init_test_env $@
. ${CONFIG:=$LUSTRE/tests/cfg/$NAME.sh}
+# also long tests: 19, 21a, 21e, 21f, 23, 27
+# 1 2.5 4 4 (min)"
+[ "$SLOW" = "no" ] && EXCEPT_SLOW="17 26b 50 51 57"
+
build_test_filter
# Allow us to override the setup if we already have a mounted system by
CLEANUP=${CLEANUP:-""}
cleanup_and_setup_lustre
-rm -rf $DIR/${TESTSUITE}/[df][0-9]* # bug 13798 new t-f tdir staff
rm -rf $DIR/[df][0-9]*
+SAMPLE_NAME=recovery-small.junk
+SAMPLE_FILE=$TMP/$SAMPLE_NAME
+# make this big, else test 9 doesn't wait for bulk -- bz 5595
+dd if=/dev/urandom of=$SAMPLE_FILE bs=1M count=4
+
test_1() {
drop_request "mcreate $MOUNT/1" || return 1
drop_reint_reply "mcreate $MOUNT/2" || return 2
#bug 1420
test_9() {
pause_bulk "cp /etc/profile $MOUNT/$tfile" || return 1
- do_facet client "cp /etc/termcap $MOUNT/${tfile}.2" || return 2
+ do_facet client "cp ${SAMPLE_FILE} $MOUNT/${tfile}.2" || return 2
do_facet client "sync"
do_facet client "rm $MOUNT/$tfile $MOUNT/${tfile}.2" || return 3
}
}
run_test 15 "failed open (-ENOMEM)"
-READ_AHEAD=`cat $LPROC/llite/*/max_read_ahead_mb | head -n 1`
+READ_AHEAD=`lctl get_param -n llite.*.max_read_ahead_mb | head -n 1`
stop_read_ahead() {
- for f in $LPROC/llite/*/max_read_ahead_mb; do
- echo 0 > $f
- done
+ lctl set_param -n llite.*.max_read_ahead_mb 0
}
start_read_ahead() {
- for f in $LPROC/llite/*/max_read_ahead_mb; do
- echo $READ_AHEAD > $f
- done
+ lctl set_param -n llite.*.max_read_ahead_mb $READ_AHEAD
}
test_16() {
pgcache_empty || return 1
# 1 stripe on ost2
- lfs setstripe $f $((128 * 1024)) 1 1
+ lfs setstripe $f -s $((128 * 1024)) -i 1 -c 1
- do_facet client cp /etc/termcap $f
+ do_facet client cp $SAMPLE_FILE $f
sync
- local osc2dev=`grep ${ost2_svc}-osc- $LPROC/devices | egrep -v 'MDT' | awk '{print $1}'`
+ local osc2dev=`lctl get_param -n devices | grep ${ost2_svc}-osc- | egrep -v 'MDT' | awk '{print $1}'`
$LCTL --device $osc2dev deactivate || return 3
# my understanding is that there should be nothing in the page
# cache after the client reconnects?
pgcache_empty || return 1
# shouldn't have to set stripe size of count==1
- lfs setstripe $f $((128 * 1024)) 0 1
- lfs setstripe $f2 $((128 * 1024)) 0 1
+ lfs setstripe $f -s $((128 * 1024)) -i 0 -c 1
+ lfs setstripe $f2 -s $((128 * 1024)) -i 0 -c 1
- do_facet client cp /etc/termcap $f
+ do_facet client cp $SAMPLE_FILE $f
sync
ost_evict_client
# allow recovery to complete
}
run_test 18b "eviction and reconnect clears page cache (2766)"
+test_18c() {
+ do_facet client mkdir -p $MOUNT/$tdir
+ f=$MOUNT/$tdir/$tfile
+ f2=$MOUNT/$tdir/${tfile}-2
+
+ cancel_lru_locks osc
+ pgcache_empty || return 1
+
+ # shouldn't have to set stripe size of count==1
+ lfs setstripe $f -s $((128 * 1024)) -i 0 -c 1
+ lfs setstripe $f2 -s $((128 * 1024)) -i 0 -c 1
+
+ do_facet client cp $SAMPLE_FILE $f
+ sync
+ ost_evict_client
+
+ # OBD_FAIL_OST_CONNECT_NET2
+ # lost reply to connect request
+ do_facet ost1 sysctl -w lustre.fail_loc=0x80000225
+ # force reconnect
+ df $MOUNT > /dev/null 2>&1
+ sleep 2
+ # my understanding is that there should be nothing in the page
+ # cache after the client reconnects?
+ rc=0
+ pgcache_empty || rc=2
+ rm -f $f $f2
+ return $rc
+}
+run_test 18c "Dropped connect reply after eviction handing (14755)"
+
test_19a() {
f=$MOUNT/$tfile
do_facet client mcreate $f || return 1
# this test can only run from a client on a separate node.
remote_ost || skip "local OST" && return
remote_mds || skip "local MDS" && return
- OST_FILE=$LPROC/obdfilter/${ost1_svc}/num_exports
- OST_EXP="`do_facet ost1 cat $OST_FILE`"
+ OST_FILE=obdfilter.${ost1_svc}.num_exports
+ OST_EXP="`do_facet ost1 lctl get_param -n $OST_FILE`"
OST_NEXP1=`echo $OST_EXP | cut -d' ' -f2`
echo starting with $OST_NEXP1 OST exports
# OBD_FAIL_PTLRPC_DROP_RPC 0x505
# might have to wait for the next ping.
echo Waiting for $(($TIMEOUT * 4)) secs
sleep $(($TIMEOUT * 4))
- OST_EXP="`do_facet ost1 cat $OST_FILE`"
+ OST_EXP="`do_facet ost1 lctl get_param -n $OST_FILE`"
OST_NEXP2=`echo $OST_EXP | cut -d' ' -f2`
echo ending with $OST_NEXP2 OST exports
do_facet client sysctl -w lustre.fail_loc=0x0
client_df
zconf_mount `hostname` $MOUNT2 || error "Failed to mount $MOUNT2"
sleep 1 # wait connections being established
- MDS_FILE=$LPROC/mdt/${mds1_svc}/num_exports
- MDS_NEXP1="`do_facet $SINGLEMDS cat $MDS_FILE | cut -d' ' -f2`"
- OST_FILE=$LPROC/obdfilter/${ost1_svc}/num_exports
- OST_NEXP1="`do_facet ost1 cat $OST_FILE | cut -d' ' -f2`"
+ MDS_FILE=mdt.${mds1_svc}.num_exports
+ MDS_NEXP1="`do_facet $SINGLEMDS lctl get_param -n $MDS_FILE | cut -d' ' -f2`"
+ OST_FILE=obdfilter.${ost1_svc}.num_exports
+ OST_NEXP1="`do_facet ost1 lctl get_param -n $OST_FILE | cut -d' ' -f2`"
echo starting with $OST_NEXP1 OST and $MDS_NEXP1 MDS exports
zconf_umount `hostname` $MOUNT2 -f
# evictor takes up to 2.25x to evict. But if there's a
# might have to wait for the next ping.
echo Waiting for $(($TIMEOUT * 4)) secs
sleep $(($TIMEOUT * 4))
- OST_NEXP2="`do_facet ost1 cat $OST_FILE | cut -d' ' -f2`"
- MDS_NEXP2="`do_facet $SINGLEMDS cat $MDS_FILE | cut -d' ' -f2`"
+ OST_NEXP2="`do_facet ost1 lctl get_param -n $OST_FILE | cut -d' ' -f2`"
+ MDS_NEXP2="`do_facet $SINGLEMDS lctl get_param -n $MDS_FILE | cut -d' ' -f2`"
echo ending with $OST_NEXP2 OST and $MDS_NEXP2 MDS exports
[ $OST_NEXP1 -le $OST_NEXP2 ] && error "client not evicted from OST"
[ $MDS_NEXP1 -le $MDS_NEXP2 ] && error "client not evicted from MDS"
test_57_helper() {
# no oscs means no client or mdt
- while [ -e $LPROC/osc ]; do
- for f in `find $LPROC -type f`; do
- cat $f > /dev/null 2>&1
- done
+ while lctl get_param osc.*.* > /dev/null 2>&1; do
+ : # loop until proc file is removed
done
}
test_59() { # bug 10589
zconf_mount `hostname` $MOUNT2 || error "Failed to mount $MOUNT2"
+ echo $DIR2 | grep -q $MOUNT2 || error "DIR2 is not set properly: $DIR2"
#define OBD_FAIL_LDLM_CANCEL_EVICT_RACE 0x311
sysctl -w lustre.fail_loc=0x311
- writes=`dd if=/dev/zero of=$DIR2/$tfile count=1 2>&1 | awk 'BEGIN { FS="+" } /out/ {print $1}'`
+ writes=$(LANG=C dd if=/dev/zero of=$DIR2/$tfile count=1 2>&1)
+ [ $? = 0 ] || error "dd write failed"
+ writes=$(echo $writes | awk -F '+' '/out/ {print $1}')
sysctl -w lustre.fail_loc=0
sync
- zconf_umount `hostname` $DIR2 -f
- reads=`dd if=$DIR/$tfile of=/dev/null 2>&1 | awk 'BEGIN { FS="+" } /in/ {print $1}'`
- [ $reads -eq $writes ] || error "read" $reads "blocks, must be" $writes
+ zconf_umount `hostname` $MOUNT2 -f
+ reads=$(LANG=C dd if=$DIR/$tfile of=/dev/null 2>&1)
+ [ $? = 0 ] || error "dd read failed"
+ reads=$(echo $reads | awk -F '+' '/in/ {print $1}')
+ [ "$reads" -eq "$writes" ] || error "read" $reads "blocks, must be" $writes
}
run_test 59 "Read cancel race on client eviction"
git://git.whamcloud.com / fs / lustre-release.git / blobdiff