stop_ost
mount_client $MOUNT
# check_mount will block trying to contact ost
+ mcreate $DIR/$tfile || return 40
+ rm -f $DIR/$tfile || return 42
umount_client $MOUNT
pass
}
test_32a() {
- # this test is totally useless on a client-only system
- [ -n "$CLIENTONLY" -o -n "$CLIENTMODSONLY" ] && skip "client only testing" && return 0
+ client_only && skip "client only testing" && return 0
[ "$NETTYPE" = "tcp" ] || { skip "NETTYPE != tcp" && return 0; }
[ -z "$TUNEFS" ] && skip_env "No tunefs" && return 0
local NID=$($LCTL list_nids | head -1)
- echo "OSC changes should return err:"
+ echo "OSC changes should succeed:"
$LCTL conf_param lustre-OST0000.osc.max_dirty_mb=15 || return 7
$LCTL conf_param lustre-OST0000.failover.node=$NID || return 8
-
echo "ok."
+
echo "MDC changes should succeed:"
$LCTL conf_param lustre-MDT0000.mdc.max_rpcs_in_flight=9 || return 9
$LCTL conf_param lustre-MDT0000.failover.node=$NID || return 10
echo "ok."
+ echo "LOV changes should succeed:"
+ $LCTL pool_new lustre.interop || return 11
+ $LCTL conf_param lustre-MDT0000.lov.stripesize=4M || return 12
+ echo "ok."
+
cleanup_32
# mount a second time to make sure we didnt leave upgrade flag on
run_test 32a "Upgrade from 1.8 (not live)"
test_32b() {
- # this test is totally useless on a client-only system
- [ -n "$CLIENTONLY" -o -n "$CLIENTMODSONLY" ] && skip "client only testing" && return 0
+ client_only && skip "client only testing" && return 0
[ "$NETTYPE" = "tcp" ] || { skip "NETTYPE != tcp" && return 0; }
[ -z "$TUNEFS" ] && skip_env "No tunefs" && return
local NID=$($LCTL list_nids | head -1)
echo "OSC changes should succeed:"
-
$LCTL conf_param ${NEWNAME}-OST0000.osc.max_dirty_mb=15 || return 7
$LCTL conf_param ${NEWNAME}-OST0000.failover.node=$NID || return 8
-
echo "ok."
+
echo "MDC changes should succeed:"
$LCTL conf_param ${NEWNAME}-MDT0000.mdc.max_rpcs_in_flight=9 || return 9
+ $LCTL conf_param lustre-MDT0000.failover.node=$NID || return 10
+ echo "ok."
+
+ echo "LOV changes should succeed:"
+ $LCTL pool_new lustre.interop || return 11
+ $LCTL conf_param lustre-MDT0000.lov.stripesize=4M || return 12
echo "ok."
# MDT and OST should have registered with new nids, so we should have
}
run_test 34c "force umount with failed ost should be normal"
-test_35() { # bug 12459
+test_35a() { # bug 12459
setup
DBG_SAVE="`lctl get_param -n debug`"
log "Wait for RECONNECT_INTERVAL seconds (10s)"
sleep 10
- MSG="conf-sanity.sh test_35 `date +%F%kh%Mm%Ss`"
+ MSG="conf-sanity.sh test_35a `date +%F%kh%Mm%Ss`"
$LCTL clear
log "$MSG"
log "Stopping the MDT:"
[ "$NEXTCONN" != "0" ] && log "The client didn't try to reconnect to the last active server (tried ${NEXTCONN} instead)" && return 7
cleanup
}
-run_test 35 "Reconnect to the last active server first"
+run_test 35a "Reconnect to the last active server first"
+
+test_35b() { # bug 18674
+ remote_mds || { skip "local MDS" && return 0; }
+ setup
+
+ debugsave
+ $LCTL set_param debug="ha"
+ $LCTL clear
+ MSG="conf-sanity.sh test_35b `date +%F%kh%Mm%Ss`"
+ log "$MSG"
+
+ log "Set up a fake failnode for the MDS"
+ FAKENID="127.0.0.2"
+ local device=$(do_facet mds "$LCTL get_param -n devices" | \
+ awk '($3 ~ "mdt" && $4 ~ "MDT") { print $4 }' | head -1)
+ do_facet mds "$LCTL conf_param ${device}.failover.node=$FAKENID" || \
+ return 1
+
+ local at_max_saved=0
+ # adaptive timeouts may prevent seeing the issue
+ if at_is_enabled; then
+ at_max_saved=$(at_max_get mds)
+ at_max_set 0 mds client
+ fi
+
+ mkdir -p $MOUNT/testdir
+ touch $MOUNT/testdir/test
+
+ log "Injecting EBUSY on MDS"
+ # Setting OBD_FAIL_MDS_RESEND=0x136
+ do_facet mds "$LCTL set_param fail_loc=0x80000136" || return 2
+
+ log "Stat on a test file"
+ stat $MOUNT/testdir/test
+
+ log "Stop injecting EBUSY on MDS"
+ do_facet mds "$LCTL set_param fail_loc=0" || return 3
+ rm -f $MOUNT/testdir/test
+
+ log "done"
+ # restore adaptive timeout
+ [ $at_max_saved -ne 0 ] && at_max_set $at_max_saved mds client
+
+ $LCTL dk $TMP/lustre-log-$TESTNAME.log
+
+ # retrieve from the log if the client has ever tried to
+ # contact the fake server after the loss of connection
+ FAILCONN=`awk "BEGIN {ret = 0;}
+ /import_select_connection.*${FSNAME}-MDT0000-mdc.* using connection/ {
+ ret = 1;
+ if (\\\$NF ~ /$FAKENID/) {
+ ret = 2;
+ exit;
+ }
+ }
+ END {print ret}" $TMP/lustre-log-$TESTNAME.log`
+
+ [ "$FAILCONN" == "0" ] && \
+ log "ERROR: The client reconnection has not been triggered" && \
+ return 4
+ [ "$FAILCONN" == "2" ] && \
+ log "ERROR: The client tried to reconnect to the failover server while the primary was busy" && \
+ return 5
+
+ cleanup
+}
+run_test 35b "Continue reconnection retries, if the active server is busy"
test_36() { # 12743
local rc
run_test 36 "df report consistency on OSTs with different block size"
test_37() {
- [ -n "$CLIENTONLY" -o -n "$CLIENTMODSONLY" ] && skip "client only testing" && return 0
+ client_only && skip "client only testing" && return 0
LOCAL_MDSDEV="$TMP/mdt.img"
SYM_MDSDEV="$TMP/sym_mdt.img"
cleanup_46a() {
trap 0
local rc=0
- local count=5
+ local count=$1
umount_client $MOUNT2 || rc=$?
umount_client $MOUNT || rc=$?
}
test_46a() {
- [ $OSTCOUNT -lt 5 ] && skip_env "too few OSTs" && return
+ echo "Testing with $OSTCOUNT OSTs"
reformat
start_mds || return 1
#first client should see only one ost
wait_osc_import_state mds ost FULL
#start_client
mount_client $MOUNT || return 3
- trap cleanup_46a EXIT ERR
-
- start_ost2 || return 4
- start ost3 `ostdevname 3` $OST_MOUNT_OPTS || return 5
- start ost4 `ostdevname 4` $OST_MOUNT_OPTS || return 6
- start ost5 `ostdevname 5` $OST_MOUNT_OPTS || return 7
- # wait until ost2-5 is sync
- # ping_interval + 1
- wait_osc_import_state mds ost2 FULL
- wait_osc_import_state mds ost3 FULL
- wait_osc_import_state mds ost4 FULL
- wait_osc_import_state mds ost5 FULL
+ trap "cleanup_46a $OSTCOUNT" EXIT ERR
+
+ local i
+ for (( i=2; i<=$OSTCOUNT; i++ )); do
+ start ost$i `ostdevname $i` $OST_MOUNT_OPTS || return $((i+2))
+ done
+
+ # wait until osts in sync
+ for (( i=2; i<=$OSTCOUNT; i++ )); do
+ wait_osc_import_state mds ost$i FULL
+ done
+
+
#second client see all ost's
mount_client $MOUNT2 || return 8
# will be deadlock
stat $MOUNT/widestripe || return 12
- cleanup_46a || { echo "cleanup_46a failed!" && return 13; }
+ cleanup_46a $OSTCOUNT || { echo "cleanup_46a failed!" && return 13; }
return 0
}
run_test 46a "handle ost additional - wide striped file"
}
run_test 50f "normal statfs one server in down =========================="
+test_50g() {
+ [ "$OSTCOUNT" -lt "2" ] && skip_env "$OSTCOUNT < 2, skipping" && return
+ setup
+ start_ost2 || error "Unable to start OST2"
+
+ local PARAM="${FSNAME}-OST0001.osc.active"
+
+ $LFS setstripe -c -1 $DIR/$tfile || error "Unable to lfs setstripe"
+ do_facet mgs $LCTL conf_param $PARAM=0 || error "Unable to deactivate OST"
+
+ umount_client $MOUNT || error "Unable to unmount client"
+ mount_client $MOUNT || error "Unable to mount client"
+ # This df should not cause a panic
+ df -k $MOUNT
+
+ do_facet mgs $LCTL conf_param $PARAM=1 || error "Unable to activate OST"
+ rm -f $DIR/$tfile
+ umount_client $MOUNT || error "Unable to unmount client"
+ stop_ost2 || error "Unable to stop OST2"
+ stop_ost || error "Unable to stop OST1"
+ stop_mds || error "Unable to stop MDS"
+ writeconf
+}
+run_test 50g "deactivated OST should not cause panic====================="
+
test_51() {
local LOCAL_TIMEOUT=20