ceph/src/test/test_lost.sh

   1 #!/bin/bash -x
   2
   3 #
   4 # Test the lost object logic
   5 #
   6
   7 # Includes
   8 source "`dirname $0`/test_common.sh"
   9
  10 TEST_POOL=rbd
  11
  12 # Functions
  13 setup() {
  14         export CEPH_NUM_OSD=$1
  15         vstart_config=$2
  16
  17         # Start ceph
  18         ./stop.sh
  19
  20         # set recovery start to a really long time to ensure that we don't start recovery
  21         ./vstart.sh -d -n -o "$vstart_config" || die "vstart failed"
  22
  23         # for exiting pools set size not greater than number of OSDs,
  24         # so recovery from degraded ps is possible
  25         local changed=0
  26         for pool in `./ceph osd pool ls`; do
  27             local size=`./ceph osd pool get ${pool} size | awk '{print $2}'`
  28             if [ "${size}" -gt "${CEPH_NUM_OSD}" ]; then
  29                 ./ceph osd pool set ${pool} size ${CEPH_NUM_OSD}
  30                 changed=1
  31             fi
  32         done
  33         if [ ${changed} -eq 1 ]; then
  34             # XXX: When a pool has degraded pgs due to size greater than number
  35             # of OSDs, after decreasing the size the recovery still could stuck
  36             # and requires an additional kick.
  37             ./ceph osd out 0
  38             ./ceph osd in 0
  39         fi
  40
  41         poll_cmd "./ceph health" HEALTH_OK 1 30
  42 }
  43
  44 recovery1_impl() {
  45         # Write lots and lots of objects
  46         write_objects 1 1 200 4000 $TEST_POOL
  47
  48         # Take down osd1
  49         stop_osd 1
  50
  51         # Continue writing a lot of objects
  52         write_objects 2 2 200 4000 $TEST_POOL
  53
  54         # Bring up osd1
  55         restart_osd 1
  56
  57         # Finish peering.
  58         sleep 15
  59
  60         # Stop osd0.
  61         # At this point we have peered, but *NOT* recovered.
  62         # Objects should be lost.
  63         stop_osd 0
  64
  65         poll_cmd "./ceph pg debug degraded_pgs_exist" TRUE 3 120
  66         [ $? -eq 1 ] || die "Failed to see degraded PGs."
  67         poll_cmd "./ceph pg debug unfound_objects_exist" TRUE 3 120
  68         [ $? -eq 1 ] || die "Failed to see unfound objects."
  69         echo "Got unfound objects."
  70
  71         restart_osd 0
  72         sleep 20
  73         start_recovery 2
  74
  75         # Turn on recovery and wait for it to complete.
  76         poll_cmd "./ceph pg debug unfound_objects_exist" FALSE 3 120
  77         [ $? -eq 1 ] || die "Failed to recover unfound objects."
  78         poll_cmd "./ceph pg debug degraded_pgs_exist" FALSE 3 120
  79         [ $? -eq 1 ] || die "Recovery never finished."
  80 }
  81
  82 recovery1() {
  83         setup 2 'osd recovery delay start = 10000'
  84         recovery1_impl
  85 }
  86
  87 lost1_impl() {
  88         local flags="$@"
  89         local lost_action=delete
  90         local pgs_unfound pg
  91
  92         if is_set revert_lost $flags; then
  93             lost_action=revert
  94         fi
  95
  96         # Write lots and lots of objects
  97         write_objects 1 1 20 8000 $TEST_POOL
  98
  99         # Take down osd1
 100         stop_osd 1
 101
 102         # Continue writing a lot of objects
 103         write_objects 2 2 20 8000 $TEST_POOL
 104
 105         # Bring up osd1
 106         restart_osd 1
 107
 108         # Finish peering.
 109         sleep 15
 110
 111         # Stop osd0.
 112         # At this point we have peered, but *NOT* recovered.
 113         # Objects should be lost.
 114         stop_osd 0
 115
 116         # Since recovery can't proceed, stuff should be unfound.
 117         poll_cmd "./ceph pg debug unfound_objects_exist" TRUE 3 120
 118         [ $? -eq 1 ] || die "Failed to see unfound objects."
 119
 120         pgs_unfound=`./ceph health detail |awk '$1 = "pg" && /[0-9] unfound$/ {print $2}'`
 121
 122         [ -n "$pgs_unfound" ] || die "no pg with unfound objects"
 123
 124         for pg in $pgs_unfound; do
 125             ./ceph pg $pg mark_unfound_lost revert &&
 126             die "mark_unfound_lost unexpectedly succeeded for pg $pg"
 127         done
 128
 129         if ! is_set mark_osd_lost $flags && ! is_set rm_osd $flags; then
 130             return
 131         fi
 132
 133         if is_set try_to_fetch_unfound $flags; then
 134           # Ask for an object while it's still unfound, and
 135           # verify we get woken to an error when it's declared lost.
 136           echo "trying to get one of the unfound objects"
 137           (
 138           ./rados -c ./ceph.conf -p $TEST_POOL get obj02 $TEMPDIR/obj02 &&\
 139             die "expected radostool error"
 140           ) &
 141         fi
 142
 143         if is_set mark_osd_lost $flags; then
 144           ./ceph osd lost 0 --yes-i-really-mean-it
 145         fi
 146
 147         if is_set rm_osd $flags; then
 148             ./ceph osd rm 0
 149         fi
 150
 151         if ! is_set auto_mark_unfound_lost $flags; then
 152             for pg in $pgs_unfound; do
 153                 ./ceph pg $pg mark_unfound_lost ${lost_action} ||
 154                   die "mark_unfound_lost failed for pg $pg"
 155             done
 156         fi
 157
 158         start_recovery 2
 159
 160         # Unfound objects go away and are turned into lost objects.
 161         poll_cmd "./ceph pg debug unfound_objects_exist" FALSE 3 120
 162         [ $? -eq 1 ] || die "Unfound objects didn't go away."
 163
 164         for pg in `ceph pg ls | awk '/^[0-9]/ {print $1}'`; do
 165             ./ceph pg $pg mark_unfound_lost revert 2>&1 |
 166               grep 'pg has no unfound objects' ||
 167               die "pg $pg has unfound objects"
 168         done
 169
 170         # Reading from a lost object gives back an error code.
 171         # TODO: check error code
 172         ./rados -c ./ceph.conf -p $TEST_POOL get obj01 $TEMPDIR/obj01
 173         if [ lost_action = delete -a $? -eq 0 ]; then
 174           die "expected radostool error"
 175         elif [ lost_action = revert -a $? -ne 0 ]; then
 176           die "unexpected radostool error"
 177         fi
 178
 179         if is_set try_to_fetch_unfound $flags; then
 180           echo "waiting for the try_to_fetch_unfound \
 181 radostool instance to finish"
 182           wait
 183         fi
 184 }
 185
 186 lost1() {
 187         setup 2 'osd recovery delay start = 10000'
 188         lost1_impl mark_osd_lost revert_lost
 189 }
 190
 191 lost2() {
 192         setup 2 'osd recovery delay start = 10000'
 193         lost1_impl mark_osd_lost try_to_fetch_unfound
 194 }
 195
 196 lost3() {
 197         setup 2 'osd recovery delay start = 10000'
 198         lost1_impl rm_osd
 199 }
 200
 201 lost4() {
 202         setup 2 'osd recovery delay start = 10000'
 203         lost1_impl mark_osd_lost rm_osd
 204 }
 205
 206 lost5() {
 207         setup 2 'osd recovery delay start = 10000'
 208         lost1_impl mark_osd_lost auto_mark_unfound_lost
 209 }
 210
 211 all_osds_die_impl() {
 212         poll_cmd "./ceph osd stat" '3 up, 3 in' 20 240
 213         [ $? -eq 1 ] || die "didn't start 3 osds"
 214
 215         stop_osd 0
 216         stop_osd 1
 217         stop_osd 2
 218
 219         # wait for the MOSDPGStat timeout
 220         poll_cmd "./ceph osd stat" '0 up' 20 240
 221         [ $? -eq 1 ] || die "all osds weren't marked as down"
 222 }
 223
 224 all_osds_die() {
 225         setup 3 'osd mon report interval max = 60
 226         osd mon report interval min = 3
 227         mon osd report timeout = 60'
 228
 229         all_osds_die_impl
 230 }
 231
 232 run() {
 233         recovery1 || die "test failed"
 234
 235         lost1 || die "test failed"
 236
 237         # XXX: try_to_fetch_unfound test currently hangs on "waiting for the
 238         # try_to_fetch_unfound radostool instance to finish"
 239         #lost2 || die "test failed"
 240
 241         lost3 || die "test failed"
 242
 243         lost4 || die "test failed"
 244
 245         # XXX: automatically marking lost is not implemented
 246         #lost5 || die "test failed"
 247
 248         all_osds_die || die "test failed"
 249 }
 250
 251 if [ -z "$@" ]; then
 252         run
 253         echo OK
 254         exit 0
 255 fi
 256
 257 $@