#!/bin/bash
-# vim: expandtab tabstop=4 softtabstop=4 shiftwidth=4 textwidth=80
+# -*- mode: Bash; tab-width: 4; indent-tabs-mode: t; -*-
+# vim:shiftwidth=4:softtabstop=4:tabstop=4:
#
# NAME
#
# -u SECONDS
# Define a duration for the test. 86400 seconds if not specified.
#
+# -p SECONDS
+# Define a max failover period. 10 minutes if not set.
+#
# -w
# Only run the workloads; no failure will be introduced.
+# -v, -s are ignored in this case.
+# -r
+# Workloads dry run for several seconds; no failures will be introduced.
+# This option is useful to verify the loads.
+# -u is ignored in this case
+# -m
+# Reboot victim nodes simultaneously.
+#
#
# ASSUMPTIONS
#
# Each target has a failnode, so that workloads can continue after a power
# failure.
#
+# CRM could be configured by 2 ways:
+# 1.
# Targets are automatically failed back when their primary node is back. This
# assumption avoids calling CRM-specific commands to trigger failbacks, making
# this script more CRM-neural.
+# 2.
+# Targets are not automatically failed back when their primary node is back.
+# CRM-specific command is executed to trigger failbacks.
#
# A crash dump mechanism is configured to catch LBUGs, panics, etc.
#
# WORKLOADS
#
-# Each client runs the same set of MPI and non-MPI workloads. These
+# Each client runs set of MPI and non-MPI workloads. These
# applications are run in short loops so that their exit status can be waited
# for and checked within reasonable time by ha_wait_loads.
+# The set of MPI and non-MPI workloads are configurable by parameters:
+# ha_nonmpi_loads
+# default set: dd, tar, iozone
+# ha_mpi_loads
+# default set: ior, simul, mdtest
+#
+# The number of clients run MPI loads is configured by parameter
+# ha_mpi_instances. Only one client runs MPI workloads by default.
#
# PROCESS STRUCTURE AND IPC
#
# ~ mpirun IOR
# ~ ha.sh (ha_repeat_mpi_load simul)
# ~ mpirun simul
+# ~ ha.sh (ha_repeat_mpi_load mdtest)
+# ~ mpirun mdtest
# ~ ... (one for each MPI load)
#
# ~ ha.sh (ha_repeat_nonmpi_load client2 dbench)
# IPC is done by files in the temporary directory.
#
+#set -x
+
+SIMUL=${SIMUL:-$(which simul 2> /dev/null || true)}
+IOR=${IOR:-$(which IOR 2> /dev/null || true)}
+MDTEST=${MDTEST:-$(which mdtest 2> /dev/null || true)}
+
+ior_blockSize=${ior_blockSize:-6g}
+mpi_threads_per_client=${mpi_threads_per_client:-2}
+
+iozone_SIZE=${iozone_SIZE:-262144} # 256m
+
+mpirun=${MPIRUN:-$(which mpirun)}
+LFS=${LFS:-$(which lfs)}
+
+ha_check_env()
+{
+ for ((load = 0; load < ${#ha_mpi_load_tags[@]}; load++)); do
+ local tag=${ha_mpi_load_tags[$load]}
+ local bin=$(echo $tag | tr '[:lower:]' '[:upper:]')
+ if [ x${!bin} = x ]; then
+ ha_error ha_mpi_loads: ${ha_mpi_loads}, $bin is not set
+ exit 1
+ fi
+ done
+}
+
ha_info()
{
- echo "$0: $(date +%s):" "$@"
+ echo "$0: $(date +%H:%M:%S' '%s):" "$@"
+}
+
+ha_log()
+{
+ local nodes=${1// /,}
+ shift
+ ha_on $nodes "lctl mark $*"
}
ha_error()
declare ha_status_file_prefix=$ha_tmp_dir/status
declare -a ha_status_files
declare ha_machine_file=$ha_tmp_dir/machine_file
-declare ha_power_down_cmd=${POWER_DOWN:-pm -0}
-declare ha_power_up_cmd=${POWER_UP:-pm -1}
+declare ha_power_down_cmd=${POWER_DOWN:-"pm -0"}
+declare ha_power_up_cmd=${POWER_UP:-"pm -1"}
+declare ha_power_delay=${POWER_DELAY:-60}
+declare ha_failback_delay=${DELAY:-5}
+declare ha_failback_cmd=${FAILBACK:-""}
+declare ha_stripe_params=${STRIPEPARAMS:-"-c 0"}
declare -a ha_clients
declare -a ha_servers
declare -a ha_victims
declare ha_test_dir=/mnt/lustre/$(basename $0)-$$
declare ha_start_time=$(date +%s)
declare ha_expected_duration=$((60 * 60 * 24))
+declare ha_max_failover_period=10
declare ha_nr_loops=0
declare ha_stop_signals="SIGINT SIGTERM SIGHUP"
declare ha_load_timeout=$((60 * 10))
declare ha_workloads_only=false
-declare -a ha_mpi_load_tags=(
- ior
- simul
-)
-declare -a ha_mpi_load_cmds=(
- "/testsuite/tests/x86_64/rhel5/IOR/src/C/IOR -b 256m -o {}/f.ior -t 2m
- -w -W -T 1"
- "/testsuite/tests/x86_64/rhel5/simul/simul -d {}"
-)
-declare -a ha_nonmpi_load_tags=(
- dd
- tar
+declare ha_workloads_dry_run=false
+declare ha_simultaneous=false
+
+declare ha_mpi_instances=${ha_mpi_instances:-1}
+
+declare ha_mpi_loads=${ha_mpi_loads="ior simul mdtest"}
+declare -a ha_mpi_load_tags=($ha_mpi_loads)
+
+declare ha_ior_params=${IORP:-'" -b $ior_blockSize -t 2m -w -W -T 1"'}
+declare ha_simul_params=${SIMULP:-'" -n 10"'}
+declare ha_mdtest_params=${MDTESTP:-'" -i 1 -n 1000"'}
+declare ha_mpirun_options=${MPIRUN_OPTIONS:-""}
+
+eval ha_params_ior=($ha_ior_params)
+eval ha_params_simul=($ha_simul_params)
+eval ha_params_mdtest=($ha_mdtest_params)
+
+declare ha_nparams_ior=${#ha_params_ior[@]}
+declare ha_nparams_simul=${#ha_params_simul[@]}
+declare ha_nparams_mdtest=${#ha_params_mdtest[@]}
+
+declare -A ha_mpi_load_cmds=(
+ [ior]="$IOR -o {}/f.ior {params}"
+ [simul]="$SIMUL {params} -d {}"
+ [mdtest]="$MDTEST {params} -d {}"
)
+
+declare ha_nonmpi_loads=${ha_nonmpi_loads="dd tar iozone"}
+declare -a ha_nonmpi_load_tags=($ha_nonmpi_loads)
declare -a ha_nonmpi_load_cmds=(
- "dd if=/dev/zero of={}/f.dd bs=1M count=256"
- "tar cf - /etc/fonts | tar xf - -C {}"
+ "dd if=/dev/zero of={}/f.dd bs=1M count=256"
+ "tar cf - /etc | tar xf - -C {}"
+ "iozone -a -e -+d -s $iozone_SIZE {}/f.iozone"
)
ha_usage()
{
local opt
- while getopts hc:s:v:d:u:w opt; do
+ while getopts hc:s:v:d:p:u:wrm opt; do
case $opt in
h)
ha_usage
u)
ha_expected_duration=$OPTARG
;;
+ p)
+ ha_max_failover_period=$OPTARG
+ ;;
w)
- ha_workloads_only=true
- ;;
+ ha_workloads_only=true
+ ;;
+ r)
+ ha_workloads_dry_run=true
+ ;;
+ m)
+ ha_simultaneous=true
+ ;;
\?)
ha_usage
exit 1
esac
done
- if [ -z "${ha_clients[*]}" ] || \
- [ -z "${ha_servers[*]}" ] || \
- [ -z "${ha_victims[*]}" ]; then
- ha_error "-c, -s, and -v are all mandatory"
- ha_usage
- exit 1
- fi
+ if [ -z "${ha_clients[*]}" ]; then
+ ha_error "-c is mandatory"
+ ha_usage
+ exit 1
+ fi
+ if ! ($ha_workloads_dry_run ||
+ $ha_workloads_only) &&
+ ([ -z "${ha_servers[*]}" ] ||
+ [ -z "${ha_victims[*]}" ]); then
+ ha_error "-s, and -v are all mandatory"
+ ha_usage
+ exit 1
+ fi
}
ha_on()
{
- local nodes=$1
- local rc=0
-
- shift
- pdsh -w $nodes PATH=/usr/kerberos/sbin:/usr/kerberos/bin:/usr/local/sbin:/usr/local/bin:/sbin:/bin:/usr/sbin:/usr/bin "$@" || rc=$?
- return $rc
+ local nodes=$1
+ local rc=0
+
+ shift
+
+ #
+ # -S is to be used here to track the
+ # remote command return values
+ #
+ pdsh -S -w $nodes PATH=/usr/local/sbin:/usr/local/bin:/sbin:\
+/bin:/usr/sbin:/usr/bin "$@" ||
+ rc=$?
+ return $rc
}
ha_trap_exit()
{
- if [ -e "$ha_fail_file" ]; then
- ha_info "Test directory $ha_test_dir not removed"
- ha_info "Temporary directory $ha_tmp_dir not removed"
- else
- ha_on ${ha_clients[0]} rm -rf "$ha_test_dir"
- rm -rf "$ha_tmp_dir"
- fi
+ touch "$ha_stop_file"
+ trap 0
+ if [ -e "$ha_fail_file" ]; then
+ ha_info "Test directory $ha_test_dir not removed"
+ ha_info "Temporary directory $ha_tmp_dir not removed"
+ else
+ ha_on ${ha_clients[0]} rm -rf "$ha_test_dir"
+ ha_info "Please find the results in the directory $ha_tmp_dir"
+ fi
}
ha_trap_stop_signals()
ha_lock "$lock"
ha_info "Dumping lctl log to $file"
- ha_on $nodes "lctl dk >$file" || true
- ha_unlock "$lock"
+
+ #
+ # some nodes could crash, so
+ # do not exit with error if not all logs are dumped
+ #
+ ha_on $nodes "lctl dk >$file" ||
+ ha_error "not all logs are dumped! Some nodes are unreachable."
+ ha_unlock "$lock"
}
ha_repeat_mpi_load()
{
- local load=$1
- local status=$2
- local tag=${ha_mpi_load_tags[$load]}
- local cmd=${ha_mpi_load_cmds[$load]}
- local dir=$ha_test_dir/$tag
- local log=$ha_tmp_dir/$tag
- local rc=0
- local nr_loops=0
- local start_time=$(date +%s)
-
- cmd=${cmd//"{}"/$dir}
-
- ha_info "Starting $tag"
-
- while [ ! -e "$ha_stop_file" ] && ((rc == 0)); do
- {
- ha_on ${ha_clients[0]} mkdir -p "$dir" && \
- mpirun -np ${#ha_clients[@]} -machinefile "$ha_machine_file" \
- $cmd && \
- ha_on ${ha_clients[0]} rm -rf "$dir"
- } >>"$log" 2>&1 || rc=$?
-
- if ((rc != 0)); then
- ha_dump_logs "${ha_clients[*]} ${ha_servers[*]}"
- touch "$ha_fail_file"
- touch "$ha_stop_file"
- fi
- echo $rc >"$status"
-
- nr_loops=$((nr_loops + 1))
- done
-
- avg_loop_time=$((($(date +%s) - start_time) / nr_loops))
-
- ha_info "$tag stopped: rc $rc avg loop time $avg_loop_time"
+ local client=$1
+ local load=$2
+ local status=$3
+ local parameter=$4
+ local tag=${ha_mpi_load_tags[$load]}
+ local cmd=${ha_mpi_load_cmds[$tag]}
+ local dir=$ha_test_dir/$client-$tag
+ local log=$ha_tmp_dir/$client-$tag
+ local rc=0
+ local nr_loops=0
+ local start_time=$(date +%s)
+
+ cmd=${cmd//"{}"/$dir}
+ cmd=${cmd//"{params}"/$parameter}
+
+ ha_info "Starting $tag"
+
+ local machines="-machinefile $ha_machine_file"
+ while [ ! -e "$ha_stop_file" ] && ((rc == 0)); do
+ {
+ ha_on $client mkdir -p "$dir" &&
+ ha_on $client chmod a+xwr $dir &&
+ ha_on $client "su mpiuser sh -c \" $mpirun $ha_mpirun_options \
+ -np $((${#ha_clients[@]} * mpi_threads_per_client )) \
+ $machines $cmd \" " &&
+ ha_on $client rm -rf "$dir";
+ } >>"$log" 2>&1 || rc=$?
+
+ ha_info rc=$rc
+
+ if ((rc != 0)); then
+ touch "$ha_fail_file"
+ touch "$ha_stop_file"
+ ha_dump_logs "${ha_clients[*]} ${ha_servers[*]}"
+ fi
+ echo $rc >"$status"
+
+ nr_loops=$((nr_loops + 1))
+ done
+
+ avg_loop_time=$((($(date +%s) - start_time) / nr_loops))
+
+ ha_info "$tag stopped: rc $rc avg loop time $avg_loop_time"
}
ha_start_mpi_loads()
{
- local client
- local load
- local tag
- local status
-
- for client in ${ha_clients[@]}; do
- echo $client >>"$ha_machine_file"
- done
-
- for ((load = 0; load < ${#ha_mpi_load_tags[@]}; load++)); do
- tag=${ha_mpi_load_tags[$load]}
- status=$ha_status_file_prefix-$tag
- ha_repeat_mpi_load $load $status &
- ha_status_files+=("$status")
- done
+ local client
+ local load
+ local tag
+ local status
+ local n
+ local nparam
+
+ for client in ${ha_clients[@]}; do
+ ha_info ha_machine_file=$ha_machine_file
+ echo $client >> $ha_machine_file
+ done
+ local dirname=$(dirname $ha_machine_file)
+ for client in ${ha_clients[@]}; do
+ ha_on $client mkdir -p $dirname
+ scp $ha_machine_file $client:$ha_machine_file
+ done
+
+ # ha_mpi_instances defines the number of
+ # clients start mpi loads; should be <= ${#ha_clients[@]}
+ local inst=$ha_mpi_instances
+ (( inst <= ${#ha_clients[@]} )) || inst=${#ha_clients[@]}
+
+ for ((n = 0; n < $inst; n++)); do
+ client=${ha_clients[n]}
+ for ((load = 0; load < ${#ha_mpi_load_tags[@]}; load++)); do
+ tag=${ha_mpi_load_tags[$load]}
+ status=$ha_status_file_prefix-$tag-$client
+ # ha_nparams_ior
+ # ha_nparams_simul
+ local num=ha_nparams_$tag
+ nparam=$((n % num))
+ local aref=ha_params_$tag[nparam]
+ local parameter=${!aref}
+ ha_repeat_mpi_load $client $load $status "$parameter" &
+ ha_status_files+=("$status")
+ done
+ done
}
ha_repeat_nonmpi_load()
ha_info "Starting $tag on $client"
- while [ ! -e "$ha_stop_file" ] && ((rc == 0)); do
- ha_on $client "mkdir -p $dir && \
- $cmd && \
- rm -rf $dir" >>"$log" 2>&1 || rc=$?
+ while [ ! -e "$ha_stop_file" ] && ((rc == 0)); do
+ ha_on $client "mkdir -p $dir && \
+ $cmd && \
+ rm -rf $dir" >>"$log" 2>&1 || rc=$?
- if ((rc != 0)); then
- ha_dump_logs "$client ${ha_servers[*]}"
- touch "$ha_fail_file"
- touch "$ha_stop_file"
- fi
- echo $rc >"$status"
+ if ((rc != 0)); then
+ ha_dump_logs "${ha_clients[*]} ${ha_servers[*]}"
+ touch "$ha_fail_file"
+ touch "$ha_stop_file"
+ fi
+ echo $rc >"$status"
- nr_loops=$((nr_loops + 1))
- done
+ nr_loops=$((nr_loops + 1))
+ done
avg_loop_time=$((($(date +%s) - start_time) / nr_loops))
ha_info "Waiting for workload status"
rm -f "${ha_status_files[@]}"
- for file in "${ha_status_files[@]}"; do
- until [ -e "$ha_stop_file" ] ||
- [ -e "$file" ]; do
- if (($(date +%s) >= end)); then
- ha_info "Timed out while waiting for load status file $file"
- touch "$ha_fail_file"
- return 1
- fi
- ha_sleep 1 >/dev/null
- done
- done
+
+ #
+ # return immediately if ha_stop_file exists,
+ # all status_files not needed to be checked
+ #
+ for file in "${ha_status_files[@]}"; do
+ if [ -e "$ha_stop_file" ]; then
+ ha_info "$ha_stop_file found! Stop."
+ break
+ fi
+ #
+ # Wait status file created during ha_load_timeout.
+ # Existing file guarantees that some application
+ # is completed. If no status file was created
+ # this function guarantees that we allow
+ # applications to continue after/before
+ # failover/failback during ha_load_timeout time.
+ #
+ until [ -e "$file" ] || (($(date +%s) >= end)); do
+ #
+ # check ha_stop_file again, it could appear
+ # during ha_load_timeout
+ #
+ if [ -e "$ha_stop_file" ]; then
+ ha_info "$ha_stop_file found! Stop."
+ break
+ fi
+ ha_sleep 1 >/dev/null
+ done
+ done
}
ha_power_down()
{
- local node=$1
+ local nodes=$1
+ local rc=1
+ local i
+
+ ha_info "Powering down $nodes"
+ for i in $(seq 1 5); do
+ $ha_power_down_cmd $nodes && rc=0 && break
+ sleep $ha_power_delay
+ done
- ha_info "Powering down $node"
- $ha_power_down_cmd $node
+ [ $rc -eq 0 ] || ha_info "Failed Powering down in $i attempts"
}
ha_power_up()
{
- local node=$1
+ local nodes=$1
+ local rc=1
+ local i
- ha_info "Powering up $node"
- $ha_power_up_cmd $node
+ ha_info "Powering up $nodes"
+ for i in $(seq 1 5); do
+ $ha_power_up_cmd $nodes && rc=0 && break
+ sleep $ha_power_delay
+ done
+
+ [ $rc -eq 0 ] || ha_info "Failed Powering up in $i attempts"
}
#
ha_aim()
{
- local i=$(ha_rand ${#ha_victims[@]})
+ local i
+ local nodes
+
+ if $ha_simultaneous ; then
+ nodes=$(echo ${ha_victims[@]})
+ nodes=${nodes// /,}
+ else
+ i=$(ha_rand ${#ha_victims[@]})
+ nodes=${ha_victims[$i]}
+ fi
+
+ echo -n $nodes
+}
- echo -n ${ha_victims[$i]}
+ha_wait_nodes()
+{
+ local nodes=$1
+ local end=$(($(date +%s) + 10 * 60))
+
+ ha_info "Waiting for $nodes to boot up"
+ until ha_on $nodes hostname >/dev/null 2>&1 ||
+ [ -e "$ha_stop_file" ] ||
+ (($(date +%s) >= end)); do
+ ha_sleep 1 >/dev/null
+ done
}
-ha_wait_node()
+ha_failback()
{
- local node=$1
- local end=$(($(date +%s) + 5 * 60))
+ local nodes=$1
+ ha_info "Failback resources on $nodes in $ha_failback_delay sec"
- ha_info "Waiting for $node to boot up"
- until pdsh -w $node -S hostname >/dev/null 2>&1 ||
- [ -e "$ha_stop_file" ] ||
- (($(date +%s) >= end)); do
- ha_sleep 1 >/dev/null
- done
+ ha_sleep $ha_failback_delay
+ [ "$ha_failback_cmd" ] ||
+ {
+ ha_info "No failback command set, skiping"
+ return 0
+ }
+
+ $ha_failback_cmd $nodes
}
ha_summarize()
ha_killer()
{
- local node
-
- while (($(date +%s) < ha_start_time + ha_expected_duration)) &&
- [ ! -e "$ha_stop_file" ]; do
- ha_info "---------------8<---------------"
-
- node=$(ha_aim)
-
- ha_info "Failing $node"
- ha_sleep $(ha_rand 10)
- ha_power_down $node
- ha_sleep 10
- ha_wait_loads || break
-
- if [ -e $ha_stop_file ]; then
- ha_power_up $node
- break
- fi
-
- ha_info "Bringing $node back"
- ha_sleep $(ha_rand 10)
- ha_power_up $node
- ha_wait_node $node
- #
- # Wait for the failback to start.
- #
- ha_sleep 60
- ha_wait_loads || break
-
- ha_sleep $(ha_rand 20)
-
- ha_nr_loops=$((ha_nr_loops + 1))
- ha_info "Loop $ha_nr_loops done"
- done
- ha_summarize
+ local nodes
+
+ while (($(date +%s) < ha_start_time + ha_expected_duration)) &&
+ [ ! -e "$ha_stop_file" ]; do
+ ha_info "---------------8<---------------"
+
+ $ha_workloads_only || nodes=$(ha_aim)
+
+ ha_info "Failing $nodes"
+ $ha_workloads_only && ha_info " is skipped: workload only..."
+
+ ha_sleep $(ha_rand $ha_max_failover_period)
+ $ha_workloads_only || ha_power_down $nodes
+ ha_sleep 10
+ ha_wait_loads || return
+
+ if [ -e $ha_stop_file ]; then
+ $ha_workloads_only || ha_power_up $nodes
+ break
+ fi
+
+ ha_info "Bringing $nodes back"
+ ha_sleep $(ha_rand 10)
+ $ha_workloads_only ||
+ {
+ ha_power_up $nodes
+ ha_wait_nodes $nodes
+ ha_failback $nodes
+ }
+
+ #
+ # Wait for the failback to start.
+ #
+ ha_sleep 60
+ ha_wait_loads || return
+
+ ha_sleep $(ha_rand 20)
+
+ ha_nr_loops=$((ha_nr_loops + 1))
+ ha_info "Loop $ha_nr_loops done"
+ done
+ ha_summarize
}
ha_main()
{
- ha_process_arguments "$@"
-
- trap ha_trap_exit EXIT
- mkdir "$ha_tmp_dir"
- ha_on ${ha_clients[0]} mkdir "$ha_test_dir"
-
- ha_start_loads
- if ha_wait_loads; then
- if $ha_workloads_only; then
- ha_sleep $((60 * 60))
- else
- ha_killer
- fi
- fi
- ha_dump_logs "${ha_clients[*]} ${ha_servers[*]}"
- ha_stop_loads
-
- if [ -e "$ha_fail_file" ]; then
- exit 1
- else
- exit 0
- fi
+ ha_process_arguments "$@"
+ ha_check_env
+
+ ha_log "${ha_clients[*]} ${ha_servers[*]}" \
+ "START: $0: $(date +%H:%M:%S' '%s)"
+ trap ha_trap_exit EXIT
+ mkdir "$ha_tmp_dir"
+ ha_on ${ha_clients[0]} mkdir "$ha_test_dir"
+ ha_on ${ha_clients[0]} " \
+ $LFS setstripe $ha_stripe_params $ha_test_dir"
+
+ ha_start_loads
+ ha_wait_loads
+
+ if $ha_workloads_dry_run; then
+ ha_sleep 5
+ else
+ ha_killer
+ ha_dump_logs "${ha_clients[*]} ${ha_servers[*]}"
+ fi
+
+ ha_stop_loads
+
+ if [ -e "$ha_fail_file" ]; then
+ exit 1
+ else
+ ha_log "${ha_clients[*]} ${ha_servers[*]}" \
+ "END: $0: $(date +%H:%M:%S' '%s)"
+ exit 0
+ fi
}
ha_main "$@"
git://git.whamcloud.com / fs / lustre-release.git / blobdiff