X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Ftests%2Ftest-framework.sh;h=ff194e575b06b69df601b9a3adb95616be529f44;hp=c2962498e779ab03dec01b78675559179cf7c9d2;hb=8ad0cb4fdc1cdf876add1d40512112d75af88e06;hpb=df669064eeedea424c3a18f74c9c14281f76ceb3 diff --git a/lustre/tests/test-framework.sh b/lustre/tests/test-framework.sh index c296249..ff194e5 100644 --- a/lustre/tests/test-framework.sh +++ b/lustre/tests/test-framework.sh @@ -15,8 +15,8 @@ export GSS_KRB5=false export GSS_PIPEFS=false export IDENTITY_UPCALL=default export QUOTA_AUTO=1 -export JOBSTATS_AUTO=${JOBSTATS_AUTO:-1} -export JOBID_VAR=${JOBID_VAR:-"procname_uid"} +# specify environment variable containing batch job name for server statistics +export JOBID_VAR=${JOBID_VAR:-"procname_uid"} # or "existing" or "disable" # LOAD_LLOOP: LU-409: only load llite_lloop module if kernel < 2.6.32 or # LOAD_LLOOP is true. LOAD_LLOOP is false by default. @@ -118,19 +118,67 @@ print_summary () { } init_test_env() { - 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} - export TUNE2FS=${TUNE2FS:-tune2fs} - export E2LABEL=${E2LABEL:-e2label} - export DUMPE2FS=${DUMPE2FS:-dumpe2fs} - export E2FSCK=${E2FSCK:-e2fsck} - export LFSCK_BIN=${LFSCK_BIN:-lfsck} + export LUSTRE=$(absolute_path $LUSTRE) + export TESTSUITE=$(basename $0 .sh) + export TEST_FAILED=false + export FAIL_ON_SKIP_ENV=${FAIL_ON_SKIP_ENV:-false} + export RPC_MODE=${RPC_MODE:-false} + + export MKE2FS=$MKE2FS + if [ -z "$MKE2FS" ]; then + if which mkfs.ldiskfs >/dev/null 2>&1; then + export MKE2FS=mkfs.ldiskfs + else + export MKE2FS=mke2fs + fi + fi + export DEBUGFS=$DEBUGFS + if [ -z "$DEBUGFS" ]; then + if which debugfs.ldiskfs >/dev/null 2>&1; then + export DEBUGFS=debugfs.ldiskfs + else + export DEBUGFS=debugfs + fi + fi + + export TUNE2FS=$TUNE2FS + if [ -z "$TUNE2FS" ]; then + if which tunefs.ldiskfs >/dev/null 2>&1; then + export TUNE2FS=tunefs.ldiskfs + else + export TUNE2FS=tune2fs + fi + fi + + export E2LABEL=$E2LABEL + if [ -z "$E2LABEL" ]; then + if which label.ldiskfs >/dev/null 2>&1; then + export E2LABEL=label.ldiskfs + else + export E2LABEL=e2label + fi + fi + + export DUMPE2FS=$DUMPE2FS + if [ -z "$DUMPE2FS" ]; then + if which dumpfs.ldiskfs >/dev/null 2>&1; then + export DUMPE2FS=dumpfs.ldiskfs + else + export DUMPE2FS=dumpe2fs + fi + fi + + export E2FSCK=$E2FSCK + if [ -z "$E2FSCK" ]; then + if which fsck.ldiskfs >/dev/null 2>&1; then + export E2FSCK=fsck.ldiskfs + else + export E2FSCK=e2fsck + fi + fi + + export LFSCK_BIN=${LFSCK_BIN:-lfsck} export LFSCK_ALWAYS=${LFSCK_ALWAYS:-"no"} # check fs after each test suite export FSCK_MAX_ERR=4 # File system errors left uncorrected @@ -190,11 +238,11 @@ init_test_env() { fi fi export LL_DECODE_FILTER_FID=${LL_DECODE_FILTER_FID:-"$LUSTRE/utils/ll_decode_filter_fid"} - [ ! -f "$LL_DECODE_FILTER_FID" ] && export LL_DECODE_FILTER_FID=$(which ll_decode_filter_fid) + [ ! -f "$LL_DECODE_FILTER_FID" ] && export LL_DECODE_FILTER_FID="ll_decode_filter_fid" export MKFS=${MKFS:-"$LUSTRE/utils/mkfs.lustre"} - [ ! -f "$MKFS" ] && export MKFS=$(which mkfs.lustre) + [ ! -f "$MKFS" ] && export MKFS="mkfs.lustre" export TUNEFS=${TUNEFS:-"$LUSTRE/utils/tunefs.lustre"} - [ ! -f "$TUNEFS" ] && export TUNEFS=$(which tunefs.lustre) + [ ! -f "$TUNEFS" ] && export TUNEFS="tunefs.lustre" export CHECKSTAT="${CHECKSTAT:-"checkstat -v"} " export LUSTRE_RMMOD=${LUSTRE_RMMOD:-$LUSTRE/scripts/lustre_rmmod} [ ! -f "$LUSTRE_RMMOD" ] && @@ -235,19 +283,21 @@ init_test_env() { 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/$(uname -r)/kernel/fs/lustre/mds.ko -a \ - ! -f /lib/modules/$(uname -r)/updates/kernel/fs/lustre/mds.ko -a \ - ! -f `dirname $0`/../mds/mds.ko ]; then + if [ ! -f /lib/modules/$(uname -r)/kernel/fs/lustre/mdt.ko -a \ + ! -f /lib/modules/$(uname -r)/updates/kernel/fs/lustre/mdt.ko -a \ + ! -f `dirname $0`/../mdt/mdt.ko ]; then export CLIENTMODSONLY=yes fi export SHUTDOWN_ATTEMPTS=${SHUTDOWN_ATTEMPTS:-3} + export OSD_TRACK_DECLARES_LBUG=${OSD_TRACK_DECLARES_LBUG:-"yes"} # command line @@ -264,10 +314,12 @@ init_test_env() { shift $((OPTIND - 1)) ONLY=${ONLY:-$*} - # print the durations of each test if "true" - DDETAILS=${DDETAILS:-false} - [ "$TESTSUITELOG" ] && rm -f $TESTSUITELOG || true - rm -f $TMP/*active + # print the durations of each test if "true" + DDETAILS=${DDETAILS:-false} + [ "$TESTSUITELOG" ] && rm -f $TESTSUITELOG || true + if ! $RPC_MODE; then + rm -f $TMP/*active + fi } check_cpt_number() { @@ -288,7 +340,7 @@ version_code() { # split arguments like "1.8.6-wc3" into "1", "8", "6", "wc3" eval set -- $(tr "[:punct:]" " " <<< $*) - echo -n $((($1 << 16) | ($2 << 8) | $3)) + echo -n "$((($1 << 16) | ($2 << 8) | $3))" } export LINUX_VERSION=$(uname -r | sed -e "s/[-.]/ /3" -e "s/ .*//") @@ -323,20 +375,24 @@ load_module() { optvar="MODOPTS_$(basename $module | tr a-z A-Z)" eval set -- \$$optvar if [ $# -eq 0 -a -n "$MODPROBECONF" ]; then - # Nothing in $MODOPTS_; try modprobe.conf - set -- $(grep -P "^options\\s+${BASE}" $MODPROBECONF) - # Get rid of "options $module" - (($# > 0)) && shift 2 - - # Ensure we have accept=all for lnet - if [ $(basename $module) = lnet ]; then - # OK, this is a bit wordy... - local arg accept_all_present=false - for arg in "$@"; do - [ "$arg" = accept=all ] && accept_all_present=true - done - $accept_all_present || set -- "$@" accept=all - fi + # Nothing in $MODOPTS_; try modprobe.conf + local opt + opt=$(awk -v var="^options $BASE" '$0 ~ var \ + {gsub("'"options $BASE"'",""); print}' $MODPROBECONF) + set -- $(echo -n $opt) + + # Ensure we have accept=all for lnet + if [ $(basename $module) = lnet ]; then + # OK, this is a bit wordy... + local arg accept_all_present=false + + for arg in "$@"; do + [ "$arg" = accept=all ] && \ + accept_all_present=true + done + $accept_all_present || set -- "$@" accept=all + fi + export $optvar="$*" fi fi @@ -375,22 +431,6 @@ llite_lloop_enabled() { } load_modules_local() { - if [ "$USE_OFD" == yes ]; then - if module_loaded obdfilter; then - if ! $LUSTRE_RMMOD ldiskfs; then - echo "$HOSTNAME may still be using obdfilter.ko" - return 1 - fi - fi - else - if module_loaded ofd; then - if ! $LUSTRE_RMMOD ldiskfs; then - echo "$HOSTNAME may still be using ofd.ko" - return 1 - fi - fi - fi - if [ -n "$MODPROBE" ]; then # use modprobe echo "Using modprobe to load modules" @@ -415,7 +455,6 @@ load_modules_local() { fi fi - local ncpts=0 # if there is only one CPU core, libcfs can only create one partition # if there is more than 4 CPU cores, libcfs should create multiple CPU # partitions. So we just force libcfs to create 2 partitions for @@ -423,12 +462,12 @@ load_modules_local() { if [ $ncpus -le 4 ] && [ $ncpus -gt 1 ]; then # force to enable multiple CPU partitions echo "Force libcfs to create 2 CPU partitions" - ncpts=2 + MODOPTS_LIBCFS="cpu_npartitions=2 $MODOPTS_LIBCFS" else echo "libcfs will create CPU partition based on online CPUs" fi - load_module ../libcfs/libcfs/libcfs cpu_npartitions=$ncpts + load_module ../libcfs/libcfs/libcfs [ "$PTLDEBUG" ] && lctl set_param debug="$PTLDEBUG" [ "$SUBSYSTEM" ] && lctl set_param subsystem_debug="${SUBSYSTEM# }" @@ -454,10 +493,10 @@ load_modules_local() { grep -q -w jbd2 $SYMLIST || { modprobe jbd2 2>/dev/null || true; } [ "$LQUOTA" != "no" ] && load_module quota/lquota $LQUOTAOPTS if [[ $(node_fstypes $HOSTNAME) == *zfs* ]]; then + modprobe zfs load_module osd-zfs/osd_zfs fi load_module mgs/mgs - load_module mds/mds load_module mdd/mdd if [[ $(node_fstypes $HOSTNAME) == *ldiskfs* ]]; then # @@ -471,13 +510,11 @@ load_modules_local() { load_module osd-ldiskfs/osd_ldiskfs fi load_module mdt/mdt - load_module cmm/cmm load_module ost/ost - if [ "$USE_OFD" == yes ]; then - load_module ofd/ofd - else - load_module obdfilter/obdfilter - fi + load_module lod/lod + load_module osp/osp + load_module ofd/ofd + load_module osp/osp fi @@ -501,10 +538,12 @@ load_modules () { # bug 19124 # load modules on remote nodes optionally # lustre-tests have to be installed on these nodes - if $LOAD_MODULES_REMOTE ; then + if $LOAD_MODULES_REMOTE; then local list=$(comma_list $(remote_nodes_list)) - echo loading modules on $list - do_rpc_nodes $list load_modules_local + if [ -n "$list" ]; then + echo "loading modules on: '$list'" + do_rpc_nodes "$list" load_modules_local + fi fi } @@ -526,12 +565,12 @@ unload_modules() { $LUSTRE_RMMOD ldiskfs || return 2 - if $LOAD_MODULES_REMOTE ; then + if $LOAD_MODULES_REMOTE; then local list=$(comma_list $(remote_nodes_list)) - if [ ! -z $list ]; then - echo unloading modules on $list - do_rpc_nodes $list $LUSTRE_RMMOD ldiskfs - do_rpc_nodes $list check_mem_leak + if [ -n "$list" ]; then + echo "unloading modules on: '$list'" + do_rpc_nodes "$list" $LUSTRE_RMMOD ldiskfs + do_rpc_nodes "$list" check_mem_leak fi fi @@ -547,6 +586,18 @@ unload_modules() { return 0 } +fs_log_size() { + local facet=${1:-$SINGLEMDS} + local fstype=$(facet_fstype $facet) + local size=0 + case $fstype in + ldiskfs) size=50;; # largest seen is 44, leave some headroom + zfs) size=256;; + esac + + echo -n $size +} + check_gss_daemon_nodes() { local list=$1 dname=$2 @@ -662,7 +713,7 @@ cleanup_gss() { facet_type() { local facet=$1 - echo -n $facet | sed -e 's/^fs[0-9]\+//' -e 's/[0-9]\+//' | + echo -n $facet | sed -e 's/^fs[0-9]\+//' -e 's/[0-9_]\+//' | tr '[:lower:]' '[:upper:]' } @@ -697,6 +748,11 @@ facet_fstype() { return fi + if [[ $facet == mgs ]] && combined_mgs_mds; then + facet_fstype mds1 + return + fi + return 1 } @@ -753,7 +809,7 @@ ostdevlabel() { } # -# This and set_obdfilter_param() shall be used to access OSD parameters +# This and set_osd_param() shall be used to access OSD parameters # once existed under "obdfilter": # # mntdev @@ -761,7 +817,7 @@ ostdevlabel() { # read_cache_enable # writethrough_cache_enable # -get_obdfilter_param() { +get_osd_param() { local nodes=$1 local device=${2:-$FSNAME-OST*} local name=$3 @@ -770,7 +826,7 @@ get_obdfilter_param() { osd-*.$device.$name 2>&1" | grep -v 'Found no match' } -set_obdfilter_param() { +set_osd_param() { local nodes=$1 local device=${2:-$FSNAME-OST*} local name=$3 @@ -809,15 +865,15 @@ set_default_debug () { } set_default_debug_nodes () { - local nodes=$1 + local nodes="$1" - if [[ ,$nodes, = *,$HOSTNAME,* ]]; then - nodes=$(exclude_items_from_list "$nodes" "$HOSTNAME") - set_default_debug - fi + if [[ ,$nodes, = *,$HOSTNAME,* ]]; then + nodes=$(exclude_items_from_list "$nodes" "$HOSTNAME") + set_default_debug + fi - [[ -n $nodes ]] && do_rpc_nodes $nodes set_default_debug \ - \\\"$PTLDEBUG\\\" \\\"$SUBSYSTEM\\\" $DEBUG_SIZE || true + do_rpc_nodes "$nodes" set_default_debug \ + \\\"$PTLDEBUG\\\" \\\"$SUBSYSTEM\\\" $DEBUG_SIZE || true } set_default_debug_facet () { @@ -980,6 +1036,10 @@ stop() { # save quota version (both administrative and operational quotas) # add an additional parameter if mountpoint is ever different from $MOUNT +# +# XXX This function is kept for interoperability with old server (< 2.3.50), +# it should be removed whenever we drop the interoperability for such +# server. quota_save_version() { local fsname=${2:-$FSNAME} local spec=$1 @@ -999,32 +1059,42 @@ quota_save_version() { done } -# client could mount several lustre -quota_type () { - local fsname=${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 $rc +# client could mount several lustre +# +# XXX This function is kept for interoperability with old server (< 2.3.50), +# it should be removed whenever we drop the interoperability for such +# server. +quota_type() { + local fsname=${1:-$FSNAME} + local rc=0 + do_facet $SINGLEMDS lctl get_param mdd.${fsname}-MDT*.quota_type || + rc=$? + do_nodes $(comma_list $(osts_nodes)) \ + lctl get_param obdfilter.${fsname}-OST*.quota_type || rc=$? + return $rc +} + +# XXX This function is kept for interoperability with old server (< 2.3.50), +# it should be removed whenever we drop the interoperability for such +# server. +restore_quota_old() { + local mntpt=${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 $old_QUOTA_TYPE } -restore_quota_type () { - local mntpt=${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 -} +# XXX This function is kept for interoperability with old server (< 2.3.50), +# it should be removed whenever we drop the interoperability for such +# server. +setup_quota_old(){ + local mntpt=$1 -setup_quota(){ - local mntpt=$1 - - if [ "$USE_OFD" = "yes" ]; then - $LFS quotacheck $mntpt || error "quotacheck failed" - return - fi + # no quota enforcement for now and accounting works out of the box + return # We need save the original quota_type params, and restore them after testing @@ -1043,11 +1113,11 @@ setup_quota(){ local quota_usrs=$QUOTA_USERS # get_filesystem_size - local disksz=$(lfs df $mntpt | grep "filesystem summary:" | awk '{print $3}') + local disksz=$(lfs_df $mntpt | grep "summary" | awk '{print $2}') local blk_soft=$((disksz + 1024)) local blk_hard=$((blk_soft + blk_soft / 20)) # Go 5% over - local Inodes=$(lfs df -i $mntpt | grep "filesystem summary:" | awk '{print $3}') + local Inodes=$(lfs_df -i $mntpt | grep "summary" | awk '{print $2}') local i_soft=$Inodes local i_hard=$((i_soft + i_soft / 20)) @@ -1068,6 +1138,98 @@ setup_quota(){ done } +# get mdt quota type +mdt_quota_type() { + local varsvc=${SINGLEMDS}_svc + do_facet $SINGLEMDS $LCTL get_param -n \ + osd-$(facet_fstype $SINGLEMDS).${!varsvc}.quota_slave.enabled +} + +# get ost quota type +ost_quota_type() { + # All OSTs should have same quota type + local varsvc=ost1_svc + do_facet ost1 $LCTL get_param -n \ + osd-$(facet_fstype ost1).${!varsvc}.quota_slave.enabled +} + +# restore old quota type settings +restore_quota() { + if [ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.3.50) ]; then + restore_quota_old + return + fi + + if [ "$old_MDT_QUOTA_TYPE" ]; then + do_facet mgs $LCTL conf_param \ + $FSNAME.quota.mdt=$old_MDT_QUOTA_TYPE + fi + if [ "$old_OST_QUOTA_TYPE" ]; then + do_facet mgs $LCTL conf_param \ + $FSNAME.quota.ost=$old_OST_QUOTA_TYPE + fi +} + +# Handle the case when there is a space in the lfs df +# "filesystem summary" line the same as when there is no space. +# This will allow fixing the "lfs df" summary line in the future. +lfs_df() { + $LFS df $* | sed -e 's/filesystem /filesystem_/' +} + +setup_quota(){ + if [ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.3.50) ]; then + setup_quota_old $1 + return + fi + + local mntpt=$1 + + # save old quota type & set new quota type + local mdt_qtype=$(mdt_quota_type) + local ost_qtype=$(ost_quota_type) + + echo "[HOST:$HOSTNAME] [old_mdt_qtype:$mdt_qtype]" \ + "[old_ost_qtype:$ost_qtype] [new_qtype:$QUOTA_TYPE]" + + export old_MDT_QUOTA_TYPE=$mdt_qtype + export old_OST_QUOTA_TYPE=$ost_qtype + + do_facet mgs $LCTL conf_param $FSNAME.quota.mdt=$QUOTA_TYPE || + error "set mdt quota type failed" + do_facet mgs $LCTL conf_param $FSNAME.quota.ost=$QUOTA_TYPE || + error "set ost quota type failed" + + local quota_usrs=$QUOTA_USERS + + # get_filesystem_size + local disksz=$(lfs_df $mntpt | grep "summary" | awk '{print $2}') + local blk_soft=$((disksz + 1024)) + local blk_hard=$((blk_soft + blk_soft / 20)) # Go 5% over + + local inodes=$(lfs_df -i $mntpt | grep "summary" | awk '{print $2}') + local i_soft=$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 $blk_soft" + cmd="$cmd -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 client=$1 local mnt=$2 @@ -1277,12 +1439,12 @@ facets_on_host () { echo $(comma_list $affected) } -facet_up () { - local facet=$1 - local host=${2:-$(facet_host $facet)} +facet_up() { + local facet=$1 + local host=${2:-$(facet_host $facet)} - local label=$(convert_facet2label $facet) - do_node $host lctl dl | awk '{print $4}' | grep -q $label + local label=$(convert_facet2label $facet) + do_node $host $LCTL dl | awk '{print $4}' | grep -q -x $label } facets_up_on_host () { @@ -1365,10 +1527,10 @@ _check_progs_installed () { } check_progs_installed () { - local nodes=$1 - shift + local nodes=$1 + shift - do_rpc_nodes $nodes _check_progs_installed $@ + do_rpc_nodes "$nodes" _check_progs_installed $@ } # recovery-scale functions @@ -1390,6 +1552,7 @@ TESTLOG_PREFIX=$TESTLOG_PREFIX \ TESTNAME=$TESTNAME \ DBENCH_LIB=$DBENCH_LIB \ DBENCH_SRC=$DBENCH_SRC \ +LFS=$LFS \ run_${load}.sh" & local ppid=$! log "Started client load: ${load} on $client" @@ -1585,15 +1748,16 @@ cleanup_check() { check_mem_leak || exit 204 - [ "`lctl dl 2> /dev/null | wc -l`" -gt 0 ] && lctl dl && \ - echo "$0: lustre didn't clean up..." 1>&2 && return 202 || true + [ "`lctl dl 2> /dev/null | wc -l`" -gt 0 ] && lctl dl && + echo "$TESTSUITE: lustre didn't clean up..." 1>&2 && + return 202 || true - if module_loaded lnet || module_loaded libcfs; then - echo "$0: modules still loaded..." 1>&2 - /sbin/lsmod 1>&2 - return 203 - fi - return 0 + if module_loaded lnet || module_loaded libcfs; then + echo "$TESTSUITE: modules still loaded..." 1>&2 + /sbin/lsmod 1>&2 + return 203 + fi + return 0 } wait_update () { @@ -1623,33 +1787,66 @@ wait_update () { return 3 } -wait_update_facet () { - local facet=$1 - wait_update $(facet_active_host $facet) "$@" +wait_update_facet() { + local facet=$1 + shift + wait_update $(facet_active_host $facet) "$@" } sync_all_data() { - do_node $(osts_nodes) "lctl set_param -n osd*.*OS*.force_sync 1" 2>&1 | + do_nodes $(comma_list $(mdts_nodes)) \ + "lctl set_param -n osd*.*MDT*.force_sync 1" + do_nodes $(comma_list $(osts_nodes)) \ + "lctl set_param -n osd*.*OS*.force_sync 1" 2>&1 | grep -v 'Found no match' } -wait_delete_completed () { - local TOTALPREV=`lctl get_param -n osc.*.kbytesavail | \ - awk 'BEGIN{total=0}; {total+=$1}; END{print total}'` +wait_delete_completed_mds() { + local MAX_WAIT=${1:-20} + local mds2sync="" + local stime=`date +%s` + local etime + local node + local changes + + # find MDS with pending deletions + for node in $(mdts_nodes); do + changes=$(do_node $node "lctl get_param -n osc.*MDT*.sync_*" \ + 2>/dev/null | calc_sum) + if [ -z "$changes" ] || [ $changes -eq 0 ]; then + continue + fi + mds2sync="$mds2sync $node" + done + if [ "$mds2sync" == "" ]; then + return + fi + mds2sync=$(comma_list $mds2sync) + + # sync MDS transactions + do_nodes $mds2sync "lctl set_param -n osd*.*MD*.force_sync 1" + + # wait till all changes are sent and commmitted by OSTs + # for ldiskfs space is released upon execution, but DMU + # do this upon commit + + local WAIT=0 + while [ "$WAIT" -ne "$MAX_WAIT" ]; do + changes=$(do_nodes $mds2sync "lctl get_param -n osc.*MDT*.sync_*" \ + | calc_sum) + #echo "$node: $changes changes on all" + if [ "$changes" -eq "0" ]; then + etime=`date +%s` + #echo "delete took $((etime - stime)) seconds" + return + fi + sleep 1 + WAIT=$(( WAIT + 1)) + done - local WAIT=0 - local MAX_WAIT=20 - while [ "$WAIT" -ne "$MAX_WAIT" ]; do - sleep 1 - TOTAL=`lctl get_param -n osc.*.kbytesavail | \ - awk 'BEGIN{total=0}; {total+=$1}; END{print total}'` - [ "$TOTAL" -eq "$TOTALPREV" ] && return 0 - echo "Waiting delete completed ... prev: $TOTALPREV current: $TOTAL " - TOTALPREV=$TOTAL - WAIT=$(( WAIT + 1)) - done - echo "Delete is not completed in $MAX_WAIT sec" - return 1 + etime=`date +%s` + echo "Delete is not completed in $((etime - stime)) seconds" + do_nodes $mds2sync "lctl get_param osc.*MDT*.sync_*" } wait_for_host() { @@ -1705,66 +1902,87 @@ wait_recovery_complete () { 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" + # 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 $MAX - done + local host=$(facet_active_host $facet) + do_rpc_nodes "$host" _wait_recovery_complete $param $MAX + done } wait_mds_ost_sync () { - # just because recovery is done doesn't mean we've finished - # orphan cleanup. Wait for llogs to get synchronized. - echo "Waiting for orphan cleanup..." - # MAX value includes time needed for MDS-OST reconnection - local MAX=$(( 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 - local i - 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. $STATUS" - return 1 + # just because recovery is done doesn't mean we've finished + # orphan cleanup. Wait for llogs to get synchronized. + echo "Waiting for orphan cleanup..." + # MAX value includes time needed for MDS-OST reconnection + local MAX=$(( TIMEOUT * 2 )) + local WAIT=0 + local new_wait=true + local list=$(comma_list $(mdts_nodes)) + local cmd="$LCTL get_param -n osp.*osc*.old_sync_processed" + if ! do_facet $SINGLEMDS \ + "$LCTL list_param osp.*osc*.old_sync_processed 2> /dev/null" + then + # old way, use mds_sync + new_wait=false + list=$(comma_list $(osts_nodes)) + cmd="$LCTL get_param -n obdfilter.*.mds_sync" + fi + while [ $WAIT -lt $MAX ]; do + local -a sync=($(do_nodes $list "$cmd")) + local con=1 + local i + for ((i=0; i<${#sync[@]}; i++)); do + if $new_wait; then + [ ${sync[$i]} -eq 1 ] && continue + else + [ ${sync[$i]} -eq 0 ] && continue + fi + # 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 + echo "Waiting for local destroys to complete" + # 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 $MAX ]; do + local -a RPCs=($($LCTL get_param -n osc.*.destroys_in_flight)) + local con=1 + local i + + for ((i=0; i<${#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 ${WAIT}s for local destroys to complete" + WAIT=$((WAIT + 1)) + done + echo "Local destroys weren't done in $MAX sec." + return 1 +} + +wait_delete_completed() { + wait_delete_completed_mds $1 || return $? + wait_destroy_complete } wait_exit_ST () { @@ -1871,32 +2089,58 @@ affected_facets () { } facet_failover() { - local facet=$1 - local sleep_time=$2 - local host=$(facet_active_host $facet) - - echo "Failing $facet on node $host" - - local affected=$(affected_facets $facet) + local facets=$1 + local sleep_time=$2 + local -a affecteds + local facet + local total=0 + local index=0 + local skip + + #Because it will only get up facets, we need get affected + #facets before shutdown + #For HARD Failure mode, it needs make sure facets on the same + #HOST will only be shutdown and reboot once + for facet in ${facets//,/ }; do + local affected_facet + skip=0 + #check whether facet has been included in other affected facets + for ((index=0; index<$total; index++)); do + [[ *,$facet,* == ,${affecteds[index]}, ]] && skip=1 + done - shutdown_facet $facet + if [ $skip -eq 0 ]; then + affecteds[$total]=$(affected_facets $facet) + total=$((total+1)) + fi + done - echo affected facets: $affected + for ((index=0; index<$total; index++)); do + facet=$(echo ${affecteds[index]} | tr -s " " | cut -d"," -f 1) + local host=$(facet_active_host $facet) + echo "Failing ${affecteds[index]} on $host" + shutdown_facet $facet + done - [ -n "$sleep_time" ] && sleep $sleep_time + for ((index=0; index<$total; index++)); do + facet=$(echo ${affecteds[index]} | tr -s " " | cut -d"," -f 1) + echo reboot facets: ${affecteds[index]} - reboot_facet $facet + reboot_facet $facet - change_active $affected + change_active ${affecteds[index]} - 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) - mount_facets $affected + wait_for_facet ${affecteds[index]} + # start mgs first if it is affected + if ! combined_mgs_mds && + list_member ${affecteds[index]} mgs; then + mount_facet mgs || error "Restart of mgs failed" + fi + # FIXME; has to be changed to mount all facets concurrently + affected=$(exclude_items_from_list ${affecteds[index]} mgs) + echo mount facets: ${affecteds[index]} + mount_facets ${affecteds[index]} + done } obd_name() { @@ -1904,40 +2148,42 @@ obd_name() { } replay_barrier() { - local facet=$1 - do_facet $facet "sync; sync; sync" - df $MOUNT + local facet=$1 + do_facet $facet "sync; sync; sync" + df $MOUNT - # make sure there will be no seq change - local clients=${CLIENTS:-$HOSTNAME} - do_nodes $clients "f=${MOUNT}/fsa-\\\$(hostname); mcreate \\\$f; rm \\\$f" + # make sure there will be no seq change + local clients=${CLIENTS:-$HOSTNAME} + local f=fsa-\\\$\(hostname\) + do_nodes $clients "mcreate $MOUNT/$f; rm $MOUNT/$f" + do_nodes $clients "if [ -d $MOUNT2 ]; then mcreate $MOUNT2/$f; rm $MOUNT2/$f; fi" - local svc=${facet}_svc - 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 ${!svc}" + local svc=${facet}_svc + 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}" } replay_barrier_nodf() { - local facet=$1 echo running=${running} - do_facet $facet "sync; sync; sync" - local svc=${facet}_svc - echo Replay barrier on ${!svc} - 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} + do_facet $facet "sync; sync; sync" + local svc=${facet}_svc + echo Replay barrier on ${!svc} + 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}" } replay_barrier_nosync() { - local facet=$1 echo running=${running} - local svc=${facet}_svc - echo Replay barrier on ${!svc} - 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} 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}" } mds_evict_client() { @@ -1951,8 +2197,12 @@ ost_evict_client() { } fail() { - facet_failover $* || error "failover: $?" - clients_up || error "post-failover df: $?" + local facets=$1 + local clients=${CLIENTS:-$HOSTNAME} + + facet_failover $* || error "failover: $?" + wait_clients_import_state "$clients" "$facets" FULL + clients_up || error "post-failover df: $?" } fail_nodf() { @@ -2118,21 +2368,26 @@ hostlist_expand() { } facet_host() { - local facet=$1 - - [ "$facet" == client ] && echo -n $HOSTNAME && return - varname=${facet}_HOST - if [ -z "${!varname}" ]; then - if [ "${facet:0:3}" == "ost" ]; then - eval ${facet}_HOST=${ost_HOST} - fi - fi - echo -n ${!varname} + local facet=$1 + local varname + + [ "$facet" == client ] && echo -n $HOSTNAME && return + varname=${facet}_HOST + if [ -z "${!varname}" ]; then + if [ "${facet:0:3}" == "ost" ]; then + eval export ${facet}_HOST=${ost_HOST} + elif [ "${facet:0:3}" == "mdt" -o \ + "${facet:0:3}" == "mds" -o \ + "${facet:0:3}" == "mgs" ]; then + eval export ${facet}_HOST=${mds_HOST} + fi + fi + echo -n ${!varname} } facet_failover_host() { local facet=$1 - local var + local varname var=${facet}failover_HOST if [ -n "${!var}" ]; then @@ -2140,12 +2395,18 @@ facet_failover_host() { return fi + if [ "${facet:0:3}" == "mdt" -o "${facet:0:3}" == "mds" -o \ + "${facet:0:3}" == "mgs" ]; then + + eval export ${facet}failover_host=${mds_HOST} + echo ${mds_HOST} + return + fi + if [[ $facet == ost* ]]; then - var=ostfailover_HOST - if [ -n "${!var}" ]; then - echo ${!var} - return - fi + eval export ${facet}failover_host=${ost_HOST} + echo ${ost_HOST} + return fi } @@ -2175,6 +2436,24 @@ facet_active_host() { fi } +# Get the passive failover partner host of facet. +facet_passive_host() { + local facet=$1 + [[ $facet = client ]] && return + + local host=${facet}_HOST + local failover_host=${facet}failover_HOST + local active_host=$(facet_active_host $facet) + + [[ -z ${!failover_host} || ${!failover_host} = ${!host} ]] && return + + if [[ $active_host = ${!host} ]]; then + echo -n ${!failover_host} + else + echo -n ${!host} + fi +} + change_active() { local facetlist=$1 local facet @@ -2268,8 +2547,6 @@ get_env_vars() { echo -n " ${var}=\"$value\"" done - echo -n " USE_OFD=$USE_OFD" - for facet in ${facets//,/ }; do var=${facet}_FSTYPE if [ -n "${!var}" ]; then @@ -2452,38 +2729,50 @@ mdsvdevname() { } mgsdevname() { - DEVNAME=MGSDEV - - local fstype=$(facet_fstype mds$num) + local DEVPTR + local fstype=$(facet_fstype mgs) case $fstype in - ldiskfs ) - #if $MGSDEV isn't defined, default is $MDSDEV1 - eval DEVPTR=${!DEVNAME:=${MDSDEV1}};; - zfs ) - #dataset name is independent of vdev device names - eval DEVPTR=${FSNAME}-mgs/mgs;; - * ) - error "unknown fstype!";; + ldiskfs ) + if [ $(facet_host mgs) = $(facet_host mds1) ] && + ( [ -z "$MGSDEV" ] || [ $MGSDEV = $(mdsdevname 1) ] ); then + DEVPTR=$(mdsdevname 1) + else + DEVPTR=$MGSDEV + fi;; + zfs ) + if [ $(facet_host mgs) = $(facet_host mds1) ] && + ( [ -z "$MGSDEV" ] || [ $MGSDEV = $(mdsvdevname 1) ] ); then + DEVPTR=$(mdsdevname 1) + else + DEVPTR=${FSNAME}-mgs/mgs + fi;; + * ) + error "unknown fstype!";; esac - echo -n $DEVPTR + echo -n $DEVPTR } mgsvdevname() { + local VDEVPTR DEVNAME=MGSDEV - local fstype=$(facet_fstype mds$num) + local fstype=$(facet_fstype mgs) case $fstype in - ldiskfs ) - # vdevs are not supported by ldiskfs - eval VDEVPTR="";; - zfs ) - #if $MGSDEV isn't defined, default is $MGSDEV1 - eval VDEVPTR=${!DEVNAME:=${MDSDEV1}};; - * ) - error "unknown fstype!";; + ldiskfs ) + # vdevs are not supported by ldiskfs + ;; + zfs ) + if [ $(facet_host mgs) = $(facet_host mds1) ] && + ( [ -z "$MGSDEV" ] || [ $MGSDEV = $(mdsvdevname 1) ] ); then + VDEVPTR=$(mdsvdevname 1) + else + VDEVPTR=$MGSDEV + fi;; + * ) + error "unknown fstype!";; esac echo -n $VDEVPTR @@ -2558,7 +2847,8 @@ cleanupall() { } combined_mgs_mds () { - [[ $MDSDEV1 = $MGSDEV ]] && [[ $mds1_HOST = $mgs_HOST ]] + [[ "$(mdsdevname 1)" = "$(mgsdevname)" ]] && + [[ "$(facet_host mds1)" = "$(facet_host mgs)" ]] } lower() { @@ -2571,9 +2861,12 @@ upper() { mkfs_opts() { local facet=$1 + local dev=$2 + local fsname=${3:-"$FSNAME"} local type=$(facet_type $facet) local index=$(($(facet_number $facet) - 1)) local fstype=$(facet_fstype $facet) + local host=$(facet_host $facet) local opts local fs_mkfs_opts local var @@ -2582,14 +2875,17 @@ mkfs_opts() { return 1 fi - if [ $type == MGS ] || ( [ $type == MDS ] && combined_mgs_mds ); then + if [ $type == MGS ] || ( [ $type == MDS ] && + [ "$dev" == $(mgsdevname) ] && + [ "$host" == "$(facet_host mgs)" ] ); then opts="--mgs" else opts="--mgsnode=$MGSNID" fi if [ $type != MGS ]; then - opts+=" --fsname=$FSNAME --$(lower ${type/MDS/MDT}) --index=$index" + opts+=" --fsname=$fsname --$(lower ${type/MDS/MDT}) \ + --index=$index" fi var=${facet}failover_HOST @@ -2608,9 +2904,10 @@ mkfs_opts() { opts+=${L_GETIDENTITY:+" --param=mdt.identity_upcall=$L_GETIDENTITY"} if [ $fstype == ldiskfs ]; then - opts+=${IAMDIR:+" --iam-dir"} - fs_mkfs_opts+=${MDSJOURNALSIZE:+" -J size=$MDSJOURNALSIZE"} + if [ ! -z $EJOURNAL ]; then + fs_mkfs_opts+=${MDSJOURNALSIZE:+" device=$EJOURNAL"} + fi fs_mkfs_opts+=${MDSISIZE:+" -i $MDSISIZE"} fi fi @@ -2661,23 +2958,24 @@ formatall() { echo Formatting mgs, mds, osts if ! combined_mgs_mds ; then echo "Format mgs: $(mgsdevname)" - add mgs $(mkfs_opts mgs) --reformat $(mgsdevname) \ - $(mgsvdevname) ${quiet:+>/dev/null} || exit 10 - fi + add mgs $(mkfs_opts mgs $(mgsdevname)) --reformat \ + $(mgsdevname) $(mgsvdevname) ${quiet:+>/dev/null} || + exit 10 + fi - for num in `seq $MDSCOUNT`; do - echo "Format mds$num: $(mdsdevname $num)" - add mds$num $(mkfs_opts mds$num) --reformat \ - $(mdsdevname $num) $(mdsvdevname $num) \ + for num in $(seq $MDSCOUNT); do + echo "Format mds$num: $(mdsdevname $num)" + add mds$num $(mkfs_opts mds$num $(mdsdevname ${num})) \ + --reformat $(mdsdevname $num) $(mdsvdevname $num) \ ${quiet:+>/dev/null} || exit 10 - done + done - for num in `seq $OSTCOUNT`; do - echo "Format ost$num: $(ostdevname $num)" - add ost$num $(mkfs_opts ost$num) --reformat \ - $(ostdevname $num) $(ostvdevname ${num}) \ + for num in $(seq $OSTCOUNT); do + echo "Format ost$num: $(ostdevname $num)" + add ost$num $(mkfs_opts ost$num $(ostdevname ${num})) \ + --reformat $(ostdevname $num) $(ostvdevname ${num}) \ ${quiet:+>/dev/null} || exit 10 - done + done } mount_client() { @@ -2725,23 +3023,31 @@ remount_client() zconf_mount `hostname` $1 || error "mount failed" } -writeconf_facet () { - local facet=$1 - local dev=$2 +writeconf_facet() { + local facet=$1 + local dev=$2 - do_facet $facet "$TUNEFS --writeconf $dev" + stop ${facet} -f + rm -f $TMP/${facet}active + do_facet ${facet} "$TUNEFS --quiet --writeconf $dev" || return 1 + return 0 } writeconf_all () { - for num in `seq $MDSCOUNT`; do - DEVNAME=$(mdsdevname $num) - writeconf_facet mds$num $DEVNAME - done + local mdt_count=${1:-$MDSCOUNT} + local ost_count=${2:-$OSTCOUNT} + local rc=0 - for num in `seq $OSTCOUNT`; do - DEVNAME=$(ostdevname $num) - writeconf_facet ost$num $DEVNAME - done + for num in $(seq $mdt_count); do + DEVNAME=$(mdsdevname $num) + writeconf_facet mds$num $DEVNAME || rc=$? + done + + for num in $(seq $ost_count); do + DEVNAME=$(ostdevname $num) + writeconf_facet ost$num $DEVNAME || rc=$? + done + return $rc } setupall() { @@ -2827,65 +3133,95 @@ mounted_lustre_filesystems() { } init_facet_vars () { - [ "$CLIENTONLY" ] && return 0 - local facet=$1 - shift - local device=$1 + [ "$CLIENTONLY" ] && return 0 + local facet=$1 + shift + local device=$1 - shift + shift + + eval export ${facet}_dev=${device} + eval export ${facet}_opt=\"$@\" + + local dev=${facet}_dev + + # We need to loop for the label + # in case its not initialized yet. + for wait_time in {0,1,3,5,10}; do + + if [ $wait_time -gt 0 ]; then + echo "${!dev} not yet initialized,"\ + "waiting ${wait_time} seconds." + sleep $wait_time + fi + + local label=$(devicelabel ${facet} ${!dev}) + + # Check to make sure the label does + # not include ffff at the end of the label. + # This indicates it has not been initialized yet. + + if [[ $label =~ [f|F]{4}$ ]]; then + # label is not initialized, unset the result + # and either try again or fail + unset label + else + break + fi + done - eval export ${facet}_dev=${device} - eval export ${facet}_opt=\"$@\" + [ -z "$label" ] && echo no label for ${!dev} && exit 1 - local dev=${facet}_dev - local label=$(devicelabel ${facet} ${!dev}) - [ -z "$label" ] && echo no label for ${!dev} && exit 1 + eval export ${facet}_svc=${label} - eval export ${facet}_svc=${label} + local varname=${facet}failover_HOST + if [ -z "${!varname}" ]; then + eval export $varname=$(facet_host $facet) + fi - local varname=${facet}failover_HOST - if [ -z "${!varname}" ]; then - eval $varname=$(facet_host $facet) - fi + varname=${facet}_HOST + if [ -z "${!varname}" ]; then + eval export $varname=$(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 + # ${facet}failover_dev is set in cfg file + varname=${facet}failover_dev + if [ -n "${!varname}" ] ; then + eval export ${facet}failover_dev=${!varname} + else + eval export ${facet}failover_dev=$device + fi - # get mount point of already mounted device - # is facet_dev is already mounted then use the real - # mount point of this facet; otherwise use $(facet_mntpt $facet) - # i.e. ${facet}_MOUNT if specified by user or default - local mntpt=$(do_facet ${facet} cat /proc/mounts | \ - awk '"'${!dev}'" == $1 && $3 == "lustre" { print $2 }') - if [ -z $mntpt ]; then - mntpt=$(facet_mntpt $facet) - fi - eval export ${facet}_MOUNT=$mntpt + # get mount point of already mounted device + # is facet_dev is already mounted then use the real + # mount point of this facet; otherwise use $(facet_mntpt $facet) + # i.e. ${facet}_MOUNT if specified by user or default + local mntpt=$(do_facet ${facet} cat /proc/mounts | \ + awk '"'${!dev}'" == $1 && $3 == "lustre" { print $2 }') + if [ -z $mntpt ]; then + mntpt=$(facet_mntpt $facet) + fi + eval export ${facet}_MOUNT=$mntpt } init_facets_vars () { - local DEVNAME + local DEVNAME - 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 + 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 combined_mgs_mds || init_facet_vars mgs $(mgsdevname) $MGS_MOUNT_OPTS - remote_ost_nodsh && return - - for num in `seq $OSTCOUNT`; do - DEVNAME=`ostdevname $num` - init_facet_vars ost$num $DEVNAME $OST_MOUNT_OPTS - done + if ! remote_ost_nodsh; then + for num in $(seq $OSTCOUNT); do + DEVNAME=$(ostdevname $num) + init_facet_vars ost$num $DEVNAME $OST_MOUNT_OPTS + done + fi } osc_ensure_active () { @@ -2907,6 +3243,25 @@ osc_ensure_active () { [ $period -lt $timeout ] || log "$count OST are inactive after $timeout seconds, give up" } +set_conf_param_and_check() { + local myfacet=$1 + local TEST=$2 + local PARAM=$3 + local ORIG=$(do_facet $myfacet "$TEST") + if [ $# -gt 3 ]; then + local FINAL=$4 + else + local -i FINAL + FINAL=$((ORIG + 5)) + fi + echo "Setting $PARAM from $ORIG to $FINAL" + do_facet mgs "$LCTL conf_param $PARAM='$FINAL'" || + error "conf_param $PARAM failed" + + wait_update $(facet_host $myfacet) "$TEST" "$FINAL" || + error "check $PARAM failed!" +} + init_param_vars () { remote_mds_nodsh || TIMEOUT=$(do_facet $SINGLEMDS "lctl get_param -n timeout") @@ -2916,26 +3271,20 @@ init_param_vars () { osc_ensure_active $SINGLEMDS $TIMEOUT osc_ensure_active client $TIMEOUT - local jobid_var - if [ -z "$(lctl get_param -n mdc.*.connect_flags | grep jobstats)" ]; then - jobid_var="none" - elif [ $JOBSTATS_AUTO -ne 0 ]; then - echo "enable jobstats, set job scheduler as $JOBID_VAR" - jobid_var=$JOBID_VAR - else - jobid_var=`$LCTL get_param -n jobid_var` - if [ $jobid_var != "disable" ]; then - echo "disable jobstats as required" - jobid_var="disable" - else - jobid_var="none" - fi - fi + if [ -n "$(lctl get_param -n mdc.*.connect_flags|grep jobstats)" ]; then + local current_jobid_var=$($LCTL get_param -n jobid_var) + + if [ $JOBID_VAR = "existing" ]; then + echo "keeping jobstats as $current_jobid_var" + elif [ $current_jobid_var != $JOBID_VAR ]; then + echo "seting jobstats to $JOBID_VAR" - if [ $jobid_var == $JOBID_VAR -o $jobid_var == "disable" ]; then - do_facet mgs $LCTL conf_param $FSNAME.sys.jobid_var=$jobid_var - wait_update $HOSTNAME "$LCTL get_param -n jobid_var" \ - $jobid_var || return 1 + set_conf_param_and_check client \ + "$LCTL get_param -n jobid_var" \ + "$FSNAME.sys.jobid_var" $JOBID_VAR + fi + else + echo "jobstats not supported by server" fi if [ $QUOTA_AUTO -ne 0 ]; then @@ -2944,7 +3293,7 @@ init_param_vars () { setup_quota $MOUNT || return 2 else echo "disable quota as required" - $LFS quotaoff -ug $MOUNT > /dev/null 2>&1 + # $LFS quotaoff -ug $MOUNT > /dev/null 2>&1 fi fi return 0 @@ -3011,15 +3360,14 @@ check_config_client () { } check_config_clients () { - local clients=${CLIENTS:-$HOSTNAME} - local mntpt=$1 + local clients=${CLIENTS:-$HOSTNAME} + local mntpt=$1 - nfs_client_mode && return + nfs_client_mode && return - do_rpc_nodes $clients check_config_client $mntpt + do_rpc_nodes "$clients" check_config_client $mntpt - sanity_mount_check || - error "environments are insane!" + sanity_mount_check || error "environments are insane!" } check_timeout () { @@ -3040,17 +3388,17 @@ is_mounted () { } is_empty_dir() { - [ $(find $1 -maxdepth 1 -print | wc -l) = 1 ] && return 0 - return 1 + [ $(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 + [ $(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 1 + [ ! -d $1/.lustre ] || is_empty_dir $1/.lustre || return 1 + return 0 } check_and_setup_lustre() { @@ -3111,6 +3459,24 @@ check_and_setup_lustre() { set_default_debug_nodes $(comma_list $(nodes_list)) fi + if [ $(lower $OSD_TRACK_DECLARES_LBUG) == 'yes' ] ; then + local facets="$(get_facets OST),$(get_facets MDS),mgs" + local facet + local nodes + local node + for facet in ${facets//,/ }; do + if [ $(facet_fstype $node) == "ldiskfs" ] ; then + node=$(facet_host ${facet}) + nodes="$nodes $node" + fi + done + if [ -n "$nodes" ] ; then + nodes=$(for i in $nodes; do echo $i; done | sort -u) + do_nodes $(comma_list $nodes) "$LCTL set_param \ + osd-ldiskfs.track_declares_assert=1" + fi + fi + init_gss if $GSS; then set_flavor_all $SEC @@ -3154,7 +3520,7 @@ get_mnt_devs() { local dev if [ "$type" == ost ]; then - devs=$(get_obdfilter_param $node "" mntdev) + devs=$(get_osd_param $node "" mntdev) else devs=$(do_node $node \ "lctl get_param -n osd-*.$FSNAME-M*.mntdev") @@ -3190,7 +3556,7 @@ run_e2fsck() { local extra_opts=$3 df > /dev/null # update statfs data on disk - local cmd="$E2FSCK -d -v -t -t -f -n $extra_opts $target_dev" + local cmd="$E2FSCK -d -v -t -t -f $extra_opts $target_dev" echo $cmd local rc=0 do_node $node $cmd || rc=$? @@ -3201,12 +3567,12 @@ run_e2fsck() { # verify a directory is shared among nodes. check_shared_dir() { - local dir=$1 + local dir=$1 - [ -z "$dir" ] && return 1 - do_rpc_nodes $(comma_list $(nodes_list)) check_logdir $dir - check_write_access $dir || return 1 - return 0 + [ -z "$dir" ] && return 1 + do_rpc_nodes "$(comma_list $(nodes_list))" check_logdir $dir + check_write_access $dir || return 1 + return 0 } # Run e2fsck on MDT and OST(s) to generate databases used for lfsck. @@ -3226,14 +3592,14 @@ generate_db() { [ $MDSCOUNT -eq 1 ] || error "CMD is not supported" - run_e2fsck $(mdts_nodes) $MDTDEV "--mdsdb $MDSDB" + run_e2fsck $(mdts_nodes) $MDTDEV "-n --mdsdb $MDSDB" i=0 ostidx=0 OSTDB_LIST="" for node in $(osts_nodes); do for dev in ${OSTDEVS[i]}; do - run_e2fsck $node $dev "--mdsdb $MDSDB --ostdb $OSTDB-$ostidx" + run_e2fsck $node $dev "-n --mdsdb $MDSDB --ostdb $OSTDB-$ostidx" OSTDB_LIST="$OSTDB_LIST $OSTDB-$ostidx" ostidx=$((ostidx + 1)) done @@ -3241,17 +3607,55 @@ generate_db() { done } +# Run lfsck on server node if lfsck can't be found on client (LU-2571) +run_lfsck_remote() { + local cmd="$LFSCK_BIN -c -l --mdsdb $MDSDB --ostdb $OSTDB_LIST $MOUNT" + local client=$1 + local mounted=true + local rc=0 + + #Check if lustre is already mounted + do_rpc_nodes $client is_mounted $MOUNT || mounted=false + if ! $mounted; then + zconf_mount $client $MOUNT || + error "failed to mount Lustre on $client" + fi + #Run lfsck + echo $cmd + do_node $node $cmd || rc=$? + #Umount if necessary + if ! $mounted; then + zconf_umount $client $MOUNT || + error "failed to unmount Lustre on $client" + fi + + [ $rc -le $FSCK_MAX_ERR ] || + error "$cmd returned $rc, should be <= $FSCK_MAX_ERR" + echo "lfsck finished with rc=$rc" + + return $rc +} + run_lfsck() { - local cmd="$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" + local facets="client $SINGLEMDS" + local found=false + local facet + local node + local rc=0 - rm -rvf $MDSDB* $OSTDB* || true - return 0 + for facet in $facets; do + node=$(facet_active_host $facet) + if check_progs_installed $node $LFSCK_BIN; then + found=true + break + fi + done + ! $found && error "None of \"$facets\" supports lfsck" + + run_lfsck_remote $node || rc=$? + + rm -rvf $MDSDB* $OSTDB* || true + return $rc } check_and_cleanup_lustre() { @@ -3261,11 +3665,11 @@ check_and_cleanup_lustre() { run_lfsck fi - if is_mounted $MOUNT; then - [ -n "$DIR" ] && rm -rf $DIR/[Rdfs][0-9]* || - error "remove sub-test dirs failed" - [ "$ENABLE_QUOTA" ] && restore_quota_type || true - fi + if is_mounted $MOUNT; then + [ -n "$DIR" ] && rm -rf $DIR/[Rdfs][0-9]* || + error "remove sub-test dirs failed" + [ "$ENABLE_QUOTA" ] && restore_quota || true + fi if [ "$I_UMOUNTED2" = "yes" ]; then restore_mount $MOUNT2 || error "restore $MOUNT2 failed" @@ -3330,10 +3734,12 @@ no_dsh() { eval $@ } +# Convert a space-delimited list to a comma-delimited list. If the input is +# only whitespace, ensure the output is empty (i.e. "") so [ -n $list ] works comma_list() { - # the sed converts spaces to commas, but leaves the last space - # alone, so the line doesn't end with a comma. - echo "$*" | tr -s " " "\n" | sort -b -u | tr "\n" " " | sed 's/ \([^$]\)/,\1/g' + # echo is used to convert newlines to spaces, since it doesn't + # introduce a trailing space as using "tr '\n' ' '" does + echo $(tr -s " " "\n" <<< $* | sort -b -u) | tr ' ' ',' } list_member () { @@ -3490,6 +3896,17 @@ drop_reint_reply() { return $RC } +drop_update_reply() { +# OBD_FAIL_UPDATE_OBJ_NET_REP + local index=$1 + shift 1 + RC=0 + do_facet mds${index} lctl set_param fail_loc=0x1701 + do_facet client "$@" || RC=$? + do_facet mds${index} lctl set_param fail_loc=0 + return $RC +} + pause_bulk() { #define OBD_FAIL_OST_BRW_PAUSE_BULK 0x214 RC=0 @@ -3501,7 +3918,7 @@ pause_bulk() { } drop_ldlm_cancel() { -#define OBD_FAIL_LDLM_CANCEL 0x304 +#define OBD_FAIL_LDLM_CANCEL_NET 0x304 local RC=0 local list=$(comma_list $(mdts_nodes) $(osts_nodes)) do_nodes $list lctl set_param fail_loc=0x304 @@ -3513,12 +3930,12 @@ drop_ldlm_cancel() { } drop_bl_callback() { -#define OBD_FAIL_LDLM_BL_CALLBACK 0x305 - RC=0 - do_facet client lctl set_param fail_loc=0x305 - do_facet client "$@" || RC=$? - do_facet client lctl set_param fail_loc=0 - return $RC +#define OBD_FAIL_LDLM_BL_CALLBACK_NET 0x305 + RC=0 + do_facet client lctl set_param fail_loc=0x80000305 + do_facet client "$@" || RC=$? + do_facet client lctl set_param fail_loc=0 + return $RC } drop_ldlm_reply() { @@ -3632,27 +4049,33 @@ log_trace_dump() { ################################## error_noexit() { - local TYPE=${TYPE:-"FAIL"} + local TYPE=${TYPE:-"FAIL"} - local dump=true - # do not dump logs if $1=false - if [ "x$1" = "xfalse" ]; then - shift - dump=false - fi + local dump=true + # do not dump logs if $1=false + if [ "x$1" = "xfalse" ]; then + shift + dump=false + fi - log " ${TESTSUITE} ${TESTNAME}: @@@@@@ ${TYPE}: $@ " - log_trace_dump - mkdir -p $LOGDIR - # We need to dump the logs on all nodes - if $dump; then - gather_logs $(comma_list $(nodes_list)) - fi + log " ${TESTSUITE} ${TESTNAME}: @@@@@@ ${TYPE}: $@ " + log_trace_dump - debugrestore - [ "$TESTSUITELOG" ] && echo "$0: ${TYPE}: $TESTNAME $@" >> $TESTSUITELOG - echo "$@" > $LOGDIR/err + mkdir -p $LOGDIR + # We need to dump the logs on all nodes + if $dump; then + gather_logs $(comma_list $(nodes_list)) + fi + + debugrestore + [ "$TESTSUITELOG" ] && + echo "$TESTSUITE: $TYPE: $TESTNAME $@" >> $TESTSUITELOG + if [ -z "$*" ]; then + echo "error() without useful message, please fix" > $LOGDIR/err + else + echo "$@" > $LOGDIR/err + fi } exit_status () { @@ -3664,25 +4087,31 @@ exit_status () { } error() { - error_noexit "$@" - exit 1 + error_noexit "$@" + exit 1 } error_exit() { - error "$@" + error "$@" } # use only if we are ignoring failures for this test, bugno required. # (like ALWAYS_EXCEPT, but run the test and ignore the results.) # e.g. error_ignore 5494 "your message" error_ignore() { - local TYPE="IGNORE (bz$1)" - shift - error_noexit "$@" + local TYPE="IGNORE (bz$1)" + shift + error_noexit "$@" +} + +error_and_remount() { + error_noexit "$@" + remount_client $MOUNT + exit 1 } skip_env () { - $FAIL_ON_SKIP_ENV && error false $@ || skip $@ + $FAIL_ON_SKIP_ENV && error false $@ || skip $@ } skip() { @@ -3762,7 +4191,6 @@ run_test() { return $? fi LAST_SKIPPED="y" - echo -n "." return 0 fi @@ -3806,10 +4234,6 @@ run_test() { return $? } -equals_msg() { - banner "$*" -} - log() { echo "$*" module_loaded lnet || load_modules @@ -3836,9 +4260,11 @@ trace() { } complete () { - equals_msg $1 test complete, duration $2 sec + local duration=$1 + + banner test complete, duration $duration sec [ -f "$TESTSUITELOG" ] && egrep .FAIL $TESTSUITELOG || true - echo duration $2 >>$TESTSUITELOG + echo duration $duration >>$TESTSUITELOG } pass() { @@ -3905,7 +4331,10 @@ run_one() { reset_fail_loc check_grant ${testnum} || error "check_grant $testnum failed with $?" check_catastrophe || error "LBUG/LASSERT detected" - ps auxww | grep -v grep | grep -q multiop && error "multiop still running" + if [ "$PARALLEL" != "yes" ]; then + ps auxww | grep -v grep | grep -q multiop && + error "multiop still running" + fi unset TESTNAME unset tdir umask $SAVE_UMASK @@ -4040,9 +4469,45 @@ mdtuuid_from_index() $LFS mdts $2 | sed -ne "/^$1: /s/.* \(.*\) .*$/\1/p" } +# Description: +# Return unique identifier for given hostname +host_id() { + local host_name=$1 + echo $host_name | md5sum | cut -d' ' -f1 +} + +# Description: +# Returns list of ip addresses for each interface +local_addr_list() { + ip addr | awk '/inet\ / {print $2}' | awk -F\/ '{print $1}' +} + +is_local_addr() { + local addr=$1 + # Cache address list to avoid mutiple execution of local_addr_list + LOCAL_ADDR_LIST=${LOCAL_ADDR_LIST:-$(local_addr_list)} + local i + for i in $LOCAL_ADDR_LIST ; do + [[ "$i" == "$addr" ]] && return 0 + done + return 1 +} + +local_node() { + local host_name=$1 + local is_local="IS_LOCAL_$(host_id $host_name)" + if [ -z "${!is_local-}" ] ; then + eval $is_local=0 + local host_ip=$($LUSTRE/tests/resolveip $host_name) + is_local_addr "$host_ip" && eval $is_local=1 + fi + [[ "${!is_local}" == "1" ]] +} + remote_node () { - local node=$1 - [ "$node" != "$(hostname)" ] + local node=$1 + local_node $node && return 1 + return 0 } remote_mds () @@ -4102,66 +4567,121 @@ local_mode () $(single_local_node $(comma_list $(nodes_list))) } -mdts_nodes () { - local MDSNODES - local NODES_sort - for num in `seq $MDSCOUNT`; do - MDSNODES="$MDSNODES $(facet_host mds$num)" - done - NODES_sort=$(for i in $MDSNODES; do echo $i; done | sort -u) - - echo $NODES_sort -} - remote_servers () { remote_ost && remote_mds } +# Get the active nodes for facets. facets_nodes () { - local facets=$1 - local nodes - local NODES_sort + local facets=$1 + local facet + local nodes + local nodes_sort + local i - 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 $nodes; do echo $i; done | sort -u) + for facet in ${facets//,/ }; do + nodes="$nodes $(facet_active_host $facet)" + done - echo $NODES_sort + nodes_sort=$(for i in $nodes; do echo $i; done | sort -u) + echo -n $nodes_sort } -osts_nodes () { - local facets=$(get_facets OST) - local nodes=$(facets_nodes $facets) +# Get all of the active MDS nodes. +mdts_nodes () { + echo -n $(facets_nodes $(get_facets MDS)) +} - echo $nodes +# Get all of the active OSS nodes. +osts_nodes () { + echo -n $(facets_nodes $(get_facets OST)) } +# Get all of the client nodes and active server nodes. nodes_list () { - # FIXME. We need a list of clients - local myNODES=$HOSTNAME - local myNODES_sort + local nodes=$HOSTNAME + local nodes_sort + local i - # CLIENTS (if specified) contains the local client - [ -n "$CLIENTS" ] && myNODES=${CLIENTS//,/ } - - if [ "$PDSH" -a "$PDSH" != "no_dsh" ]; then - myNODES="$myNODES $(facets_nodes $(get_facets))" - fi + # CLIENTS (if specified) contains the local client + [ -n "$CLIENTS" ] && nodes=${CLIENTS//,/ } - myNODES_sort=$(for i in $myNODES; do echo $i; done | sort -u) + if [ "$PDSH" -a "$PDSH" != "no_dsh" ]; then + nodes="$nodes $(facets_nodes $(get_facets))" + fi - echo $myNODES_sort + nodes_sort=$(for i in $nodes; do echo $i; done | sort -u) + echo -n $nodes_sort } +# Get all of the remote client nodes and remote active server nodes. remote_nodes_list () { - local rnodes=$(nodes_list) - rnodes=$(echo " $rnodes " | sed -re "s/\s+$HOSTNAME\s+/ /g") - echo $rnodes + echo -n $(nodes_list) | sed -re "s/\<$HOSTNAME\>//g" +} + +# Get all of the MDS nodes, including active and passive nodes. +all_mdts_nodes () { + local host + local failover_host + local nodes + local nodes_sort + local i + + for i in $(seq $MDSCOUNT); do + host=mds${i}_HOST + failover_host=mds${i}failover_HOST + nodes="$nodes ${!host} ${!failover_host}" + done + + nodes_sort=$(for i in $nodes; do echo $i; done | sort -u) + echo -n $nodes_sort +} + +# Get all of the OSS nodes, including active and passive nodes. +all_osts_nodes () { + local host + local failover_host + local nodes + local nodes_sort + local i + + for i in $(seq $OSTCOUNT); do + host=ost${i}_HOST + failover_host=ost${i}failover_HOST + nodes="$nodes ${!host} ${!failover_host}" + done + + nodes_sort=$(for i in $nodes; do echo $i; done | sort -u) + echo -n $nodes_sort +} + +# Get all of the server nodes, including active and passive nodes. +all_server_nodes () { + local nodes + local nodes_sort + local i + + nodes="$mgs_HOST $mgsfailover_HOST $(all_mdts_nodes) $(all_osts_nodes)" + + nodes_sort=$(for i in $nodes; do echo $i; done | sort -u) + echo -n $nodes_sort +} + +# Get all of the client and server nodes, including active and passive nodes. +all_nodes () { + local nodes=$HOSTNAME + local nodes_sort + local i + + # CLIENTS (if specified) contains the local client + [ -n "$CLIENTS" ] && nodes=${CLIENTS//,/ } + + if [ "$PDSH" -a "$PDSH" != "no_dsh" ]; then + nodes="$nodes $(all_server_nodes)" + fi + + nodes_sort=$(for i in $nodes; do echo $i; done | sort -u) + echo -n $nodes_sort } init_clients_lists () { @@ -4242,10 +4762,11 @@ generate_machine_file() { } get_stripe () { - local file=$1/stripe - touch $file - $LFS getstripe -v $file || error - rm -f $file + local file=$1/stripe + + touch $file + $LFS getstripe -v $file || error "getstripe $file failed" + rm -f $file } setstripe_nfsserver () { @@ -4462,33 +4983,53 @@ calc_osc_kbytes () { $LCTL get_param -n osc.*[oO][sS][cC][-_][0-9a-f]*.$1 | calc_sum } -# save_lustre_params(node, parameter_mask) -# generate a stream of formatted strings ( =) +# save_lustre_params(comma separated facet list, parameter_mask) +# generate a stream of formatted strings ( =) save_lustre_params() { - local s - do_nodesv $1 "lctl get_param $2 | while read s; do echo \\\$s; done" + local facets=$1 + local facet + local nodes + local node + + for facet in ${facets//,/ }; do + node=$(facet_active_host $facet) + [[ *\ $node\ * = " $nodes " ]] && continue + nodes="$nodes $node" + + do_node $node "$LCTL get_param $2 | + while read s; do echo $facet \\\$s; done" + done } # restore lustre parameters from input stream, produces by save_lustre_params restore_lustre_params() { - local node - local name - local val - while IFS=" =" read node name val; do - do_node ${node//:/} "lctl set_param -n $name $val" - done + local facet + local name + local val + + while IFS=" =" read facet name val; do + do_facet $facet "$LCTL set_param -n $name $val" + done } check_catastrophe() { - local rnodes=${1:-$(comma_list $(remote_nodes_list))} - local C=$CATASTROPHE - [ -f $C ] && [ $(cat $C) -ne 0 ] && return 1 - - if [ $rnodes ]; then - do_nodes $rnodes "rc=\\\$([ -f $C ] && echo \\\$(< $C) || echo 0); -if [ \\\$rc -ne 0 ]; then echo \\\$(hostname): \\\$rc; fi -exit \\\$rc;" - fi + local rnodes=${1:-$(comma_list $(remote_nodes_list))} + local C=$CATASTROPHE + [ -f $C ] && [ $(cat $C) -ne 0 ] && return 1 + + [ -z "$rnodes" ] && return 0 + + local data + data=$(do_nodes "$rnodes" "rc=\\\$([ -f $C ] && + echo \\\$(< $C) || echo 0); + if [ \\\$rc -ne 0 ]; then echo \\\$(hostname): \\\$rc; fi + exit \\\$rc") + local rc=$? + if [ -n "$data" ]; then + echo $data + return $rc + fi + return 0 } # CMD: determine mds index where directory inode presents @@ -4519,10 +5060,11 @@ get_mds_dir () { } mdsrate_cleanup () { - if [ -d $4 ]; then - mpi_run -np $1 -machinefile $2 ${MDSRATE} --unlink --nfiles $3 --dir $4 --filefmt $5 $6 - rmdir $4 - fi + if [ -d $4 ]; then + mpi_run -np $1 ${MACHINEFILE_OPTION} $2 ${MDSRATE} --unlink \ + --nfiles $3 --dir $4 --filefmt $5 $6 + rmdir $4 + fi } delayed_recovery_enabled () { @@ -4549,7 +5091,7 @@ convert_facet2label() { } get_clientosc_proc_path() { - echo "${1}-osc-[^M]*" + echo "${1}-osc-*" } get_lustre_version () { @@ -4613,7 +5155,7 @@ _wait_import_state () { local CONN_STATE local i=0 - CONN_STATE=$($LCTL get_param -n $CONN_PROC 2>/dev/null | cut -f2) + CONN_STATE=$($LCTL get_param -n $CONN_PROC 2>/dev/null | cut -f2 | uniq) while [ "${CONN_STATE}" != "${expected}" ]; do if [ "${expected}" == "DISCONN" ]; then # for disconn we can check after proc entry is removed @@ -4626,7 +5168,8 @@ _wait_import_state () { 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 $CONN_PROC 2>/dev/null | cut -f2) + # Add uniq for multi-mount case + CONN_STATE=$($LCTL get_param -n $CONN_PROC 2>/dev/null | cut -f2 | uniq) i=$(($i + 1)) done @@ -4645,6 +5188,14 @@ wait_import_state() { done } +wait_import_state_mount() { + if ! is_mounted $MOUNT && ! is_mounted $MOUNT2; then + return 0 + fi + + wait_import_state $* +} + # One client request could be timed out because server was not ready # when request was sent by client. # The request timeout calculation details : @@ -4685,25 +5236,62 @@ request_timeout () { echo $(( init_connect_timeout + at_min )) } -wait_osc_import_state() { +_wait_osc_import_state() { local facet=$1 local ost_facet=$2 local expected=$3 local ost=$(get_osc_import_name $facet $ost_facet) - local param="osc.${ost}.ost_server_uuid" + local param="osc.${ost}.ost_server_uuid" + local i=0 # 1. wait the deadline of client 1st request (it could be skipped) # 2. wait the deadline of client 2nd request local maxtime=$(( 2 * $(request_timeout $facet))) - if ! do_rpc_nodes $(facet_host $facet) \ - _wait_import_state $expected $param $maxtime; then - error "import is not in ${expected} state" - return 1 - fi + #During setup time, the osc might not be setup, it need wait + #until list_param can return valid value. And also if there + #are mulitple osc entries we should list all of them before + #go to wait. + local params=$($LCTL list_param $param 2>/dev/null || true) + while [ -z "$params" ]; do + if [ $i -ge $maxtime ]; then + echo "can't get $param by list_param in $maxtime secs" + if [[ $facet != client* ]]; then + echo "Go with $param directly" + params=$param + break + else + return 1 + fi + fi + sleep 1 + i=$((i + 1)) + params=$($LCTL list_param $param 2>/dev/null || true) + done - return 0 + if ! do_rpc_nodes "$(facet_active_host $facet)" \ + wait_import_state $expected "$params" $maxtime; then + error "import is not in ${expected} state" + return 1 + fi + + return 0 +} + +wait_osc_import_state() { + local facet=$1 + local ost_facet=$2 + local expected=$3 + local num + + if [[ $facet = mds ]]; then + for num in $(seq $MDSCOUNT); do + _wait_osc_import_state mds$num "$ost_facet" "$expected" + done + else + _wait_osc_import_state "$facet" "$ost_facet" "$expected" + fi } get_clientmdc_proc_path() { @@ -4711,12 +5299,16 @@ get_clientmdc_proc_path() { } do_rpc_nodes () { - local list=$1 - shift + local list=$1 + shift + + [ -z "$list" ] && return 0 - # Add paths to lustre tests for 32 and 64 bit systems. - local RPATH="PATH=$RLUSTRE/tests:/usr/lib/lustre/tests:/usr/lib64/lustre/tests:$PATH" - do_nodesv $list "${RPATH} NAME=${NAME} sh rpc.sh $@ " + # Add paths to lustre tests for 32 and 64 bit systems. + local LIBPATH="/usr/lib/lustre/tests:/usr/lib64/lustre/tests:" + local TESTPATH="$RLUSTRE/tests:" + local RPATH="PATH=${TESTPATH}${LIBPATH}${PATH}:/sbin:/bin:/usr/sbin:" + do_nodesv $list "${RPATH} NAME=${NAME} sh rpc.sh $@ " } wait_clients_import_state () { @@ -4741,10 +5333,10 @@ wait_clients_import_state () { local params=$(expand_list $params $proc_path) done - if ! do_rpc_nodes $list wait_import_state $expected $params; then - error "import is not in ${expected} state" - return 1 - fi + if ! do_rpc_nodes "$list" wait_import_state_mount $expected $params; then + error "import is not in ${expected} state" + return 1 + fi } oos_full() { @@ -5333,16 +5925,45 @@ run_llverfs() { local dir=$1 local llverfs_opts=$2 + local use_partial_arg=$3 local partial_arg="" local size=$(df -B G $dir |tail -n 1 |awk '{print $2}' |sed 's/G//') #GB # Run in partial (fast) mode if the size # of a partition > 1 GB - [ $size -gt 1 ] && partial_arg="-p" + [ "x$use_partial_arg" != "xno" ] && [ $size -gt 1 ] && partial_arg="-p" llverfs $partial_arg $llverfs_opts $dir } +#Remove objects from OST +remove_ost_objects() { + shift + local ostdev=$1 + local group=$2 + shift 2 + local objids="$@" + local facet=ost$((OSTIDX + 1)) + local mntpt=$(facet_mntpt $facet) + local opts=$OST_MOUNT_OPTS + local i + local rc + + echo "removing objects from $ostdev on $facet: $objids" + if ! do_facet $facet test -b $ostdev; then + opts=$(csa_add "$opts" -o loop) + fi + mount -t $(facet_fstype $facet) $opts $ostdev $mntpt || + return $? + rc=0; + for i in $objids; do + rm $mntpt/O/$group/d$((i % 32))/$i || { rc=$?; break; } + done + umount -f $mntpt || return $? + return $rc +} + +#Remove files from MDT remove_mdt_files() { local facet=$1 local mdtdev=$2 @@ -5353,7 +5974,7 @@ remove_mdt_files() { echo "removing files from $mdtdev on $facet: $files" if [ $(facet_fstype $facet) == ldiskfs ] && - ! do_facet $facet test -b ${!dev}; then + ! do_facet $facet test -b $mdtdev; then opts=$(csa_add "$opts" -o loop) fi mount -t $(facet_fstype $facet) $opts $mdtdev $mntpt || @@ -5377,7 +5998,7 @@ duplicate_mdt_files() { echo "duplicating files on $mdtdev on $facet: $files" mkdir -p $mntpt || return $? if [ $(facet_fstype $facet) == ldiskfs ] && - ! do_facet $facet test -b ${!dev}; then + ! do_facet $facet test -b $mdtdev; then opts=$(csa_add "$opts" -o loop) fi mount -t $(facet_fstype $facet) $opts $mdtdev $mntpt || @@ -5459,10 +6080,12 @@ get_block_size() { # Check whether the "large_xattr" feature is enabled or not. large_xattr_enabled() { - local mds_dev=$(mdsdevname ${SINGLEMDS//mds/}) + [[ $(facet_fstype $SINGLEMDS) == zfs ]] && return 0 - do_facet $SINGLEMDS "$DUMPE2FS -h $mds_dev 2>&1 | grep -q large_xattr" - return ${PIPESTATUS[0]} + local mds_dev=$(mdsdevname ${SINGLEMDS//mds/}) + + do_facet $SINGLEMDS "$DUMPE2FS -h $mds_dev 2>&1 | grep -q large_xattr" + return ${PIPESTATUS[0]} } # Get the maximum xattr size supported by the filesystem. @@ -5516,6 +6139,8 @@ mds_backup_restore() { local metaea=${TMP}/backup_restore.ea local metadata=${TMP}/backup_restore.tgz local opts=${MDS_MOUNT_OPTS} + local svc=${SINGLEMDS}_svc + local igif=$1 if ! ${rcmd} test -b ${devname}; then opts=$(csa_add "$opts" -o loop) @@ -5529,6 +6154,10 @@ mds_backup_restore() { ${rcmd} rm -f $metaea $metadata # step 3: mount dev ${rcmd} mount -t ldiskfs $opts $devname $mntpt || return 1 + if [ ! -z $igif ]; then + # step 3.5: rm .lustre + ${rcmd} rm -rf $mntpt/ROOT/.lustre || return 1 + fi # step 4: backup metaea echo "backup EA" ${rcmd} "cd $mntpt && getfattr -R -d -m '.*' -P . > $metaea && cd -" || @@ -5542,8 +6171,9 @@ mds_backup_restore() { reformat_external_journal || return 5 # step 8: reformat dev echo "reformat new device" - add ${SINGLEMDS} $(mkfs_opts ${SINGLEMDS}) --backfstype ldiskfs \ - --reformat $devname > /dev/null || return 6 + add ${SINGLEMDS} $(mkfs_opts ${SINGLEMDS} ${devname}) --backfstype \ + ldiskfs --reformat ${devname} $(mdsvdevname 1) > /dev/null || + exit 6 # step 9: mount dev ${rcmd} mount -t ldiskfs $opts $devname $mntpt || return 7 # step 10: restore metadata @@ -5559,6 +6189,8 @@ mds_backup_restore() { ${rcmd} umount -d $mntpt || return 10 # step 14: cleanup tmp backup ${rcmd} rm -f $metaea $metadata + # step 15: reset device label - it's not virgin on + ${rcmd} e2label $devname ${!svc} } # remove OI files @@ -5596,3 +6228,72 @@ mds_remove_ois() { ${rcmd} umount -d $mntpt || return 2 # OI files will be recreated when mounted as lustre next time. } + +# generate maloo upload-able log file name +# \param logname specify unique part of file name +generate_logname() { + local logname=${1:-"default_logname"} + + echo "$TESTLOG_PREFIX.$TESTNAME.$logname.$(hostname -s).log" +} + +# mkdir directory on different MDTs +test_mkdir() { + local option + local parent + local child + local path + local dir + local rc=0 + + if [ $# -eq 2 ]; then + option=$1 + path=$2 + else + path=$1 + fi + + child=${path##*/} + parent=${path%/*} + + if [ "$parent" == "$child" ]; then + parent=$(pwd) + fi + + if [ "$option" == "-p" -a -d ${parent}/${child} ]; then + return $rc + fi + + # it needs to check whether there is further / in child + dir=$(echo $child | awk -F '/' '{print $2}') + if [ ! -z "$dir" ]; then + local subparent=$(echo $child | awk -F '/' '{ print $1 }') + parent=${parent}"/"${subparent} + child=$dir + fi + + if [ ! -d ${parent} ]; then + if [ "$option" == "-p" ]; then + mkdir -p ${parent} + else + return 1 + fi + fi + + if [ $MDSCOUNT -le 1 ]; then + mkdir $option ${parent}/${child} || rc=$? + else + local mdt_idx=$($LFS getstripe -M $parent) + + if [ "$mdt_idx" -ne 0 ]; then + mkdir $option ${parent}/${child} || rc=$? + return $rc + fi + + local test_num=$(echo $testnum | sed -e 's/[^0-9]*//g') + local mdt_idx=$((test_num % MDSCOUNT)) + echo "mkdir $mdt_idx for ${parent}/${child}" + $LFS setdirstripe -i $mdt_idx ${parent}/${child} || rc=$? + fi + return $rc +}