lustre/tests/large-scale.sh

   1 #!/bin/bash
   2 # -*- mode: Bash; tab-width: 4; indent-tabs-mode: t; -*-
   3 # vim:shiftwidth=4:softtabstop=4:tabstop=4:
   4
   5 set -e
   6
   7 # bug number:
   8 ALWAYS_EXCEPT="$LARGE_SCALE_EXCEPT"
   9
  10 SAVE_PWD=$PWD
  11 PTLDEBUG=${PTLDEBUG:--1}
  12 LUSTRE=${LUSTRE:-`dirname $0`/..}
  13 SETUP=${SETUP:-""}
  14 CLEANUP=${CLEANUP:-""}
  15 . $LUSTRE/tests/test-framework.sh
  16
  17 init_test_env $@
  18
  19 . ${CONFIG:=$LUSTRE/tests/cfg/$NAME.sh}
  20 init_logging
  21
  22 remote_mds_nodsh && skip "remote MDS with nodsh"
  23
  24 [ -z "$CLIENTS" ] && skip_env "$TESTSUITE: Need two or more clients"
  25 [ $CLIENTCOUNT -lt 2 ] &&
  26         skip_env "$TESTSUITE: Need 2+ clients, have only $CLIENTCOUNT"
  27
  28 [ "$SLOW" = "no" ] && EXCEPT_SLOW=""
  29
  30 MOUNT_2=""
  31 build_test_filter
  32
  33 check_and_setup_lustre
  34 rm -rf $DIR/[df][0-9]*
  35
  36 get_mpiuser_id $MPI_USER
  37 MPI_RUNAS=${MPI_RUNAS:-"runas -u $MPI_USER_UID -g $MPI_USER_GID"}
  38 $GSS_KRB5 && refresh_krb5_tgt $MPI_USER_UID $MPI_USER_GID $MPI_RUNAS
  39
  40 [ "$DAEMONFILE" ] && $LCTL debug_daemon start $DAEMONFILE $DAEMONSIZE
  41
  42 test_3a() {
  43     assert_env CLIENTS MDSRATE MPIRUN
  44
  45     local -a nodes=(${CLIENTS//,/ })
  46
  47     # INCREMENT is a number of clients
  48     # a half of clients by default
  49     increment=${INCREMENT:-$(( CLIENTCOUNT / 2 ))}
  50
  51     machinefile=${MACHINEFILE:-$TMP/$TESTSUITE.machines}
  52     local LOG=$TMP/${TESTSUITE}_$tfile
  53
  54     local var=${SINGLEMDS}_svc
  55     local procfile="*.${!var}.recovery_status"
  56     local iters=${ITERS:-3}
  57     local nfiles=${NFILES:-50000}
  58     local nthreads=${THREADS_PER_CLIENT:-3}
  59
  60     local IFree=$(inodes_available)
  61     [ $IFree -gt $nfiles ] || nfiles=$IFree
  62
  63     local dir=$DIR/d0.$TESTNAME
  64     mkdir -p $dir
  65     chmod 0777 $dir
  66
  67     local pid
  68     local list
  69     local -a res
  70
  71     local num=$increment
  72
  73         while [ $num -le $CLIENTCOUNT ]; do
  74                 list=$(comma_list ${nodes[@]:0:$num})
  75
  76                 generate_machine_file $list $machinefile ||
  77                         { error "can not generate machinefile"; exit 1; }
  78
  79                 for i in $(seq $iters); do
  80                         mdsrate_cleanup $num $machinefile $nfiles $dir 'f%%d' \
  81                                 --ignore
  82
  83                         COMMAND="${MDSRATE} --create --nfiles $nfiles --dir
  84                                  $dir --filefmt 'f%%d'"
  85                         mpi_run ${MACHINEFILE_OPTION} $machinefile \
  86                                 -np $((num * nthreads)) ${COMMAND} | tee ${LOG}&
  87
  88                         pid=$!
  89                         echo "pid=$pid"
  90
  91                         # 2 threads 100000 creates 117 secs
  92                         sleep 20
  93
  94                         log "$i : Starting failover on $SINGLEMDS"
  95                         facet_failover $SINGLEMDS
  96                         if ! wait_recovery_complete $SINGLEMDS \
  97                              $((TIMEOUT * 10)); then
  98                                 echo "$SINGLEMDS recovery is not completed!"
  99                                 kill -9 $pid
 100                                 exit 7
 101                         fi
 102
 103                         duration=$(do_facet $SINGLEMDS lctl get_param -n \
 104                                 $procfile | grep recovery_duration)
 105
 106                         res=( "${res[@]}" "$num" )
 107                         res=( "${res[@]}" "$duration" )
 108                         echo "RECOVERY TIME: NFILES=$nfiles number of clients: $num $duration"
 109                         wait $pid
 110                 done
 111                 num=$((num + increment))
 112         done
 113
 114     mdsrate_cleanup $num $machinefile $nfiles $dir 'f%%d' --ignore
 115
 116     i=0
 117     while [ $i -lt ${#res[@]} ]; do
 118         echo "RECOVERY TIME: NFILES=$nfiles number of clients: ${res[i]}  ${res[i+1]}"
 119         i=$((i+2))
 120     done
 121 }
 122
 123 run_test 3a "recovery time, $CLIENTCOUNT clients"
 124
 125 complete $SECONDS
 126 check_and_cleanup_lustre
 127 exit_status