#!/bin/bash
# vim:expandtab:shiftwidth=4:softtabstop=4:tabstop=4:
-trap 'print_summary && echo "test-framework exiting on error"' ERR
+trap 'print_summary && touch $TF_FAIL && \
+ echo "test-framework exiting on error"' ERR
set -e
#set -x
-
export REFORMAT=${REFORMAT:-""}
export WRITECONF=${WRITECONF:-""}
export VERBOSE=false
-export GMNALNID=${GMNALNID:-/usr/sbin/gmlndnid}
export CATASTROPHE=${CATASTROPHE:-/proc/sys/lnet/catastrophe}
export GSS=false
export GSS_KRB5=false
export E2LABEL=${E2LABEL:-e2label}
export DUMPE2FS=${DUMPE2FS:-dumpe2fs}
export E2FSCK=${E2FSCK:-e2fsck}
+ export LFSCK_BIN=${LFSCK_BIN:-lfsck}
+ export LFSCK_ALWAYS=${LFSCK_ALWAYS:-"no"} # check filesystem after each test suit
+ export SKIP_LFSCK=${SKIP_LFSCK:-"yes"} # bug 13698, change to "no" when fixed
+ export SHARED_DIRECTORY=${SHARED_DIRECTORY:-"/tmp"}
+ export FSCK_MAX_ERR=4 # File system errors left uncorrected
+ if [ "$SKIP_LFSCK" == "no" ]; then
+ if [ ! -x `which $LFSCK_BIN` ]; then
+ log "$($E2FSCK -V)"
+ error_exit "$E2FSCK does not support lfsck"
+ fi
+ export MDSDB=${MDSDB:-$SHARED_DIRECTORY/mdsdb}
+ export OSTDB=${OSTDB:-$SHARED_DIRECTORY/ostdb}
+ export MDSDB_OPT="--mdsdb $MDSDB"
+ export OSTDB_OPT="--ostdb $OSTDB-\$ostidx"
+ fi
#[ -d /r ] && export ROOT=${ROOT:-/r}
export TMP=${TMP:-$ROOT/tmp}
export TESTSUITELOG=${TMP}/${TESTSUITE}.log
export PATH=$PATH:$LUSTRE/utils/gss
fi
if ! echo $PATH | grep -q $LUSTRE/tests; then
- export PATH=$PATH:$LUSTRE/tests
+ export PATH=$LUSTRE/tests:$PATH
fi
export LST=${LST:-"$LUSTRE/../lnet/utils/lst"}
[ ! -f "$LST" ] && export LST=$(which lst)
+ export SGPDDSURVEY=${SGPDDSURVEY:-$(which sgpdd-survey)}
export MDSRATE=${MDSRATE:-"$LUSTRE/tests/mpi/mdsrate"}
[ ! -f "$MDSRATE" ] && export MDSRATE=$(which mdsrate 2> /dev/null)
if ! echo $PATH | grep -q $LUSTRE/tests/racer; then
export L_GETIDENTITY=NONE
fi
fi
+ export LL_DECODE_FILTER_FID=${LL_DECODE_FILTER_FID:-"$LUSTRE/utils/ll_decode_filter_fid"}
+ [ ! -f "$LL_DECODE_FILTER_FID" ] && export LL_DECODE_FILTER_FID=$(which ll_decode_filter_fid)
export MKFS=${MKFS:-"$LUSTRE/utils/mkfs.lustre"}
[ ! -f "$MKFS" ] && export MKFS=$(which mkfs.lustre)
export TUNEFS=${TUNEFS:-"$LUSTRE/utils/tunefs.lustre"}
export RPWD=${RPWD:-$PWD}
export I_MOUNTED=${I_MOUNTED:-"no"}
if [ ! -f /lib/modules/$(uname -r)/kernel/fs/lustre/mds.ko -a \
+ ! -f /lib/modules/$(uname -r)/updates/kernel/fs/lustre/mds.ko -a \
! -f `dirname $0`/../mds/mds.ko ]; then
export CLIENTMODSONLY=yes
fi
shift
BASE=`basename $module $EXT`
+ module_loaded ${BASE} && return
+
# If no module arguments were passed, get them from $MODOPTS_<MODULE>, else from
# modprobe.conf
if [ $# -eq 0 ]; then
# Nothing in $MODOPTS_<MODULE>; try modprobe.conf
set -- $(grep "^options\\s*\<${module}\>" $MODPROBECONF)
# Get rid of "options $module"
- (($# > 0)) && shift 2
+ (($# > 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
[ $# -gt 0 ] && echo "${module} options: '$*'"
- module_loaded ${BASE} && return
-
# Note that insmod will ignore anything in modprobe.conf, which is why we're
# passing options on the command-line.
if [ "$BASE" == "lnet_selftest" ] && \
echo "Using modprobe to load modules"
return 0
fi
- if [ "$HAVE_MODULES" = true ]; then
- # we already loaded
- echo "Modules already loaded"
- return 0
- fi
- HAVE_MODULES=true
echo Loading modules from $LUSTRE
load_module ../libcfs/libcfs/libcfs
load_module obdclass/obdclass
load_module ptlrpc/ptlrpc
load_module ptlrpc/gss/ptlrpc_gss
- [ "$USE_QUOTA" = "yes" -a "$LQUOTA" != "no" ] && load_module quota/lquota
+ [ "$USE_QUOTA" = "yes" -a "$LQUOTA" != "no" ] && load_module quota/lquota $LQUOTAOPTS
load_module fld/fld
load_module fid/fid
load_module lmv/lmv
load_module mdt/mdt
load_module lvfs/fsfilt_$FSTYPE
load_module cmm/cmm
- load_module osd/osd
+ load_module osd-ldiskfs/osd_ldiskfs
load_module ost/ost
load_module obdfilter/obdfilter
fi
fi
fi
- HAVE_MODULES=false
-
check_mem_leak || return 254
echo "modules unloaded."
fi"
}
-shudown_node_hard () {
+shutdown_node () {
+ local node=$1
+ echo + $POWER_DOWN $node
+ $POWER_DOWN $node
+}
+
+shutdown_node_hard () {
local host=$1
local attempts=3
for i in $(seq $attempts) ; do
- $POWER_DOWN $host
+ shutdown_node $host
sleep 1
ping -w 3 -c 1 $host > /dev/null 2>&1 || return 0
echo "waiting for $host to fail attempts=$attempts"
local attempts=3
if [ "$FAILURE_MODE" = HARD ]; then
- shudown_node_hard $client
+ shutdown_node_hard $client
else
zconf_umount_clients $client $mnt -f
fi
}
+facets_on_host () {
+ local host=$1
+ local facets="$(get_facets OST),$(get_facets MDS)"
+ local affected
+
+ combined_mgs_mds || facets="$facets,mgs"
+
+ for facet in ${facets//,/ }; do
+ if [ $(facet_active_host $facet) == $host ]; then
+ affected="$affected $facet"
+ fi
+ done
+
+ echo $(comma_list $affected)
+}
+
shutdown_facet() {
local facet=$1
+
if [ "$FAILURE_MODE" = HARD ]; then
- shudown_node_hard $(facet_active_host $facet)
- elif [ "$FAILURE_MODE" = SOFT ]; then
+ shutdown_node_hard $(facet_active_host $facet)
+ else
stop $facet
fi
}
+reboot_node() {
+ local node=$1
+ echo + $POWER_UP $node
+ $POWER_UP $node
+}
+
+remount_facet() {
+ local facet=$1
+
+ stop $facet
+ mount_facet $facet
+}
+
reboot_facet() {
- facet=$1
+ local facet=$1
if [ "$FAILURE_MODE" = HARD ]; then
- $POWER_UP `facet_active_host $facet`
+ reboot_node $(facet_active_host $facet)
else
sleep 10
fi
boot_node() {
local node=$1
if [ "$FAILURE_MODE" = HARD ]; then
- $POWER_UP $node
+ reboot_node $node
wait_for_host $node
fi
}
}
wait_for_host() {
- local host=$1
- check_network "$host" 900
- while ! do_node $host hostname > /dev/null; do sleep 5; done
+ local hostlist=$1
+
+ # we can use "for" here because we are waiting the slowest
+ for host in ${hostlist//,/ }; do
+ check_network "$host" 900
+ done
+ while ! do_nodes $hostlist hostname > /dev/null; do sleep 5; done
}
-wait_for() {
- local facet=$1
- local host=`facet_active_host $facet`
- wait_for_host $host
+wait_for_facet() {
+ local facetlist=$1
+ local hostlist
+
+ for facet in ${facetlist//,/ }; do
+ hostlist=$(expand_list $hostlist $(facet_active_host $facet))
+ done
+ wait_for_host $hostlist
}
-wait_recovery_complete () {
- local facet=$1
+_wait_recovery_complete () {
+ local param=$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:-$(( TIMEOUT * 4 ))}
+ local MAX=${2:-$(max_recovery_time)}
- local var_svc=${facet}_svc
- local procfile="*.${!var_svc}.recovery_status"
local WAIT=0
local STATUS=
while [ $WAIT -lt $MAX ]; do
- STATUS=$(do_facet $facet lctl get_param -n $procfile | grep status)
- [[ $STATUS = "status: COMPLETE" ]] && return 0
+ STATUS=$(lctl get_param -n $param | grep status)
+ echo $param $STATUS
+ [[ $STATUS = "status: COMPLETE" || $STATUS = "status: INACTIVE" ]] && return 0
sleep 5
WAIT=$((WAIT + 5))
- echo "Waiting $((MAX - WAIT)) secs for $facet recovery done. $STATUS"
+ echo "Waiting $((MAX - WAIT)) secs for $param recovery done. $STATUS"
done
- echo "$facet recovery not done in $MAX sec. $STATUS"
+ echo "$param recovery not done in $MAX sec. $STATUS"
return 1
}
+wait_recovery_complete () {
+ local facet=$1
+
+ # with an assumption that at_max is the same on all nodes
+ local MAX=${2:-$(max_recovery_time)}
+
+ local facets=$facet
+ if [ "$FAILURE_MODE" = HARD ]; then
+ facets=$(facets_on_host $(facet_active_host $facet))
+ fi
+ echo affected facets: $facets
+
+ # we can use "for" here because we are waiting the slowest
+ for facet in ${facets//,/ }; do
+ local var_svc=${facet}_svc
+ local param="*.${!var_svc}.recovery_status"
+
+ local host=$(facet_active_host $facet)
+ do_rpc_nodes $host _wait_recovery_complete $param $MAX
+ done
+}
+
wait_mds_ost_sync () {
# just because recovery is done doesn't mean we've finished
# orphan cleanup. Wait for llogs to get synchronized.
local -a sync=($(do_nodes $(comma_list $(osts_nodes)) \
"$LCTL get_param -n obdfilter.*.mds_sync"))
local con=1
+ local i
for ((i=0; i<${#sync[@]}; i++)); do
[ ${sync[$i]} -eq 0 ] && continue
# there is a not finished MDS-OST synchronization
facet_failover() {
local facet=$1
local sleep_time=$2
- echo "Failing $facet on node `facet_active_host $facet`"
+ local host=$(facet_active_host $facet)
+
+ echo "Failing $facet on node $host"
+
+ local affected=$facet
+
+ if [ "$FAILURE_MODE" = HARD ]; then
+ affected=$(facets_on_host $host)
+ fi
+
shutdown_facet $facet
+
+ echo affected facets: $affected
+
[ -n "$sleep_time" ] && sleep $sleep_time
+
reboot_facet $facet
- change_active $facet
- local TO=`facet_active_host $facet`
- echo "Failover $facet to $TO"
- wait_for $facet
- mount_facet $facet || error "Restart of $facet failed"
+
+ change_active $affected
+
+ wait_for_facet $affected
+ # start mgs first if it is affected
+ if ! combined_mgs_mds && list_member $affected mgs; then
+ mount_facet mgs || error "Restart of mgs failed"
+ fi
+ # FIXME; has to be changed to mount all facets concurrently
+ affected=$(exclude_items_from_list $affected mgs)
+ for facet in ${affected//,/ }; do
+ mount_facet $facet || error "Restart of $facet on node $host failed!"
+ done
}
obd_name() {
exit 1
}
-h2gm () {
- if [ "$1" = "client" -o "$1" = "'*'" ]; then echo \'*\'; else
- ID=`$PDSH $1 $GMNALNID -l | cut -d\ -f2`
- echo $ID"@gm"
- fi
-}
-
h2name_or_ip() {
if [ "$1" = "client" -o "$1" = "'*'" ]; then echo \'*\'; else
echo $1"@$2"
}
declare -fx h2elan
-h2openib() {
- h2name_or_ip "$1" "openib"
-}
-declare -fx h2openib
-
h2o2ib() {
h2name_or_ip "$1" "o2ib"
}
}
change_active() {
- local facet=$1
+ local facetlist=$1
+ local facet
+
+ facetlist=$(exclude_items_from_list $facetlist mgs)
+
+ for facet in ${facetlist//,/ }; do
local failover=${facet}failover
- host=`facet_host $failover`
+ local host=`facet_host $failover`
[ -z "$host" ] && return
+
local curactive=`facet_active $facet`
if [ -z "${curactive}" -o "$curactive" == "$failover" ] ; then
eval export ${facet}active=$facet
# save the active host for this facet
local activevar=${facet}active
echo "$activevar=${!activevar}" > $TMP/$activevar
+ local TO=`facet_active_host $facet`
+ echo "Failover $facet to $TO"
+ done
}
do_node() {
$myPDSH $rnodes $LCTL mark "$@" > /dev/null 2>&1 || :
fi
- if $verbose ; then
+ # do not replace anything from pdsh output if -N is used
+ # -N Disable hostname: prefix on lines of output.
+ if $verbose || [[ $myPDSH = *-N* ]]; then
$myPDSH $rnodes "(PATH=\$PATH:$RLUSTRE/utils:$RLUSTRE/tests:/sbin:/usr/sbin; cd $RPWD; LUSTRE=\"$RLUSTRE\" $(get_env_vars) sh -c \"$@\")"
else
- $myPDSH $rnodes "(PATH=\$PATH:$RLUSTRE/utils:$RLUSTRE/tests:/sbin:/usr/sbin; cd $RPWD; LUSTRE=\"$RLUSTRE\" $(get_env_vars) sh -c \"$@\")" | sed -re "s/\w+:\s//g"
+ $myPDSH $rnodes "(PATH=\$PATH:$RLUSTRE/utils:$RLUSTRE/tests:/sbin:/usr/sbin; cd $RPWD; LUSTRE=\"$RLUSTRE\" $(get_env_vars) sh -c \"$@\")" | sed -re "s/^[^:]*: //g"
fi
return ${PIPESTATUS[0]}
}
return 0
}
+cleanup_echo_devs () {
+ local devs=$($LCTL dl | grep echo | awk '{print $4}')
+
+ for dev in $devs; do
+ $LCTL --device $dev cleanup
+ $LCTL --device $dev detach
+ done
+}
+
cleanupall() {
nfs_client_mode && return
stopall $*
+ cleanup_echo_devs
+
unload_modules
cleanup_gss
}
-mdsmkfsopts()
-{
- local nr=$1
- test $nr = 1 && echo -n $MDS_MKFS_OPTS || echo -n $MDSn_MKFS_OPTS
-}
-
combined_mgs_mds () {
[[ $MDSDEV1 = $MGSDEV ]] && [[ $mds1_HOST = $mgs_HOST ]]
}
+mkfs_opts () {
+ local facet=$1
+
+ local tgt=$(echo $facet | tr -d [:digit:] | tr "[:lower:]" "[:upper:]")
+ local optvar=${tgt}_MKFS_OPTS
+ local opt=${!optvar}
+
+ # FIXME: ! combo mgs/mds + mgsfailover is not supported yet
+ [[ $facet = mgs ]] && echo $opt && return
+
+ # 1.
+ # --failnode options
+ local var=${facet}failover_HOST
+ if [ x"${!var}" != x ] && [ x"${!var}" != x$(facet_host $facet) ] ; then
+ local failnode=$(h2$NETTYPE ${!var})
+ failnode="--failnode=$failnode"
+ # options does not contain
+ # or contains wrong --failnode=
+ if [[ $opt != *${failnode}* ]]; then
+ opt=$(echo $opt | sed 's/--failnode=.* / /')
+ opt="$opt $failnode"
+ fi
+ fi
+
+ # 2.
+ # --mgsnode options
+ # no additional mkfs mds "--mgsnode" option for this configuration
+ if [[ $facet = mds ]] && combined_mgs_mds; then
+ echo $opt
+ return
+ fi
+
+ # additional mkfs "--mgsnode"
+ local mgsnode="--mgsnode=$MGSNID"
+ opt=${opt//$mgsnode }
+ for nid in ${MGSNID//:/ }; do
+ local mgsnode="--mgsnode=$nid"
+ # options does not contain
+ # --mgsnode=$nid
+ if [[ $opt != *${mgsnode}" "* ]]; then
+ opt="$opt --mgsnode=$nid"
+ fi
+ done
+
+ echo $opt
+}
+
formatall() {
if [ "$IAMDIR" == "yes" ]; then
MDS_MKFS_OPTS="$MDS_MKFS_OPTS --iam-dir"
- MDSn_MKFS_OPTS="$MDSn_MKFS_OPTS --iam-dir"
fi
[ "$FSTYPE" ] && FSTYPE_OPT="--backfstype $FSTYPE"
[ "$CLIENTONLY" ] && return
echo Formatting mgs, mds, osts
if ! combined_mgs_mds ; then
- add mgs $mgs_MKFS_OPTS $FSTYPE_OPT --reformat $MGSDEV || exit 10
+ add mgs $(mkfs_opts mgs) $FSTYPE_OPT --reformat $MGSDEV || exit 10
fi
for num in `seq $MDSCOUNT`; do
echo "Format mds$num: $(mdsdevname $num)"
if $VERBOSE; then
- add mds$num `mdsmkfsopts $num` $FSTYPE_OPT --reformat `mdsdevname $num` || exit 9
+ add mds$num $(mkfs_opts mds) $FSTYPE_OPT --reformat $(mdsdevname $num) || exit 10
else
- add mds$num `mdsmkfsopts $num` $FSTYPE_OPT --reformat `mdsdevname $num` > /dev/null || exit 9
+ add mds$num $(mkfs_opts mds) $FSTYPE_OPT --reformat $(mdsdevname $num) > /dev/null || exit 10
fi
done
+ # the ost-s could have different OST_MKFS_OPTS
+ # because of different failnode-s
for num in `seq $OSTCOUNT`; do
echo "Format ost$num: $(ostdevname $num)"
if $VERBOSE; then
- add ost$num $OST_MKFS_OPTS --reformat `ostdevname $num` || exit 10
+ add ost$num $(mkfs_opts ost${num}) $FSTYPE_OPT --reformat `ostdevname $num` || exit 10
else
- add ost$num $OST_MKFS_OPTS --reformat `ostdevname $num` > /dev/null || exit 10
+ add ost$num $(mkfs_opts ost${num}) $FSTYPE_OPT --reformat `ostdevname $num` > /dev/null || exit 10
fi
done
}
do_facet mds$num "lctl set_param -n mdt.$MDT.identity_upcall \"NONE\""
fi
- do_facet mds$num "lctl set_param -n mdt/$MDT/identity_flush \"-1\""
+ do_facet mds$num "lctl set_param -n mdt/$MDT/identity_flush=-1"
if [ $old = "NONE" ]; then
return 1
echo $WRITECONF | grep -q "writeconf" && \
writeconf_all
if ! combined_mgs_mds ; then
- start mgs $MGSDEV $mgs_MOUNT_OPTS
+ start mgs $MGSDEV $MGS_MOUNT_OPTS
fi
for num in `seq $MDSCOUNT`; do
done
fi
+ combined_mgs_mds || init_facet_vars mgs $MGSDEV $MGS_MOUNT_OPTS
+
remote_ost_nodsh && return
for num in `seq $OSTCOUNT`; do
osc_ensure_active () {
local facet=$1
- local type=$2
- local timeout=$3
+ local timeout=$2
local period=0
while [ $period -lt $timeout ]; do
- count=$(do_facet $facet "lctl dl | grep '${FSNAME}-OST.*-osc-${type}' | grep ' IN ' 2>/dev/null | wc -l")
+ count=$(do_facet $facet "lctl dl | grep ' IN osc ' 2>/dev/null | wc -l")
if [ $count -eq 0 ]; then
break
fi
log "Using TIMEOUT=$TIMEOUT"
- osc_ensure_active $SINGLEMDS M $TIMEOUT
- osc_ensure_active client c $TIMEOUT
+ osc_ensure_active $SINGLEMDS $TIMEOUT
+ osc_ensure_active client $TIMEOUT
if [ $QUOTA_AUTO -ne 0 ]; then
if [ "$ENABLE_QUOTA" ]; then
is_mounted () {
local mntpt=$1
+ [ -z $mntpt ] && return 1
local mounted=$(mounted_lustre_filesystems)
echo $mounted' ' | grep -w -q $mntpt' '
}
+is_empty_dir() {
+ [ $(find $1 -maxdepth 1 -print | wc -l) = 1 ] && return 0
+ return 1
+}
+
+# empty lustre filesystem may have empty directories lost+found and .lustre
+is_empty_fs() {
+ [ $(find $1 -maxdepth 1 -name lost+found -o -name .lustre -prune -o \
+ -print | wc -l) = 1 ] || return 1
+ [ ! -d $1/lost+found ] || is_empty_dir $1/lost+found && return 0
+ [ ! -d $1/.lustre ] || is_empty_dir $1/.lustre && return 0
+ return 1
+}
+
check_and_setup_lustre() {
nfs_client_mode && return
check_and_setup_lustre
}
+# Get all of the server target devices from a given server node and type.
+get_mnt_devs() {
+ local node=$1
+ local type=$2
+ local obd_type
+ local devs
+ local dev
+
+ case $type in
+ mdt) obd_type="osd" ;;
+ ost) obd_type="obdfilter" ;; # needs to be fixed when OST also uses an OSD
+ *) echo "invalid server type" && return 1 ;;
+ esac
+
+ devs=$(do_node $node "lctl get_param -n $obd_type*.*.mntdev")
+ for dev in $devs; do
+ case $dev in
+ *loop*) do_node $node "losetup $dev" | \
+ sed -e "s/.*(//" -e "s/).*//" ;;
+ *) echo $dev ;;
+ esac
+ done
+}
+
+# Get all of the server target devices.
+get_svr_devs() {
+ local i
+
+ # MDT device
+ MDTDEV=$(get_mnt_devs $(mdts_nodes) mdt)
+
+ # OST devices
+ i=0
+ for node in $(osts_nodes); do
+ OSTDEVS[i]=$(get_mnt_devs $node ost)
+ i=$((i + 1))
+ done
+}
+
+# Run e2fsck on MDT or OST device.
+run_e2fsck() {
+ local node=$1
+ local target_dev=$2
+ local ostidx=$3
+ local ostdb_opt=$4
+
+ df > /dev/null # update statfs data on disk
+ local cmd="$E2FSCK -d -v -f -n $MDSDB_OPT $ostdb_opt $target_dev"
+ echo $cmd
+ local rc=0
+ do_node $node $cmd || rc=$?
+ [ $rc -le $FSCK_MAX_ERR ] || \
+ error "$cmd returned $rc, should be <= $FSCK_MAX_ERR"
+ return 0
+}
+
+# Run e2fsck on MDT and OST(s) to generate databases used for lfsck.
+generate_db() {
+ local i
+ local ostidx
+ local dev
+ local tmp_file
+
+ [ $MDSCOUNT -eq 1 ] || error "CMD is not supported"
+ tmp_file=$(mktemp -p $SHARED_DIRECTORY ||
+ error "fail to create file in $SHARED_DIRECTORY")
+
+ # make sure everything gets to the backing store
+ local list=$(comma_list $CLIENTS $(facet_host $SINGLEMDS) $(osts_nodes))
+ do_nodes $list "sync; sleep 2; sync"
+
+ do_nodes $list ls $tmp_file || \
+ error "$SHARED_DIRECTORY is not a shared directory"
+ rm $tmp_file
+
+ run_e2fsck $(mdts_nodes) $MDTDEV
+
+ i=0
+ ostidx=0
+ OSTDB_LIST=""
+ for node in $(osts_nodes); do
+ for dev in ${OSTDEVS[i]}; do
+ local ostdb_opt=`eval echo $OSTDB_OPT`
+ run_e2fsck $node $dev $ostidx "$ostdb_opt"
+ OSTDB_LIST="$OSTDB_LIST $OSTDB-$ostidx"
+ ostidx=$((ostidx + 1))
+ done
+ i=$((i + 1))
+ done
+}
+
+run_lfsck() {
+ local cmd="$LFSCK_BIN -c -l --mdsdb $MDSDB --ostdb $OSTDB_LIST $MOUNT"
+ echo $cmd
+ local rc=0
+ eval $cmd || rc=$?
+ [ $rc -le $FSCK_MAX_ERR ] || \
+ error "$cmd returned $rc, should be <= $FSCK_MAX_ERR"
+ echo "lfsck finished with rc=$rc"
+
+ rm -rvf $MDSDB* $OSTDB* || true
+ return 0
+}
+
check_and_cleanup_lustre() {
+ if [ "$LFSCK_ALWAYS" = "yes" ]; then
+ get_svr_devs
+ generate_db
+ if [ "$SKIP_LFSCK" == "no" ]; then
+ run_lfsck
+ else
+ echo "skip lfsck"
+ fi
+ fi
+
if is_mounted $MOUNT; then
[ -n "$DIR" ] && rm -rf $DIR/[Rdfs][0-9]*
[ "$ENABLE_QUOTA" ] && restore_quota_type || true
echo "$*" | tr -s " " "\n" | sort -b -u | tr "\n" " " | sed 's/ \([^$]\)/,\1/g'
}
+list_member () {
+ local list=$1
+ local item=$2
+ echo $list | grep -qw $item
+}
+
# list, excluded are the comma separated lists
exclude_items_from_list () {
local list=$1
for item in ${excluded//,/ }; do
list=$(echo " $list " | sed -re "s/\s+$item\s+/ /g")
done
- echo $(comma_list $list)
+ echo $(comma_list $list)
}
# list, expand are the comma separated lists
}
get_facets () {
- local name=$(echo $1 | tr "[:upper:]" "[:lower:]")
- local type=$(echo $1 | tr "[:lower:]" "[:upper:]")
+ local types=${1:-"OST MDS MGS"}
local list=""
- local count=${type}COUNT
- for ((i=1; i<=${!count}; i++)) do
- list="$list ${name}$i"
+
+ for entry in $types; do
+ local name=$(echo $entry | tr "[:upper:]" "[:lower:]")
+ local type=$(echo $entry | tr "[:lower:]" "[:upper:]")
+
+ case $type in
+ MGS ) list="$list $name";;
+ MDS|OST ) local count=${type}COUNT
+ for ((i=1; i<=${!count}; i++)) do
+ list="$list ${name}$i"
+ done;;
+ * ) error "Invalid facet type"
+ exit 1;;
+ esac
done
echo $(comma_list $list)
}
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() {
# 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)")"
}
#
rm -rf $LOGDIR/err
echo
- run_one $1 "$2" 2>&1 | tee $test_log
+ log_sub_test_begin test_${1}
+ (run_one $1 "$2") 2>&1 | tee $test_log
local RC=${PIPESTATUS[0]}
[ $RC -ne 0 ] && [ ! -f $LOGDIR/err ] && \
duration=$((`date +%s` - $BEFORE))
pass "(${duration}s)"
[ -f $LOGDIR/err ] && TEST_ERROR=$(cat $LOGDIR/err)
- log_sub_test test_${1} $TEST_STATUS $duration "$RC" "$TEST_ERROR"
+ log_sub_test_end $TEST_STATUS $duration "$RC" "$TEST_ERROR"
if [ -f $LOGDIR/err ]; then
$FAIL_ON_ERROR && exit $RC
remote_ost && remote_mds
}
-osts_nodes () {
- local OSTNODES=$(facet_host ost1)
+facets_nodes () {
+ local facets=$1
+ local nodes
local NODES_sort
- for num in `seq $OSTCOUNT`; do
- local myOST=$(facet_host ost$num)
- OSTNODES="$OSTNODES $myOST"
+ for facet in ${facets//,/ }; do
+ if [ "$FAILURE_MODE" = HARD ]; then
+ nodes="$nodes $(facet_active_host $facet)"
+ else
+ nodes="$nodes $(facet_host $facet)"
+ fi
done
- NODES_sort=$(for i in $OSTNODES; do echo $i; done | sort -u)
+ NODES_sort=$(for i in $nodes; do echo $i; done | sort -u)
echo $NODES_sort
}
+osts_nodes () {
+ local facets=$(get_facets OST)
+ local nodes=$(facets_nodes $facets)
+
+ echo $nodes
+}
+
nodes_list () {
# FIXME. We need a list of clients
local myNODES=$HOSTNAME
[ -n "$CLIENTS" ] && myNODES=${CLIENTS//,/ }
if [ "$PDSH" -a "$PDSH" != "no_dsh" ]; then
- myNODES="$myNODES $(osts_nodes) $(mdts_nodes)"
+ myNODES="$myNODES $(facets_nodes $(get_facets))"
fi
myNODES_sort=$(for i in $myNODES; do echo $i; done | sort -u)
get_clientosc_proc_path() {
local ost=$1
- echo "{$1}-osc-*"
+ echo "${1}-osc-*"
}
get_lustre_version () {
- local node=${1:-"mds"}
- do_facet $node $LCTL get_param -n version | awk '/^lustre:/ {print $2}'
+ local facet=${1:-"$SINGLEMDS"}
+ do_facet $facet $LCTL get_param -n version | awk '/^lustre:/ {print $2}'
}
get_mds_version_major () {
- local version=$(get_lustre_version mds)
+ local facet=${1:-"$SINGLEMDS"}
+ local version=$(get_lustre_version $facet)
echo $version | awk -F. '{print $1}'
}
get_mds_version_minor () {
- local version=$(get_lustre_version mds)
+ local facet=${1:-"$SINGLEMDS"}
+ local version=$(get_lustre_version $facet)
echo $version | awk -F. '{print $2}'
}
+# If the 2.0 MDS was mounted on 1.8 device, then the OSC and LOV names
+# used by MDT would not be changed.
+# mdt lov: fsname-mdtlov
+# mdt osc: fsname-OSTXXXX-osc
+mds_on_old_device() {
+ local mds=${1:-"$SINGLEMDS"}
+ local major=$(get_mds_version_major $mds)
+ local minor=$(get_mds_version_minor $mds)
+
+ if [ $major -ge 2 ] || [ $major -eq 1 -a $minor -gt 8 ]; then
+ do_facet $mds "lctl list_param osc.$FSNAME-OST*-osc \
+ > /dev/null 2>&1" && return 0
+ fi
+ return 1
+}
+
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"
+ local mds_facet=$1
+ local ost_label=${2:-"*OST*"}
+
+ [ "$mds_facet" = "mds" ] && mds_facet=$SINGLEMDS
+ local mdt_label=$(convert_facet2label $mds_facet)
+ local mdt_index=$(echo $mdt_label | sed -e 's/^.*-//')
+
+ local major=$(get_mds_version_major $mds_facet)
+ local minor=$(get_mds_version_minor $mds_facet)
+ if [ $major -le 1 -a $minor -le 8 ] || mds_on_old_device $mds_facet; then
+ echo "${ost_label}-osc"
else
- echo "${ost}-osc-MDT0000"
+ echo "${ost_label}-osc-${mdt_index}"
fi
}
local ost=$2
local label=$(convert_facet2label $ost)
- if [ "$facet" == "mds" ]; then
- get_mdtosc_proc_path $label
+ if [ "${facet:0:3}" = "mds" ]; then
+ get_mdtosc_proc_path $facet $label
return 0
fi
return 0
}
-wait_import_state () {
+_wait_import_state () {
local expected=$1
local CONN_PROC=$2
+ local maxtime=${3:-max_recovery_time}
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
+ [ $i -ge $maxtime ] && \
+ error "can't put import for $CONN_PROC into ${expected} state after $i sec, have ${CONN_STATE}" && \
+ return 1
sleep 1
CONN_STATE=$($LCTL get_param -n $CONN_PROC 2>/dev/null | cut -f2)
i=$(($i + 1))
done
- log "$CONN_PROC now in ${CONN_STATE} state"
+ log "$CONN_PROC in ${CONN_STATE} state after $i sec"
return 0
}
+wait_import_state() {
+ local state=$1
+ local params=$2
+ local maxtime=${3:-max_recovery_time}
+ local param
+
+ for param in ${params//,/ }; do
+ _wait_import_state $state $param $maxtime || return
+ done
+}
wait_osc_import_state() {
local facet=$1
local ost_facet=$2
local list=$1
local facet=$2
local expected=$3
- shift
+ local facets=$facet
+
+ if [ "$FAILURE_MODE" = HARD ]; then
+ facets=$(facets_on_host $(facet_active_host $facet))
+ fi
+
+ for facet in ${facets//,/ }; do
local label=$(convert_facet2label $facet)
local proc_path
case $facet in
mds* ) proc_path="mdc.$(get_clientmdc_proc_path $label).mds_server_uuid" ;;
*) error "unknown facet!" ;;
esac
+ local params=$(expand_list $params $proc_path)
+ done
- if ! do_rpc_nodes $list wait_import_state $expected $proc_path; then
+ if ! do_rpc_nodes $list wait_import_state $expected $params; then
error "import is not in ${expected} state"
return 1
fi
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`}
R=0
else
N=$((N + 1))
- while [ $N -le $M ]; do
+ while [ $N -lt $M ]; do
R=$((R * N))
N=$((N + 1))
done
{
local flavor=$1
local cnt=0
+ local mdtosc
for num in `seq $MDSCOUNT`; do
- output=`do_facet mds$num lctl get_param -n osc.*OST*-osc-MDT*.$PROC_CLI 2>/dev/null`
+ mdtosc=$(get_mdtosc_proc_path mds$num)
+ mdtosc=${mdtosc/-MDT*/-MDT\*}
+ output=$(do_facet mds$num lctl get_param -n \
+ osc.$mdtosc.$PROC_CLI 2>/dev/null)
tmpcnt=`count_flvr "$output" $flavor`
cnt=$((cnt + tmpcnt))
done
return 0
else
echo "found $res $flavor connections of $dir, not ready ($expect)"
+ return 0
sleep 4
fi
done
yml_log_test $1 >> $YAML_LOG
}
-log_sub_test() {
- yml_log_sub_test $@ >> $YAML_LOG
+log_test_status() {
+ yml_log_test_status $@ >> $YAML_LOG
+}
+
+log_sub_test_begin() {
+ yml_log_sub_test_begin $@ >> $YAML_LOG
+}
+
+log_sub_test_end() {
+ yml_log_sub_test_end $@ >> $YAML_LOG
}
run_llverdev()
local dev=$1
local devname=$(basename $1)
local size=$(grep "$devname"$ /proc/partitions | awk '{print $3}')
+ # loop devices aren't in /proc/partitions
+ [ "x$size" == "x" ] && local size=$(ls -l $dev | awk '{print $5}')
+
size=$(($size / 1024 / 1024)) # Gb
local partial_arg=""
llverdev --force $partial_arg $dev
}
+
+remove_mdt_files() {
+ local facet=$1
+ local mdtdev=$2
+ shift 2
+ local files="$@"
+ local mntpt=${MOUNT%/*}/$facet
+
+ echo "removing files from $mdtdev on $facet: $files"
+ mount -t $FSTYPE $MDS_MOUNT_OPTS $mdtdev $mntpt || return $?
+ rc=0;
+ for f in $files; do
+ rm $mntpt/ROOT/$f || { rc=$?; break; }
+ done
+ umount -f $mntpt || return $?
+ return $rc
+}
+
+duplicate_mdt_files() {
+ local facet=$1
+ local mdtdev=$2
+ shift 2
+ local files="$@"
+ local mntpt=${MOUNT%/*}/$facet
+
+ echo "duplicating files on $mdtdev on $facet: $files"
+ mkdir -p $mntpt || return $?
+ mount -t $FSTYPE $MDS_MOUNT_OPTS $mdtdev $mntpt || return $?
+
+ do_umount() {
+ trap 0
+ popd > /dev/null
+ rm $tmp
+ umount -f $mntpt
+ }
+ trap do_umount EXIT
+
+ tmp=$(mktemp $TMP/setfattr.XXXXXXXXXX)
+ pushd $mntpt/ROOT > /dev/null || return $?
+ rc=0
+ for f in $files; do
+ touch $f.bad || return $?
+ getfattr -n trusted.lov $f | sed "s#$f#&.bad#" > $tmp
+ rc=${PIPESTATUS[0]}
+ [ $rc -eq 0 ] || return $rc
+ setfattr --restore $tmp || return $?
+ done
+ do_umount
+}
+
+run_sgpdd () {
+ local devs=${1//,/ }
+ shift
+ local params=$@
+ local rslt=$TMP/sgpdd_survey
+
+ # sgpdd-survey cleanups ${rslt}.* files
+
+ local cmd="rslt=$rslt $params scsidevs=\"$devs\" $SGPDDSURVEY"
+ echo + $cmd
+ eval $cmd
+ cat ${rslt}.detail
+}
+