Viewing: recovery-double-scale.sh
#!/usr/bin/bash
#
# All pairwise combinations of node failures.
# Was cmd3-17
#
# Author: Chris Cooper <ccooper@clusterfs.com>
#
# Script fails pair of nodes:
# -- in parallel by default
# -- in series if SERIAL is set
set -e
ONLY=${ONLY:-"$*"}
LUSTRE=${LUSTRE:-$(dirname $0)/..}
. $LUSTRE/tests/test-framework.sh
init_test_env "$@"
init_logging
# bug number for skipped test:
ALWAYS_EXCEPT="$RECOVERY_DOUBLE_SCALE_EXCEPT"
# UPDATE THE COMMENT ABOVE WITH BUG NUMBERS WHEN CHANGING ALWAYS_EXCEPT!
build_test_filter
remote_mds_nodsh && skip_env "remote MDS with nodsh"
remote_ost_nodsh && skip_env "remote OST with nodsh"
(( CLIENTCOUNT >= 3 )) ||
skip_env "need three or more clients"
if [ -z "$SHARED_DIRECTORY" ] || ! check_shared_dir $SHARED_DIRECTORY; then
skip_env "SHARED_DIRECTORY should be specified with a shared directory \
which is accessable on all of the nodes"
exit 0
fi
[[ $FAILURE_MODE = SOFT ]] && \
log "WARNING: $0 is not functional with FAILURE_MODE = SOFT, bz22797"
# Set SERIAL to serialize the failure through a recovery of the first failure.
SERIAL=${SERIAL:-""}
ERRORS_OK="yes"
[ "$SERIAL" ] && ERRORS_OK=""
FAILOVER_PERIOD=${FAILOVER_PERIOD:-$((60 * 5))} # 5 minutes
END_RUN_FILE=${END_RUN_FILE:-$SHARED_DIRECTORY/end_run_file}
LOAD_PID_FILE=${LOAD_PID_FILE:-$TMP/client-load.pid}
init_stripe_dir_params RECOVERY_SCALE_ENABLE_REMOTE_DIRS \
RECOVERY_SCALE_ENABLE_STRIPED_DIRS
reboot_recover_node () {
# item var contains a pair of clients if nodetype=clients
# I would prefer to have a list here
local item=$1
local nodetype=$2
local c
# MDS, OST item contains the facet
case $nodetype in
MDS|OST ) facet_failover $item
[ "$SERIAL" ] && wait_recovery_complete $item || true
;;
clients) for c in ${item//,/ }; do
# make sure the client loads die
stop_process $c $LOAD_PID_FILE
shutdown_client $c
boot_node $c
echo "Reintegrating $c"
zconf_mount $c $MOUNT ||
error "mount $MOUNT on $c failed"
client_up $c || error "start client on $c failed"
done
start_client_loads $item
;;
* ) echo "ERROR: invalid nodetype=$nodetype." \
"Must be one of 'MDS', 'OST', or 'clients'."
exit 1;;
esac
}
get_item_type () {
local type=$1
local excluded=${2:-""}
local list
case $type in
MDS ) list=$MDTS;;
OST ) list=$OSTS;;
clients) list=$NODES_TO_USE;;
* ) echo "ERROR: invalid type=$type." \
"Must be one of 'MDS', 'OST', or 'clients'."
exit 1;;
esac
[ "$excluded" ] && list=$(exclude_items_from_list $list $excluded)
# empty list
if [ ! "$(echo $list)" ]; then
echo
return
fi
local item=$(get_random_entry $list)
if [ "$type" = "clients" ]; then
item="$item $(get_random_entry $(exclude_items_from_list $list $item))"
item=$(comma_list $item)
fi
echo $item
}
# failover_pair
#
# for the two nodetypes specified, chooses a random node(s) from each
# class, reboots the nodes sequentially, and then restarts lustre on
# the nodes.
failover_pair() {
local type1=$1
local type2=$2
local title=$3
local client_nodes=""
local item1=
local item2=
local client1=
local client2=
log "
==== START === $title"
item1=$(get_item_type $type1)
[ "$item1" ] || \
{ echo "type1=$type1 item1 is empty" && return 0; }
item2=$(get_item_type $type2 $item1)
[ "$item2" ] || \
{ echo "type1=$type1 item1=$item1 type2=$type2 item2=$item2 is empty" \
&& return 0; }
# Check that our client loads are still running. If any have died,
# that means they have died outside of recovery, which is unacceptable.
log "==== Checking the clients loads BEFORE failover -- failure NOT OK"
# FIXME. need print summary on exit
check_client_loads $NODES_TO_USE || exit $?
log "Done checking client loads. Failing type1=$type1 item1=$item1 ... "
reboot_recover_node $item1 $type1 || exit $?
# Hendrix test17 description:
# Introduce a failure, wait at
# least 5 minutes (for recovery),
# introduce a 2nd
# failure, and wait another 5
# minutes
# reboot_recover_node waits recovery in according to
# SERIAL value.
# We have a "double failures" if SERIAL is not set,
# do not need a sleep between failures for "double failures"
log " Failing type2=$type2 item2=$item2 ... "
reboot_recover_node $item2 $type2 || exit $?
# Client loads are allowed to die while in recovery, so we just
# restart them.
log "==== Checking the clients loads AFTER failovers -- ERRORS_OK=$ERRORS_OK"
restart_client_loads $NODES_TO_USE $ERRORS_OK || exit $?
log "Done checking / re-starting client loads. PASS"
return 0
}
summary_and_cleanup () {
local rc=$?
trap 0
CURRENT_TS=$(date +%s)
ELAPSED=$((CURRENT_TS - START_TS))
# Having not empty END_RUN_FILE means the failed loads only
if [ -s $END_RUN_FILE ]; then
print_end_run_file $END_RUN_FILE
rc=1
fi
echo $(date +'%F %H:%M:%S') Terminating clients loads ...
echo "$0" >> $END_RUN_FILE
local result=PASS
[ $rc -eq 0 ] || result=FAIL
log "
Server failover period: $FAILOVER_PERIOD seconds
Exited after: $ELAPSED seconds
Status: $result: rc=$rc"
# stop the client loads
stop_client_loads $NODES_TO_USE $LOAD_PID_FILE
if [ $rc -ne 0 ]; then
# we are interested in only on failed clients and servers
local failedclients=$(cat $END_RUN_FILE | grep -v $0)
gather_logs $(comma_list $(all_server_nodes) $failedclients)
fi
exit $rc
}
################################## Main Flow ###################################
check_and_setup_lustre
rm -rf $DIR/[Rdfs][0-9]*
check_timeout || exit 1
# The test node needs to be insulated from a lustre failure as much as possible,
# so not even loading the lustre modules is ideal.
# -- umount lustre
# -- remove hostname from clients list
zconf_umount $HOSTNAME $MOUNT
NODES_TO_USE=${NODES_TO_USE:-$CLIENTS}
log "Using NODES_TO_USE: $NODES_TO_USE and CLIENTS: $CLIENTS"
NODES_TO_USE=$(exclude_items_from_list $NODES_TO_USE $HOSTNAME)
log "Using remote NODES_TO_USE: $NODES_TO_USE HOSTNAME=$HOSTNAME"
check_progs_installed $NODES_TO_USE "${CLIENT_LOADS[@]}"
MDTS=$(get_facets MDS)
OSTS=$(get_facets OST)
ELAPSED=0
START_TS=$(date +%s)
CURRENT_TS=$START_TS
# Every pairwise combination of client failures (2 clients),
# MDS failure, and OST failure will be tested.
test_pairwise_fail() {
trap summary_and_cleanup EXIT TERM INT
# Start client loads.
rm -f $END_RUN_FILE
start_client_loads $NODES_TO_USE
echo clients load pids:
do_nodesv $NODES_TO_USE "cat $LOAD_PID_FILE" || exit 3
# FIXME: Do we want to have an initial sleep period where the clients
# just run before introducing a failure?
sleep $FAILOVER_PERIOD
# CMD_TEST_NUM=17.1
failover_pair MDS OST "test 1: failover MDS, then OST =========="
sleep $FAILOVER_PERIOD
# CMD_TEST_NUM=17.2
failover_pair MDS clients "test 2: failover MDS, then 2 clients ===="
sleep $FAILOVER_PERIOD
# CMD_TEST_NUM=17.3
if [ $MDSCOUNT -gt 1 ]; then
failover_pair MDS MDS "test 3: failover MDS, then another MDS =="
sleep $FAILOVER_PERIOD
else
skip_env "has less than 2 MDTs, test 3 skipped"
fi
# CMD_TEST_NUM=17.4
if [ $OSTCOUNT -gt 1 ]; then
failover_pair OST OST "test 4: failover OST, then another OST =="
sleep $FAILOVER_PERIOD
else
skip_env "has less than 2 OSTs, test 4 skipped"
fi
# CMD_TEST_NUM=17.5
failover_pair OST clients "test 5: failover OST, then 2 clients ===="
sleep $FAILOVER_PERIOD
# CMD_TEST_NUM=17.6
failover_pair OST MDS "test 6: failover OST, then MDS =========="
sleep $FAILOVER_PERIOD
# CMD_TEST_NUM=17.7
failover_pair clients MDS "test 7: failover 2 clients, then MDS ===="
sleep $FAILOVER_PERIOD
# CMD_TEST_NUM=17.8
failover_pair clients OST "test 8: failover 2 clients, then OST ===="
sleep $FAILOVER_PERIOD
# CMD_TEST_NUM=17.9
if [ $CLIENTCOUNT -gt 4 ]; then
failover_pair clients clients \
"test 9: failover 2 clients, then 2 different clients =="
sleep $FAILOVER_PERIOD
else
skip_env "has less than 5 Clients, test 9 skipped"
fi
log "==== Checking the clients loads AFTER all failovers -- failure NOT OK"
if ! check_client_loads $NODES_TO_USE; then
log "Client load failed after failover. Exiting..."
exit 5
fi
exit 0
}
run_test pairwise_fail "pairwise combination of clients, MDS, and OST failures"
zconf_mount $HOSTNAME $MOUNT || error "mount $MOUNT on $HOSTNAME failed"
client_up || error "start client on $HOSTNAME failed"
complete_test $SECONDS
check_and_cleanup_lustre
exit_status
