export GSS_KRB5=false
export GSS_PIPEFS=false
export IDENTITY_UPCALL=default
+export QUOTA_AUTO=1
#export PDSH="pdsh -S -Rssh -w"
# function used by scripts run on remote nodes
LUSTRE=${LUSTRE:-$(cd $(dirname $0)/..; echo $PWD)}
. $LUSTRE/tests/functions.sh
+. $LUSTRE/tests/yaml.sh
+
+LUSTRE_TESTS_CFG_DIR=${LUSTRE_TESTS_CFG_DIR:-${LUSTRE}/tests/cfg}
+
+EXCEPT_LIST_FILE=${EXCEPT_LIST_FILE:-${LUSTRE_TESTS_CFG_DIR}/tests-to-skip.sh}
+
+if [ -f "$EXCEPT_LIST_FILE" ]; then
+ echo "Reading test skip list from $EXCEPT_LIST_FILE"
+ cat $EXCEPT_LIST_FILE
+ . $EXCEPT_LIST_FILE
+fi
+
+[ -z "$MODPROBECONF" -a -f /etc/modprobe.conf ] && MODPROBECONF=/etc/modprobe.conf
+[ -z "$MODPROBECONF" -a -f /etc/modprobe.d/Lustre ] && MODPROBECONF=/etc/modprobe.d/Lustre
assert_DIR () {
local failed=""
}
print_summary () {
- [ "$TESTSUITE" == "lfscktest" ] && return 0
+ trap 0
+ [ "$TESTSUITE" == "lfsck" ] && return 0
[ -n "$ONLY" ] && echo "WARNING: ONLY is set to ${ONLY}."
local form="%-13s %-17s %s\n"
printf "$form" "status" "script" "skipped tests E(xcluded) S(low)"
echo "------------------------------------------------------------------------------------"
- for O in $TESTSUITE_LIST; do
+ for O in $DEFAULT_SUITES; do
local skipped=""
local slow=""
- local o=$(echo $O | tr "[:upper:]" "[:lower:]")
+ O=$(echo $O | tr "-" "_" | tr "[:lower:]" "[:upper:]")
+ local o=$(echo $O | tr "[:upper:]" "[:lower:]")
o=${o//_/-}
- o=${o//tyn/tyN}
local log=${TMP}/${o}.log
[ -f $log ] && skipped=$(grep excluded $log | awk '{ printf " %s", $3 }' | sed 's/test_//g')
[ -f $log ] && slow=$(grep SLOW $log | awk '{ printf " %s", $3 }' | sed 's/test_//g')
done
- for O in $TESTSUITE_LIST; do
- [ "${!O}" = "no" ] && \
- printf "$form" "Skipped" "$O" ""
+ for O in $DEFAULT_SUITES; do
+ O=$(echo $O | tr "-" "_" | tr "[:lower:]" "[:upper:]")
+ if [ "${!O}" = "no" ]; then
+ # FIXME.
+ # only for those tests suits which are run directly from acc-sm script:
+ # bonnie, iozone, etc.
+ if [ -f "$TESTSUITELOG" ] && grep FAIL $TESTSUITELOG | grep -q ' '$O ; then
+ printf "$form" "UNFINISHED" "$O" ""
+ else
+ printf "$form" "Skipped" "$O" ""
+ fi
+ fi
done
- for O in $TESTSUITE_LIST; do
+ for O in $DEFAULT_SUITES; do
+ O=$(echo $O | tr "-" "_" | tr "[:lower:]" "[:upper:]")
[ "${!O}" = "done" -o "${!O}" = "no" ] || \
printf "$form" "UNFINISHED" "$O" ""
done
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}
#[ -d /r ] && export ROOT=${ROOT:-/r}
export TMP=${TMP:-$ROOT/tmp}
export TESTSUITELOG=${TMP}/${TESTSUITE}.log
+ if [[ -z $LOGDIRSET ]]; then
+ export LOGDIR=${LOGDIR:-${TMP}/test_logs/}/$(date +%s)
+ export LOGDIRSET=true
+ fi
export HOSTNAME=${HOSTNAME:-`hostname`}
if ! echo $PATH | grep -q $LUSTRE/utils; then
export PATH=$PATH:$LUSTRE/utils
export PATH=$PATH:$LUSTRE/utils/gss
fi
if ! echo $PATH | grep -q $LUSTRE/tests; then
- export PATH=$PATH:$LUSTRE/tests
+ export PATH=$PATH:$LUSTRE/tests
fi
+ export LST=${LST:-"$LUSTRE/../lnet/utils/lst"}
+ [ ! -f "$LST" ] && export LST=$(which lst)
export MDSRATE=${MDSRATE:-"$LUSTRE/tests/mpi/mdsrate"}
[ ! -f "$MDSRATE" ] && export MDSRATE=$(which mdsrate 2> /dev/null)
if ! echo $PATH | grep -q $LUSTRE/tests/racer; then
- export PATH=$PATH:$LUSTRE/tests/racer
+ export PATH=$LUSTRE/tests/racer:$PATH:
fi
if ! echo $PATH | grep -q $LUSTRE/tests/mpi; then
export PATH=$PATH:$LUSTRE/tests/mpi
fi
+ export RSYNC_RSH=${RSYNC_RSH:-rsh}
export LCTL=${LCTL:-"$LUSTRE/utils/lctl"}
[ ! -f "$LCTL" ] && export LCTL=$(which lctl)
export LFS=${LFS:-"$LUSTRE/utils/lfs"}
export TUNEFS=${TUNEFS:-"$LUSTRE/utils/tunefs.lustre"}
[ ! -f "$TUNEFS" ] && export TUNEFS=$(which tunefs.lustre)
export CHECKSTAT="${CHECKSTAT:-"checkstat -v"} "
+ export LUSTRE_RMMOD=${LUSTRE_RMMOD:-$LUSTRE/scripts/lustre_rmmod}
+ [ ! -f "$LUSTRE_RMMOD" ] && export LUSTRE_RMMOD=$(which lustre_rmmod 2> /dev/null)
export FSTYPE=${FSTYPE:-"ldiskfs"}
export NAME=${NAME:-local}
export LGSSD=${LGSSD:-"$LUSTRE/utils/gss/lgssd"}
xkrb5*)
echo "Using GSS/krb5 ptlrpc security flavor"
which lgss_keyring > /dev/null 2>&1 || \
- error "built with gss disabled! SEC=$SEC"
+ error_exit "built with gss disabled! SEC=$SEC"
GSS=true
GSS_KRB5=true
;;
IDENTITY_UPCALL=false
;;
esac
+ export LOAD_MODULES_REMOTE=${LOAD_MODULES_REMOTE:-false}
# Paths on remote nodes, if different
export RLUSTRE=${RLUSTRE:-$LUSTRE}
export RPWD=${RPWD:-$PWD}
export I_MOUNTED=${I_MOUNTED:-"no"}
+ if [ ! -f /lib/modules/$(uname -r)/kernel/fs/lustre/mds.ko -a \
+ ! -f `dirname $0`/../mds/mds.ko ]; then
+ export CLIENTMODSONLY=yes
+ fi
# command line
[ "$TESTSUITELOG" ] && rm -f $TESTSUITELOG || true
rm -f $TMP/*active
-
}
case `uname -r` in
*) EXT=".ko"; USE_QUOTA=yes;;
esac
+
+module_loaded () {
+ /sbin/lsmod | grep -q $1
+}
+
+# Load a module on the system where this is running.
+#
+# Synopsis: load_module module_name [module arguments for insmod/modprobe]
+#
+# If module arguments are not given but MODOPTS_<MODULE> is set, then its value
+# will be used as the arguments. Otherwise arguments will be obtained from
+# /etc/modprobe.conf, from /etc/modprobe.d/Lustre, or else none will be used.
+#
load_module() {
+ local optvar
EXT=".ko"
module=$1
shift
BASE=`basename $module $EXT`
- lsmod | grep -q ${BASE} || \
- if [ -f ${LUSTRE}/${module}${EXT} ]; then
- insmod ${LUSTRE}/${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
+ # $MODOPTS_<MODULE>; we could use associative arrays, but that's not in
+ # Bash until 4.x, so we resort to eval.
+ optvar="MODOPTS_$(basename $module | tr a-z A-Z)"
+ eval set -- \$$optvar
+ if [ $# -eq 0 -a -n "$MODPROBECONF" ]; then
+ # Nothing in $MODOPTS_<MODULE>; try modprobe.conf
+ set -- $(grep "^options\\s*\<${module}\>" $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
+ fi
+ fi
+
+ [ $# -gt 0 ] && echo "${module} options: '$*'"
+
+ # Note that insmod will ignore anything in modprobe.conf, which is why we're
+ # passing options on the command-line.
+ if [ "$BASE" == "lnet_selftest" ] && \
+ [ -f ${LUSTRE}/../lnet/selftest/${module}${EXT} ]; then
+ insmod ${LUSTRE}/../lnet/selftest/${module}${EXT}
+ elif [ -f ${LUSTRE}/${module}${EXT} ]; then
+ insmod ${LUSTRE}/${module}${EXT} "$@"
else
# must be testing a "make install" or "rpm" installation
# note failed to load ptlrpc_gss is considered not fatal
if [ "$BASE" == "ptlrpc_gss" ]; then
- modprobe $BASE $@ 2>/dev/null || echo "gss/krb5 is not supported"
+ modprobe $BASE "$@" 2>/dev/null || echo "gss/krb5 is not supported"
else
- modprobe $BASE $@
+ modprobe $BASE "$@"
fi
fi
}
-load_modules() {
+load_modules_local() {
if [ -n "$MODPROBE" ]; then
# use modprobe
- return 0
- fi
- if [ "$HAVE_MODULES" = true ]; then
- # we already loaded
+ echo "Using modprobe to load modules"
return 0
fi
- HAVE_MODULES=true
echo Loading modules from $LUSTRE
load_module ../libcfs/libcfs/libcfs
[ "$PTLDEBUG" ] && lctl set_param debug="$PTLDEBUG"
[ "$SUBSYSTEM" ] && lctl set_param subsystem_debug="${SUBSYSTEM# }"
- local MODPROBECONF=
- [ -f /etc/modprobe.conf ] && MODPROBECONF=/etc/modprobe.conf
- [ ! "$MODPROBECONF" -a -d /etc/modprobe.d ] && MODPROBECONF=/etc/modprobe.d/Lustre
- [ -z "$LNETOPTS" -a "$MODPROBECONF" ] && \
- LNETOPTS=$(awk '/^options lnet/ { print $0}' $MODPROBECONF | sed 's/^options lnet //g')
- echo $LNETOPTS | grep -q "accept=all" || LNETOPTS="$LNETOPTS accept=all";
- echo "lnet options: '$LNETOPTS'"
- # note that insmod will ignore anything in modprobe.conf
- load_module ../lnet/lnet/lnet $LNETOPTS
+ load_module ../lnet/lnet/lnet
LNETLND=${LNETLND:-"socklnd/ksocklnd"}
load_module ../lnet/klnds/$LNETLND
load_module lvfs/lvfs
load_module osc/osc
load_module lov/lov
load_module mgc/mgc
- if [ -z "$CLIENTONLY" ] && [ -z "$CLIENTMODSONLY" ]; then
+ if ! client_only; then
grep -q crc16 /proc/kallsyms || { modprobe crc16 2>/dev/null || true; }
grep -q jbd /proc/kallsyms || { modprobe jbd 2>/dev/null || true; }
[ "$FSTYPE" = "ldiskfs" ] && load_module ../ldiskfs/ldiskfs/ldiskfs
[ -f $LUSTRE/utils/mount.lustre ] && cp $LUSTRE/utils/mount.lustre /sbin/. || true
}
-RMMOD=rmmod
-if [ `uname -r | cut -c 3` -eq 4 ]; then
- RMMOD="modprobe -r"
-fi
-
-wait_for_lnet() {
- local UNLOADED=0
- local WAIT=0
- local MAX=60
- MODULES=$($LCTL modules | awk '{ print $2 }')
- while [ -n "$MODULES" ]; do
- sleep 5
- $RMMOD $MODULES > /dev/null 2>&1 || true
- MODULES=$($LCTL modules | awk '{ print $2 }')
- if [ -z "$MODULES" ]; then
- return 0
- else
- WAIT=$((WAIT + 5))
- echo "waiting, $((MAX - WAIT)) secs left"
- fi
- if [ $WAIT -eq $MAX ]; then
- echo "LNET modules $MODULES will not unload"
- lsmod
- return 3
- fi
- done
-}
-
-unload_dep_module() {
- #lsmod output
- #libcfs 107852 17 llite_lloop,lustre,obdfilter,ost,...
- local MODULE=$1
- local DEPS=$(lsmod | awk '($1 == "'$MODULE'") { print $4 }' | tr ',' ' ')
- for SUBMOD in $DEPS; do
- unload_dep_module $SUBMOD
- done
- [ "$MODULE" = "libcfs" ] && $LCTL dk $TMP/debug || true
- $RMMOD $MODULE || true
+load_modules () {
+ load_modules_local
+ # bug 19124
+ # load modules on remote nodes optionally
+ # lustre-tests have to be installed on these nodes
+ if $LOAD_MODULES_REMOTE ; then
+ local list=$(comma_list $(remote_nodes_list))
+ echo loading modules on $list
+ do_rpc_nodes $list load_modules
+ fi
}
check_mem_leak () {
unload_modules() {
wait_exit_ST client # bug 12845
- lsmod | grep libcfs > /dev/null && $LCTL dl
- unload_dep_module $FSTYPE
- unload_dep_module libcfs
-
- local MODULES=$($LCTL modules | awk '{ print $2 }')
- if [ -n "$MODULES" ]; then
- echo "Modules still loaded: "
- echo $MODULES
- if [ "$(lctl dl)" ]; then
- echo "Lustre still loaded"
- lctl dl || true
- lsmod
- return 2
- else
- echo "Lustre stopped but LNET is still loaded, waiting..."
- wait_for_lnet || return 3
- fi
+ $LUSTRE_RMMOD $FSTYPE || return 2
+ if $LOAD_MODULES_REMOTE ; then
+ local list=$(comma_list $(remote_nodes_list))
+ if [ ! -z $list ]; then
+ echo unloading modules on $list
+ do_rpc_nodes $list $LUSTRE_RMMOD $FSTYPE
+ do_rpc_nodes $list check_mem_leak
+ fi
fi
- HAVE_MODULES=false
check_mem_leak || return 254
return 0
}
+check_gss_daemon_nodes() {
+ local list=$1
+ dname=$2
+
+ do_nodesv $list "num=\\\$(ps -o cmd -C $dname | grep $dname | wc -l);
+if [ \\\"\\\$num\\\" -ne 1 ]; then
+ echo \\\$num instance of $dname;
+ exit 1;
+fi; "
+}
+
check_gss_daemon_facet() {
facet=$1
dname=$2
}
send_sigint() {
- local facet=$1
+ local list=$1
shift
- do_facet $facet "killall -2 $@ 2>/dev/null || true"
+ echo Stopping $@ on $list
+ do_nodes $list "killall -2 $@ 2>/dev/null || true"
}
+# start gss daemons on all nodes, or
+# "daemon" on "list" if set
start_gss_daemons() {
- # starting on MDT
- for num in `seq $MDSCOUNT`; do
- do_facet mds$num "$LSVCGSSD -v"
- if $GSS_PIPEFS; then
- do_facet mds$num "$LGSSD -v"
- fi
- done
- # starting on OSTs
- for num in `seq $OSTCOUNT`; do
- do_facet ost$num "$LSVCGSSD -v"
- done
- # starting on client
- # FIXME: is "client" the right facet name?
+ local list=$1
+ local daemon=$2
+
+ if [ "$list" ] && [ "$daemon" ] ; then
+ echo "Starting gss daemon on nodes: $list"
+ do_nodes $list "$daemon" || return 8
+ return 0
+ fi
+
+ local list=$(comma_list $(mdts_nodes))
+
+ echo "Starting gss daemon on mds: $list"
+ do_nodes $list "$LSVCGSSD -v" || return 1
+ if $GSS_PIPEFS; then
+ do_nodes $list "$LGSSD -v" || return 2
+ fi
+
+ list=$(comma_list $(osts_nodes))
+ echo "Starting gss daemon on ost: $list"
+ do_nodes $list "$LSVCGSSD -v" || return 3
+ # starting on clients
+
+ local clients=${CLIENTS:-`hostname`}
if $GSS_PIPEFS; then
- do_facet client "$LGSSD -v"
+ echo "Starting $LGSSD on clients $clients "
+ do_nodes $clients "$LGSSD -v" || return 4
fi
# wait daemons entering "stable" status
#
# check daemons are running
#
- for num in `seq $MDSCOUNT`; do
- check_gss_daemon_facet mds$num lsvcgssd
- if $GSS_PIPEFS; then
- check_gss_daemon_facet mds$num lgssd
- fi
- done
- for num in `seq $OSTCOUNT`; do
- check_gss_daemon_facet ost$num lsvcgssd
- done
+ list=$(comma_list $(mdts_nodes) $(osts_nodes))
+ check_gss_daemon_nodes $list lsvcgssd || return 5
+ if $GSS_PIPEFS; then
+ list=$(comma_list $(mdts_nodes))
+ check_gss_daemon_nodes $list lgssd || return 6
+ fi
if $GSS_PIPEFS; then
- check_gss_daemon_facet client lgssd
+ check_gss_daemon_nodes $clients lgssd || return 7
fi
}
stop_gss_daemons() {
- for num in `seq $MDSCOUNT`; do
- send_sigint mds$num lsvcgssd lgssd
- done
- for num in `seq $OSTCOUNT`; do
- send_sigint ost$num lsvcgssd
- done
- send_sigint client lgssd
+ local list=$(comma_list $(mdts_nodes))
+
+ send_sigint $list lsvcgssd lgssd
+
+ list=$(comma_list $(osts_nodes))
+ send_sigint $list lsvcgssd
+
+ list=${CLIENTS:-`hostname`}
+ send_sigint $list lgssd
}
init_gss() {
if $GSS; then
- start_gss_daemons
+ if ! module_loaded ptlrpc_gss; then
+ load_module ptlrpc/gss/ptlrpc_gss
+ module_loaded ptlrpc_gss ||
+ error_exit "init_gss : GSS=$GSS, but gss/krb5 is not supported!"
+ fi
+ start_gss_daemons || error_exit "start gss daemon failed! rc=$?"
if [ -n "$LGSS_KEYRING_DEBUG" ]; then
echo $LGSS_KEYRING_DEBUG > /proc/fs/lustre/sptlrpc/gss/lgss_keyring/debug_level
}
# save quota version (both administrative and operational quotas)
+# add an additional parameter if mountpoint is ever different from $MOUNT
quota_save_version() {
local fsname=${2:-$FSNAME}
- do_facet mgs "lctl conf_param ${fsname}-MDT*.mdd.quota_type=$1"
+ local spec=$1
+ local ver=$(tr -c -d "123" <<< $spec)
+ local type=$(tr -c -d "ug" <<< $spec)
+
+ [ -n "$ver" -a "$ver" != "3" ] && error "wrong quota version specifier"
+
+ [ -n "$type" ] && { $LFS quotacheck -$type $MOUNT || error "quotacheck has failed"; }
+
+ do_facet mgs "lctl conf_param ${fsname}-MDT*.mdd.quota_type=$spec"
local varsvc
local osts=$(get_facets OST)
for ost in ${osts//,/ }; do
varsvc=${ost}_svc
- do_facet mgs "lctl conf_param ${!varsvc}.ost.quota_type=$1"
+ do_facet mgs "lctl conf_param ${!varsvc}.ost.quota_type=$spec"
done
}
if [ ! "$old_QUOTA_TYPE" ] || [ "$quota_type" = "$old_QUOTA_TYPE" ]; then
return
fi
- $LFS quotaoff $mntpt
quota_save_version $old_QUOTA_TYPE
- $LFS quotacheck -ug $mntpt
}
setup_quota(){
local mntpt=$1
- # We need:
- # 1. run quotacheck only if quota is off
- # 2. save the original quota_type params, restore them after testing
+ # We need save the original quota_type params, and restore them after testing
# Suppose that quota type the same on mds and ost
local quota_type=$(quota_type | grep MDT | cut -d "=" -f2)
[ ${PIPESTATUS[0]} -eq 0 ] || error "quota_type failed!"
+ echo "[HOST:$HOSTNAME] [old_quota_type:$quota_type] [new_quota_type:$QUOTA_TYPE]"
if [ "$quota_type" != "$QUOTA_TYPE" ]; then
export old_QUOTA_TYPE=$quota_type
quota_save_version $QUOTA_TYPE
- $LFS quotacheck -ug $mntpt
+ else
+ qtype=$(tr -c -d "ug" <<< $QUOTA_TYPE)
+ $LFS quotacheck -$qtype $mntpt || error "quotacheck has failed for $type"
fi
local quota_usrs=$QUOTA_USERS
local cmd
for usr in $quota_usrs; do
- echo "Setting up quota on $client:$mntpt for $usr..."
+ echo "Setting up quota on $HOSTNAME:$mntpt for $usr..."
for type in u g; do
cmd="$LFS setquota -$type $usr -b $blk_soft -B $blk_hard -i $i_soft -I $i_hard $mntpt"
echo "+ $cmd"
local rc=0
for mnt in $mnts ; do
- do_nodes $nodes "set -x; running=\\\$(grep -c $mnt' ' /proc/mounts);
+ do_nodes $nodes "running=\\\$(grep -c $mnt' ' /proc/mounts);
mpts=\\\$(mount | grep -w -c $mnt);
if [ \\\$running -ne \\\$mpts ]; then
echo \\\$(hostname) env are INSANE!;
}
sanity_mount_check_servers () {
+ [ "$CLIENTONLY" ] &&
+ { echo "CLIENTONLY mode, skip mount_check_servers"; return 0; } || true
echo Checking servers environments
# FIXME: modify get_facets to display all facets wo params
- local facets="$(get_facets OST),$(get_facets MDS)"
+ local facets="$(get_facets OST),$(get_facets MDS),mgs"
local node
local mnt
local facet
# mount clients if not mouted
zconf_mount_clients() {
- local OPTIONS
local clients=$1
local mnt=$2
-
+ local OPTIONS=${3:-$MOUNTOPT}
# Only supply -o to mount if we have options
- if [ -n "$MOUNTOPT" ]; then
- OPTIONS="-o $MOUNTOPT"
+ if [ "$OPTIONS" ]; then
+ OPTIONS="-o $OPTIONS"
fi
local device=$MGSNID:/$FSNAME
if [ -z "$mnt" -o -z "$FSNAME" ]; then
echo "Starting client $clients: $OPTIONS $device $mnt"
- do_nodes $clients "set -x;
+ do_nodes $clients "
running=\\\$(mount | grep -c $mnt' ');
rc=0;
if [ \\\$running -eq 0 ] ; then
mkdir -p $mnt;
mount -t lustre $OPTIONS $device $mnt;
- rc=$?;
+ rc=\\\$?;
fi;
-exit $rc"
+exit \\\$rc" || return ${PIPESTATUS[0]}
echo "Started clients $clients: "
do_nodes $clients "mount | grep -w $mnt"
[ "$3" ] && force=-f
echo "Stopping clients: $clients $mnt (opts:$force)"
- do_nodes $clients "set -x; running=\\\$(grep -c $mnt' ' /proc/mounts);
+ do_nodes $clients "running=\\\$(grep -c $mnt' ' /proc/mounts);
if [ \\\$running -ne 0 ] ; then
-echo Stopping client \\\$(hostname) client $mnt opts:$force;
+echo Stopping client \\\$(hostname) $mnt opts:$force;
lsof -t $mnt || need_kill=no;
if [ "x$force" != "x" -a "x\\\$need_kill" != "xno" ]; then
pids=\\\$(lsof -t $mnt | sort -u);
local node=$1
if [ "$FAILURE_MODE" = HARD ]; then
$POWER_UP $node
+ wait_for_host $node
fi
}
shift
local progs=$@
- do_nodes $clients "set -x ; PATH=:$PATH status=true; for prog in $progs; do
- which \\\$prog || { echo \\\$prog missing on \\\$(hostname) && status=false; }
- done;
- eval \\\$status"
+ do_nodes $clients "PATH=:$PATH; status=true;
+for prog in $progs; do
+ if ! [ \\\"\\\$(which \\\$prog)\\\" -o \\\"\\\${!prog}\\\" ]; then
+ echo \\\$prog missing on \\\$(hostname);
+ status=false;
+ fi
+done;
+eval \\\$status"
}
client_var_name() {
testnum=$((nodenum % numloads))
start_client_load ${clients[nodenum]} ${CLIENT_LOADS[testnum]}
done
+ # bug 22169: wait the background threads to start
+ sleep 2
}
# only for remote client
local TESTLOAD=run_${!var}.sh
ps auxww | grep -v grep | grep $client | grep -q "$TESTLOAD" || return 1
-
- check_catastrophe $client || return 2
-
- # see if the load is still on the client
+
+ # bug 18914: try to connect several times not only when
+ # check ps, but while check_catastrophe also
local tries=3
local RC=254
while [ $RC = 254 -a $tries -gt 0 ]; do
let tries=$tries-1
# assume success
RC=0
+ if ! check_catastrophe $client; then
+ RC=${PIPESTATUS[0]}
+ if [ $RC -eq 254 ]; then
+ # FIXME: not sure how long we shuold sleep here
+ sleep 10
+ continue
+ fi
+ echo "check catastrophe failed: RC=$RC "
+ return $RC
+ fi
+ done
+ # We can continue try to connect if RC=254
+ # Just print the warning about this
+ if [ $RC = 254 ]; then
+ echo "got a return status of $RC from do_node while checking catastrophe on $client"
+ fi
+
+ # see if the load is still on the client
+ tries=3
+ RC=254
+ while [ $RC = 254 -a $tries -gt 0 ]; do
+ let tries=$tries-1
+ # assume success
+ RC=0
if ! do_node $client "ps auxwww | grep -v grep | grep -q $TESTLOAD"; then
RC=${PIPESTATUS[0]}
sleep 30
fi
done
if [ $RC = 254 ]; then
- echo "got a return status of $RC from do_node while checking (i.e. with 'ps') the client load on the remote system"
+ echo "got a return status of $RC from do_node while checking (catastrophe and 'ps') the client load on $client"
# see if we can diagnose a bit why this is
fi
check_client_load $client
rc=${PIPESTATUS[0]}
if [ "$rc" != 0 -a "$expectedfail" ]; then
- start_client_load $client
- echo "Restarted client load: on $client. Checking ..."
- check_client_load $client
+ local var=$(client_var_name $client)_load
+ start_client_load $client ${!var}
+ echo "Restarted client load ${!var}: on $client. Checking ..."
+ check_client_load $client
rc=${PIPESTATUS[0]}
if [ "$rc" != 0 ]; then
log "Client load failed to restart on node $client, rc=$rc"
[ "`lctl dl 2> /dev/null | wc -l`" -gt 0 ] && lctl dl && \
echo "$0: lustre didn't clean up..." 1>&2 && return 202 || true
- if [ "`/sbin/lsmod 2>&1 | egrep 'lnet|libcfs'`" ]; then
+ if module_loaded lnet || module_loaded libcfs; then
echo "$0: modules still loaded..." 1>&2
/sbin/lsmod 1>&2
return 203
sleep 1
TOTAL=`lctl get_param -n osc.*.kbytesavail | \
awk 'BEGIN{total=0}; {total+=$1}; END{print total}'`
- [ "$TOTAL" -eq "$TOTALPREV" ] && break
+ [ "$TOTAL" -eq "$TOTALPREV" ] && return 0
echo "Waiting delete completed ... prev: $TOTALPREV current: $TOTAL "
TOTALPREV=$TOTAL
WAIT=$(( WAIT + 1))
done
- echo "Delete completed."
+ echo "Delete is not completed in $MAX_WAIT sec"
+ return 1
}
wait_for_host() {
local host=$1
check_network "$host" 900
- while ! do_node $host "ls -d $LUSTRE " > /dev/null; do sleep 5; done
+ while ! do_node $host hostname > /dev/null; do sleep 5; done
}
wait_for() {
wait_recovery_complete () {
local facet=$1
- # Use default policy if $2 is not passed by caller.
- #define OBD_RECOVERY_TIMEOUT (obd_timeout * 5 / 2)
- # as we are in process of changing obd_timeout in different ways
- # let's set MAX longer than that
- local MAX=${2:-$(( TIMEOUT * 4 ))}
-
+ # Use default policy if $2 is not passed by caller.
+ local MAX=${2:-$(max_recovery_time)}
+
local var_svc=${facet}_svc
local procfile="*.${!var_svc}.recovery_status"
local WAIT=0
return 1
}
+wait_mds_ost_sync () {
+ # just because recovery is done doesn't mean we've finished
+ # orphan cleanup. Wait for llogs to get synchronized.
+ echo "Waiting for orphan cleanup..."
+ # MAX value includes time needed for MDS-OST reconnection
+ local MAX=$(( TIMEOUT * 2 ))
+ local WAIT=0
+ while [ $WAIT -lt $MAX ]; do
+ local -a sync=($(do_nodes $(comma_list $(osts_nodes)) \
+ "$LCTL get_param -n obdfilter.*.mds_sync"))
+ local con=1
+ for ((i=0; i<${#sync[@]}; i++)); do
+ [ ${sync[$i]} -eq 0 ] && continue
+ # there is a not finished MDS-OST synchronization
+ con=0
+ break;
+ done
+ sleep 2 # increase waiting time and cover statfs cache
+ [ ${con} -eq 1 ] && return 0
+ echo "Waiting $WAIT secs for $facet mds-ost sync done."
+ WAIT=$((WAIT + 2))
+ done
+ echo "$facet recovery not done in $MAX sec. $STATUS"
+ return 1
+}
+
+wait_destroy_complete () {
+ echo "Waiting for destroy to be done..."
+ # MAX value shouldn't be big as this mean server responsiveness
+ # never increase this just to make test pass but investigate
+ # why it takes so long time
+ local MAX=5
+ local WAIT=0
+ while [ $WAIT -lt $MAX ]; do
+ local -a RPCs=($($LCTL get_param -n osc.*.destroys_in_flight))
+ local con=1
+ for ((i=0; i<${#RPCs[@]}; i++)); do
+ [ ${RPCs[$i]} -eq 0 ] && continue
+ # there are still some destroy RPCs in flight
+ con=0
+ break;
+ done
+ sleep 1
+ [ ${con} -eq 1 ] && return 0 # done waiting
+ echo "Waiting $WAIT secs for destroys to be done."
+ WAIT=$((WAIT + 1))
+ done
+ echo "Destroys weren't done in $MAX sec."
+ return 1
+}
+
wait_exit_ST () {
local facet=$1
return $rc
}
-client_df() {
+clients_up() {
# not every config has many clients
+ sleep 1
if [ ! -z "$CLIENTS" ]; then
- $PDSH $CLIENTS "df $MOUNT" > /dev/null
+ $PDSH $CLIENTS "stat -f $MOUNT" > /dev/null
+ else
+ stat -f $MOUNT > /dev/null
+ fi
+}
+
+client_up() {
+ local client=$1
+ # usually checked on particular client or locally
+ sleep 1
+ if [ ! -z "$client" ]; then
+ $PDSH $client "stat -f $MOUNT" > /dev/null
else
- df $MOUNT > /dev/null
+ stat -f $MOUNT > /dev/null
fi
}
+client_evicted() {
+ ! client_up $1
+}
+
client_reconnect() {
uname -n >> $MOUNT/recon
if [ -z "$CLIENTS" ]; then
shutdown_facet $facet
[ -n "$sleep_time" ] && sleep $sleep_time
reboot_facet $facet
- client_df &
- DFPID=$!
- echo "df pid is $DFPID"
change_active $facet
local TO=`facet_active_host $facet`
echo "Failover $facet to $TO"
fail() {
facet_failover $* || error "failover: $?"
- df $MOUNT || error "post-failover df: $?"
+ clients_up || error "post-failover df: $?"
}
fail_nodf() {
stop $facet
change_active $facet
mount_facet $facet -o abort_recovery
- df $MOUNT || echo "first df failed: $?"
- sleep 1
- df $MOUNT || error "post-failover df: $?"
+ clients_up || echo "first df failed: $?"
+ clients_up || error "post-failover df: $?"
}
do_lmc() {
}
do_node() {
- HOST=$1
+ local verbose=false
+ # do not stripe off hostname if verbose, bug 19215
+ if [ x$1 = x--verbose ]; then
+ shift
+ verbose=true
+ fi
+
+ local HOST=$1
shift
local myPDSH=$PDSH
if [ "$HOST" = "$HOSTNAME" ]; then
if [ "$myPDSH" = "rsh" ]; then
# we need this because rsh does not return exit code of an executed command
- local command_status="$TMP/cs"
- rsh $HOST ":> $command_status"
- rsh $HOST "(PATH=\$PATH:$RLUSTRE/utils:$RLUSTRE/tests:/sbin:/usr/sbin;
- cd $RPWD; sh -c \"$@\") ||
- echo command failed >$command_status"
- [ -n "$($myPDSH $HOST cat $command_status)" ] && return 1 || true
+ local command_status="$TMP/cs"
+ rsh $HOST ":> $command_status"
+ rsh $HOST "(PATH=\$PATH:$RLUSTRE/utils:$RLUSTRE/tests:/sbin:/usr/sbin;
+ cd $RPWD; LUSTRE=\"$RLUSTRE\" sh -c \"$@\") ||
+ echo command failed >$command_status"
+ [ -n "$($myPDSH $HOST cat $command_status)" ] && return 1 || true
return 0
fi
- $myPDSH $HOST "(PATH=\$PATH:$RLUSTRE/utils:$RLUSTRE/tests:/sbin:/usr/sbin; cd $RPWD; sh -c \"$@\")" | sed "s/^${HOST}: //"
+
+ if $verbose ; then
+ # print HOSTNAME for myPDSH="no_dsh"
+ if [[ $myPDSH = no_dsh ]]; then
+ $myPDSH $HOST "(PATH=\$PATH:$RLUSTRE/utils:$RLUSTRE/tests:/sbin:/usr/sbin; cd $RPWD; LUSTRE=\"$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 ${PIPESTATUS[0]}
}
+do_nodev() {
+ do_node --verbose "$@"
+}
+
single_local_node () {
[ "$1" = "$HOSTNAME" ]
}
+# Outputs environment variable assignments that should be passed to remote nodes
+get_env_vars() {
+ local var
+ local value
+
+ for var in ${!MODOPTS_*}; do
+ value=${!var}
+ echo "${var}=\"$value\""
+ done
+}
+
do_nodes() {
+ local verbose=false
+ # do not stripe off hostname if verbose, bug 19215
+ if [ x$1 = x--verbose ]; then
+ shift
+ verbose=true
+ fi
+
local rnodes=$1
shift
- if $(single_local_node $rnodes); then
- do_node $rnodes $@
+ if single_local_node $rnodes; then
+ if $verbose; then
+ do_nodev $rnodes "$@"
+ else
+ do_node $rnodes "$@"
+ fi
return $?
fi
$myPDSH $rnodes $LCTL mark "$@" > /dev/null 2>&1 || :
fi
- $myPDSH $rnodes "(PATH=\$PATH:$RLUSTRE/utils:$RLUSTRE/tests:/sbin:/usr/sbin; cd $RPWD; sh -c \"$@\")" | sed -re "s/\w+:\s//g"
+ if $verbose ; then
+ $myPDSH $rnodes "(PATH=\$PATH:$RLUSTRE/utils:$RLUSTRE/tests:/sbin:/usr/sbin; cd $RPWD; LUSTRE=\"$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"
+ fi
return ${PIPESTATUS[0]}
}
do_node $HOST "$@"
}
+do_nodesv() {
+ do_nodes --verbose "$@"
+}
+
add() {
local facet=$1
shift
rm -f $TMP/ost${num}active
done
+ if ! combined_mgs_mds ; then
+ stop mgs
+ fi
+
return 0
}
cleanupall() {
+ nfs_client_mode && return
+
stopall $*
unload_modules
cleanup_gss
test $nr = 1 && echo -n $MDS_MKFS_OPTS || echo -n $MDSn_MKFS_OPTS
}
+combined_mgs_mds () {
+ [[ $MDSDEV1 = $MGSDEV ]] && [[ $mds1_HOST = $mgs_HOST ]]
+}
+
formatall() {
if [ "$IAMDIR" == "yes" ]; then
MDS_MKFS_OPTS="$MDS_MKFS_OPTS --iam-dir"
[ "$FSTYPE" ] && FSTYPE_OPT="--backfstype $FSTYPE"
- if [ ! -z $SEC ]; then
- MDS_MKFS_OPTS="$MDS_MKFS_OPTS --param srpc.flavor.default=$SEC"
- MDSn_MKFS_OPTS="$MDSn_MKFS_OPTS --param srpc.flavor.default=$SEC"
- OST_MKFS_OPTS="$OST_MKFS_OPTS --param srpc.flavor.default=$SEC"
- fi
-
stopall
# We need ldiskfs here, may as well load them all
load_modules
[ "$CLIENTONLY" ] && return
- echo "Formatting mdts, osts"
+ echo Formatting mgs, mds, osts
+ if ! combined_mgs_mds ; then
+ add mgs $mgs_MKFS_OPTS $FSTYPE_OPT --reformat $MGSDEV || exit 10
+ fi
+
for num in `seq $MDSCOUNT`; do
echo "Format mds$num: $(mdsdevname $num)"
if $VERBOSE; then
remount_client()
{
- zconf_umount `hostname` $1 || error "umount failed"
- zconf_mount `hostname` $1 || error "mount failed"
+ zconf_umount `hostname` $1 || error "umount failed"
+ zconf_mount `hostname` $1 || error "mount failed"
}
writeconf_facet () {
}
setupall() {
+ nfs_client_mode && return
+
sanity_mount_check ||
error "environments are insane!"
load_modules
- init_gss
+
if [ -z "$CLIENTONLY" ]; then
- echo "Setup mdts, osts"
+ echo Setup mgs, mdt, osts
echo $WRITECONF | grep -q "writeconf" && \
writeconf_all
+ if ! combined_mgs_mds ; then
+ start mgs $MGSDEV $mgs_MOUNT_OPTS
+ fi
+
for num in `seq $MDSCOUNT`; do
DEVNAME=$(mdsdevname $num)
start mds$num $DEVNAME $MDS_MOUNT_OPTS
eval mds${num}failover_HOST=$(facet_host mds$num)
fi
- if [ $IDENTITY_UPCALL != "default" ]; then
+ if [ $IDENTITY_UPCALL != "default" ]; then
switch_identity $num $IDENTITY_UPCALL
- fi
+ fi
done
for num in `seq $OSTCOUNT`; do
DEVNAME=$(ostdevname $num)
done
fi
+
+ init_gss
+
# wait a while to allow sptlrpc configuration be propogated to targets,
# only needed when mounting new target devices.
- $GSS && sleep 10
+ if $GSS; then
+ sleep 10
+ fi
[ "$DAEMONFILE" ] && $LCTL debug_daemon start $DAEMONFILE $DAEMONSIZE
mount_client $MOUNT
[ -n "$CLIENTS" ] && zconf_mount_clients $CLIENTS $MOUNT
if [ "$MOUNT_2" ]; then
- mount_client $MOUNT2
+ mount_client $MOUNT2
[ -n "$CLIENTS" ] && zconf_mount_clients $CLIENTS $MOUNT2
fi
# by a context negotiation rpc with $TIMEOUT.
# FIXME better by monitoring import status.
if $GSS; then
+ set_flavor_all $SEC
sleep $((TIMEOUT + 5))
else
sleep 5
}
mounted_lustre_filesystems() {
- awk '($3 ~ "lustre" && $1 ~ ":") { print $2 }' /proc/mounts
+ awk '($3 ~ "lustre" && $1 ~ ":") { print $2 }' /proc/mounts
}
init_facet_vars () {
+ [ "$CLIENTONLY" ] && return 0
local facet=$1
shift
local device=$1
done
}
+osc_ensure_active () {
+ local facet=$1
+ local type=$2
+ local timeout=$3
+ local period=0
+
+ while [ $period -lt $timeout ]; do
+ count=$(do_facet $facet "lctl dl | grep '${FSNAME}-OST.*-osc-${type}' | grep ' IN ' 2>/dev/null | wc -l")
+ if [ $count -eq 0 ]; then
+ break
+ fi
+
+ echo "There are $count OST are inactive, wait $period seconds, and try again"
+ sleep 3
+ period=$((period+3))
+ done
+
+ [ $period -lt $timeout ] || log "$count OST are inactive after $timeout seconds, give up"
+}
+
init_param_vars () {
if ! remote_ost_nodsh && ! remote_mds_nodsh; then
export MDSVER=$(do_facet $SINGLEMDS "lctl get_param version" | cut -d. -f1,2)
log "Using TIMEOUT=$TIMEOUT"
- if [ "$ENABLE_QUOTA" ]; then
- setup_quota $MOUNT || return 2
+ osc_ensure_active $SINGLEMDS M $TIMEOUT
+ osc_ensure_active client c $TIMEOUT
+
+ if [ $QUOTA_AUTO -ne 0 ]; then
+ if [ "$ENABLE_QUOTA" ]; then
+ echo "enable quota as required"
+ setup_quota $MOUNT || return 2
+ else
+ echo "disable quota as required"
+ $LFS quotaoff -ug $MOUNT > /dev/null 2>&1
+ fi
+ fi
+
+ return 0
+}
+
+nfs_client_mode () {
+ if [ "$NFSCLIENT" ]; then
+ echo "NFSCLIENT mode: setup, cleanup, check config skipped"
+ local clients=$CLIENTS
+ [ -z $clients ] && clients=$(hostname)
+
+ # FIXME: remove hostname when 19215 fixed
+ do_nodes $clients "echo \\\$(hostname); grep ' '$MOUNT' ' /proc/mounts"
+ declare -a nfsexport=(`grep ' '$MOUNT' ' /proc/mounts | awk '{print $1}' | awk -F: '{print $1 " " $2}'`)
+ do_nodes ${nfsexport[0]} "echo \\\$(hostname); df -T ${nfsexport[1]}"
+ return
fi
+ return 1
}
-check_config () {
+check_config_client () {
local mntpt=$1
+
+ local mounted=$(mount | grep " $mntpt ")
+ if [ "$CLIENTONLY" ]; then
+ # bug 18021
+ # CLIENTONLY should not depend on *_HOST settings
+ local mgc=$($LCTL device_list | awk '/MGC/ {print $4}')
+ # in theory someone could create a new,
+ # client-only config file that assumed lustre was already
+ # configured and didn't set the MGSNID. If MGSNID is not set,
+ # then we should use the mgs nid currently being used
+ # as the default value. bug 18021
+ [[ x$MGSNID = x ]] &&
+ MGSNID=${mgc//MGC/}
+
+ if [[ x$mgc != xMGC$MGSNID ]]; then
+ if [ "$mgs_HOST" ]; then
+ local mgc_ip=$(ping -q -c1 -w1 $mgs_HOST | grep PING | awk '{print $3}' | sed -e "s/(//g" -e "s/)//g")
+# [[ x$mgc = xMGC$mgc_ip@$NETTYPE ]] ||
+# error_exit "MGSNID=$MGSNID, mounted: $mounted, MGC : $mgc"
+ fi
+ fi
+ return 0
+ fi
+
local myMGS_host=$mgs_HOST
if [ "$NETTYPE" = "ptl" ]; then
myMGS_host=$(h2ptl $mgs_HOST | sed -e s/@ptl//)
local mgshost=$(mount | grep " $mntpt " | awk -F@ '{print $1}')
mgshost=$(echo $mgshost | awk -F: '{print $1}')
- if [ "$mgshost" != "$myMGS_host" ]; then
- FAIL_ON_ERROR=true \
- error "Bad config file: lustre is mounted with mgs $mgshost, but mgs_HOST=$mgs_HOST, NETTYPE=$NETTYPE
- Please use correct config or set mds_HOST correctly!"
- fi
+# if [ "$mgshost" != "$myMGS_host" ]; then
+# log "Bad config file: lustre is mounted with mgs $mgshost, but mgs_HOST=$mgs_HOST, NETTYPE=$NETTYPE
+# Please use correct config or set mds_HOST correctly!"
+# fi
+
+}
+
+check_config_clients () {
+ local clients=${CLIENTS:-$HOSTNAME}
+ local mntpt=$1
+
+ nfs_client_mode && return
+
+ do_rpc_nodes $clients check_config_client $mntpt
sanity_mount_check ||
error "environments are insane!"
fi
}
+is_mounted () {
+ local mntpt=$1
+ local mounted=$(mounted_lustre_filesystems)
+
+ echo $mounted' ' | grep -w -q $mntpt' '
+}
+
check_and_setup_lustre() {
+ nfs_client_mode && return
+
local MOUNTED=$(mounted_lustre_filesystems)
- if [ -z "$MOUNTED" ] || ! $(echo $MOUNTED | grep -w -q $MOUNT); then
+
+ local do_check=true
+ # 1.
+ # both MOUNT and MOUNT2 are not mounted
+ if ! is_mounted $MOUNT && ! is_mounted $MOUNT2; then
[ "$REFORMAT" ] && formatall
+ # setupall mounts both MOUNT and MOUNT2 (if MOUNT_2 is set)
setupall
- MOUNTED=$(mounted_lustre_filesystems | head -1)
- [ -z "$MOUNTED" ] && error "NAME=$NAME not mounted"
+ is_mounted $MOUNT || error "NAME=$NAME not mounted"
export I_MOUNTED=yes
- else
- check_config $MOUNT
+ do_check=false
+ # 2.
+ # MOUNT2 is mounted
+ elif is_mounted $MOUNT2; then
+ # 3.
+ # MOUNT2 is mounted, while MOUNT_2 is not set
+ if ! [ "$MOUNT_2" ]; then
+ cleanup_mount $MOUNT2
+ export I_UMOUNTED2=yes
+
+ # 4.
+ # MOUNT2 is mounted, MOUNT_2 is set
+ else
+ # FIXME: what to do if check_config failed?
+ # i.e. if:
+ # 1) remote client has mounted other Lustre fs ?
+ # 2) it has insane env ?
+ # let's try umount MOUNT2 on all clients and mount it again:
+ if ! check_config_clients $MOUNT2; then
+ cleanup_mount $MOUNT2
+ restore_mount $MOUNT2
+ export I_MOUNTED2=yes
+ fi
+ fi
+
+ # 5.
+ # MOUNT is mounted MOUNT2 is not mounted
+ elif [ "$MOUNT_2" ]; then
+ restore_mount $MOUNT2
+ export I_MOUNTED2=yes
+ fi
+
+ if $do_check; then
+ # FIXME: what to do if check_config failed?
+ # i.e. if:
+ # 1) remote client has mounted other Lustre fs?
+ # 2) lustre is mounted on remote_clients atall ?
+ check_config_clients $MOUNT
init_facets_vars
init_param_vars
+
+ do_nodes $(comma_list $(nodes_list)) "lctl set_param debug=\\\"$PTLDEBUG\\\";
+ lctl set_param subsystem_debug=\\\"${SUBSYSTEM# }\\\";
+ lctl set_param debug_mb=${DEBUG_SIZE};
+ sync"
fi
+
+ init_gss
+ set_flavor_all $SEC
+
if [ "$ONLY" == "setup" ]; then
exit 0
fi
}
+restore_mount () {
+ local clients=${CLIENTS:-$HOSTNAME}
+ local mntpt=$1
+
+ zconf_mount_clients $clients $mntpt
+}
+
+cleanup_mount () {
+ local clients=${CLIENTS:-$HOSTNAME}
+ local mntpt=$1
+
+ zconf_umount_clients $clients $mntpt
+}
+
cleanup_and_setup_lustre() {
if [ "$ONLY" == "cleanup" -o "`mount | grep $MOUNT`" ]; then
lctl set_param debug=0 || true
cleanupall
if [ "$ONLY" == "cleanup" ]; then
- exit 0
+ exit 0
fi
fi
check_and_setup_lustre
}
check_and_cleanup_lustre() {
- if [ "`mount | grep $MOUNT`" ]; then
+ if is_mounted $MOUNT; then
[ -n "$DIR" ] && rm -rf $DIR/[Rdfs][0-9]*
[ "$ENABLE_QUOTA" ] && restore_quota_type || true
fi
+
+ if [ "$I_UMOUNTED2" = "yes" ]; then
+ restore_mount $MOUNT2 || error "restore $MOUNT2 failed"
+ fi
+
+ if [ "$I_MOUNTED2" = "yes" ]; then
+ cleanup_mount $MOUNT2
+ fi
+
if [ "$I_MOUNTED" = "yes" ]; then
cleanupall -f || error "cleanup failed"
+ unset I_MOUNTED
fi
- unset I_MOUNTED
}
#######
local WAIT=0
local MAX=$2
while [ $NETWORK -eq 0 ]; do
- ping -c 1 -w 3 $1 > /dev/null
- if [ $? -eq 0 ]; then
+ if ping -c 1 -w 3 $1 > /dev/null; then
NETWORK=1
else
WAIT=$((WAIT + 5))
echo $(comma_list $expanded)
}
+testslist_filter () {
+ local script=$LUSTRE/tests/${TESTSUITE}.sh
+
+ [ -f $script ] || return 0
+
+ local start_at=$START_AT
+ local stop_at=$STOP_AT
+
+ local var=${TESTSUITE//-/_}_START_AT
+ [ x"${!var}" != x ] && start_at=${!var}
+ var=${TESTSUITE//-/_}_STOP_AT
+ [ x"${!var}" != x ] && stop_at=${!var}
+
+ sed -n 's/^test_\([^ (]*\).*/\1/p' $script | \
+ awk ' BEGIN { if ("'${start_at:-0}'" != 0) flag = 1 }
+ /^'${start_at}'$/ {flag = 0}
+ {if (flag == 1) print $0}
+ /^'${stop_at}'$/ { flag = 1 }'
+}
+
absolute_path() {
(cd `dirname $1`; echo $PWD/`basename $1`)
}
##################################
# Adaptive Timeouts funcs
-at_is_valid() {
- if [ -z "$AT_MAX_PATH" ]; then
- AT_MAX_PATH=$(do_facet $SINGLEMDS "find /sys/ -name at_max")
- [ -z "$AT_MAX_PATH" ] && echo "missing /sys/.../at_max " && return 1
- fi
- return 0
-}
-
at_is_enabled() {
- at_is_valid || error "invalid call"
-
# only check mds, we assume at_max is the same on all nodes
- local at_max=$(do_facet $SINGLEMDS "cat $AT_MAX_PATH")
+ local at_max=$(do_facet $SINGLEMDS "lctl get_param -n at_max")
if [ $at_max -eq 0 ]; then
return 1
else
at_max_get() {
local facet=$1
- at_is_valid || error "invalid call"
-
# suppose that all ost-s has the same at_max set
if [ $facet == "ost" ]; then
- do_facet ost1 "cat $AT_MAX_PATH"
+ do_facet ost1 "lctl get_param -n at_max"
else
- do_facet $facet "cat $AT_MAX_PATH"
+ do_facet $facet "lctl get_param -n at_max"
fi
}
local at_max=$1
shift
- at_is_valid || error "invalid call"
-
local facet
for facet in $@; do
if [ $facet == "ost" ]; then
for i in `seq $OSTCOUNT`; do
- do_facet ost$i "echo $at_max > $AT_MAX_PATH"
+ do_facet ost$i "lctl set_param at_max=$at_max"
+
done
elif [ $facet == "mds" ]; then
for i in `seq $MDSCOUNT`; do
- do_facet mds$i "echo $at_max > $AT_MAX_PATH"
+ do_facet mds$i "lctl set_param at_max=$at_max"
done
else
- do_facet $facet "echo $at_max > $AT_MAX_PATH"
+ do_facet $facet "lctl set_param at_max=$at_max"
fi
done
}
}
set_nodes_failloc () {
- local nodes=$1
- local node
-
- for node in $nodes ; do
- do_node $node lctl set_param fail_loc=$2
- done
+ do_nodes $(comma_list $1) lctl set_param fail_loc=$2
}
cancel_lru_locks() {
}
debugrestore() {
- [ -n "$DEBUGSAVE" ] && lctl set_param debug="${DEBUGSAVE}"
+ [ -n "$DEBUGSAVE" ] && \
+ do_nodes $(comma_list $(nodes_list)) "$LCTL set_param debug=\\\"${DEBUGSAVE}\\\";"
DEBUGSAVE=""
}
+debug_size_save() {
+ DEBUG_SIZE_SAVED="$(lctl get_param -n debug_mb)"
+}
+
+debug_size_restore() {
+ [ -n "$DEBUG_SIZE_SAVED" ] && \
+ do_nodes $(comma_list $(nodes_list)) "$LCTL set_param debug_mb=$DEBUG_SIZE_SAVED"
+ DEBUG_SIZE_SAVED=""
+}
+
+start_full_debug_logging() {
+ debugsave
+ debug_size_save
+
+ local FULLDEBUG=-1
+ local DEBUG_SIZE=150
+
+ do_nodes $(comma_list $(nodes_list)) "$LCTL set_param debug_mb=$DEBUG_SIZE"
+ do_nodes $(comma_list $(nodes_list)) "$LCTL set_param debug=$FULLDEBUG;"
+}
+
+stop_full_debug_logging() {
+ debug_size_restore
+ debugrestore
+}
+
##################################
# Test interface
##################################
error_noexit() {
local TYPE=${TYPE:-"FAIL"}
- local ERRLOG
- lctl set_param fail_loc=0 2>/dev/null || true
- log " ${TESTSUITE} ${TESTNAME}: @@@@@@ ${TYPE}: $@ "
- ERRLOG=$TMP/lustre_${TESTSUITE}_${TESTNAME}.$(date +%s)
- echo "Dumping lctl log to $ERRLOG"
- # We need to dump the logs on all nodes
- local NODES=${NODES:-$(nodes_list)}
- for NODE in $NODES; do
- do_node $NODE $LCTL dk $ERRLOG
- done
+
+ local dump=true
+ # do not dump logs if $1=false
+ if [ "x$1" = "xfalse" ]; then
+ shift
+ dump=false
+ fi
+
+ log " ${TESTSUITE} ${TESTNAME}: @@@@@@ ${TYPE}: $@ "
+
+ # We need to dump the logs on all nodes
+ if $dump; then
+ gather_logs $(comma_list $(nodes_list))
+ fi
+
debugrestore
[ "$TESTSUITELOG" ] && echo "$0: ${TYPE}: $TESTNAME $@" >> $TESTSUITELOG
- TEST_FAILED=true
+ echo "$@" > $LOGDIR/err
}
error() {
error_noexit "$@"
- $FAIL_ON_ERROR && exit 1 || true
+ exit 1
}
error_exit() {
- error_noexit "$@"
- exit 1
+ error "$@"
}
# use only if we are ignoring failures for this test, bugno required.
error_noexit "$@"
}
+skip_env () {
+ $FAIL_ON_SKIP_ENV && error false $@ || skip $@
+}
+
skip () {
echo
log " SKIP: ${TESTSUITE} ${TESTNAME} $@"
}
build_test_filter() {
+ EXCEPT="$EXCEPT $(testslist_filter)"
+
[ "$ONLY" ] && log "only running test `echo $ONLY`"
for O in $ONLY; do
eval ONLY_${O}=true
done
for G in $GRANT_CHECK_LIST; do
eval GCHECK_ONLY_${G}=true
- done
-}
-
-_basetest() {
- echo $*
+ done
}
basetest() {
- IFS=abcdefghijklmnopqrstuvwxyz _basetest $1
+ if [[ $1 = [a-z]* ]]; then
+ echo $1
+ else
+ echo ${1%%[a-z]*}
+ fi
}
# print a newline if the last test was skipped
export LAST_SKIPPED=
+#
+# Main entry into test-framework. This is called with the name and
+# description of a test. The name is used to find the function to run
+# the test using "test_$name".
+#
+# This supports a variety of methods of specifying specific test to
+# run or not run. These need to be documented...
+#
run_test() {
assert_DIR
testname=ONLY_$1
if [ ${!testname}x != x ]; then
[ "$LAST_SKIPPED" ] && echo "" && LAST_SKIPPED=
- run_one $1 "$2"
+ run_one_logged $1 "$2"
return $?
fi
testname=ONLY_$base
if [ ${!testname}x != x ]; then
[ "$LAST_SKIPPED" ] && echo "" && LAST_SKIPPED=
- run_one $1 "$2"
+ run_one_logged $1 "$2"
return $?
fi
LAST_SKIPPED="y"
fi
LAST_SKIPPED=
- run_one $1 "$2"
+ run_one_logged $1 "$2"
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() {
echo "$*"
- lsmod | grep lnet > /dev/null || load_modules
+ module_loaded lnet || load_modules
local MSG="$*"
# Get rid of '
MSG=${MSG//\>/\\\>}
MSG=${MSG//\</\\\<}
MSG=${MSG//\//\\\/}
- local NODES=$(nodes_list)
- for NODE in $NODES; do
- do_node $NODE $LCTL mark "$MSG" 2> /dev/null || true
- done
+ do_nodes $(comma_list $(nodes_list)) $LCTL mark "$MSG" 2> /dev/null || true
}
trace() {
- log "STARTING: $*"
- strace -o $TMP/$1.strace -ttt $*
- RC=$?
- log "FINISHED: $*: rc $RC"
- return 1
+ log "STARTING: $*"
+ strace -o $TMP/$1.strace -ttt $*
+ RC=$?
+ log "FINISHED: $*: rc $RC"
+ return 1
}
pass() {
- $TEST_FAILED && echo -n "FAIL " || echo -n "PASS "
- echo $@
+ # Set TEST_STATUS here; will be used for logging the result
+ if [ -f $LOGDIR/err ]; then
+ TEST_STATUS="FAIL"
+ else
+ TEST_STATUS="PASS"
+ fi
+ echo $TEST_STATUS " " $@
}
check_mds() {
}
reset_fail_loc () {
- local myNODES=$(nodes_list)
- local NODE
-
echo -n "Resetting fail_loc on all nodes..."
- for NODE in $myNODES; do
- do_node $NODE "lctl set_param -n fail_loc=0 2>/dev/null || true"
- done
+ do_nodes $(comma_list $(nodes_list)) "lctl set_param -n fail_loc=0 2>/dev/null || true"
echo done.
}
+
+#
+# Log a message (on all nodes) padded with "=" before and after.
+# Also appends a timestamp and prepends the testsuite name.
+#
+
+EQUALS="===================================================================================================="
+banner() {
+ msg="== ${TESTSUITE} $*"
+ last=${msg: -1:1}
+ [[ $last != "=" && $last != " " ]] && msg="$msg "
+ msg=$(printf '%s%.*s' "$msg" $((${#EQUALS} - ${#msg})) $EQUALS )
+ # always include at least == after the message
+ log "$msg== $(date +"%H:%M:%S (%s)")"
+}
+
+#
+# Run a single test function and cleanup after it.
+#
+# This function should be run in a subshell so the test func can
+# exit() without stopping the whole script.
+#
run_one() {
- testnum=$1
- message=$2
+ local testnum=$1
+ local message=$2
tfile=f${testnum}
export tdir=d0.${TESTSUITE}/d${base}
-
+ export TESTNAME=test_$testnum
local SAVE_UMASK=`umask`
umask 0022
- local BEFORE=`date +%s`
- echo
- log "== test $testnum: $message == `date +%H:%M:%S` ($BEFORE)"
- #check_mds
- export TESTNAME=test_$testnum
- TEST_FAILED=false
+ banner "test $testnum: $message"
test_${testnum} || error "test_$testnum failed with $?"
- #check_mds
cd $SAVE_PWD
reset_fail_loc
check_grant ${testnum} || error "check_grant $testnum failed with $?"
check_catastrophe || error "LBUG/LASSERT detected"
ps auxww | grep -v grep | grep -q multiop && error "multiop still running"
- pass "($((`date +%s` - $BEFORE))s)"
- TEST_FAILED=false
unset TESTNAME
unset tdir
umask $SAVE_UMASK
+ return 0
+}
+
+#
+# Wrapper around run_one to ensure:
+# - test runs in subshell
+# - output of test is saved to separate log file for error reporting
+# - test result is saved to data file
+#
+run_one_logged() {
+ local BEFORE=`date +%s`
+ local TEST_ERROR
+ local name=${TESTSUITE}.test_${1}.test_log.$(hostname).log
+ local test_log=$LOGDIR/$name
+ rm -rf $LOGDIR/err
+
+ echo
+ log_sub_test_begin test_${1}
+ (run_one $1 "$2") 2>&1 | tee $test_log
+ local RC=${PIPESTATUS[0]}
+
+ [ $RC -ne 0 ] && [ ! -f $LOGDIR/err ] && \
+ echo "test_$1 returned $RC" | tee $LOGDIR/err
+
+ duration=$((`date +%s` - $BEFORE))
+ pass "(${duration}s)"
+ [ -f $LOGDIR/err ] && TEST_ERROR=$(cat $LOGDIR/err)
+ log_sub_test_end $TEST_STATUS $duration "$RC" "$TEST_ERROR"
+
+ if [ -f $LOGDIR/err ]; then
+ $FAIL_ON_ERROR && exit $RC
+ fi
+
+ return 0
}
canonical_path() {
sync_clients() {
[ -d $DIR1 ] && cd $DIR1 && sync; sleep 1; sync
[ -d $DIR2 ] && cd $DIR2 && sync; sleep 1; sync
- cd $SAVE_PWD
+ cd $SAVE_PWD
}
check_grant() {
export base=`basetest $1`
[ "$CHECK_GRANT" == "no" ] && return 0
- testname=GCHECK_ONLY_${base}
+ testname=GCHECK_ONLY_${base}
[ ${!testname}x == x ] && return 0
echo -n "checking grant......"
- cd $SAVE_PWD
- # write some data to sync client lost_grant
- rm -f $DIR1/${tfile}_check_grant_* 2>&1
- for i in `seq $OSTCOUNT`; do
- $LFS setstripe $DIR1/${tfile}_check_grant_$i -i $(($i -1)) -c 1
- dd if=/dev/zero of=$DIR1/${tfile}_check_grant_$i bs=4k \
- count=1 > /dev/null 2>&1
- done
- # sync all the data and make sure no pending data on server
- sync_clients
-
- #get client grant and server grant
- client_grant=0
+ cd $SAVE_PWD
+ # write some data to sync client lost_grant
+ rm -f $DIR1/${tfile}_check_grant_* 2>&1
+ for i in `seq $OSTCOUNT`; do
+ $LFS setstripe $DIR1/${tfile}_check_grant_$i -i $(($i -1)) -c 1
+ dd if=/dev/zero of=$DIR1/${tfile}_check_grant_$i bs=4k \
+ count=1 > /dev/null 2>&1
+ done
+ # sync all the data and make sure no pending data on server
+ sync_clients
+
+ #get client grant and server grant
+ client_grant=0
for d in `lctl get_param -n osc.*.cur_grant_bytes`; do
- client_grant=$((client_grant + $d))
- done
- server_grant=0
- for d in `lctl get_param -n obdfilter.*.tot_granted`; do
- server_grant=$((server_grant + $d))
- done
-
- # cleanup the check_grant file
- for i in `seq $OSTCOUNT`; do
- rm $DIR1/${tfile}_check_grant_$i
- done
-
- #check whether client grant == server grant
- if [ $client_grant != $server_grant ]; then
- echo "failed: client:${client_grant} server: ${server_grant}"
- return 1
- else
- echo "pass"
- fi
+ client_grant=$((client_grant + $d))
+ done
+ server_grant=0
+ for d in `lctl get_param -n obdfilter.*.tot_granted`; do
+ server_grant=$((server_grant + $d))
+ done
+
+ # cleanup the check_grant file
+ for i in `seq $OSTCOUNT`; do
+ rm $DIR1/${tfile}_check_grant_$i
+ done
+
+ #check whether client grant == server grant
+ if [ $client_grant != $server_grant ]; then
+ echo "failed: client:${client_grant} server: ${server_grant}"
+ return 1
+ else
+ echo "pass"
+ fi
}
########################
remote_mds_nodsh()
{
+ [ "$CLIENTONLY" ] && return 0 || true
remote_mds && [ "$PDSH" = "no_dsh" -o -z "$PDSH" -o -z "$mds_HOST" ]
}
+require_dsh_mds()
+{
+ remote_mds_nodsh && echo "SKIP: $TESTSUITE: remote MDS with nodsh" && \
+ MSKIPPED=1 && return 1
+ return 0
+}
+
remote_ost ()
{
local node
remote_ost_nodsh()
{
+ [ "$CLIENTONLY" ] && return 0 || true
remote_ost && [ "$PDSH" = "no_dsh" -o -z "$PDSH" -o -z "$ost_HOST" ]
}
+require_dsh_ost()
+{
+ remote_ost_nodsh && echo "SKIP: $TESTSUITE: remote OST with nodsh" && \
+ OSKIPPED=1 && return 1
+ return 0
+}
+
remote_mgs_nodsh()
{
local MGS
remote_node $MGS && [ "$PDSH" = "no_dsh" -o -z "$PDSH" -o -z "$ost_HOST" ]
}
+local_mode ()
+{
+ remote_mds_nodsh || remote_ost_nodsh || \
+ $(single_local_node $(comma_list $(nodes_list)))
+}
+
mdts_nodes () {
local MDSNODES
local NODES_sort
echo ${nodes[i]}
}
+client_only () {
+ [ "$CLIENTONLY" ] || [ "$CLIENTMODSONLY" = yes ]
+}
+
is_patchless ()
{
lctl get_param version | grep -q patchless
generate_machine_file() {
local nodes=${1//,/ }
local machinefile=$2
- rm -f $machinefile || error "can't rm $machinefile"
+ rm -f $machinefile
for node in $nodes; do
- echo $node >>$machinefile
+ echo $node >>$machinefile || \
+ { echo "can not generate machinefile $machinefile" && return 1; }
done
}
rm -f $file
}
+setstripe_nfsserver () {
+ local dir=$1
+
+ local nfsserver=$(awk '"'$dir'" ~ $2 && $3 ~ "nfs" && $2 != "/" \
+ { print $1 }' /proc/mounts | cut -f 1 -d : | head -1)
+
+ [ -z $nfsserver ] && echo "$dir is not nfs mounted" && return 1
+
+ do_nodev $nfsserver lfs setstripe "$@"
+}
+
check_runas_id_ret() {
local myRC=0
local myRUNAS_UID=$1
add user $myRUNAS_UID:$myRUNAS_GID on these nodes."
}
+# obtain the UID/GID for MPI_USER
+get_mpiuser_id() {
+ local mpi_user=$1
+
+ MPI_USER_UID=$(do_facet client "getent passwd $mpi_user | cut -d: -f3;
+exit \\\${PIPESTATUS[0]}") || error_exit "failed to get the UID for $mpi_user"
+
+ MPI_USER_GID=$(do_facet client "getent passwd $mpi_user | cut -d: -f4;
+exit \\\${PIPESTATUS[0]}") || error_exit "failed to get the GID for $mpi_user"
+}
+
+# obtain and cache Kerberos ticket-granting ticket
+refresh_krb5_tgt() {
+ local myRUNAS_UID=$1
+ local myRUNAS_GID=$2
+ shift 2
+ local myRUNAS=$@
+ if [ -z "$myRUNAS" ]; then
+ error_exit "myRUNAS command must be specified for refresh_krb5_tgt"
+ fi
+
+ CLIENTS=${CLIENTS:-$HOSTNAME}
+ do_nodes $CLIENTS "set -x
+if ! $myRUNAS krb5_login.sh; then
+ echo "Failed to refresh Krb5 TGT for UID/GID $myRUNAS_UID/$myRUNAS_GID."
+ exit 1
+fi"
+}
+
# Run multiop in the background, but wait for it to print
# "PAUSING" to its stdout before returning from this function.
multiop_bg_pause() {
return 0
}
+do_and_time () {
+ local cmd=$1
+ local rc
+
+ SECONDS=0
+ eval '$cmd'
+
+ [ ${PIPESTATUS[0]} -eq 0 ] || rc=1
+
+ echo $SECONDS
+ return $rc
+}
+
inodes_available () {
local IFree=$($LFS df -i $MOUNT | grep ^$FSNAME | awk '{print $4}' | sort -un | head -1) || return 1
echo $IFree
}
+mdsrate_inodes_available () {
+ echo $(($(inodes_available) - 1))
+}
+
# reset llite stat counters
clear_llite_stats(){
lctl set_param -n llite.*.stats 0
# generate a stream of formatted strings (<node> <param name>=<param value>)
save_lustre_params() {
local s
- do_node $1 "lctl get_param $2" | while read s; do echo "$1 $s"; done
+ do_nodesv $1 "lctl get_param $2 | while read s; do echo \\\$s; done"
}
# restore lustre parameters from input stream, produces by save_lustre_params
local name
local val
while IFS=" =" read node name val; do
- do_node $node "lctl set_param -n $name $val"
+ do_node ${node//:/} "lctl set_param -n $name $val"
done
}
# $2 file
# $3 $RUNAS
get_stripe_info() {
- local tmp_file
+ local tmp_file
- stripe_size=0
- stripe_count=0
- stripe_index=0
- tmp_file=$(mktemp)
+ stripe_size=0
+ stripe_count=0
+ stripe_index=0
+ tmp_file=$(mktemp)
- do_facet $1 $3 lfs getstripe -v $2 > $tmp_file
+ do_facet $1 $3 lfs getstripe -v $2 > $tmp_file
- stripe_size=`awk '$1 ~ /size/ {print $2}' $tmp_file`
- stripe_count=`awk '$1 ~ /count/ {print $2}' $tmp_file`
- stripe_index=`awk '/obdidx/ {start = 1; getline; print $1; exit}' $tmp_file`
- rm -f $tmp_file
+ stripe_size=`awk '$1 ~ /size/ {print $2}' $tmp_file`
+ stripe_count=`awk '$1 ~ /count/ {print $2}' $tmp_file`
+ stripe_index=`awk '$1 ~ /stripe_offset/ {print $2}' $tmp_file`
+ rm -f $tmp_file
}
# CMD: determine mds index where directory inode presents
local dir=$1
local file=$dir/f0.get_mds_dir_tmpfile
+ mkdir -p $dir
rm -f $file
sleep 1
local iused=$(lfs df -i $dir | grep MDT | awk '{print $3}')
- local oldused=($iused)
+ local -a oldused=($iused)
- touch $file
+ openfile -f O_CREAT:O_LOV_DELAY_CREATE -m 0644 $file > /dev/null
sleep 1
iused=$(lfs df -i $dir | grep MDT | awk '{print $3}')
- local newused=($iused)
+ local -a newused=($iused)
local num=0
for ((i=0; i<${#newused[@]}; i++)); do
error "mdt-s : inodes count OLD ${oldused[@]} NEW ${newused[@]}"
}
-mpi_run () {
- local mpirun="$MPIRUN $MPIRUN_OPTIONS"
- local command="$mpirun $@"
- local mpilog=$TMP/mpi.log
- local rc
-
- if [ "$MPI_USER" != root -a $mpirun ]; then
- echo "+ chmod 0777 $MOUNT"
- chmod 0777 $MOUNT
- command="su $MPI_USER sh -c \"$command \""
- fi
-
- ls -ald $MOUNT
- echo "+ $command"
- eval $command 2>&1 > $mpilog || true
-
- rc=${PIPESTATUS[0]}
- if [ $rc -eq 0 ] && grep -q "p4_error: : [^0]" $mpilog ; then
- rc=1
- fi
- cat $mpilog
- return $rc
-}
-
mdsrate_cleanup () {
- mpi_run -np $1 -machinefile $2 ${MDSRATE} --unlink --nfiles $3 --dir $4 --filefmt $5 $6
+ if [ -d $4 ]; then
+ mpi_run -np $1 -machinefile $2 ${MDSRATE} --unlink --nfiles $3 --dir $4 --filefmt $5 $6
+ rmdir $4
+ fi
}
delayed_recovery_enabled () {
}
########################
-convert_facet2name() {
- case "$1" in
- "ost" ) echo "OST0000" ;;
- "ost1") echo "OST0000" ;;
- "ost2") echo "OST0001" ;;
- "ost3") echo "OST0002" ;;
- "ost4") echo "OST0003" ;;
- "ost5") echo "OST0004" ;;
- *) error "unknown facet!" ;;
- esac
+
+convert_facet2label() {
+ local facet=$1
+
+ if [ x$facet = xost ]; then
+ facet=ost1
+ fi
+
+ local varsvc=${facet}_svc
+
+ if [ -n ${!varsvc} ]; then
+ echo ${!varsvc}
+ else
+ error "No lablel for $facet!"
+ fi
}
get_clientosc_proc_path() {
}
get_osc_import_name() {
- local node=$1
+ local facet=$1
local ost=$2
- local name=$(convert_facet2name $ost)
+ local label=$(convert_facet2label $ost)
- if [ "$node" == "mds" ]; then
- get_mdtosc_proc_path $name
+ if [ "$facet" == "mds" ]; then
+ get_mdtosc_proc_path $label
return 0
fi
- get_clientosc_proc_path $name
+ get_clientosc_proc_path $label
+ return 0
+}
+
+wait_import_state () {
+ local expected=$1
+ local CONN_PROC=$2
+ local CONN_STATE
+ local i=0
+
+ CONN_STATE=$($LCTL get_param -n $CONN_PROC 2>/dev/null | cut -f2)
+ while [ "${CONN_STATE}" != "${expected}" ]; do
+ if [ "${expected}" == "DISCONN" ]; then
+ # for disconn we can check after proc entry is removed
+ [ "x${CONN_STATE}" == "x" ] && return 0
+ # with AT we can have connect request timeout ~ reconnect timeout
+ # and test can't see real disconnect
+ [ "${CONN_STATE}" == "CONNECTING" ] && return 0
+ fi
+ # disconnect rpc should be wait not more obd_timeout
+ [ $i -ge $(($TIMEOUT * 3 / 2)) ] && \
+ error "can't put import for $CONN_PROC into ${expected} state" && return 1
+ sleep 1
+ CONN_STATE=$($LCTL get_param -n $CONN_PROC 2>/dev/null | cut -f2)
+ i=$(($i + 1))
+ done
+
+ log "$CONN_PROC now in ${CONN_STATE} state"
return 0
}
wait_osc_import_state() {
- local node=$1
+ local facet=$1
local ost_facet=$2
local expected=$3
- local ost=$(get_osc_import_name $node $ost_facet)
+ local ost=$(get_osc_import_name $facet $ost_facet)
local CONN_PROC
local CONN_STATE
local i=0
- CONN_PROC="osc.${FSNAME}-${ost}.ost_server_uuid"
- CONN_STATE=$(do_facet $node lctl get_param -n $CONN_PROC | cut -f2)
+ CONN_PROC="osc.${ost}.ost_server_uuid"
+ CONN_STATE=$(do_facet $facet lctl get_param -n $CONN_PROC 2>/dev/null | cut -f2)
while [ "${CONN_STATE}" != "${expected}" ]; do
- # for disconn we can check after proc entry is removed
- [ "x${CONN_STATE}" == "x" -a "${expected}" == "DISCONN" ] && return 0
+ if [ "${expected}" == "DISCONN" ]; then
+ # for disconn we can check after proc entry is removed
+ [ "x${CONN_STATE}" == "x" ] && return 0
+ # with AT we can have connect request timeout ~ reconnect timeout
+ # and test can't see real disconnect
+ [ "${CONN_STATE}" == "CONNECTING" ] && return 0
+ fi
# disconnect rpc should be wait not more obd_timeout
[ $i -ge $(($TIMEOUT * 3 / 2)) ] && \
error "can't put import for ${ost}(${ost_facet}) into ${expected} state" && return 1
sleep 1
- CONN_STATE=$(do_facet $node lctl get_param -n $CONN_PROC | cut -f2)
+ CONN_STATE=$(do_facet $facet lctl get_param -n $CONN_PROC 2>/dev/null | cut -f2)
i=$(($i + 1))
done
log "${ost_facet} now in ${CONN_STATE} state"
return 0
}
+get_clientmdc_proc_path() {
+ echo "${1}-mdc-*"
+}
+
+do_rpc_nodes () {
+ local list=$1
+ shift
+
+ # Add paths to lustre tests for 32 and 64 bit systems.
+ local RPATH="$RLUSTRE/tests:/usr/lib/lustre/tests:/usr/lib64/lustre/tests:$PATH"
+ do_nodesv $list "PATH=$RPATH sh rpc.sh $@ "
+}
+
+wait_clients_import_state () {
+ local list=$1
+ local facet=$2
+ local expected=$3
+ shift
+
+ local label=$(convert_facet2label $facet)
+ local proc_path
+ case $facet in
+ ost* ) proc_path="osc.$(get_clientosc_proc_path $label).ost_server_uuid" ;;
+ mds* ) proc_path="mdc.$(get_clientmdc_proc_path $label).mds_server_uuid" ;;
+ *) error "unknown facet!" ;;
+ esac
+
+ if ! do_rpc_nodes $list wait_import_state $expected $proc_path; then
+ error "import is not in ${expected} state"
+ return 1
+ fi
+}
+
+oos_full() {
+ local -a AVAILA
+ local -a GRANTA
+ local OSCFULL=1
+ AVAILA=($(do_nodes $(comma_list $(osts_nodes)) \
+ $LCTL get_param obdfilter.*.kbytesavail))
+ GRANTA=($(do_nodes $(comma_list $(osts_nodes)) \
+ $LCTL get_param -n obdfilter.*.tot_granted))
+ for ((i=0; i<${#AVAILA[@]}; i++)); do
+ local -a AVAIL1=(${AVAILA[$i]//=/ })
+ GRANT=$((${GRANTA[$i]}/1024))
+ echo -n $(echo ${AVAIL1[0]} | cut -d"." -f2) avl=${AVAIL1[1]} grnt=$GRANT diff=$((AVAIL1[1] - GRANT))
+ [ $((AVAIL1[1] - GRANT)) -lt 400 ] && OSCFULL=0 && echo " FULL" || echo
+ done
+ return $OSCFULL
+}
+
+pool_list () {
+ do_facet mgs lctl pool_list $1
+}
+
+create_pool() {
+ local fsname=${1%%.*}
+ local poolname=${1##$fsname.}
+
+ do_facet mgs lctl pool_new $1
+ local RC=$?
+ # get param should return err unless pool is created
+ [[ $RC -ne 0 ]] && return $RC
+
+ wait_update $HOSTNAME "lctl get_param -n lov.$fsname-*.pools.$poolname \
+ 2>/dev/null || echo foo" "" || RC=1
+ if [[ $RC -eq 0 ]]; then
+ add_pool_to_list $1
+ else
+ error "pool_new failed $1"
+ fi
+ return $RC
+}
+
+add_pool_to_list () {
+ local fsname=${1%%.*}
+ local poolname=${1##$fsname.}
+
+ local listvar=${fsname}_CREATED_POOLS
+ eval export ${listvar}=$(expand_list ${!listvar} $poolname)
+}
+
+remove_pool_from_list () {
+ local fsname=${1%%.*}
+ local poolname=${1##$fsname.}
+
+ local listvar=${fsname}_CREATED_POOLS
+ eval export ${listvar}=$(exclude_items_from_list ${!listvar} $poolname)
+}
+
+destroy_pool_int() {
+ local ost
+ local OSTS=$(do_facet $SINGLEMDS lctl pool_list $1 | \
+ awk '$1 !~ /^Pool:/ {print $1}')
+ for ost in $OSTS; do
+ do_facet mgs lctl pool_remove $1 $ost
+ done
+ do_facet mgs lctl pool_destroy $1
+}
+
+# <fsname>.<poolname> or <poolname>
+destroy_pool() {
+ local fsname=${1%%.*}
+ local poolname=${1##$fsname.}
+
+ [[ x$fsname = x$poolname ]] && fsname=$FSNAME
+
+ local RC
+
+ pool_list $fsname.$poolname || return $?
+
+ destroy_pool_int $fsname.$poolname
+ RC=$?
+ [[ $RC -ne 0 ]] && return $RC
+
+ wait_update $HOSTNAME "lctl get_param -n lov.$fsname-*.pools.$poolname \
+ 2>/dev/null || echo foo" "foo" || RC=1
+
+ if [[ $RC -eq 0 ]]; then
+ remove_pool_from_list $fsname.$poolname
+ else
+ error "destroy pool failed $1"
+ fi
+ return $RC
+}
+
+destroy_pools () {
+ local fsname=${1:-$FSNAME}
+ local poolname
+ local listvar=${fsname}_CREATED_POOLS
+
+ pool_list $fsname
+
+ [ x${!listvar} = x ] && return 0
+
+ echo destroy the created pools: ${!listvar}
+ for poolname in ${!listvar//,/ }; do
+ destroy_pool $fsname.$poolname
+ done
+}
+
+cleanup_pools () {
+ local fsname=${1:-$FSNAME}
+ trap 0
+ destroy_pools $fsname
+}
+
+gather_logs () {
+ local list=$1
+
+ local ts=$(date +%s)
+
+ # bug 20237, comment 11
+ # It would also be useful to provide the option
+ # of writing the file to an NFS directory so it doesn't need to be copied.
+ local tmp=$TMP
+ local docp=true
+ [ -f $LOGDIR/shared ] && docp=false
+
+ # dump lustre logs, dmesg
+
+ prefix="$LOGDIR/${TESTSUITE}.${TESTNAME}"
+ suffix="$ts.log"
+ echo "Dumping lctl log to ${prefix}.*.${suffix}"
+
+ if [ "$CLIENTONLY" -o "$PDSH" == "no_dsh" ]; then
+ echo "Dumping logs only on local client."
+ $LCTL dk > ${prefix}.debug_log.$(hostname).${suffix}
+ dmesg > ${prefix}.dmesg.$(hostname).${suffix}
+ return
+ fi
+
+ do_nodesv $list \
+ "$LCTL dk > ${prefix}.debug_log.\\\$(hostname).${suffix};
+ dmesg > ${prefix}.dmesg.\\\$(hostname).${suffix}"
+ if [ ! -f $LOGDIR/shared ]; then
+ do_nodes $list rsync -az "${prefix}.*.${suffix}" $HOSTNAME:$LOGDIR
+ fi
+
+ local archive=$LOGDIR/${TESTSUITE}-$ts.tar.bz2
+ tar -jcf $archive $LOGDIR/*$ts* $LOGDIR/*${TESTSUITE}*
+
+ echo $archive
+}
+
+cleanup_logs () {
+ local list=${1:-$(comma_list $(nodes_list))}
+
+ [ -n ${TESTSUITE} ] && do_nodes $list "rm -f $TMP/*${TESTSUITE}*" || true
+}
+
+do_ls () {
+ local mntpt_root=$1
+ local num_mntpts=$2
+ local dir=$3
+ local i
+ local cmd
+ local pids
+ local rc=0
+
+ for i in $(seq 0 $num_mntpts); do
+ cmd="ls -laf ${mntpt_root}$i/$dir"
+ echo + $cmd;
+ $cmd > /dev/null &
+ pids="$pids $!"
+ done
+ echo pids=$pids
+ for pid in $pids; do
+ wait $pid || rc=$?
+ done
+
+ return $rc
+}
+
+# target_start_and_reset_recovery_timer()
+# service_time = at_est2timeout(service_time);
+# service_time += 2 * (CONNECTION_SWITCH_MAX + CONNECTION_SWITCH_INC +
+# INITIAL_CONNECT_TIMEOUT);
+# CONNECTION_SWITCH_MAX : min(25U, max(CONNECTION_SWITCH_MIN,obd_timeout))
+#define CONNECTION_SWITCH_INC 1
+#define INITIAL_CONNECT_TIMEOUT max(CONNECTION_SWITCH_MIN,obd_timeout/20)
+#define CONNECTION_SWITCH_MIN 5U
+
+max_recovery_time () {
+ local init_connect_timeout=$(( 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`}
+
+ # we need to take into account the clients mounts and
+ # exclude mds/ost mounts if any;
+ do_nodes $clients cat /proc/mounts | grep lustre | grep $MOUNT | wc -l
+}
+
+# gss functions
+PROC_CLI="srpc_info"
+
+combination()
+{
+ local M=$1
+ local N=$2
+ local R=1
+
+ if [ $M -lt $N ]; then
+ R=0
+ else
+ N=$((N + 1))
+ while [ $N -le $M ]; do
+ R=$((R * N))
+ N=$((N + 1))
+ done
+ fi
+
+ echo $R
+ return 0
+}
+
+calc_connection_cnt() {
+ local dir=$1
+
+ # MDT->MDT = 2 * C(M, 2)
+ # MDT->OST = M * O
+ # CLI->OST = C * O
+ # CLI->MDT = C * M
+ comb_m2=$(combination $MDSCOUNT 2)
+
+ local num_clients=$(get_clients_mount_count)
+
+ local cnt_mdt2mdt=$((comb_m2 * 2))
+ local cnt_mdt2ost=$((MDSCOUNT * OSTCOUNT))
+ local cnt_cli2ost=$((num_clients * OSTCOUNT))
+ local cnt_cli2mdt=$((num_clients * MDSCOUNT))
+ local cnt_all2ost=$((cnt_mdt2ost + cnt_cli2ost))
+ local cnt_all2mdt=$((cnt_mdt2mdt + cnt_cli2mdt))
+ local cnt_all2all=$((cnt_mdt2ost + cnt_mdt2mdt + cnt_cli2ost + cnt_cli2mdt))
+
+ local var=cnt_$dir
+ local res=${!var}
+
+ echo $res
+}
+
+set_rule()
+{
+ local tgt=$1
+ local net=$2
+ local dir=$3
+ local flavor=$4
+ local cmd="$tgt.srpc.flavor"
+
+ if [ $net == "any" ]; then
+ net="default"
+ fi
+ cmd="$cmd.$net"
+
+ if [ $dir != "any" ]; then
+ cmd="$cmd.$dir"
+ fi
+
+ cmd="$cmd=$flavor"
+ log "Setting sptlrpc rule: $cmd"
+ do_facet mgs "$LCTL conf_param $cmd"
+}
+
+count_flvr()
+{
+ local output=$1
+ local flavor=$2
+ local count=0
+
+ rpc_flvr=`echo $flavor | awk -F - '{ print $1 }'`
+ bulkspec=`echo $flavor | awk -F - '{ print $2 }'`
+
+ count=`echo "$output" | grep "rpc flavor" | grep $rpc_flvr | wc -l`
+
+ if [ "x$bulkspec" != "x" ]; then
+ algs=`echo $bulkspec | awk -F : '{ print $2 }'`
+
+ if [ "x$algs" != "x" ]; then
+ bulk_count=`echo "$output" | grep "bulk flavor" | grep $algs | wc -l`
+ else
+ bulk=`echo $bulkspec | awk -F : '{ print $1 }'`
+ if [ $bulk == "bulkn" ]; then
+ bulk_count=`echo "$output" | grep "bulk flavor" \
+ | grep "null/null" | wc -l`
+ elif [ $bulk == "bulki" ]; then
+ bulk_count=`echo "$output" | grep "bulk flavor" \
+ | grep "/null" | grep -v "null/" | wc -l`
+ else
+ bulk_count=`echo "$output" | grep "bulk flavor" \
+ | grep -v "/null" | grep -v "null/" | wc -l`
+ fi
+ fi
+
+ [ $bulk_count -lt $count ] && count=$bulk_count
+ fi
+
+ echo $count
+}
+
+flvr_cnt_cli2mdt()
+{
+ local flavor=$1
+ local cnt
+
+ local clients=${CLIENTS:-`hostname`}
+
+ for c in ${clients//,/ }; do
+ output=`do_node $c lctl get_param -n mdc.*-MDT*-mdc-*.$PROC_CLI 2>/dev/null`
+ tmpcnt=`count_flvr "$output" $flavor`
+ cnt=$((cnt + tmpcnt))
+ done
+ echo $cnt
+}
+
+flvr_cnt_cli2ost()
+{
+ local flavor=$1
+ local cnt
+
+ local clients=${CLIENTS:-`hostname`}
+
+ for c in ${clients//,/ }; do
+ output=`do_node $c lctl get_param -n osc.*OST*-osc-[^M][^D][^T]*.$PROC_CLI 2>/dev/null`
+ tmpcnt=`count_flvr "$output" $flavor`
+ cnt=$((cnt + tmpcnt))
+ done
+ echo $cnt
+}
+
+flvr_cnt_mdt2mdt()
+{
+ local flavor=$1
+ local cnt=0
+
+ if [ $MDSCOUNT -le 1 ]; then
+ echo 0
+ return
+ fi
+
+ for num in `seq $MDSCOUNT`; do
+ output=`do_facet mds$num lctl get_param -n mdc.*-MDT*-mdc[0-9]*.$PROC_CLI 2>/dev/null`
+ tmpcnt=`count_flvr "$output" $flavor`
+ cnt=$((cnt + tmpcnt))
+ done
+ echo $cnt;
+}
+
+flvr_cnt_mdt2ost()
+{
+ local flavor=$1
+ local cnt=0
+
+ for num in `seq $MDSCOUNT`; do
+ output=`do_facet mds$num lctl get_param -n osc.*OST*-osc-MDT*.$PROC_CLI 2>/dev/null`
+ tmpcnt=`count_flvr "$output" $flavor`
+ cnt=$((cnt + tmpcnt))
+ done
+ echo $cnt;
+}
+
+flvr_cnt_mgc2mgs()
+{
+ local flavor=$1
+
+ output=`do_facet client lctl get_param -n mgc.*.$PROC_CLI 2>/dev/null`
+ count_flvr "$output" $flavor
+}
+
+do_check_flavor()
+{
+ local dir=$1 # from to
+ local flavor=$2 # flavor expected
+ local res=0
+
+ if [ $dir == "cli2mdt" ]; then
+ res=`flvr_cnt_cli2mdt $flavor`
+ elif [ $dir == "cli2ost" ]; then
+ res=`flvr_cnt_cli2ost $flavor`
+ elif [ $dir == "mdt2mdt" ]; then
+ res=`flvr_cnt_mdt2mdt $flavor`
+ elif [ $dir == "mdt2ost" ]; then
+ res=`flvr_cnt_mdt2ost $flavor`
+ elif [ $dir == "all2ost" ]; then
+ res1=`flvr_cnt_mdt2ost $flavor`
+ res2=`flvr_cnt_cli2ost $flavor`
+ res=$((res1 + res2))
+ elif [ $dir == "all2mdt" ]; then
+ res1=`flvr_cnt_mdt2mdt $flavor`
+ res2=`flvr_cnt_cli2mdt $flavor`
+ res=$((res1 + res2))
+ elif [ $dir == "all2all" ]; then
+ res1=`flvr_cnt_mdt2ost $flavor`
+ res2=`flvr_cnt_cli2ost $flavor`
+ res3=`flvr_cnt_mdt2mdt $flavor`
+ res4=`flvr_cnt_cli2mdt $flavor`
+ res=$((res1 + res2 + res3 + res4))
+ fi
+
+ echo $res
+}
+
+wait_flavor()
+{
+ local dir=$1 # from to
+ local flavor=$2 # flavor expected
+ local expect=${3:-$(calc_connection_cnt $dir)} # number expected
+
+ local res=0
+
+ for ((i=0;i<20;i++)); do
+ echo -n "checking..."
+ res=$(do_check_flavor $dir $flavor)
+ if [ $res -eq $expect ]; then
+ echo "found $res $flavor connections of $dir, OK"
+ return 0
+ else
+ echo "found $res $flavor connections of $dir, not ready ($expect)"
+ sleep 4
+ fi
+ done
+
+ echo "Error checking $flavor of $dir: expect $expect, actual $res"
+ return 1
+}
+
+restore_to_default_flavor()
+{
+ local proc="mgs.MGS.live.$FSNAME"
+
+ echo "restoring to default flavor..."
+
+ nrule=`do_facet mgs lctl get_param -n $proc 2>/dev/null | grep ".srpc.flavor." | wc -l`
+
+ # remove all existing rules if any
+ if [ $nrule -ne 0 ]; then
+ echo "$nrule existing rules"
+ for rule in `do_facet mgs lctl get_param -n $proc 2>/dev/null | grep ".srpc.flavor."`; do
+ echo "remove rule: $rule"
+ spec=`echo $rule | awk -F = '{print $1}'`
+ do_facet mgs "$LCTL conf_param $spec="
+ done
+ fi
+
+ # verify no rules left
+ nrule=`do_facet mgs lctl get_param -n $proc 2>/dev/null | grep ".srpc.flavor." | wc -l`
+ [ $nrule -ne 0 ] && error "still $nrule rules left"
+
+ # wait for default flavor to be applied
+ # currently default flavor for all connections are 'null'
+ wait_flavor all2all null
+ echo "now at default flavor settings"
+}
+
+set_flavor_all()
+{
+ local flavor=${1:-null}
+
+ echo "setting all flavor to $flavor"
+
+ # FIXME need parameter to this fn
+ # and remove global vars
+ local cnt_all2all=$(calc_connection_cnt all2all)
+
+ local res=$(do_check_flavor all2all $flavor)
+ if [ $res -eq $cnt_all2all ]; then
+ echo "already have total $res $flavor connections"
+ return
+ fi
+
+ echo "found $res $flavor out of total $cnt_all2all connections"
+ restore_to_default_flavor
+
+ [[ $flavor = null ]] && return 0
+
+ set_rule $FSNAME any any $flavor
+ wait_flavor all2all $flavor
+}
+
+
+check_logdir() {
+ local dir=$1
+ # Checking for shared logdir
+ if [ ! -d $dir ]; then
+ # Not found. Create local logdir
+ mkdir -p $dir
+ else
+ touch $dir/node.$(hostname).yml
+ fi
+ return 0
+}
+
+check_write_access() {
+ local dir=$1
+ for node in $(nodes_list); do
+ if [ ! -f "$dir/node.${node}.yml" ]; then
+ # Logdir not accessible/writable from this node.
+ return 1
+ fi
+ done
+ return 0
+}
+
+init_logging() {
+ if [[ -n $YAML_LOG ]]; then
+ return
+ fi
+ export YAML_LOG=${LOGDIR}/results.yml
+ mkdir -p $LOGDIR
+ init_clients_lists
+
+ do_rpc_nodes $(comma_list $(nodes_list)) check_logdir $LOGDIR
+ if check_write_access $LOGDIR; then
+ touch $LOGDIR/shared
+ echo "Logging to shared log directory: $LOGDIR"
+ else
+ echo "Logging to local directory: $LOGDIR"
+ fi
+
+ yml_nodes_file $LOGDIR
+ yml_results_file >> $YAML_LOG
+}
+
+log_test() {
+ yml_log_test $1 >> $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=""
+ # Run in partial (fast) mode if the size
+ # of a partition > 10 GB
+ [ $size -gt 10 ] && partial_arg="-p"
+
+ llverdev --force $partial_arg $dev
+}
git://git.whamcloud.com / fs / lustre-release.git / blobdiff