FAIL_CLIENTS=$(echo " $FAIL_CLIENTS " | sed -re "s/\s+$LIVE_CLIENT\s+/ /g")
DIR=${DIR:-$MOUNT}
+TESTDIR=$DIR/d0.$(basename $0 .sh)
#####
# fail clients round robin
echo "fail $FAIL_CLIENT, next is $FAIL_NEXT"
}
-shutdown_client() {
- client=$1
- if [ "$FAILURE_MODE" = HARD ]; then
- $POWER_DOWN $client
- while ping -w 3 -c 1 $client > /dev/null 2>&1; do
- echo "waiting for node $client to fail"
- sleep 1
- done
- elif [ "$FAILURE_MODE" = SOFT ]; then
- zconf_umount $client $MOUNT -f
- fi
-}
-
fail_clients() {
num=$1
file=$1
for c in $LIVE_CLIENT $FAIL_CLIENTS; do
if echo $DOWN_CLIENTS | grep -q $c; then continue; fi
- $PDSH $c touch $MOUNT/${c}_$file || return 1
+ $PDSH $c touch $TESTDIR/${c}_$file || return 1
done
}
client_rm() {
file=$1
for c in $LIVE_CLIENT $FAIL_CLIENTS; do
- $PDSH $c rm $MOUNT/${c}_$file
+ $PDSH $c rm $TESTDIR/${c}_$file
done
}
client_mkdirs() {
for c in $LIVE_CLIENT $FAIL_CLIENTS; do
- echo "$c mkdir $MOUNT/$c"
- $PDSH $c "mkdir $MOUNT/$c"
- $PDSH $c "ls -l $MOUNT/$c"
+ echo "$c mkdir $TESTDIR/$c"
+ $PDSH $c "mkdir $TESTDIR/$c && ls -l $TESTDIR/$c"
done
}
client_rmdirs() {
for c in $LIVE_CLIENT $FAIL_CLIENTS; do
- echo "rmdir $MOUNT/$c"
- $PDSH $LIVE_CLIENT "rmdir $MOUNT/$c"
+ echo "rmdir $TESTDIR/$c"
+ $PDSH $LIVE_CLIENT "rmdir $TESTDIR/$c"
done
}
check_and_setup_lustre
+rm -rf $TESTDIR
+mkdir -p $TESTDIR
+
# 9 Different Failure Modes Combinations
echo "Starting Test 17 at `date`"
test_0() {
- facet_failover $SINGLEMDS
- echo "Waiting for df pid: $DFPID"
- wait $DFPID || { echo "df returned $?" && return 1; }
+ fail $SINGLEMDS
for i in $(seq $OSTCOUNT) ; do
- facet_failover ost$i || return 4
- echo "Waiting for df pid: $DFPID"
- wait $DFPID || { echo "df returned $?" && return 3; }
+ fail ost$i
done
return 0
}
[ -z "$(mounted_lustre_filesystems)" ] && error "Lustre is not running"
#MDS Portion
- facet_failover $SINGLEMDS
- wait $DFPID || echo df failed: $?
+ fail $SINGLEMDS
#Check FS
echo "Test Lustre stability after MDS failover"
############### Fifth Failure Mode ###############
test_5() {
- [ $OSTCOUNT -lt 2 ] && skip "$OSTCOUNT < 2, not enough OSTs" && return 0
+ [ $OSTCOUNT -lt 2 ] && skip_env "$OSTCOUNT < 2, not enough OSTs" && return 0
echo "Fifth Failure Mode: OST/OST `date`"
#Check FS
echo "Test Lustre stability after CLIENTs failure"
client_df
- $PDSH $LIVE_CLIENT "ls -l $MOUNT"
- $PDSH $LIVE_CLIENT "rm -f $MOUNT/*_testfile"
+ $PDSH $LIVE_CLIENT "ls -l $TESTDIR"
+ $PDSH $LIVE_CLIENT "rm -f $TESTDIR/*_testfile"
#Sleep
echo "Wait 1 minutes"
client_rm testfile
#MDS Portion
- facet_failover $SINGLEMDS
+ fail $SINGLEMDS
- #Check FS
- echo "Test Lustre stability after MDS failover"
- wait $DFPID || echo "df on down clients fails " || return 1
- $PDSH $LIVE_CLIENT "ls -l $MOUNT"
- $PDSH $LIVE_CLIENT "rm -f $MOUNT/*_testfile"
+ $PDSH $LIVE_CLIENT "ls -l $TESTDIR"
+ $PDSH $LIVE_CLIENT "rm -f $TESTDIR/*_testfile"
#Reintegration
echo "Reintegrating CLIENTs"
#Check FS
echo "Test Lustre stability after CLIENTs failure"
client_df
- $PDSH $LIVE_CLIENT "ls -l $MOUNT"
- $PDSH $LIVE_CLIENT "rm -f $MOUNT/*_testfile"
+ $PDSH $LIVE_CLIENT "ls -l $TESTDIR"
+ $PDSH $LIVE_CLIENT "rm -f $TESTDIR/*_testfile"
#Sleep
echo "Wait 1 minutes"
DFPID=$!
sleep 5
#non-failout hangs forever here
- #$PDSH $LIVE_CLIENT "ls -l $MOUNT"
- #$PDSH $LIVE_CLIENT "rm -f $MOUNT/*_testfile"
+ #$PDSH $LIVE_CLIENT "ls -l $TESTDIR"
+ #$PDSH $LIVE_CLIENT "rm -f $TESTDIR/*_testfile"
#Reintegration
echo "Reintegrating CLIENTs/OST"
#Check FS
echo "Test Lustre stability after CLIENTs failure"
client_df
- $PDSH $LIVE_CLIENT "ls -l $MOUNT" || return 1
- $PDSH $LIVE_CLIENT "rm -f $MOUNT/*_testfile" || return 2
+ $PDSH $LIVE_CLIENT "ls -l $TESTDIR" || return 1
+ $PDSH $LIVE_CLIENT "rm -f $TESTDIR/*_testfile" || return 2
#Sleep
echo "Wait 1 minutes"
#Check FS
echo "Test Lustre stability after CLIENTs failure"
client_df
- $PDSH $LIVE_CLIENT "ls -l $MOUNT" || return 5
- $PDSH $LIVE_CLIENT "rm -f $MOUNT/*_testfile" || return 6
+ $PDSH $LIVE_CLIENT "ls -l $TESTDIR" || return 5
+ $PDSH $LIVE_CLIENT "rm -f $TESTDIR/*_testfile" || return 6
#Reintegration
echo "Reintegrating CLIENTs/CLIENTs"