ceph/qa/standalone/scrub/osd-scrub-snaps.sh

   1 #! /bin/bash
   2 #
   3 # Copyright (C) 2015 Red Hat <contact@redhat.com>
   4 #
   5 # Author: David Zafman <dzafman@redhat.com>
   6 #
   7 # This program is free software; you can redistribute it and/or modify
   8 # it under the terms of the GNU Library Public License as published by
   9 # the Free Software Foundation; either version 2, or (at your option)
  10 # any later version.
  11 #
  12 # This program is distributed in the hope that it will be useful,
  13 # but WITHOUT ANY WARRANTY; without even the implied warranty of
  14 # MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  15 # GNU Library Public License for more details.
  16 #
  17 source $CEPH_ROOT/qa/standalone/ceph-helpers.sh
  18
  19 function run() {
  20     local dir=$1
  21     shift
  22
  23     export CEPH_MON="127.0.0.1:7121" # git grep '\<7121\>' : there must be only one
  24     export CEPH_ARGS
  25     CEPH_ARGS+="--fsid=$(uuidgen) --auth-supported=none "
  26     CEPH_ARGS+="--mon-host=$CEPH_MON "
  27
  28     local funcs=${@:-$(set | sed -n -e 's/^\(TEST_[0-9a-z_]*\) .*/\1/p')}
  29     for func in $funcs ; do
  30         $func $dir || return 1
  31     done
  32 }
  33
  34 function TEST_scrub_snaps() {
  35     local dir=$1
  36     local poolname=test
  37
  38     TESTDATA="testdata.$$"
  39
  40     setup $dir || return 1
  41     run_mon $dir a --osd_pool_default_size=1 || return 1
  42     run_mgr $dir x || return 1
  43     run_osd $dir 0 || return 1
  44
  45     create_rbd_pool || return 1
  46     wait_for_clean || return 1
  47
  48     # Create a pool with a single pg
  49     create_pool $poolname 1 1
  50     wait_for_clean || return 1
  51     poolid=$(ceph osd dump | grep "^pool.*[']test[']" | awk '{ print $2 }')
  52
  53     dd if=/dev/urandom of=$TESTDATA bs=1032 count=1
  54     for i in `seq 1 15`
  55     do
  56         rados -p $poolname put obj${i} $TESTDATA
  57     done
  58
  59     SNAP=1
  60     rados -p $poolname mksnap snap${SNAP}
  61     dd if=/dev/urandom of=$TESTDATA bs=256 count=${SNAP}
  62     rados -p $poolname put obj1 $TESTDATA
  63     rados -p $poolname put obj5 $TESTDATA
  64     rados -p $poolname put obj3 $TESTDATA
  65     for i in `seq 6 14`
  66      do rados -p $poolname put obj${i} $TESTDATA
  67     done
  68
  69     SNAP=2
  70     rados -p $poolname mksnap snap${SNAP}
  71     dd if=/dev/urandom of=$TESTDATA bs=256 count=${SNAP}
  72     rados -p $poolname put obj5 $TESTDATA
  73
  74     SNAP=3
  75     rados -p $poolname mksnap snap${SNAP}
  76     dd if=/dev/urandom of=$TESTDATA bs=256 count=${SNAP}
  77     rados -p $poolname put obj3 $TESTDATA
  78
  79     SNAP=4
  80     rados -p $poolname mksnap snap${SNAP}
  81     dd if=/dev/urandom of=$TESTDATA bs=256 count=${SNAP}
  82     rados -p $poolname put obj5 $TESTDATA
  83     rados -p $poolname put obj2 $TESTDATA
  84
  85     SNAP=5
  86     rados -p $poolname mksnap snap${SNAP}
  87     SNAP=6
  88     rados -p $poolname mksnap snap${SNAP}
  89     dd if=/dev/urandom of=$TESTDATA bs=256 count=${SNAP}
  90     rados -p $poolname put obj5 $TESTDATA
  91
  92     SNAP=7
  93     rados -p $poolname mksnap snap${SNAP}
  94
  95     rados -p $poolname rm obj4
  96     rados -p $poolname rm obj2
  97
  98     kill_daemons $dir TERM osd || return 1
  99
 100     # Don't need to ceph_objectstore_tool function because osd stopped
 101
 102     JSON="$(ceph-objectstore-tool --data-path $dir/0 --journal-path $dir/0/journal --head --op list obj1)"
 103     ceph-objectstore-tool --data-path $dir/0 --journal-path $dir/0/journal "$JSON" --force remove
 104
 105     JSON="$(ceph-objectstore-tool --data-path $dir/0 --journal-path $dir/0/journal --op list obj5 | grep \"snapid\":2)"
 106     ceph-objectstore-tool --data-path $dir/0 --journal-path $dir/0/journal "$JSON" remove
 107
 108     JSON="$(ceph-objectstore-tool --data-path $dir/0 --journal-path $dir/0/journal --op list obj5 | grep \"snapid\":1)"
 109     OBJ5SAVE="$JSON"
 110     ceph-objectstore-tool --data-path $dir/0 --journal-path $dir/0/journal "$JSON" remove
 111
 112     JSON="$(ceph-objectstore-tool --data-path $dir/0 --journal-path $dir/0/journal --op list obj5 | grep \"snapid\":4)"
 113     dd if=/dev/urandom of=$TESTDATA bs=256 count=18
 114     ceph-objectstore-tool --data-path $dir/0 --journal-path $dir/0/journal "$JSON" set-bytes $TESTDATA
 115
 116     JSON="$(ceph-objectstore-tool --data-path $dir/0 --journal-path $dir/0/journal --head --op list obj3)"
 117     dd if=/dev/urandom of=$TESTDATA bs=256 count=15
 118     ceph-objectstore-tool --data-path $dir/0 --journal-path $dir/0/journal "$JSON" set-bytes $TESTDATA
 119
 120     JSON="$(ceph-objectstore-tool --data-path $dir/0 --journal-path $dir/0/journal --op list obj4 | grep \"snapid\":7)"
 121     ceph-objectstore-tool --data-path $dir/0 --journal-path $dir/0/journal "$JSON" remove
 122
 123     JSON="$(ceph-objectstore-tool --data-path $dir/0 --journal-path $dir/0/journal --head --op list obj2)"
 124     ceph-objectstore-tool --data-path $dir/0 --journal-path $dir/0/journal "$JSON" rm-attr snapset
 125
 126     # Create a clone which isn't in snapset and doesn't have object info
 127     JSON="$(echo "$OBJ5SAVE" | sed s/snapid\":1/snapid\":7/)"
 128     dd if=/dev/urandom of=$TESTDATA bs=256 count=7
 129     ceph-objectstore-tool --data-path $dir/0 --journal-path $dir/0/journal "$JSON" set-bytes $TESTDATA
 130
 131     rm -f $TESTDATA
 132
 133     JSON="$(ceph-objectstore-tool --data-path $dir/0 --journal-path $dir/0/journal --head --op list obj6)"
 134     ceph-objectstore-tool --data-path $dir/0 --journal-path $dir/0/journal "$JSON" clear-snapset
 135     JSON="$(ceph-objectstore-tool --data-path $dir/0 --journal-path $dir/0/journal --head --op list obj7)"
 136     ceph-objectstore-tool --data-path $dir/0 --journal-path $dir/0/journal "$JSON" clear-snapset corrupt
 137     JSON="$(ceph-objectstore-tool --data-path $dir/0 --journal-path $dir/0/journal --head --op list obj8)"
 138     ceph-objectstore-tool --data-path $dir/0 --journal-path $dir/0/journal "$JSON" clear-snapset seq
 139     JSON="$(ceph-objectstore-tool --data-path $dir/0 --journal-path $dir/0/journal --head --op list obj9)"
 140     ceph-objectstore-tool --data-path $dir/0 --journal-path $dir/0/journal "$JSON" clear-snapset clone_size
 141     JSON="$(ceph-objectstore-tool --data-path $dir/0 --journal-path $dir/0/journal --head --op list obj10)"
 142     ceph-objectstore-tool --data-path $dir/0 --journal-path $dir/0/journal "$JSON" clear-snapset clone_overlap
 143     JSON="$(ceph-objectstore-tool --data-path $dir/0 --journal-path $dir/0/journal --head --op list obj11)"
 144     ceph-objectstore-tool --data-path $dir/0 --journal-path $dir/0/journal "$JSON" clear-snapset clones
 145     JSON="$(ceph-objectstore-tool --data-path $dir/0 --journal-path $dir/0/journal --head --op list obj12)"
 146     ceph-objectstore-tool --data-path $dir/0 --journal-path $dir/0/journal "$JSON" clear-snapset head
 147     JSON="$(ceph-objectstore-tool --data-path $dir/0 --journal-path $dir/0/journal --head --op list obj13)"
 148     ceph-objectstore-tool --data-path $dir/0 --journal-path $dir/0/journal "$JSON" clear-snapset snaps
 149     JSON="$(ceph-objectstore-tool --data-path $dir/0 --journal-path $dir/0/journal --head --op list obj14)"
 150     ceph-objectstore-tool --data-path $dir/0 --journal-path $dir/0/journal "$JSON" clear-snapset size
 151
 152     echo "garbage" > $dir/bad
 153     JSON="$(ceph-objectstore-tool --data-path $dir/0 --journal-path $dir/0/journal --head --op list obj15)"
 154     ceph-objectstore-tool --data-path $dir/0 --journal-path $dir/0/journal "$JSON" set-attr snapset $dir/bad
 155     rm -f $dir/bad
 156
 157     run_osd $dir 0 || return 1
 158     create_rbd_pool || return 1
 159     wait_for_clean || return 1
 160
 161     local pgid="${poolid}.0"
 162     if ! pg_scrub "$pgid" ; then
 163         cat $dir/osd.0.log
 164         return 1
 165     fi
 166     grep 'log_channel' $dir/osd.0.log
 167
 168     rados list-inconsistent-pg $poolname > $dir/json || return 1
 169     # Check pg count
 170     test $(jq '. | length' $dir/json) = "1" || return 1
 171     # Check pgid
 172     test $(jq -r '.[0]' $dir/json) = $pgid || return 1
 173
 174     rados list-inconsistent-snapset $pgid > $dir/json || return 1
 175     test $(jq '.inconsistents | length' $dir/json) = "21" || return 1
 176
 177     local jqfilter='.inconsistents'
 178     local sortkeys='import json; import sys ; JSON=sys.stdin.read() ; ud = json.loads(JSON) ; print json.dumps(ud, sort_keys=True, indent=2)'
 179
 180     jq "$jqfilter" << EOF | python -c "$sortkeys" > $dir/checkcsjson
 181 {
 182   "inconsistents": [
 183     {
 184       "errors": [
 185         "headless"
 186       ],
 187       "snap": 1,
 188       "locator": "",
 189       "nspace": "",
 190       "name": "obj1"
 191     },
 192     {
 193       "errors": [
 194         "size_mismatch"
 195       ],
 196       "snap": 1,
 197       "locator": "",
 198       "nspace": "",
 199       "name": "obj10"
 200     },
 201     {
 202       "errors": [
 203         "headless"
 204       ],
 205       "snap": 1,
 206       "locator": "",
 207       "nspace": "",
 208       "name": "obj11"
 209     },
 210     {
 211       "errors": [
 212         "size_mismatch"
 213       ],
 214       "snap": 1,
 215       "locator": "",
 216       "nspace": "",
 217       "name": "obj14"
 218     },
 219     {
 220       "errors": [
 221         "headless"
 222       ],
 223       "snap": 1,
 224       "locator": "",
 225       "nspace": "",
 226       "name": "obj6"
 227     },
 228     {
 229       "errors": [
 230         "headless"
 231       ],
 232       "snap": 1,
 233       "locator": "",
 234       "nspace": "",
 235       "name": "obj7"
 236     },
 237     {
 238       "errors": [
 239         "size_mismatch"
 240       ],
 241       "snap": 1,
 242       "locator": "",
 243       "nspace": "",
 244       "name": "obj9"
 245     },
 246     {
 247       "errors": [
 248         "headless"
 249       ],
 250       "snap": 4,
 251       "locator": "",
 252       "nspace": "",
 253       "name": "obj2"
 254     },
 255     {
 256       "errors": [
 257         "size_mismatch"
 258       ],
 259       "snap": 4,
 260       "locator": "",
 261       "nspace": "",
 262       "name": "obj5"
 263     },
 264     {
 265       "errors": [
 266         "headless"
 267       ],
 268       "snap": 7,
 269       "locator": "",
 270       "nspace": "",
 271       "name": "obj2"
 272     },
 273     {
 274       "errors": [
 275         "oi_attr_missing",
 276         "headless"
 277       ],
 278       "snap": 7,
 279       "locator": "",
 280       "nspace": "",
 281       "name": "obj5"
 282     },
 283     {
 284       "extra clones": [
 285         1
 286       ],
 287       "errors": [
 288         "extra_clones"
 289       ],
 290       "snap": "head",
 291       "locator": "",
 292       "nspace": "",
 293       "name": "obj11"
 294     },
 295     {
 296       "errors": [
 297         "head_mismatch"
 298       ],
 299       "snap": "head",
 300       "locator": "",
 301       "nspace": "",
 302       "name": "obj12"
 303     },
 304     {
 305       "errors": [
 306         "ss_attr_corrupted"
 307       ],
 308       "snap": "head",
 309       "locator": "",
 310       "nspace": "",
 311       "name": "obj15"
 312     },
 313     {
 314       "extra clones": [
 315         7,
 316         4
 317       ],
 318       "errors": [
 319         "ss_attr_missing",
 320         "extra_clones"
 321       ],
 322       "snap": "head",
 323       "locator": "",
 324       "nspace": "",
 325       "name": "obj2"
 326     },
 327     {
 328       "errors": [
 329         "size_mismatch"
 330       ],
 331       "snap": "head",
 332       "locator": "",
 333       "nspace": "",
 334       "name": "obj3"
 335     },
 336     {
 337       "missing": [
 338         7
 339       ],
 340       "errors": [
 341         "clone_missing"
 342       ],
 343       "snap": "head",
 344       "locator": "",
 345       "nspace": "",
 346       "name": "obj4"
 347     },
 348     {
 349       "missing": [
 350         2,
 351         1
 352       ],
 353       "extra clones": [
 354         7
 355       ],
 356       "errors": [
 357         "extra_clones",
 358         "clone_missing"
 359       ],
 360       "snap": "head",
 361       "locator": "",
 362       "nspace": "",
 363       "name": "obj5"
 364     },
 365     {
 366       "extra clones": [
 367         1
 368       ],
 369       "errors": [
 370         "extra_clones"
 371       ],
 372       "snap": "head",
 373       "locator": "",
 374       "nspace": "",
 375       "name": "obj6"
 376     },
 377     {
 378       "extra clones": [
 379         1
 380       ],
 381       "errors": [
 382         "head_mismatch",
 383         "extra_clones"
 384       ],
 385       "snap": "head",
 386       "locator": "",
 387       "nspace": "",
 388       "name": "obj7"
 389     },
 390     {
 391       "errors": [
 392         "snapset_mismatch"
 393       ],
 394       "snap": "head",
 395       "locator": "",
 396       "nspace": "",
 397       "name": "obj8"
 398     }
 399   ],
 400   "epoch": 20
 401 }
 402 EOF
 403
 404     jq "$jqfilter" $dir/json | python -c "$sortkeys" > $dir/csjson
 405     diff ${DIFFCOLOPTS} $dir/checkcsjson $dir/csjson || return 1
 406
 407     if which jsonschema > /dev/null;
 408     then
 409       jsonschema -i $dir/json $CEPH_ROOT/doc/rados/command/list-inconsistent-snap.json || return 1
 410     fi
 411
 412     for i in `seq 1 7`
 413     do
 414         rados -p $poolname rmsnap snap$i
 415     done
 416
 417     ERRORS=0
 418
 419     pidfile=$(find $dir 2>/dev/null | grep $name_prefix'[^/]*\.pid')
 420     pid=$(cat $pidfile)
 421     if ! kill -0 $pid
 422     then
 423         echo "OSD crash occurred"
 424         tail -100 $dir/osd.0.log
 425         ERRORS=$(expr $ERRORS + 1)
 426     fi
 427
 428     kill_daemons $dir || return 1
 429
 430     declare -a err_strings
 431     err_strings[0]="log_channel[(]cluster[)] log [[]ERR[]] : scrub [0-9]*[.]0 .*::obj10:.* is missing in clone_overlap"
 432     err_strings[1]="log_channel[(]cluster[)] log [[]ERR[]] : scrub [0-9]*[.]0 .*::obj5:7 no '_' attr"
 433     err_strings[2]="log_channel[(]cluster[)] log [[]ERR[]] : scrub [0-9]*[.]0 .*::obj5:7 is an unexpected clone"
 434     err_strings[3]="log_channel[(]cluster[)] log [[]ERR[]] : scrub [0-9]*[.]0 .*::obj5:4 on disk size [(]4608[)] does not match object info size [(]512[)] adjusted for ondisk to [(]512[)]"
 435     err_strings[4]="log_channel[(]cluster[)] log [[]ERR[]] : scrub [0-9]*[.]0 .*:::obj5:head expected clone .*:::obj5:2"
 436     err_strings[5]="log_channel[(]cluster[)] log [[]ERR[]] : scrub [0-9]*[.]0 .*:::obj5:head expected clone .*:::obj5:1"
 437     err_strings[6]="log_channel[(]cluster[)] log [[]INF[]] : scrub [0-9]*[.]0 .*:::obj5:head 2 missing clone[(]s[)]"
 438     err_strings[7]="log_channel[(]cluster[)] log [[]ERR[]] : scrub [0-9]*[.]0 .*:::obj12:head snapset.head_exists=false, but head exists"
 439     err_strings[8]="log_channel[(]cluster[)] log [[]ERR[]] : scrub [0-9]*[.]0 .*:::obj8:head snaps.seq not set"
 440     err_strings[9]="log_channel[(]cluster[)] log [[]ERR[]] : scrub [0-9]*[.]0 .*:::obj7:head snapset.head_exists=false, but head exists"
 441     err_strings[10]="log_channel[(]cluster[)] log [[]ERR[]] : scrub [0-9]*[.]0 .*:::obj7:1 is an unexpected clone"
 442     err_strings[11]="log_channel[(]cluster[)] log [[]ERR[]] : scrub [0-9]*[.]0 .*:::obj3:head on disk size [(]3840[)] does not match object info size [(]768[)] adjusted for ondisk to [(]768[)]"
 443     err_strings[12]="log_channel[(]cluster[)] log [[]ERR[]] : scrub [0-9]*[.]0 .*:::obj6:1 is an unexpected clone"
 444     err_strings[13]="log_channel[(]cluster[)] log [[]ERR[]] : scrub [0-9]*[.]0 .*:::obj2:head no 'snapset' attr"
 445     err_strings[14]="log_channel[(]cluster[)] log [[]ERR[]] : scrub [0-9]*[.]0 .*:::obj2:7 clone ignored due to missing snapset"
 446     err_strings[15]="log_channel[(]cluster[)] log [[]ERR[]] : scrub [0-9]*[.]0 .*:::obj2:4 clone ignored due to missing snapset"
 447     err_strings[16]="log_channel[(]cluster[)] log [[]ERR[]] : scrub [0-9]*[.]0 .*:::obj4:head expected clone .*:::obj4:7"
 448     err_strings[17]="log_channel[(]cluster[)] log [[]INF[]] : scrub [0-9]*[.]0 .*:::obj4:head 1 missing clone[(]s[)]"
 449     err_strings[18]="log_channel[(]cluster[)] log [[]ERR[]] : scrub [0-9]*[.]0 .*:::obj1:1 is an unexpected clone"
 450     err_strings[19]="log_channel[(]cluster[)] log [[]ERR[]] : scrub [0-9]*[.]0 .*:::obj9:1 is missing in clone_size"
 451     err_strings[20]="log_channel[(]cluster[)] log [[]ERR[]] : scrub [0-9]*[.]0 .*:::obj11:1 is an unexpected clone"
 452     err_strings[21]="log_channel[(]cluster[)] log [[]ERR[]] : scrub [0-9]*[.]0 .*:::obj14:1 size 1032 != clone_size 1033"
 453     err_strings[22]="log_channel[(]cluster[)] log [[]ERR[]] : [0-9]*[.]0 scrub 22 errors"
 454     err_strings[23]="log_channel[(]cluster[)] log [[]ERR[]] : scrub [0-9]*[.]0 .*:::obj15:head can't decode 'snapset' attr buffer"
 455
 456     for err_string in "${err_strings[@]}"
 457     do
 458         if ! grep "$err_string" $dir/osd.0.log > /dev/null;
 459         then
 460             echo "Missing log message '$err_string'"
 461             ERRORS=$(expr $ERRORS + 1)
 462         fi
 463     done
 464
 465     teardown $dir || return 1
 466
 467     if [ $ERRORS != "0" ];
 468     then
 469         echo "TEST FAILED WITH $ERRORS ERRORS"
 470         return 1
 471     fi
 472
 473     echo "TEST PASSED"
 474     return 0
 475 }
 476
 477 main osd-scrub-snaps "$@"
 478
 479 # Local Variables:
 480 # compile-command: "cd ../.. ; make -j4 && \
 481 #    test/osd/osd-scrub-snaps.sh"