[ ! -f "$LCTL" ] && export LCTL=$(which lctl)
export LFS=${LFS:-"$LUSTRE/utils/lfs"}
[ ! -f "$LFS" ] && export LFS=$(which lfs)
+ export KSOCKLND_CONFIG=${KSOCKLND_CONFIG:-"$LUSTRE/scripts/ksocklnd-config"}
+ [ ! -f "$KSOCKLND_CONFIG" ] &&
+ export KSOCKLND_CONFIG=$(which ksocklnd-config 2> /dev/null)
- export PERM_CMD=${PERM_CMD:-"$LCTL conf_param"}
+ export PERM_CMD=$(echo ${PERM_CMD:-"$LCTL conf_param"})
export L_GETIDENTITY=${L_GETIDENTITY:-"$LUSTRE/utils/l_getidentity"}
if [ ! -f "$L_GETIDENTITY" ]; then
fi
load_module ../lnet/klnds/$LNETLND
load_module obdclass/obdclass
+ MODOPTS_PTLRPC=${MODOPTS_PTLRPC:-"lbug_on_grant_miscount=1"}
load_module ptlrpc/ptlrpc
load_module ptlrpc/gss/ptlrpc_gss
load_module fld/fld
}
load_modules () {
+ local facets
+ local facet
+ local failover
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 $(remote_nodes_list))
+
+ # include failover nodes in case they are not in the list yet
+ facets=$(get_facets)
+ for facet in ${facets//,/ }; do
+ failover=$(facet_failover_host $facet)
+ [ -n "$list" ] && [[ ! "$list" =~ "$failover" ]] &&
+ list="$list,$failover"
+ done
+
if [ -n "$list" ]; then
echo "loading modules on: '$list'"
do_rpc_nodes "$list" load_modules_local
local device=${2:-$FSNAME-OST*}
local name=$3
- do_nodes $nodes "$LCTL get_param -n obdfilter.$device.$name \
- osd-*.$device.$name 2>&1" | grep -v 'error:'
+ do_nodes $nodes "$LCTL get_param -n osd-*.$device.$name"
}
set_osd_param() {
local name=$3
local value=$4
- do_nodes $nodes "$LCTL set_param -n obdfilter.$device.$name=$value \
- osd-*.$device.$name=$value 2>&1" | grep -v 'error:'
+ do_nodes $nodes "$LCTL set_param -n osd-*.$device.$name=$value"
}
set_debug_size () {
local devicelabel
local dm_dev=${!dev}
+ [[ $dev == "mgsfailover_dev" ]] && combined_mgs_mds &&
+ dev=mds1failover_dev
+
module_loaded lustre || load_modules
case $fstype in
eval export ${dev_alias}_dev=${device}
eval export ${facet}_opt=\"$@\"
+ combined_mgs_mds && [[ ${dev_alias} == mds1 ]] &&
+ eval export mgs_dev=${device}
+
local varname=${dev_alias}failover_dev
if [ -n "${!varname}" ] ; then
eval export ${dev_alias}failover_dev=${!varname}
else
eval export ${dev_alias}failover_dev=$device
+ combined_mgs_mds && [[ ${dev_alias} == mds1 ]] &&
+ eval export mgsfailover_dev=${device}
+
fi
local mntpt=$(facet_mntpt $facet)
reboot_facet() {
local facet=$1
local node=$(facet_active_host $facet)
+ local sleep_time=${2:-10}
if [ "$FAILURE_MODE" = HARD ]; then
boot_node $node
else
- sleep 10
+ sleep $sleep_time
fi
}
fi
echo affected facets: $facets
- # we can use "for" here because we are waiting the slowest
- for facet in ${facets//,/ }; do
+ facets=${facets//,/ }
+ # We can use "for" here because we are waiting the slowest.
+ # The mgs not having the recovery_status proc entry, exclude it
+ # from the facet list.
+ for facet in ${facets//mgs/ }; do
local var_svc=${facet}_svc
local param="*.${!var_svc}.recovery_status"
facet=$(echo ${affecteds[index]} | tr -s " " | cut -d"," -f 1)
echo reboot facets: ${affecteds[index]}
- reboot_facet $facet
+ reboot_facet $facet $sleep_time
change_active ${affecteds[index]}
if ! combined_mgs_mds &&
list_member ${affecteds[index]} mgs; then
mount_facet mgs || error "Restart of mgs failed"
+ affecteds[index]=$(exclude_items_from_list \
+ ${affecteds[index]} mgs)
fi
# FIXME; has to be changed to mount all facets concurrently
- affected=$(exclude_items_from_list ${affecteds[index]} mgs)
- echo mount facets: ${affecteds[index]}
- mount_facets ${affecteds[index]}
+ if [ -n "${affecteds[index]}" ]; then
+ echo mount facets: ${affecteds[index]}
+ mount_facets ${affecteds[index]}
+ fi
if $GSS_SK; then
do_nodes $(comma_list $(all_nodes)) \
"keyctl show | grep lustre | cut -c1-11 |
elif [ "${facet:0:3}" == "mdt" -o \
"${facet:0:3}" == "mds" -o \
"${facet:0:3}" == "mgs" ]; then
- eval export ${facet}_HOST=${mds_HOST}
+ local temp
+ if [ "${facet}" == "mgsfailover" ] &&
+ [ -n "$mds1failover_HOST" ]; then
+ temp=$mds1failover_HOST
+ else
+ temp=${mds_HOST}
+ fi
+ eval export ${facet}_HOST=$temp
fi
fi
echo -n ${!varname}
return
fi
+ if combined_mgs_mds && [ $facet == "mgs" ] &&
+ [ -z $mds1failover_HOST ]; then
+ temp=mds1failover_HOST
+ echo ${!temp}
+ return
+ fi
+
if [ "${facet:0:3}" == "mdt" -o "${facet:0:3}" == "mds" -o \
"${facet:0:3}" == "mgs" ]; then
local facetlist=$1
local facet
- facetlist=$(exclude_items_from_list $facetlist mgs)
-
for facet in ${facetlist//,/ }; do
local failover=${facet}failover
local host=`facet_host $failover`
eval $myPDSH $HOST "(PATH=\$PATH:$RLUSTRE/utils:$RLUSTRE/tests;
PATH=\$PATH:/sbin:/usr/sbin;
cd $RPWD;
- LUSTRE=\"$RLUSTRE\" sh -c \"$@\") ||
+ LUSTRE=\"$RLUSTRE\" bash -c \"$@\") ||
echo command failed >$command_status"
[[ -n "$($myPDSH $HOST cat $command_status)" ]] && return 1 ||
return 0
fi
- if [[ -n "$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 $RPWD; LUSTRE=\"$RLUSTRE\" sh -c \"$@\")" | sed -e "s/^/${HOSTNAME}: /"
- else
- $myPDSH $HOST "(PATH=\$PATH:$RLUSTRE/utils:$RLUSTRE/tests:/sbin:/usr/sbin; cd $RPWD; LUSTRE=\"$RLUSTRE\" sh -c \"$@\")"
- fi
- else
- $myPDSH $HOST "(PATH=\$PATH:$RLUSTRE/utils:$RLUSTRE/tests:/sbin:/usr/sbin; cd $RPWD; LUSTRE=\"$RLUSTRE\" sh -c \"$@\")" | sed "s/^${HOST}: //"
- fi
- return ${PIPESTATUS[0]}
+ if [[ -n "$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 $RPWD; LUSTRE=\"$RLUSTRE\" bash -c \"$@\")" |
+ sed -e "s/^/${HOSTNAME}: /"
+ else
+ $myPDSH $HOST \
+ "(PATH=\$PATH:$RLUSTRE/utils:$RLUSTRE/tests:/sbin:/usr/sbin;\
+ cd $RPWD; LUSTRE=\"$RLUSTRE\" bash -c \"$@\")"
+ fi
+ else
+ $myPDSH $HOST \
+ "(PATH=\$PATH:$RLUSTRE/utils:$RLUSTRE/tests:/sbin:/usr/sbin;\
+ cd $RPWD; LUSTRE=\"$RLUSTRE\" bash -c \"$@\")" |
+ sed "s/^${HOST}: //"
+ fi
+ return ${PIPESTATUS[0]}
}
single_local_node () {
# do not replace anything from pdsh output if -N is used
# -N Disable hostname: prefix on lines of output.
if [[ -n "$verbose" || $myPDSH = *-N* ]]; then
- $myPDSH $rnodes "(PATH=\$PATH:$RLUSTRE/utils:$RLUSTRE/tests:/sbin:/usr/sbin; cd $RPWD; LUSTRE=\"$RLUSTRE\" $(get_env_vars) sh -c \"$@\")"
+ $myPDSH $rnodes "(PATH=\$PATH:$RLUSTRE/utils:$RLUSTRE/tests:/sbin:/usr/sbin; cd $RPWD; LUSTRE=\"$RLUSTRE\" $(get_env_vars) bash -c \"$@\")"
else
- $myPDSH $rnodes "(PATH=\$PATH:$RLUSTRE/utils:$RLUSTRE/tests:/sbin:/usr/sbin; cd $RPWD; LUSTRE=\"$RLUSTRE\" $(get_env_vars) sh -c \"$@\")" | sed -re "s/^[^:]*: //g"
+ $myPDSH $rnodes "(PATH=\$PATH:$RLUSTRE/utils:$RLUSTRE/tests:/sbin:/usr/sbin; cd $RPWD; LUSTRE=\"$RLUSTRE\" $(get_env_vars) bash -c \"$@\")" | sed -re "s/^[^:]*: //g"
fi
return ${PIPESTATUS[0]}
}
local varname=${facet}failover_HOST
if [ -z "${!varname}" ]; then
- eval export $varname=$(facet_host $facet)
+ local temp
+ if combined_mgs_mds && [ $facet == "mgs" ] &&
+ [ -n "$mds1failover_HOST" ]; then
+ temp=$mds1failover_HOST
+ else
+ temp=$(facet_host $facet)
+ fi
+ eval export $varname=$temp
fi
varname=${facet}_HOST
if ! remote_mds_nodsh; then
for num in $(seq $MDSCOUNT); do
- DEVNAME=`mdsdevname $num`
+ DEVNAME=$(mdsdevname $num)
init_facet_vars mds$num $DEVNAME $MDS_MOUNT_OPTS
done
fi
- combined_mgs_mds || init_facet_vars mgs $(mgsdevname) $MGS_MOUNT_OPTS
+ init_facet_vars mgs $(mgsdevname) $MGS_MOUNT_OPTS
if ! remote_ost_nodsh; then
for num in $(seq $OSTCOUNT); do
fi
(( MDS1_VERSION <= $(version_code 2.13.52) )) ||
- do_nodes $(comma_list $(mdts_nodes)) \
- "$LCTL set_param lod.*.mdt_hash=crush"
- do_node $(mgs_node) "$LCTL set_param -P *.*.lbug_on_grant_miscount=1"
+ do_facet mgs "$LCTL set_param -P lod.*.mdt_hash=crush"
return 0
}
# General functions
wait_for_function () {
- local quiet=""
-
- # suppress fn both stderr and stdout
- if [ "$1" = "--quiet" ]; then
- shift
- quiet=" > /dev/null 2>&1"
+ local quiet=""
- fi
+ # suppress fn both stderr and stdout
+ if [ "$1" = "--quiet" ]; then
+ shift
+ quiet=" > /dev/null 2>&1"
+ fi
- local fn=$1
- local max=${2:-900}
- local sleep=${3:-5}
+ local fn=$1
+ local max=${2:-900}
+ local sleep=${3:-5}
- local wait=0
+ local wait=0
- while true; do
+ while true; do
- eval $fn $quiet && return 0
+ eval $fn $quiet && return 0
- wait=$((wait + sleep))
- [ $wait -lt $max ] || return 1
- echo waiting $fn, $((max - wait)) secs left ...
- sleep $sleep
- done
+ [ $wait -lt $max ] || return 1
+ echo waiting $fn, $((max - wait)) secs left ...
+ wait=$((wait + sleep))
+ [ $wait -gt $max ] && ((sleep -= wait - max))
+ sleep $sleep
+ done
}
check_network() {
}
ostname_from_index() {
- local uuid=$(ostuuid_from_index $1)
+ local uuid=$(ostuuid_from_index $1 $2)
echo ${uuid/_UUID/}
}
}
_wait_import_state () {
- local expected=$1
- local CONN_PROC=$2
- local maxtime=${3:-$(max_recovery_time)}
- local error_on_failure=${4:-1}
- local CONN_STATE
- local i=0
+ local expected="$1"
+ local CONN_PROC="$2"
+ local maxtime=${3:-$(max_recovery_time)}
+ local err_on_fail=${4:-1}
+ local CONN_STATE
+ local i=0
CONN_STATE=$($LCTL get_param -n $CONN_PROC 2>/dev/null | cut -f2 | uniq)
- while ! echo "${CONN_STATE}" | egrep -q "^${expected}\$" ; do
- if [ "${expected}" == "DISCONN" ]; then
- # for disconn we can check after proc entry is removed
- [ "x${CONN_STATE}" == "x" ] && return 0
- # with AT enabled, we can have connect request timeout near of
- # reconnect timeout and test can't see real disconnect
- [ "${CONN_STATE}" == "CONNECTING" ] && return 0
- fi
- if [ $i -ge $maxtime ]; then
- [ $error_on_failure -ne 0 ] && \
- error "can't put import for $CONN_PROC into ${expected}" \
- "state after $i sec, have ${CONN_STATE}"
- return 1
- fi
- sleep 1
- # Add uniq for multi-mount case
- CONN_STATE=$($LCTL get_param -n $CONN_PROC 2>/dev/null | cut -f2 | uniq)
- i=$(($i + 1))
- done
+ while ! echo "${CONN_STATE}" | egrep -q "^${expected}\$" ; do
+ if [[ "${expected}" == "DISCONN" ]]; then
+ # for disconn we can check after proc entry is removed
+ [[ -z "${CONN_STATE}" ]] && return 0
+ # with AT, we can have connect request timeout near
+ # reconnect timeout and test can't see real disconnect
+ [[ "${CONN_STATE}" == "CONNECTING" ]] && return 0
+ fi
+ if (( $i >= $maxtime )); then
+ (( $err_on_fail != 0 )) &&
+ error "can't put import for $CONN_PROC into ${expected} state after $i sec, have ${CONN_STATE}"
+ return 1
+ fi
+ sleep 1
+ # Add uniq for multi-mount case
+ CONN_STATE=$($LCTL get_param -n $CONN_PROC 2>/dev/null |
+ cut -f2 | uniq)
+ i=$((i + 1))
+ done
- log "$CONN_PROC in ${CONN_STATE} state after $i sec"
- return 0
+ log "$CONN_PROC in ${CONN_STATE} state after $i sec"
+ return 0
}
wait_import_state() {
- local state=$1
- local params=$2
- local maxtime=${3:-$(max_recovery_time)}
- local error_on_failure=${4:-1}
- local param
-
- for param in ${params//,/ }; do
- _wait_import_state $state $param $maxtime $error_on_failure || return
- done
+ local expected="$1"
+ local params="$2"
+ local maxtime=${3:-$(max_recovery_time)}
+ local err_on_fail=${4:-1}
+ local param
+
+ for param in ${params//,/ }; do
+ _wait_import_state "$expected" "$param" $maxtime $err_on_fail ||
+ return
+ done
}
wait_import_state_mount() {
return 0
fi
- wait_import_state $*
+ wait_import_state "$@"
}
# One client request could be timed out because server was not ready
}
wait_clients_import_state () {
- local list=$1
- local facet=$2
- local expected=$3
-
- local facets=$facet
+ local list="$1"
+ local facet="$2"
+ local expected="$3"
+ local facets="$facet"
if [ "$FAILURE_MODE" = HARD ]; then
facets=$(facets_on_host $(facet_active_host $facet))
local proc_path
case $facet in
ost* ) proc_path="osc.$(get_clientosc_proc_path \
- $label).ost_server_uuid" ;;
+ $label).ost_server_uuid" ;;
mds* ) proc_path="mdc.$(get_clientmdc_proc_path \
- $label).mds_server_uuid" ;;
+ $label).mds_server_uuid" ;;
mgs* ) proc_path="mgc.$(get_clientmgc_proc_path \
- $label).mgs_server_uuid" ;;
+ $label).mgs_server_uuid" ;;
*) error "unknown facet!" ;;
esac
# ldiskfs xattrs over one block in size. Allow both the historical
# Lustre feature name (large_xattr) and the upstream name (ea_inode).
large_xattr_enabled() {
- [[ $(facet_fstype $SINGLEMDS) == zfs ]] && return 1
+ [[ $(facet_fstype $SINGLEMDS) == zfs ]] && return 0
local mds_dev=$(mdsdevname ${SINGLEMDS//mds/})
local devname=$(mdsdevname $(facet_number $facet))
local mntpt=$(facet_mntpt brpt)
local rcmd="do_facet $facet"
- local metaea=${TMP}/backup_restore.ea
local metadata=${TMP}/backup_restore.tgz
local opts=${MDS_MOUNT_FS_OPTS}
local svc=${facet}_svc
# step 1: build mount point
${rcmd} mkdir -p $mntpt
# step 2: cleanup old backup
- ${rcmd} rm -f $metaea $metadata
+ ${rcmd} rm -f $metadata
# step 3: mount dev
- ${rcmd} mount -t ldiskfs $opts $devname $mntpt || return 1
+ ${rcmd} mount -t ldiskfs $opts $devname $mntpt || return 3
if [ ! -z $igif ]; then
# step 3.5: rm .lustre
- ${rcmd} rm -rf $mntpt/ROOT/.lustre || return 1
+ ${rcmd} rm -rf $mntpt/ROOT/.lustre || return 3
fi
- # step 4: backup metaea
- echo "backup EA"
- ${rcmd} "cd $mntpt && getfattr -R -d -m '.*' -P . > $metaea && cd -" ||
- return 2
- # step 5: backup metadata
+ # step 4: backup metadata
echo "backup data"
- ${rcmd} tar zcf $metadata -C $mntpt/ . > /dev/null 2>&1 || return 3
- # step 6: umount
- ${rcmd} $UMOUNT $mntpt || return 4
- # step 8: reformat dev
+ ${rcmd} tar zcf $metadata --xattrs --xattrs-include="trusted.*" \
+ --sparse -C $mntpt/ . > /dev/null 2>&1 || return 4
+ # step 5: umount
+ ${rcmd} $UMOUNT $mntpt || return 5
+ # step 6: reformat dev
echo "reformat new device"
format_mdt $(facet_number $facet)
- # step 9: mount dev
+ # step 7: mount dev
${rcmd} mount -t ldiskfs $opts $devname $mntpt || return 7
- # step 10: restore metadata
+ # step 8: restore metadata
echo "restore data"
- ${rcmd} tar zxfp $metadata -C $mntpt > /dev/null 2>&1 || return 8
- # step 11: restore metaea
- echo "restore EA"
- ${rcmd} "cd $mntpt && setfattr --restore=$metaea && cd - " || return 9
- # step 12: remove recovery logs
+ ${rcmd} tar zxfp $metadata --xattrs --xattrs-include="trusted.*" \
+ --sparse -C $mntpt > /dev/null 2>&1 || return 8
+ # step 9: remove recovery logs
echo "remove recovery logs"
${rcmd} rm -fv $mntpt/OBJECTS/* $mntpt/CATALOGS
- # step 13: umount dev
+ # step 10: umount dev
${rcmd} $UMOUNT $mntpt || return 10
- # step 14: cleanup tmp backup
+ # step 11: cleanup tmp backup
${rcmd} rm -f $metaea $metadata
- # step 15: reset device label - it's not virgin on
+ # step 12: reset device label - it's not virgin on
${rcmd} e2label $devname ${!svc}
}
fi
echo "waited $((i - 1)) seconds for sync"
}
+
+consume_precreations() {
+ local dir=$1
+ local mfacet=$2
+ local OSTIDX=$3
+ local extra=${4:-2}
+ local OST=$(ostname_from_index $OSTIDX $dir)
+
+ test_mkdir -p $dir/${OST}
+ $LFS setstripe -i $OSTIDX -c 1 ${dir}/${OST}
+
+ # on the mdt's osc
+ local mdtosc_proc=$(get_mdtosc_proc_path $mfacet $OST)
+ local last_id=$(do_facet $mfacet $LCTL get_param -n \
+ osp.$mdtosc_proc.prealloc_last_id)
+ local next_id=$(do_facet $mfacet $LCTL get_param -n \
+ osp.$mdtosc_proc.prealloc_next_id)
+ echo "Creating to objid $last_id on ost $OST..."
+ createmany -o $dir/${OST}/f $next_id $((last_id - next_id + extra))
+}
+
+__exhaust_precreations() {
+ local OSTIDX=$1
+ local FAILLOC=$2
+ local FAILIDX=${3:-$OSTIDX}
+ local ofacet=ost$((OSTIDX + 1))
+
+ mkdir_on_mdt0 $DIR/$tdir
+ local mdtidx=$($LFS getstripe -m $DIR/$tdir)
+ local mfacet=mds$((mdtidx + 1))
+ echo OSTIDX=$OSTIDX MDTIDX=$mdtidx
+
+ local mdtosc_proc=$(get_mdtosc_proc_path $mfacet)
+ do_facet $mfacet $LCTL get_param osp.$mdtosc_proc.prealloc*
+
+#define OBD_FAIL_OST_ENOSPC 0x215
+ do_facet $ofacet $LCTL set_param fail_val=$FAILIDX fail_loc=0x215
+
+ consume_precreations $DIR/$tdir $mfacet $OSTIDX
+
+ do_facet $mfacet $LCTL get_param osp.$mdtosc_proc.prealloc*
+ do_facet $ofacet $LCTL set_param fail_loc=$FAILLOC
+}
+
+exhaust_precreations() {
+ __exhaust_precreations $1 $2 $3
+ sleep_maxage
+}
+
+exhaust_all_precreations() {
+ local i
+ for (( i=0; i < OSTCOUNT; i++ )) ; do
+ __exhaust_precreations $i $1 -1
+ done
+ sleep_maxage
+}
git://git.whamcloud.com / fs / lustre-release.git / blobdiff