Viewing: ha.sh
#!/usr/bin/bash
# -*- mode: Bash; tab-width: 4; indent-tabs-mode: t; -*-
# vim:shiftwidth=4:softtabstop=4:tabstop=4:
#
# NAME
#
# ha.sh - test Lustre HA (aka failover) configurations
#
# SYNOPSIS
#
# ha.sh [OPTIONS]
#
# DESCRIPTION
#
# ha.sh tests Lustre HA (aka failover) configurations with a CRM.
#
# OPTIONS
#
# -h
# Help.
#
# -c HOST[,...]
# Specify client nodes.
#
# -s HOST[,...]
# Specify server nodes.
#
# -v HOST[,...]
# Specify victim nodes to be rebooted.
#
# -d DIRECTORY
# Choose a parent of the test directory. "/mnt/lustre" if not specified.
#
# -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
#
# A Lustre file system is up and mounted on all client nodes. This script
# does not mount or unmount any Lustre targets or clients, let alone format
# anything.
#
# 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 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.
#
# MPI workloads can be run from several users. The list of users to use is
# configured by parameter ha_mpi_users, default is "mpiuser".
#
# PROCESS STRUCTURE AND IPC
#
# On the node where this script is run, the processes look like this:
#
# ~ ha.sh (ha_killer)
#
# ~ ha.sh (ha_repeat_mpi_load ior)
# ~ 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)
# ~ pdsh client2 dbench
# ~ ha.sh (ha_repeat_nonmpi_load client2 iozone)
# ~ pdsh client2 iozone
# ~ ha.sh (ha_repeat_nonmpi_load client5 iozone)
# ~ pdsh client5 iozone
# ~ ... (one for each non-MPI load on each client)
#
# Each tilde represents a process. Indentations imply parent-children
# relation.
#
# 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[$load]}
if [[ x$tag == xextraprog ]]; then
if ! $(which "$EXTRAPROG"); then
ha_error $tag $EXTRAPROG not found
exit 1
fi
fi
local bin=$(echo $tag | tr '[:lower:]' '[:upper:]')
if [ x${!bin} = x ]; then
ha_error ha_mpi_loads: ${ha_mpi_loads}, $bin is not set
exit 1
fi
done
}
ha_info()
{
local -a msg=("$@")
echo "$0: $(date +%H:%M:%S' '%s):" ${msg[@]}
}
ha_touch()
{
local date=$(date +%H:%M:%S' '%s)
[[ $1 =~ stop ]] &&
echo $date ${FUNCNAME[1]} $2 >> $ha_stop_file ||
true
[[ $1 =~ fail ]] &&
echo $date ${FUNCNAME[1]} $2 >> $ha_fail_file ||
true
[[ $1 =~ lfsck ]] &&
echo $date ${FUNCNAME[1]} $2 >> $ha_lfsck_stop ||
true
}
ha_recovery_status ()
{
local log
local -a nodes=(${ha_victims[*]} ${ha_victims_pair[*]})
local node
while [ ! -e "$ha_stop_file" ]; do
for ((i=0; i<${#nodes[@]}; i++)) {
node=${nodes[i]}
log=$ha_tmp_dir/${node}.recovery.status
local lock=${log}.lock
if [ ! -e $lock ]; then
ha_on $node \
"date; \
lctl get_param *.*.recovery_status" >>\
"$log" 2>&1 || true
fi
}
ha_sleep $ha_recovery_status_delay \
"recovery status each $ha_recovery_status_delay sec"
done
}
ha_log()
{
local nodes=${1// /,}
shift
ha_on $nodes "lctl mark $*"
}
declare -A ha_node_vmstat_pids
ha_start_vmstat_node()
{
local node=$1
local delay=$2
local log=$ha_vmstat_dir/${node}.vmstat
rm -f $ha_tmp_dir/${node}.vmstat.lock
local pid=$(ha_on $node "mkdir -p $ha_vmstat_dir; vmstat -t $delay >> \
$log 2>/dev/null </dev/null & echo \$!" | awk '{print $2}')
echo "VMSTAT started on $node PID: $pid, log: ${node}:$log"
ha_on $node ps aux | grep vmstat
ha_node_vmstat_pids[$node]=$pid
}
ha_start_vmstat()
{
local -a nodes=(${ha_victims[*]} ${ha_victims_pair[*]})
for ((i=0; i<${#nodes[@]}; i++)) {
ha_start_vmstat_node ${nodes[i]} $ha_vmstat_delay
}
}
ha_stop_vmstat()
{
local -a nodes=(${ha_victims[*]} ${ha_victims_pair[*]})
for ((i=0; i<${#nodes[@]}; i++)) {
node=${nodes[i]}
ha_info "Stopping vmstat on $node ... "
ha_on $node "ps aux | grep vmstat" || continue
local pid=${ha_node_vmstat_pids[$node]}
ha_on $node "kill -s TERM $pid; \
tail --pid=$pid -f /dev/null" || true
ha_info "Check is vmstat still running on $node ..."
ha_on $node "ps aux | grep vmstat" || true
}
}
ha_error()
{
ha_info "$@" >&2
}
ha_trap_err()
{
local i
ha_error "Trap ERR triggered by:"
ha_error " $BASH_COMMAND"
ha_error "Call trace:"
for ((i = 0; i < ${#FUNCNAME[@]}; i++)); do
ha_error " ${FUNCNAME[$i]} [${BASH_SOURCE[$i]}:${BASH_LINENO[$i]}]"
done
}
trap ha_trap_err ERR
set -eE
declare TMP=${TMP:-/tmp}
# Set equal to value if want to gather recovery_status info
# each "value" secs.
# 0 means "do not collect vmstat and recovery status info"
declare ha_recovery_status_delay=${RECOVERY_STATUS_DELAY:-0}
declare ha_recovery_status_pid
declare ha_vmstat_delay=${VMSTAT_DELAY:-0}
declare ha_power_down_pids
declare ha_test_subdir=$(basename $0)-$$
declare ha_tmp_dir=$TMP/$ha_test_subdir
# Useless to store vmstat results in /tmp because
# of no guarantee that files not disapeared when node crashed
declare ha_vmstat_dir=${VMSTATDIR:-$TMP}/$ha_test_subdir
declare ha_stop_file=$ha_tmp_dir/stop
declare ha_fail_file=$ha_tmp_dir/fail
declare ha_count_file=$ha_tmp_dir/count
declare ha_count_lock_file=$ha_tmp_dir/count.lock
declare ha_clients_free_file=$ha_tmp_dir/clients_free
declare ha_pm_states=$ha_tmp_dir/ha_pm_states
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_lfsck_log=$ha_tmp_dir/lfsck.log
declare ha_lfsck_lock=$ha_tmp_dir/lfsck.lock
declare ha_lfsck_stop=$ha_tmp_dir/lfsck.stop
declare ha_lfsck_bg=${LFSCK_BG:-false}
declare ha_lfsck_after=${LFSCK_AFTER:-false}
declare ha_lfsck_node=${LFSCK_NODE:-""}
declare ha_lfsck_device=${LFSCK_DEV:-""}
declare ha_lfsck_types=${LFSCK_TYPES:-"namespace layout"}
declare ha_lfsck_custom_params=${LFSCK_CUSTOM_PARAMS:-""}
declare ha_lfsck_wait=${LFSCK_WAIT:-1200}
declare ha_lfsck_fail_on_repaired=${LFSCK_FAIL_ON_REPAIRED:-false}
declare ha_power_down_cmd=${POWER_DOWN:-"pm -0"}
declare ha_power_up_cmd=${POWER_UP:-"pm -1"}
declare ha_reboot=${REBOOT:-"pm -r"}
declare ha_power_delay=${POWER_DELAY:-60}
declare ha_node_up_delay=${NODE_UP_DELAY:-10}
declare ha_wait_nodes_up=${WAIT_NODES_UP:-600}
declare ha_pm_host=${PM_HOST:-$(hostname)}
declare ha_failback_delay=${DELAY:-5}
declare ha_failback_cmd=${FAILBACK:-""}
declare ha_stripe_params=${STRIPEPARAMS:-"-c 0"}
declare ha_test_dir_stripe_count=${TDSTRIPECOUNT:-"1"}
declare ha_test_dir_mdt_index=${TDMDTINDEX:-"0"}
declare ha_test_dir_mdt_index_random=${TDMDTINDEXRAND:-false}
declare ha_dir_stripe_count=${DSTRIPECOUNT:-"1"}
declare ha_dir_stripe_count_random=${DSTRIPECOUNTRAND:-false}
declare ha_mdt_index=${MDTINDEX:-"0"}
declare ha_mdt_index_random=${MDTINDEXRAND:-false}
declare -a ha_clients
declare -a ha_servers
declare -a ha_victims
declare -a ha_victims_pair
declare ha_test_dir=/mnt/lustre/$(basename $0)-$$
declare -a ha_testdirs=(${ha_test_dirs="$ha_test_dir"})
# Run NLOOPS if set != 0; if 0 (default) the tags are
# executed defined duration for the test.
declare ha_nloops=${NLOOPS:=0}
for ((i=0; i<${#ha_testdirs[@]}; i++)); do
ha_testdirs[i]="${ha_testdirs[i]}/$(basename $0)-$$"
done
declare ha_dumplogs=${DUMPLOGS:-true}
declare ha_cleanup=${CLEANUP:-true}
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=${LOAD_TIMEOUT:-$((60 * 10))}
declare ha_workloads_only=false
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 -a ha_mpiusers=(${ha_mpi_users="mpiuser"})
declare -a ha_users
declare -A ha_mpiopts
for ((i=0; i<${#ha_mpiusers[@]}; i++)); do
u=${ha_mpiusers[i]%%:*}
o=""
# user gets empty option if ha_mpi_users does not specify it explicitly
[[ ${ha_mpiusers[i]} =~ : ]] && o=${ha_mpiusers[i]##*:}
ha_users[i]=$u
ha_mpiopts[$u]+=" $o"
done
ha_users=(${!ha_mpiopts[@]})
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_extraprog_params=${EXTRAPROGP:-'""'}
declare ha_mpirun_options=${MPIRUN_OPTIONS:-""}
declare ha_clients_stripe=${CLIENTSSTRIPE:-'"$STRIPEPARAMS"'}
declare ha_nclientsset=${NCLIENTSSET:-1}
declare ha_ninstmustfail=${NINSTMUSTFAIL:-0}
declare ha_racer_params=${RACERP:-"MDSCOUNT=1"}
eval ha_params_ior=($ha_ior_params)
eval ha_params_simul=($ha_simul_params)
eval ha_params_mdtest=($ha_mdtest_params)
eval ha_params_extraprog=($ha_extraprog_params)
eval ha_stripe_clients=($ha_clients_stripe)
declare ha_nparams_ior=${#ha_params_ior[@]}
declare ha_nparams_simul=${#ha_params_simul[@]}
declare ha_nparams_mdtest=${#ha_params_mdtest[@]}
declare ha_nparams_extraprog=${#ha_params_extraprog[@]}
declare ha_nstripe_clients=${#ha_stripe_clients[@]}
declare -A ha_mpi_load_cmds=(
[ior]="$IOR -o {}/f.ior {params}"
[simul]="$SIMUL {params} -d {}"
[mdtest]="$MDTEST {params} -d {}"
[extraprog]="$EXTRAPROG {params}"
)
declare racer=${RACER:-"$(dirname $0)/racer/racer.sh"}
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]="dd if=/dev/zero of={}/f.dd bs=1M count=256"
[tar]="tar cf - /etc | tar xf - -C {}"
[iozone]="iozone -a -e -+d -s $iozone_SIZE {}/f.iozone"
[racer]="$ha_racer_params $racer {}"
)
declare ha_check_attrs="find {} -type f -ls 2>&1 | grep -e '?'"
ha_usage()
{
ha_info "Usage: $0 -c HOST[,...] -s HOST[,...]" \
"-v HOST[,...] -f HOST[,...] [-d DIRECTORY] [-u SECONDS]"
}
ha_process_arguments()
{
local opt
while getopts hc:s:v:d:p:u:wrmf: opt; do
case $opt in
h)
ha_usage
exit 0
;;
c)
ha_clients=(${OPTARG//,/ })
;;
s)
ha_servers=(${OPTARG//,/ })
;;
v)
ha_victims=(${OPTARG//,/ })
;;
d)
ha_test_dir=$OPTARG/$(basename $0)-$$
;;
u)
ha_expected_duration=$OPTARG
;;
p)
ha_max_failover_period=$OPTARG
;;
w)
ha_workloads_only=true
;;
r)
ha_workloads_dry_run=true
;;
m)
ha_simultaneous=true
;;
f)
ha_victims_pair=(${OPTARG//,/ })
;;
\?)
ha_usage
exit 1
;;
esac
done
if [ -z "${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
#
# -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()
{
ha_touch stop
trap 0
if (( ha_vmstat_delay != 0 )); then
ha_stop_vmstat
fi
if (( ha_recovery_status_delay != 0 )); then
# the process $ha_recovery_status_pid
# could be completed by ha_stop_loads()->wait
wait $ha_recovery_status_pid || true
fi
if [ -e "$ha_fail_file" ]; then
ha_info "Test directories ${ha_testdirs[@]} not removed"
ha_info "Temporary directory $ha_tmp_dir not removed"
else
$ha_cleanup &&
ha_on ${ha_clients[0]} rm -rf ${ha_testdirs[@]} ||
ha_info "Test directories ${ha_testdirs[@]} not removed"
ha_info "Please find the results in the directory $ha_tmp_dir"
fi
}
ha_trap_stop_signals()
{
ha_info "${ha_stop_signals// /,} received"
ha_touch stop "${ha_stop_signals// /,} received"
}
ha_sleep()
{
local n=$1
local reason=$2
[[ -n $reason ]] &&
reason=", Reason: $reason"
ha_info "Sleeping for ${n}s$reason"
#
# sleep(1) could interrupted.
#
sleep $n || true
}
ha_wait_unlock()
{
local lock=$1
while [ -e $lock ]; do
sleep 1
done
}
ha_lock()
{
local lock=$1
until mkdir "$lock" >/dev/null 2>&1; do
ha_sleep 1 >/dev/null
done
}
ha_unlock()
{
local lock=$1
rm -r "$lock"
}
ha_dump_logs()
{
local nodes=${1// /,}
local file=${ha_tmp_dir}-$(date +%s).dk
local lock=$ha_tmp_dir/lock-dump-logs
local rc=0
$ha_dumplogs ||
{ echo "Requested to skip the logs dumping"; return 0; }
ha_lock "$lock"
ha_info "Dumping lctl log to $file"
#
# some nodes could crash, so
# do not exit with error if not all logs are dumped
#
ha_on $nodes "lctl dk >>$file" || rc=$?
[ $rc -eq 0 ] ||
ha_error "not all logs are dumped! Some nodes are unreachable."
ha_unlock "$lock"
}
ha_repeat_mpi_load()
{
local client=$1
local load=$2
local status=$3
local parameter=$4
local machines=$5
local stripeparams=$6
local mpiuser=$7
local mustpass=$8
local mpirunoptions=$9
local test_dir=${10}
local tag=${ha_mpi_load_tags[$load]}
local cmdrun=${ha_mpi_load_cmds[$tag]}
local dir
local log
local rc=0
local rccheck=0
local rcprepostcmd=0
local nr_loops=0
local avg_loop_time=0
local start_time=$(date +%s)
local check_attrs=${ha_check_attrs//"{}"/$dir}
cmdrun=${cmdrun//"{params}"/$parameter}
machines="-machinefile $machines"
while [ ! -e "$ha_stop_file" ] && ((rc == 0)) && ((rccheck == 0)) &&
( ((nr_loops < ${11})) || ((${11} == 0)) ); do
rcprepostcmd=0
local cur=$client-$tag-$nr_loops-$mpiuser
local log=$ha_tmp_dir/$cur
dir=$test_dir/$cur
cmd=${cmdrun//"{}"/$dir}
ha_info "$client Starts: $mpiuser: $cmd \
LOOP $nr_loops (from: ${11})" 2>&1 | \
tee -a $log
{
local mdt_index
if $ha_mdt_index_random && [ $ha_mdt_index -ne 0 ]; then
mdt_index=$(ha_rand $((ha_mdt_index + 1)) )
else
mdt_index=$ha_mdt_index
fi
local dir_stripe_count
if $ha_dir_stripe_count_random &&
[ $ha_dir_stripe_count -ne 1 ]; then
dir_stripe_count=$(($(ha_rand $ha_dir_stripe_count) + 1))
else
dir_stripe_count=$ha_dir_stripe_count
fi
if [[ -n "$ha_precmd" ]]; then
local precmd=${ha_precmd//"{}"/$dir}
ha_info "precmd: $precmd"
ha_on $client "$precmd" >>"$log" 2>&1 ||
rcprepostcmd=$?
ha_info "rcprepostcmd: $rcprepostcmd"
if (( rcprepostcmd != 0 )); then
ha_touch stop,fail $cur
ha_dump_logs "${ha_clients[*]} ${ha_servers[*]}"
(( nr_loops+=1 ))
continue
fi
fi
ha_info "$client Creates $dir with -i$mdt_index \
-c$dir_stripe_count ; stripeparams: $stripeparams "
ha_on $client $LFS mkdir -i$mdt_index -c$dir_stripe_count "$dir" &&
ha_on $client $LFS getdirstripe "$dir" &&
ha_on $client $LFS setstripe $stripeparams $dir &&
ha_on $client $LFS getstripe $dir &&
ha_on $client chmod a+xwr $dir &&
ha_on $client "su $mpiuser bash -c \" $mpirun $mpirunoptions \
-np $((${#ha_clients[@]} * mpi_threads_per_client / ha_nclientsset)) \
$machines $cmd \" " || rc=$?
ha_on ${ha_clients[0]} "$check_attrs && \
$LFS df $dir && \
$check_attrs " && rccheck=1
if [[ -n "$ha_postcmd" ]]; then
local postcmd=${ha_postcmd//"{}"/$dir}
ha_info "$postcmd"
ha_on $client "$postcmd" >>"$log" 2>&1 ||
rcprepostcmd=$?
if (( rcprepostcmd != 0 )); then
ha_touch stop,fail $cur
ha_dump_logs "${ha_clients[*]} ${ha_servers[*]}"
fi
fi
if (( ((rc == 0)) && ((rccheck == 0)) && \
(( mustpass != 0 )) )) ||
(( ((rc != 0)) && ((rccheck == 0)) && \
(( mustpass == 0 )) )); then
$ha_cleanup && ha_on $client rm -rf "$dir" ||
ha_on $client mv "$dir" "${dir}.bak"
fi;
} >>"$log" 2>&1
ha_info $client: rccheck=$rccheck mustpass=$mustpass \
nr_loops=$nr_loops
# mustpass=0 means that failure is expected
if (( rccheck != 0 )); then
ha_touch stop,fail $cur
ha_dump_logs "${ha_clients[*]} ${ha_servers[*]}"
elif (( rc !=0 )); then
if (( mustpass != 0 )); then
ha_touch stop,fail $cur
ha_dump_logs "${ha_clients[*]} ${ha_servers[*]}"
else
# Ok to fail
rc=0
fi
elif (( mustpass == 0 )); then
ha_touch stop,fail $cur
ha_dump_logs "${ha_clients[*]} ${ha_servers[*]}"
fi
echo rc=$rc rccheck=$rccheck mustpass=$mustpass >"$status"
(( nr_loops+=1 ))
done
local stop_found="No stop file found"
[[ -e "$ha_stop_file" ]] &&
stop_found="$ha_stop_file found"
[ $nr_loops -ne 0 ] &&
avg_loop_time=$((($(date +%s) - start_time) / nr_loops))
ha_info "$client $tag ended: $stop_found: rc=$rc mustpass=$mustpass \
rcprepostcmd=$rcprepostcmd \
nr_loops=$nr_loops (from ${11}) \
avg loop time $avg_loop_time"
if (( ha_nloops != 0 )); then
flock $ha_count_lock_file sh -c \
"awk -i inplace -v inc=$nr_loops \
'{print \$1-inc'} $ha_count_file"
local count=$(cat $ha_count_file)
(( count <= 0 )) &&
ha_touch stop || true
fi
echo $client >> $ha_clients_free_file
}
remove_client_from_list()
{
sed -i "/^$1$/d" $ha_clients_free_file 2>&1
ha_info "FREE clients: $(echo $(cat $ha_clients_free_file))"
}
wait_clients_free()
{
local -a clients_free=($(cat $ha_clients_free_file))
ha_info "Waiting any clients free"
while [ ! -e "$ha_stop_file" ] &&
(( ${#clients_free[@]} == 0 )); do
sleep 60
clients_free=($(cat $ha_clients_free_file))
done
}
ha_start_mpi_loads()
{
local client
local load
local tag
local status
local n
local nparam
local machines
local m
local -a mach
local mpiuser
local nmpi
local inst
# Define names for machinefiles for each client set
for (( n=0; n < $ha_nclientsset; n++ )); do
mach[$n]=$ha_machine_file$n
done
for ((n = 0; n < ${#ha_clients[@]}; n++)); do
m=$(( n % ha_nclientsset))
machines=${mach[m]}
ha_info machine_file=$machines
echo ${ha_clients[n]} >> $machines
done
local dirname=$(dirname $ha_machine_file)
for client in ${ha_clients[@]}; do
ha_on $client mkdir -p $dirname
scp $ha_machine_file* $client:$dirname
done
# ha_mpi_instances defines the number of
# clients start mpi loads; should be <= ${#ha_clients[@]} if
# NLOOPS not set. In case NLOOPS=value -- each instance executed
# "value" times and then the remaining not run yet instances started.
#
# do nothing if
# ha_mpi_instances = 0
# or
# ${#ha_mpi_load_tags[@]} =0
(( ha_mpi_instances == 0 )) || (( ${#ha_mpi_load_tags[@]} == 0 )) &&
ha_info "no mpi load to start" &&
return 0
# We do not start next instance on client already running the
# mpi intstance, we need to wait some started instances completed
# before start the remaining instances, i.e. we reuse the clients completed the
# instance.
# The list of free clients is stored in ha_clients_free_file.
local inststarted=0
while (( ha_mpi_instances > 0 )); do
wait_clients_free
local -a clients_free=($(cat $ha_clients_free_file))
inst=$ha_mpi_instances
(( inst <= ${#clients_free[@]} )) ||
inst=${#clients_free[@]}
local ndir
for ((n = 0; n < $inst; n++)); do
client=${clients_free[n]}
local k=$(( n + inststarted ))
nmpi=$(( k % ${#ha_users[@]}))
mpiuser=${ha_users[nmpi]}
# if clients_free has one client only
# the tests always will be run on ha_testdirs[0]
ndir=$((k % ${#ha_testdirs[@]}))
test_dir=${ha_testdirs[ndir]}
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_$tag
# the loads could have no params
(( num == 0 )) && num=1
nparam=$((k % num))
local aref=ha_params_$tag[nparam]
local parameter=${!aref}
local nstripe=$((k % ha_nstripe_clients))
aref=ha_stripe_clients[nstripe]
local stripe=${!aref}
local m=$(( k % ha_nclientsset))
machines=${mach[m]}
local mustpass=1
[[ $ha_ninstmustfail == 0 ]] ||
mustpass=$(( k % ha_ninstmustfail ))
ha_info "$client going to start and repeat tag $tag \
$ha_nloops loops \
instance: $(( inststarted + n )) \
(remaining $(( ha_mpi_instances - n - 1 )))"
ha_repeat_mpi_load $client $load $status "$parameter" \
$machines "$stripe" "$mpiuser" "$mustpass" \
"${ha_mpiopts[$mpiuser]} $ha_mpirun_options" \
"$test_dir" $ha_nloops &
ha_status_files+=("$status")
# get rid of duplicated status files
ha_status_files=($(echo ${ha_status_files[@]} | \
tr ' ' '\n' | sort -u ))
# remove client from the free client list
remove_client_from_list $client
ha_sleep 2
done
done
(( inststarted+=inst ))
# true needs to avoid immediately exit in case
# ha_mpi_instances = 0
(( ha_mpi_instances-=inst )) || true
done
}
ha_repeat_nonmpi_load()
{
local client=$1
local load=$2
local status=$3
local tag=${ha_nonmpi_load_tags[$load]}
local cmd=${ha_nonmpi_load_cmds[$tag]}
local test_dir=$4
local dir=$test_dir/$client-$tag
local log=$ha_tmp_dir/$client-$tag
local rc=0
local rccheck=0
local nr_loops=0
local avg_loop_time=0
local start_time=$(date +%s)
local check_attrs=${ha_check_attrs//"{}"/$dir}
cmd=${cmd//"{}"/$dir}
ha_info "Starting $tag on $client on $dir"
while [ ! -e "$ha_stop_file" ] && ((rc == 0)); do
ha_info "$client Starts: $cmd" 2>&1 | tee -a $log
ha_on $client "mkdir -p $dir && \
$cmd" >>"$log" 2>&1 || rc=$?
ha_on $client "$check_attrs && \
$LFS df $dir && \
$check_attrs " >>"$log" 2>&1 && rccheck=1 ||
ha_on $client "rm -rf $dir" >>"$log" 2>&1
ha_info rc=$rc rccheck=$rccheck
if (( (rc + rccheck) != 0 )); then
ha_dump_logs "${ha_clients[*]} ${ha_servers[*]}"
ha_touch stop,fail $client,$tag
fi
echo $rc >"$status"
nr_loops=$((nr_loops + 1))
done
[ $nr_loops -ne 0 ] &&
avg_loop_time=$((($(date +%s) - start_time) / nr_loops))
ha_info "$tag on $client stopped: rc $rc avg loop time ${avg_loop_time}s"
}
ha_start_nonmpi_loads()
{
local client
local load
local tag
local status
local n
local test_dir
local ndir
for (( n = 0; n < ${#ha_clients[@]}; n++)); do
client=${ha_clients[n]}
ndir=$((n % ${#ha_testdirs[@]}))
test_dir=${ha_testdirs[ndir]}
for ((load = 0; load < ${#ha_nonmpi_load_tags[@]}; load++)); do
tag=${ha_nonmpi_load_tags[$load]}
status=$ha_status_file_prefix-$tag-$client
ha_repeat_nonmpi_load $client $load $status $test_dir &
ha_status_files+=("$status")
done
done
}
declare ha_bgcmd=${ha_bgcmd:-""}
declare ha_bgcmd_log=$ha_tmp_dir/bgcmdlog
ha_cmd_bg () {
[[ -z "$ha_bgcmd" ]] && return 0
for ((i=0; i<${#ha_testdirs[@]}; i++)); do
ha_bgcmd=${ha_bgcmd//"{}"/${ha_testdirs[i]}}
done
ha_info "BG cmd: $ha_bgcmd"
while [ true ]; do
[ -f $ha_stop_file ] &&
ha_info "$ha_stop_file found! $ha_bgcmd no started" &&
break
eval $ha_bgcmd 2>&1 | tee -a $ha_bgcmd_log
sleep 1
done &
CMD_BG_PID=$!
ha_info CMD BG PID: $CMD_BG_PID
ps aux | grep $CMD_BG_PID
}
ha_lfsck_bg () {
rm -f $ha_lfsck_log
rm -f $ha_lfsck_stop
ha_info "LFSCK BG"
while [ true ]; do
[ -f $ha_lfsck_stop ] && ha_info "LFSCK stopped" && break
[ -f $ha_stop_file ] &&
ha_info "$ha_stop_file found! LFSCK not started" &&
break
ha_start_lfsck 2>&1 | tee -a $ha_lfsck_log
sleep 1
done &
LFSCK_BG_PID=$!
ha_info LFSCK BG PID: $LFSCK_BG_PID
}
ha_wait_lfsck_completed () {
local -a status
local -a types=($ha_lfsck_types)
local type
local s
local nodes="${ha_servers[@]}"
nodes=${nodes// /,}
# -A start LFSCK on all nodes
# -t default all
[ ${#types[@]} -eq 0 ] && types=(namespace layout)
ha_info "Waiting LFSCK completed in $ha_lfsck_wait sec: types ${types[@]}"
for type in ${types[@]}; do
eval var_$type=0
for (( i=0; i<=ha_lfsck_wait; i++)); do
status=($(ha_on $nodes lctl get_param -n *.*.lfsck_$type 2>/dev/null | \
awk '/status/ { print $3 }'))
for (( s=0; s<${#status[@]}; s++ )); do
# "partial" is expected after HARD failover
[[ "${status[s]}" = "completed" ]] ||
[[ "${status[s]}" = "partial" ]] || break
done
[[ $s -eq ${#status[@]} ]] && eval var_$type=1 && break
sleep 1
done
ha_info "LFSCK $type status in $i sec:"
ha_on $nodes lctl get_param -n *.*.lfsck_$type 2>/dev/null | grep status
done
for type in ${types[@]}; do
local var=var_$type
ha_on $nodes lctl get_param -n *.*.lfsck_$type 2>/dev/null
[[ ${!var} -eq 1 ]] ||
{ ha_info "lfsck not completed in $ha_lfsck_wait sec";
return 1; }
done
return 0
}
ha_start_lfsck()
{
local -a types=($ha_lfsck_types)
local rc=0
# -A: start LFSCK on all nodes via the specified MDT device
# (see "-M" option) by single LFSCK command
local params=" -A -r $ha_lfsck_custom_params"
# use specified device if set
[ -n "$ha_lfsck_device" ] && params="-M $ha_lfsck_device $params"
# -t: check type(s) to be performed (default all)
# check only specified types if set
if [ ${#types[@]} -ne 0 ]; then
local type="${types[@]}"
params="$params -t ${type// /,}"
fi
ha_info "LFSCK start $params"
ha_on $ha_lfsck_node "lctl lfsck_start $params" || rc=1
if [ $rc -ne 0 ]; then
if [ -e $ha_lfsck_lock ]; then
rc=0
ha_wait_unlock $ha_lfsck_lock
ha_sleep 120 "before lfsck restarting"
ha_on $ha_lfsck_node "lctl lfsck_start $params" || rc=1
fi
fi
[ $rc -eq 0 ] ||
{ ha_touch stop,fail,lfsck; return 1; }
ha_wait_lfsck_completed ||
{ ha_touch stop,fail,lfsck; return 1; }
return 0
}
ha_lfsck_repaired()
{
local n=0
n=$(cat $ha_lfsck_log | awk '/repaired/ {print $3}' |\
awk '{sum += $1} END { print sum }')
(( n == 0 )) ||
{ ha_info "Total repaired: $n";
ha_touch fail; return 1; }
return 0
}
ha_start_loads()
{
ha_cmd_bg
$ha_lfsck_bg && ha_lfsck_bg
trap ha_trap_stop_signals $ha_stop_signals
(( ha_nloops != 0 )) &&
echo $(( ha_mpi_instances * ha_nloops )) > $ha_count_file ||
true
echo ${ha_clients[@]} | sed "s/ /\n/g" > $ha_clients_free_file
ha_start_nonmpi_loads
ha_start_mpi_loads
}
ha_stop_loads()
{
ha_touch stop
[[ -n $CMD_BG_PID ]] && wait $CMD_BG_PID || true
# true because of lfsck_bg could be stopped already
$ha_lfsck_bg && wait $LFSCK_BG_PID || true
trap - $ha_stop_signals
ha_info "Waiting for workloads to stop"
wait
}
ha_wait_loads()
{
local file
local end=$(($(date +%s) + ha_load_timeout))
ha_info "Waiting $ha_load_timeout sec for workload status..."
rm -f "${ha_status_files[@]}"
#
# 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_powermanage()
{
local nodes=$1
local expected_state=$2
local state
local -a states
local i
local rc=0
# store pm -x -q $nodes results in a file to have
# more information about nodes statuses
ha_on $ha_pm_host pm -x -q $nodes | awk '{print $2 $3}' > $ha_pm_states
rc=${PIPESTATUS[0]}
echo pmrc=$rc
while IFS=": " read node state; do
[[ "$state" = "$expected_state" ]] && {
nodes=${nodes/$node/}
nodes=${nodes//,,/,}
nodes=${nodes/#,}
nodes=${nodes/%,}
}
done < $ha_pm_states
if [ -n "$nodes" ]; then
cat $ha_pm_states
return 1
fi
return 0
}
ha_power_down_cmd_fn()
{
local nodes=$1
local cmd
local pid
local rc=0
case $ha_power_down_cmd in
# format is: POWER_DOWN=sysrqcrash
sysrqcrash)
cmd="pdsh -S -w $nodes -u 120 \"echo c > /proc/sysrq-trigger\" &"
# stop grab recovery status on crashed nodes
if (( ha_recovery_status_delay != 0 )); then
for n in ${nodes//,/ }; do
touch $ha_tmp_dir/${n}.recovery.status.lock
echo $(date) \
"recovery status collection is paused: \
$n is going to power down" >> \
$ha_tmp_dir/${n}.recovery.status
done
fi
# restart vmstat after node back
if (( ha_vmstat_delay != 0 )); then
for n in ${nodes//,/ }; do
touch $ha_tmp_dir/${n}.vmstat.lock
done
fi
eval $cmd
pid=$!
ha_power_down_pids=$(echo $ha_power_down_pids $pid)
ha_info "ha_power_down_pids: $ha_power_down_pids"
[[ -z "$ha_power_down_pids" ]] ||
ps aux | grep " ${ha_power_down_pids// / \| } " ||
true
;;
*)
cmd="$ha_power_down_cmd $nodes"
eval $cmd
rc=$?
;;
esac
return $rc
}
ha_power_down()
{
local nodes=$1
local rc=1
local i
local state
case $ha_power_down_cmd in
*pm*) state=off ;;
sysrqcrash) state=off ;;
*) state=on;;
esac
if $ha_lfsck_bg && [[ ${nodes//,/ /} =~ $ha_lfsck_node ]]; then
ha_info "$ha_lfsck_node down, delay start LFSCK"
ha_lock $ha_lfsck_lock
fi
ha_info "Powering down $nodes : cmd: $ha_power_down_cmd"
ha_power_down_pids=""
for (( i=0; i<10; i++ )) {
ha_info "attempt: $i"
ha_power_down_cmd_fn $nodes || rc=1
ha_sleep $ha_power_delay "delay node status check after powerdown ..."
ha_powermanage $nodes $state && rc=0 && break
}
if [[ -n "$ha_power_down_pids" ]]; then
kill -9 $ha_power_down_pids || true
wait $ha_power_down_pids || true
fi
[ $rc -eq 0 ] || {
ha_info "Failed Powering down in $i attempts:" \
"$ha_power_down_cmd"
cat $ha_pm_states
exit 1
}
}
ha_get_pair()
{
local node=$1
local i
for ((i=0; i<${#ha_victims[@]}; i++)) {
[[ ${ha_victims[i]} == $node ]] && echo ${ha_victims_pair[i]} &&
return
}
[[ $i -ne ${#ha_victims[@]} ]] ||
ha_error "No pair found!"
}
ha_power_up_delay()
{
local nodes=$1
local end=$(($(date +%s) + ha_node_up_delay))
local rc
if [[ ${#ha_victims_pair[@]} -eq 0 ]]; then
ha_sleep $ha_node_up_delay "before node power up"
return 0
fi
# Check CRM status on failover pair
while (($(date +%s) <= end)); do
rc=0
for n in ${nodes//,/ }; do
local pair=$(ha_get_pair $n)
local status=$(ha_on $pair crm_mon -1rQ | \
grep -w $n | head -2)
ha_info "$n pair: $pair status: $status"
[[ "$status" == *OFFLINE* ]] ||
rc=$((rc + $?))
ha_info "rc: $rc"
done
if [[ $rc -eq 0 ]]; then
ha_info "CRM: Got all victims status OFFLINE"
return 0
fi
sleep 60
done
ha_info "$nodes CRM status not OFFLINE"
for n in ${nodes//,/ }; do
local pair=$(ha_get_pair $n)
ha_info "CRM --- $n"
ha_on $pair crm_mon -1rQ
done
ha_error "CRM: some of $nodes are not OFFLINE in $ha_node_up_delay sec"
exit 1
}
ha_power_up()
{
local nodes=$1
local rc=1
local i
ha_power_up_delay $nodes
ha_info "Powering up $nodes : cmd: $ha_power_up_cmd"
for (( i=0; i<10; i++ )) {
ha_info "attempt: $i"
$ha_power_up_cmd $nodes &&
ha_powermanage $nodes on && rc=0 && break
sleep $ha_power_delay
}
[ $rc -eq 0 ] || {
ha_info "Failed Powering up in $i attempts: $ha_power_up_cmd"
cat $ha_pm_states
exit 1
}
}
#
# rand MAX
#
# Print a random integer within [0, MAX).
#
ha_rand()
{
local max=$1
#
# See "5.2 Bash Variables" from "info bash".
#
echo -n $((RANDOM * max / 32768))
}
ha_aim()
{
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
}
ha_wait_nodes()
{
local nodes=$1
local end=$(($(date +%s) + $ha_wait_nodes_up))
local attempts=5
for ((i=1; i<=attempts; i++)); do
ha_info "Waiting for $nodes to boot up in \
$ha_wait_nodes_up attempt: $i"
until ha_on $nodes hostname >/dev/null 2>&1 ||
[ -e "$ha_stop_file" ] ||
(($(date +%s) >= end)); do
ha_sleep 1 >/dev/null
done
ha_info "Check where we are ..."
[ -e "$ha_stop_file" ] &&
ha_info "$ha_stop_file found!"
local -a nodes_up
local -a nodes_down
nodes_up=($(ha_on $nodes hostname | awk '{ print $2 }'))
ha_info "Nodes $nodes are up: ${nodes_up[@]}"
local -a n=(${nodes//,/ })
if [[ ${#nodes_up[@]} -ne ${#n[@]} ]]; then
nodes_down=($(echo ${n[@]} ${nodes_up[@]} |\
tr ' ' '\n' | sort | uniq -u))
ha_info "Failed boot up ${nodes_down[@]} in \
$ha_wait_nodes_up sec! attempt: $i"
if (( i == attempts )); then
ha_touch fail,stop
return 1
else
local down=${nodes_down[@]}
down=${down// /,/}
ha_info "REBOOTING $ha_reboot $down \
attempt: $i"
local cmd="$ha_reboot $down"
end=$(($(date +%s) + $ha_wait_nodes_up))
eval $cmd
continue
fi
else
break
fi
done
return 0
}
ha_failback()
{
local nodes=$1
local rc=1
local attempts=5
local i
for ((i=0; i<attempts; i++)); do
ha_info "Failback resources on $nodes in \
$ha_failback_delay sec, attempt: $i ($attempts); \
cmd: $ha_failback_cmd $nodes"
ha_sleep $ha_failback_delay "delay before failback"
[ "$ha_failback_cmd" ] ||
{
ha_info "No failback command set, skiping"
return 0
}
if $ha_failback_cmd $nodes ; then
rc=0
ha_info "Failback succesfully started: attempt: $i"
for n in ${nodes//,/ }; do
if (( ha_recovery_status_delay != 0 )); then
local lock=$ha_tmp_dir/${n}.recovery.status.lock
ls -al $lock
rm -f $lock
echo $(date) \
"recovery status collection is \
resumed" >> \
$ha_tmp_dir/${n}.recovery.status
fi
lock=$ha_tmp_dir/${n}.vmstat.lock
if (( ha_vmstat_delay != 0 )) && [[ -e $lock ]]; then
ha_start_vmstat_node $n $ha_vmstat_delay
fi
done
break
fi
done
[ -e $ha_lfsck_lock ] && ha_unlock $ha_lfsck_lock || true
ha_info "Failback status in $i attempt \
with $ha_failback_delay delay each: rc=$rc"
return $rc
}
ha_summarize()
{
ha_info "---------------8<---------------"
ha_info "Summary:"
ha_info " Duration: $(($(date +%s) - $ha_start_time))s"
ha_info " Loops: $ha_nr_loops"
}
ha_killer()
{
local nodes
while (($(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 $ha_max_failover_period) \
"random of max failover set ($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 $ha_nr_loops done"
done
ha_summarize
}
ha_run_info() {
local -a vars=($@)
local i
echo "****** Run Information ****"
for ((i=0; i<${#vars[@]}; i++)); do
local v=${vars[i]}
local -n aref=$v
echo $v: "${aref[@]}"
done
echo "***************************"
}
ha_main()
{
ha_process_arguments "$@"
ha_check_env
ha_log "${ha_clients[*]} ${ha_servers[*]}" \
"START: $0: $(date +%H:%M:%S' '%s)"
trap ha_trap_exit EXIT
ha_run_info ha_expected_duration ha_precmd ha_postcmd \
ha_mpi_instances ha_nloops \
EXTRAPROG EXTRAPROGP \
ha_testdirs ha_clients
mkdir "$ha_tmp_dir"
local mdt_index
if $ha_test_dir_mdt_index_random &&
[ $ha_test_dir_mdt_index -ne 0 ]; then
mdt_index=$(ha_rand $((ha_test_dir_mdt_index + 1)) )
else
mdt_index=$ha_test_dir_mdt_index
fi
local dir
test_dir=${ha_testdirs[0]}
ha_on ${ha_clients[0]} "$LFS mkdir -i$mdt_index \
-c$ha_test_dir_stripe_count $test_dir"
for ((i=0; i<${#ha_testdirs[@]}; i++)); do
test_dir=${ha_testdirs[i]}
ha_on ${ha_clients[0]} $LFS getdirstripe $test_dir
ha_on ${ha_clients[0]} " \
$LFS setstripe $ha_stripe_params $test_dir"
done
if (( ha_recovery_status_delay != 0 )); then
ha_info "Dumping recovery status info \
each $ha_recovery_status_delay sec"
ha_recovery_status &
ha_recovery_status_pid=$!
fi
if (( ha_vmstat_delay != 0 )); then
ha_info "Starting vmstat with delay $ha_vmstat_delay"
ha_start_vmstat
fi
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
$ha_lfsck_after && ha_start_lfsck | tee -a $ha_lfsck_log
$ha_lfsck_fail_on_repaired && ha_lfsck_repaired
if [ -e "$ha_fail_file" ]; then
exit 1
else
ha_log "${ha_clients[*]} ${ha_servers[*]}" \
"END: $0: $(date +%H:%M:%S' '%s)"
exit 0
fi
}
ha_main "$@"
