init_test_env $@
. ${CONFIG:=$LUSTRE/tests/cfg/$NAME.sh}
+init_logging
#
ALWAYS_EXCEPT="10 $INSANITY_EXCEPT"
assert_env ost_HOST OST_MKFS_OPTS OSTCOUNT
assert_env LIVE_CLIENT FSNAME
-remote_mds_nodsh && skip "remote MDS with nodsh" && exit 0
-remote_ost_nodsh && skip "remote OST with nodsh" && exit 0
+require_dsh_mds || exit 0
+require_dsh_ost || exit 0
# FAIL_CLIENTS list should not contain the LIVE_CLIENT
FAIL_CLIENTS=$(echo " $FAIL_CLIENTS " | sed -re "s/\s+$LIVE_CLIENT\s+/ /g")
DIR=${DIR:-$MOUNT}
+TESTDIR=$DIR/d0.$(basename $0 .sh)
#####
# fail clients round robin
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
}
echo "Verify Lustre filesystem is up and running"
[ -z "$(mounted_lustre_filesystems)" ] && error "Lustre is not running"
- client_df
+ clients_up
shutdown_facet $SINGLEMDS
reboot_facet $SINGLEMDS
change_active $SINGLEMDS
reboot_facet $SINGLEMDS
- client_df &
+ clients_up &
DFPID=$!
sleep 5
wait $DFPID
clients_recover_osts ost1
echo "Verify reintegration"
- client_df || return 1
+ clients_up || return 1
}
run_test 2 "Second Failure Mode: MDS/OST `date`"
[ -z "$(mounted_lustre_filesystems)" ] && error "Lustre is not running"
#MDS Portion
- facet_failover $SINGLEMDS
- wait $DFPID || echo df failed: $?
+ fail $SINGLEMDS
#Check FS
echo "Test Lustre stability after MDS failover"
- client_df
+ clients_up
#CLIENT Portion
echo "Failing 2 CLIENTS"
#Check FS
echo "Test Lustre stability after CLIENT failure"
- client_df
+ clients_up
#Reintegration
echo "Reintegrating CLIENTS"
reintegrate_clients || return 1
- client_df || return 3
+ clients_up || return 3
sleep 2 # give it a little time for fully recovered before next test
}
run_test 3 "Thirdb Failure Mode: MDS/CLIENT `date`"
#Check FS
echo "Test Lustre stability after OST failure"
- client_df &
+ clients_up &
DFPIDA=$!
sleep 5
change_active $SINGLEMDS
reboot_facet $SINGLEMDS
- client_df &
+ clients_up &
DFPIDB=$!
sleep 5
wait $DFPIDB
clients_recover_osts ost1
echo "Test Lustre stability after MDS failover"
- client_df || return 1
+ clients_up || return 1
}
run_test 4 "Fourth Failure Mode: OST/MDS `date`"
###################################################
############### 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`"
echo "Verify Lustre filesystem is up and running"
[ -z "$(mounted_lustre_filesystems)" ] && error "Lustre is not running"
- client_df
+ clients_up
#OST Portion
shutdown_facet ost1
#Check FS
echo "Test Lustre stability after OST failure"
- client_df &
+ clients_up &
DFPIDA=$!
sleep 5
#Check FS
echo "Test Lustre stability after OST failure"
- client_df &
+ clients_up &
DFPIDB=$!
sleep 5
wait $DFPIDA
wait $DFPIDB
- client_df || return 2
+ clients_up || return 2
}
run_test 5 "Fifth Failure Mode: OST/OST `date`"
###################################################
echo "Verify Lustre filesystem is up and running"
[ -z "$(mounted_lustre_filesystems)" ] && error "Lustre is not running"
- client_df
+ clients_up
client_touch testfile || return 2
#OST Portion
#Check FS
echo "Test Lustre stability after OST failure"
- client_df &
+ clients_up &
DFPIDA=$!
echo DFPIDA=$DFPIDA
sleep 5
#Check FS
echo "Test Lustre stability after CLIENTs failure"
- client_df &
+ clients_up &
DFPIDB=$!
echo DFPIDB=$DFPIDB
sleep 5
reintegrate_clients || return 1
sleep 5
- wait_remote_prog df $((TIMEOUT * 3 + 10))
+ wait_remote_prog "stat -f" $((TIMEOUT * 3 + 20))
wait $DFPIDA
wait $DFPIDB
echo "Verifying mount"
[ -z "$(mounted_lustre_filesystems)" ] && return 3
- client_df
+ clients_up
}
run_test 6 "Sixth Failure Mode: OST/CLIENT `date`"
###################################################
echo "Verify Lustre filesystem is up and running"
[ -z "$(mounted_lustre_filesystems)" ] && error "Lustre is not running"
- client_df
+ clients_up
client_touch testfile || return 1
#CLIENT Portion
#Check FS
echo "Test Lustre stability after CLIENTs failure"
- client_df
- $PDSH $LIVE_CLIENT "ls -l $MOUNT"
- $PDSH $LIVE_CLIENT "rm -f $MOUNT/*_testfile"
+ clients_up
+ $PDSH $LIVE_CLIENT "ls -l $TESTDIR"
+ $PDSH $LIVE_CLIENT "rm -f $TESTDIR/*_testfile"
#Sleep
echo "Wait 1 minutes"
echo "Verify Lustre filesystem is up and running"
[ -z "$(mounted_lustre_filesystems)" ] && return 2
- client_df
+ clients_up
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 $TESTDIR/*_testfile"
#Reintegration
echo "Reintegrating CLIENTs"
reintegrate_clients || return 2
- client_df
+ clients_up
#Sleep
echo "wait 1 minutes"
echo "Verify Lustre filesystem is up and running"
[ -z "$(mounted_lustre_filesystems)" ] && error "Lustre is not running"
- client_df
+ clients_up
client_touch testfile
#CLIENT Portion
#Check FS
echo "Test Lustre stability after CLIENTs failure"
- client_df
- $PDSH $LIVE_CLIENT "ls -l $MOUNT"
- $PDSH $LIVE_CLIENT "rm -f $MOUNT/*_testfile"
+ clients_up
+ $PDSH $LIVE_CLIENT "ls -l $TESTDIR"
+ $PDSH $LIVE_CLIENT "rm -f $TESTDIR/*_testfile"
#Sleep
echo "Wait 1 minutes"
echo "Verify Lustre filesystem is up and running"
[ -z "$(mounted_lustre_filesystems)" ] && error "Lustre is not running"
- client_df
+ clients_up
client_touch testfile
#Check FS
echo "Test Lustre stability after OST failure"
- client_df &
+ clients_up &
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"
wait_for ost1
start_ost 1
wait $DFPID
- client_df || return 1
+ clients_up || return 1
client_touch testfile2 || return 2
#Sleep
echo "Verify Lustre filesystem is up and running"
[ -z "$(mounted_lustre_filesystems)" ] && error "Lustre is not running"
- client_df
+ clients_up
client_touch testfile || return 1
#CLIENT Portion
#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
+ clients_up
+ $PDSH $LIVE_CLIENT "ls -l $TESTDIR" || return 1
+ $PDSH $LIVE_CLIENT "rm -f $TESTDIR/*_testfile" || return 2
#Sleep
echo "Wait 1 minutes"
#Create files
echo "Verify Lustre filesystem is up and running"
- $PDSH $LIVE_CLIENT df $MOUNT || return 3
+ client_up $LIVE_CLIENT || return 3
client_touch testfile || return 4
#CLIENT Portion
#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
+ clients_up
+ $PDSH $LIVE_CLIENT "ls -l $TESTDIR" || return 5
+ $PDSH $LIVE_CLIENT "rm -f $TESTDIR/*_testfile" || return 6
#Reintegration
echo "Reintegrating CLIENTs/CLIENTs"
reintegrate_clients || return 7
- client_df
+ clients_up
#Sleep
echo "Wait 1 minutes"