \""
+ if [ -n ${!varsvc} ]; then
+ echo ${!varsvc}
+ else
+ error "No lablel for $facet!"
fi
+}
- ls -ald $MOUNT
- echo "+ $command"
- eval $command
+get_clientosc_proc_path() {
+ local ost=$1
+
+ echo "{$1}-osc-*"
}
-mdsrate_cleanup () {
- mpi_run -np $1 -machinefile $2 ${MDSRATE} --unlink --nfiles $3 --dir $4 --filefmt $5
+get_lustre_version () {
+ local node=${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 facet=$1
+ local ost=$2
+ local label=$(convert_facet2label $ost)
+
+ if [ "$facet" == "mds" ]; then
+ get_mdtosc_proc_path $label
+ return 0
+ fi
+
+ get_clientosc_proc_path $label
+ return 0
+}
+
+wait_import_state () {
+ local expected=$1
+ local CONN_PROC=$2
+ local CONN_STATE
+ local i=0
+
+ CONN_STATE=$($LCTL get_param -n $CONN_PROC 2>/dev/null | cut -f2)
+ while [ "${CONN_STATE}" != "${expected}" ]; do
+ if [ "${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 $CONN_PROC into ${expected} state" && return 1
+ sleep 1
+ CONN_STATE=$($LCTL get_param -n $CONN_PROC 2>/dev/null | cut -f2)
+ i=$(($i + 1))
+ done
+
+ log "$CONN_PROC now in ${CONN_STATE} state"
+ return 0
+}
+
+wait_osc_import_state() {
+ local facet=$1
+ local ost_facet=$2
+ local expected=$3
+ local ost=$(get_osc_import_name $facet $ost_facet)
+ local CONN_PROC
+ local CONN_STATE
+ local i=0
+
+ CONN_PROC="osc.${ost}.ost_server_uuid"
+ CONN_STATE=$(do_facet $facet lctl get_param -n $CONN_PROC 2>/dev/null | cut -f2)
+ while [ "${CONN_STATE}" != "${expected}" ]; do
+ if [ "${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 ${expected} state" && return 1
+ sleep 1
+ CONN_STATE=$(do_facet $facet lctl get_param -n $CONN_PROC 2>/dev/null | cut -f2)
+ i=$(($i + 1))
+ done
+
+ log "${ost_facet} now in ${CONN_STATE} state"
+ return 0
+}
+
+get_clientmdc_proc_path() {
+ echo "${1}-mdc-*"
+}
+
+do_rpc_nodes () {
+ local list=$1
+ shift
+
+ do_nodes --verbose $list "PATH=$LUSTRE/tests/:$PATH sh rpc.sh $@ "
+}
+
+wait_clients_import_state () {
+ local list=$1
+ local facet=$2
+ local expected=$3
+ shift
+
+ local label=$(convert_facet2label $facet)
+ local proc_path
+ case $facet in
+ ost* ) proc_path="osc.$(get_clientosc_proc_path $label).ost_server_uuid" ;;
+ mds* ) proc_path="mdc.$(get_clientmdc_proc_path $label).mds_server_uuid" ;;
+ *) error "unknown facet!" ;;
+ esac
+
+ if ! do_rpc_nodes $list wait_import_state $expected $proc_path; then
+ error "import is not in ${expected} state"
+ return 1
+ fi
+}
+
+oos_full() {
+ local -a AVAILA
+ local -a GRANTA
+ local OSCFULL=1
+ AVAILA=($(do_nodes $(comma_list $(osts_nodes)) \
+ $LCTL get_param obdfilter.*.kbytesavail))
+ GRANTA=($(do_nodes $(comma_list $(osts_nodes)) \
+ $LCTL get_param -n obdfilter.*.tot_granted))
+ for ((i=0; i<${#AVAILA[@]}; i++)); do
+ local -a AVAIL1=(${AVAILA[$i]//=/ })
+ GRANT=$((${GRANTA[$i]}/1024))
+ echo -n $(echo ${AVAIL1[0]} | cut -d"." -f2) avl=${AVAIL1[1]} grnt=$GRANT diff=$((AVAIL1[1] - GRANT))
+ [ $((AVAIL1[1] - GRANT)) -lt 400 ] && OSCFULL=0 && echo " FULL" || echo
+ done
+ return $OSCFULL
+}
+
+pool_list () {
+ do_facet mgs lctl pool_list $1
+}
+
+create_pool() {
+ local fsname=${1%%.*}
+ local poolname=${1##$fsname.}
+
+ do_facet mgs lctl pool_new $1
+ local RC=$?
+ # get param should return err unless pool is created
+ [[ $RC -ne 0 ]] && return $RC
+
+ wait_update $HOSTNAME "lctl get_param -n lov.$fsname-*.pools.$poolname \
+ 2>/dev/null || echo foo" "" || RC=1
+ if [[ $RC -eq 0 ]]; then
+ add_pool_to_list $1
+ else
+ error "pool_new failed $1"
+ fi
+ return $RC
+}
+
+add_pool_to_list () {
+ local fsname=${1%%.*}
+ local poolname=${1##$fsname.}
+
+ local listvar=${fsname}_CREATED_POOLS
+ eval export ${listvar}=$(expand_list ${!listvar} $poolname)
+}
+
+remove_pool_from_list () {
+ local fsname=${1%%.*}
+ local poolname=${1##$fsname.}
+
+ local listvar=${fsname}_CREATED_POOLS
+ eval export ${listvar}=$(exclude_items_from_list ${!listvar} $poolname)
+}
+
+destroy_pool_int() {
+ local ost
+ local OSTS=$(do_facet $SINGLEMDS lctl pool_list $1 | \
+ awk '$1 !~ /^Pool:/ {print $1}')
+ for ost in $OSTS; do
+ do_facet mgs lctl pool_remove $1 $ost
+ done
+ do_facet mgs lctl pool_destroy $1
+}
+
+# <fsname>.<poolname> or <poolname>
+destroy_pool() {
+ local fsname=${1%%.*}
+ local poolname=${1##$fsname.}
+
+ [[ x$fsname = x$poolname ]] && fsname=$FSNAME
+
+ local RC
+
+ pool_list $fsname.$poolname || return $?
+
+ destroy_pool_int $fsname.$poolname
+ RC=$?
+ [[ $RC -ne 0 ]] && return $RC
+
+ wait_update $HOSTNAME "lctl get_param -n lov.$fsname-*.pools.$poolname \
+ 2>/dev/null || echo foo" "foo" || RC=1
+
+ if [[ $RC -eq 0 ]]; then
+ remove_pool_from_list $fsname.$poolname
+ else
+ error "destroy pool failed $1"
+ fi
+ return $RC
+}
+
+destroy_pools () {
+ local fsname=${1:-$FSNAME}
+ local poolname
+ local listvar=${fsname}_CREATED_POOLS
+
+ pool_list $fsname
+
+ [ x${!listvar} = x ] && return 0
+
+ echo destroy the created pools: ${!listvar}
+ for poolname in ${!listvar//,/ }; do
+ destroy_pool $fsname.$poolname
+ done
+}
+
+cleanup_pools () {
+ local fsname=${1:-$FSNAME}
+ trap 0
+ destroy_pools $fsname
+}
+
+gather_logs () {
+ local list=$1
+
+ local ts=$(date +%s)
+
+ # bug 20237, comment 11
+ # It would also be useful to provide the option
+ # of writing the file to an NFS directory so it doesn't need to be copied.
+ local tmp=$TMP
+ local docp=true
+ [ -d "$SHARED_DIR_LOGS" ] && tmp=$SHARED_DIR_LOGS && docp=false
+
+ # dump lustre logs, dmesg
+ do_nodes $list "log=$tmp/\\\$(hostname)-debug-$ts.log ;
+lctl dk \\\$log >/dev/null;
+log=$tmp/\\\$(hostname)-dmesg-$ts.log;
+dmesg > \\\$log; "
+
+ # FIXME: does it make sense to collect the logs for $ts only, but all
+ # TESTSUITE logs?
+ # rsync $TMP/*${TESTSUITE}* to gather the logs dumped by error fn
+ local logs=$TMP/'*'${TESTSUITE}'*'
+ if $docp; then
+ logs=$logs' '$tmp/'*'$ts'*'
+ fi
+ for node in ${list//,/ }; do
+ rsync -az $node:"$logs" $TMP
+ done
+
+ local archive=$TMP/${TESTSUITE}-$ts.tar.bz2
+ tar -jcf $archive $tmp/*$ts* $TMP/*${TESTSUITE}*
+
+ echo $archive
+}
+
+cleanup_logs () {
+ local list=${1:-$(comma_list $(nodes_list))}
+
+ [ -n ${TESTSUITE} ] && do_nodes $list "rm -f $TMP/*${TESTSUITE}*" || true
+}
+
+do_ls () {
+ local mntpt_root=$1
+ local num_mntpts=$2
+ local dir=$3
+ local i
+ local cmd
+ local pids
+ local rc=0
+
+ for i in $(seq 0 $num_mntpts); do
+ cmd="ls -laf ${mntpt_root}$i/$dir"
+ echo + $cmd;
+ $cmd > /dev/null &
+ pids="$pids $!"
+ done
+ echo pids=$pids
+ for pid in $pids; do
+ wait $pid || rc=$?
+ done
+
+ return $rc
}