X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Ftests%2Fsanityn.sh;h=a0bfb327959983ea280faef386b6e41e785fc4d8;hp=6b00d2f506973e120c648d05c6203f0a562643c3;hb=7fa66448c2f51775fbbd4f50d114b4ca64e59576;hpb=a880f385164547837a34df92acd6cdc8e9a9698d diff --git a/lustre/tests/sanityn.sh b/lustre/tests/sanityn.sh old mode 100644 new mode 100755 index 6b00d2f..a0bfb32 --- a/lustre/tests/sanityn.sh +++ b/lustre/tests/sanityn.sh @@ -3,8 +3,8 @@ set -e ONLY=${ONLY:-"$*"} -# bug number for skipped test: 3192 LU-1205 15528/3811 9977 15528/11549 18080 -ALWAYS_EXCEPT=" 14b 18c 19 28 29 35 $SANITYN_EXCEPT" +# bug number for skipped test: LU-1205 9977/LU-7105 LU-9452 +ALWAYS_EXCEPT=" 18c 28 29 $SANITYN_EXCEPT" # UPDATE THE COMMENT ABOVE WITH BUG NUMBERS WHEN CHANGING ALWAYS_EXCEPT! SRCDIR=$(dirname $0) @@ -38,7 +38,7 @@ if [ $(facet_fstype $SINGLEMDS) = "zfs" ]; then # LU-2829 / LU-2887 - make allowances for ZFS slowness TEST33_NFILES=${TEST33_NFILES:-1000} fi - +# 23 (min)" [ "$SLOW" = "no" ] && EXCEPT_SLOW="33a" FAIL_ON_ERROR=false @@ -61,9 +61,6 @@ check_runas_id $RUNAS_ID $RUNAS_GID $RUNAS build_test_filter -mkdir -p $MOUNT2 -mount_client $MOUNT2 - test_1a() { touch $DIR1/f1 [ -f $DIR2/f1 ] || error @@ -748,42 +745,42 @@ enable_lockless_truncate() { test_32a() { # bug 11270 local p="$TMP/$TESTSUITE-$TESTNAME.parameters" save_lustre_params client "osc.*.lockless_truncate" > $p - cancel_lru_locks osc - enable_lockless_truncate 1 - rm -f $DIR1/$tfile - lfs setstripe -c -1 $DIR1/$tfile - dd if=/dev/zero of=$DIR1/$tfile count=$OSTCOUNT bs=$STRIPE_BYTES > \ - /dev/null 2>&1 - clear_osc_stats - - log "checking cached lockless truncate" - $TRUNCATE $DIR1/$tfile 8000000 - $CHECKSTAT -s 8000000 $DIR2/$tfile || error "wrong file size" - [ $(calc_osc_stats lockless_truncate) -ne 0 ] || + cancel_lru_locks osc + enable_lockless_truncate 1 + rm -f $DIR1/$tfile + lfs setstripe -c -1 $DIR1/$tfile + dd if=/dev/zero of=$DIR1/$tfile count=$OSTCOUNT bs=$STRIPE_BYTES > \ + /dev/null 2>&1 + clear_stats osc.*.osc_stats + + log "checking cached lockless truncate" + $TRUNCATE $DIR1/$tfile 8000000 + $CHECKSTAT -s 8000000 $DIR2/$tfile || error "wrong file size" + [ $(calc_stats osc.*.osc_stats lockless_truncate) -ne 0 ] || error "cached truncate isn't lockless" - log "checking not cached lockless truncate" - $TRUNCATE $DIR2/$tfile 5000000 - $CHECKSTAT -s 5000000 $DIR1/$tfile || error "wrong file size" - [ $(calc_osc_stats lockless_truncate) -ne 0 ] || + log "checking not cached lockless truncate" + $TRUNCATE $DIR2/$tfile 5000000 + $CHECKSTAT -s 5000000 $DIR1/$tfile || error "wrong file size" + [ $(calc_stats osc.*.osc_stats lockless_truncate) -ne 0 ] || error "not cached truncate isn't lockless" - log "disabled lockless truncate" - enable_lockless_truncate 0 - clear_osc_stats - $TRUNCATE $DIR2/$tfile 3000000 - $CHECKSTAT -s 3000000 $DIR1/$tfile || error "wrong file size" - [ $(calc_osc_stats lockless_truncate) -eq 0 ] || - error "lockless truncate disabling failed" - rm $DIR1/$tfile - # restore lockless_truncate default values - restore_lustre_params < $p - rm -f $p + log "disabled lockless truncate" + enable_lockless_truncate 0 + clear_stats osc.*.osc_stats + $TRUNCATE $DIR2/$tfile 3000000 + $CHECKSTAT -s 3000000 $DIR1/$tfile || error "wrong file size" + [ $(calc_stats osc.*.osc_stats lockless_truncate) -eq 0 ] || + error "lockless truncate disabling failed" + rm $DIR1/$tfile + # restore lockless_truncate default values + restore_lustre_params < $p + rm -f $p } run_test 32a "lockless truncate" test_32b() { # bug 11270 - remote_ost_nodsh && skip "remote OST with nodsh" && return + remote_ost_nodsh && skip "remote OST with nodsh" && return local node local facets=$(get_facets OST) @@ -796,35 +793,42 @@ test_32b() { # bug 11270 "ldlm.namespaces.filter-*.contended_locks" >> $p save_lustre_params $facets \ "ldlm.namespaces.filter-*.contention_seconds" >> $p - clear_osc_stats + clear_stats osc.*.osc_stats - # agressive lockless i/o settings - for node in $(osts_nodes); do - do_node $node 'lctl set_param -n ldlm.namespaces.filter-*.max_nolock_bytes 2000000; lctl set_param -n ldlm.namespaces.filter-*.contended_locks 0; lctl set_param -n ldlm.namespaces.filter-*.contention_seconds 60' - done - lctl set_param -n osc.*.contention_seconds 60 - for i in $(seq 5); do - dd if=/dev/zero of=$DIR1/$tfile bs=4k count=1 conv=notrunc > /dev/null 2>&1 - dd if=/dev/zero of=$DIR2/$tfile bs=4k count=1 conv=notrunc > /dev/null 2>&1 - done - [ $(calc_osc_stats lockless_write_bytes) -ne 0 ] || error "lockless i/o was not triggered" - # disable lockless i/o (it is disabled by default) - for node in $(osts_nodes); do - do_node $node 'lctl set_param -n ldlm.namespaces.filter-*.max_nolock_bytes 0; lctl set_param -n ldlm.namespaces.filter-*.contended_locks 32; lctl set_param -n ldlm.namespaces.filter-*.contention_seconds 0' - done - # set contention_seconds to 0 at client too, otherwise Lustre still - # remembers lock contention - lctl set_param -n osc.*.contention_seconds 0 - clear_osc_stats - for i in $(seq 1); do - dd if=/dev/zero of=$DIR1/$tfile bs=4k count=1 conv=notrunc > /dev/null 2>&1 - dd if=/dev/zero of=$DIR2/$tfile bs=4k count=1 conv=notrunc > /dev/null 2>&1 - done - [ $(calc_osc_stats lockless_write_bytes) -eq 0 ] || - error "lockless i/o works when disabled" - rm -f $DIR1/$tfile - restore_lustre_params <$p - rm -f $p + # agressive lockless i/o settings + do_nodes $(comma_list $(osts_nodes)) \ + "lctl set_param -n ldlm.namespaces.*.max_nolock_bytes=2000000 \ + ldlm.namespaces.filter-*.contended_locks=0 \ + ldlm.namespaces.filter-*.contention_seconds=60" + lctl set_param -n osc.*.contention_seconds=60 + for i in {1..5}; do + dd if=/dev/zero of=$DIR1/$tfile bs=4k count=1 conv=notrunc > \ + /dev/null 2>&1 + dd if=/dev/zero of=$DIR2/$tfile bs=4k count=1 conv=notrunc > \ + /dev/null 2>&1 + done + [ $(calc_stats osc.*.osc_stats lockless_write_bytes) -ne 0 ] || + error "lockless i/o was not triggered" + # disable lockless i/o (it is disabled by default) + do_nodes $(comma_list $(osts_nodes)) \ + "lctl set_param -n ldlm.namespaces.filter-*.max_nolock_bytes=0 \ + ldlm.namespaces.filter-*.contended_locks=32 \ + ldlm.namespaces.filter-*.contention_seconds=0" + # set contention_seconds to 0 at client too, otherwise Lustre still + # remembers lock contention + lctl set_param -n osc.*.contention_seconds=0 + clear_stats osc.*.osc_stats + for i in {1..1}; do + dd if=/dev/zero of=$DIR1/$tfile bs=4k count=1 conv=notrunc > \ + /dev/null 2>&1 + dd if=/dev/zero of=$DIR2/$tfile bs=4k count=1 conv=notrunc > \ + /dev/null 2>&1 + done + [ $(calc_stats osc.*.osc_stats lockless_write_bytes) -eq 0 ] || + error "lockless i/o works when disabled" + rm -f $DIR1/$tfile + restore_lustre_params <$p + rm -f $p } run_test 32b "lockless i/o" @@ -876,7 +880,7 @@ test_33a() { [ $fstype = ldiskfs ] && jbdold=$(print_jbd_stat) echo "=== START createmany old: $jbdold transaction" - local elapsed=$(do_and_time "do_nodes $CLIENT1,$CLIENT2 createmany -o $DIR1/$tdir-\\\$(hostname)-$i/f- -r $DIR2/$tdir-\\\$(hostname)-$i/f- $nfiles > /dev/null 2>&1") + local elapsed=$(do_and_time "do_nodes $CLIENT1,$CLIENT2 createmany -o $DIR1/$tdir-\\\$(hostname)-$i/f- -r$DIR2/$tdir-\\\$(hostname)-$i/f- $nfiles > /dev/null 2>&1") [ $fstype = ldiskfs ] && jbdnew=$(print_jbd_stat) [ $fstype = ldiskfs ] && jbd=$(( jbdnew - jbdold )) echo "=== END createmany new: $jbdnew transaction : $jbd transactions nfiles $nfiles time $elapsed COS=$COS" @@ -932,7 +936,7 @@ test_33b() { echo "=== START createmany old: $jbdold transaction" local elapsed=$(do_and_time "do_nodes $CLIENT1,$CLIENT2\ createmany -o $DIR1/$tdir-\\\$(hostname)-$i/f- \ - -r $DIR2/$tdir-\\\$(hostname)-$i/f- $nfiles > \ + -r$DIR2/$tdir-\\\$(hostname)-$i/f- $nfiles > \ /dev/null 2>&1") jbdnew=$(print_jbd_stat) jbd=$(( jbdnew - jbdold )) @@ -1351,6 +1355,32 @@ test_39c() { } run_test 39c "check truncate mtime update ======================" +test_39d() { # LU-7310 + touch $DIR1/$tfile + touch -m -d @$TEST_39_MTIME $DIR1/$tfile + + local mtime1=$(stat -c %Y $DIR2/$tfile) + [ "$mtime1" = $TEST_39_MTIME ] || + error "mtime: $mtime1, should be $TEST_39_MTIME" + + # force sync write + # define OBD_FAIL_OSC_NO_GRANT 0x411 + $LCTL set_param fail_loc=0x411 + + local d1=$(date +%s) + echo hello >> $DIR1/$tfile + local d2=$(date +%s) + + $LCTL set_param fail_loc=0 + + cancel_lru_locks osc + + local mtime2=$(stat -c %Y $DIR2/$tfile) + [ "$mtime2" -ge "$d1" ] && [ "$mtime2" -le "$d2" ] || + error "mtime is not updated on write: $d1 <= $mtime2 <= $d2" +} +run_test 39d "sync write should update mtime" + # check that pid exists hence second operation wasn't blocked by first one # if it is so then there is no conflict, return 0 # else second operation is conflicting with first one, return 1 @@ -1391,10 +1421,10 @@ test_40a() { rmdir $DIR2/$tfile-3 check_pdo_conflict $PID1 || error "unlink is blocked" - # all operations above shouldn't wait the first one + # all operations above shouldn't wait the first one check_pdo_conflict $PID1 || error "parallel operation is blocked" wait $PID1 - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 40a "pdirops: create vs others ==============" @@ -1424,7 +1454,7 @@ test_40b() { check_pdo_conflict $PID1 || error "parallel operation is blocked" wait $PID1 - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 40b "pdirops: open|create and others ==============" @@ -1455,7 +1485,7 @@ test_40c() { # all operations above shouldn't wait the first one check_pdo_conflict $PID1 || error "parallel operation is blocked" wait $PID1 - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 40c "pdirops: link and others ==============" @@ -1514,7 +1544,7 @@ test_40e() { # all operations above shouldn't wait the first one check_pdo_conflict $PID1 || error "parallel operation is blocked" wait $PID1 - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 40e "pdirops: rename and others ==============" @@ -1528,7 +1558,7 @@ test_41a() { sleep 1 mkdir $DIR2/$tfile && error "mkdir must fail" check_pdo_conflict $PID1 && { wait $PID1; echo "mkdir isn't blocked"; } - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 41a "pdirops: create vs mkdir ==============" @@ -1541,7 +1571,7 @@ test_41b() { sleep 1 $MULTIOP $DIR2/$tfile oO_CREAT:O_EXCL:c && error "create must fail" check_pdo_conflict $PID1 && { wait $PID1; error "create isn't blocked"; } - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 41b "pdirops: create vs create ==============" @@ -1555,7 +1585,7 @@ test_41c() { sleep 1 link $DIR2/$tfile-2 $DIR2/$tfile && error "link must fail" check_pdo_conflict $PID1 && { wait $PID1; error "link isn't blocked"; } - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 41c "pdirops: create vs link ==============" @@ -1568,7 +1598,7 @@ test_41d() { sleep 1 rm $DIR2/$tfile || error "unlink must succeed" check_pdo_conflict $PID1 && { wait $PID1; error "unlink isn't blocked"; } - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 41d "pdirops: create vs unlink ==============" @@ -1582,7 +1612,7 @@ test_41e() { sleep 1 mv $DIR2/$tfile-2 $DIR2/$tfile || error "rename must succeed" check_pdo_conflict $PID1 && { wait $PID1; error "rename isn't blocked"; } - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 41e "pdirops: create and rename (tgt) ==============" @@ -1595,7 +1625,7 @@ test_41f() { sleep 1 mv $DIR2/$tfile $DIR2/$tfile-2 || error "rename must succeed" check_pdo_conflict $PID1 && { wait $PID1; error "rename isn't blocked"; } - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 41f "pdirops: create and rename (src) ==============" @@ -1608,7 +1638,7 @@ test_41g() { sleep 1 stat $DIR2/$tfile > /dev/null || error "stat must succeed" check_pdo_conflict $PID1 && { wait $PID1; error "getattr isn't blocked"; } - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 41g "pdirops: create vs getattr ==============" @@ -1621,7 +1651,7 @@ test_41h() { sleep 1 ls -lia $DIR2/ > /dev/null check_pdo_conflict $PID1 && { wait $PID1; error "readdir isn't blocked"; } - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 41h "pdirops: create vs readdir ==============" @@ -1635,7 +1665,7 @@ test_42a() { sleep 1 mkdir $DIR2/$tfile && error "mkdir must fail" check_pdo_conflict $PID1 && { wait $PID1; error "mkdir isn't blocked"; } - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 42a "pdirops: mkdir vs mkdir ==============" @@ -1648,7 +1678,7 @@ test_42b() { sleep 1 $MULTIOP $DIR2/$tfile oO_CREAT:O_EXCL:c && error "create must fail" check_pdo_conflict $PID1 && { wait $PID1; error "create isn't blocked"; } - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 42b "pdirops: mkdir vs create ==============" @@ -1662,7 +1692,7 @@ test_42c() { sleep 1 link $DIR2/$tfile-2 $DIR2/$tfile && error "link must fail" check_pdo_conflict $PID1 && { wait $PID1; error "link isn't blocked"; } - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 42c "pdirops: mkdir vs link ==============" @@ -1675,7 +1705,7 @@ test_42d() { sleep 1 rmdir $DIR2/$tfile || error "unlink must succeed" check_pdo_conflict $PID1 && { wait $PID1; error "unlink isn't blocked"; } - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 42d "pdirops: mkdir vs unlink ==============" @@ -1689,7 +1719,7 @@ test_42e() { sleep 1 mv -T $DIR2/$tfile-2 $DIR2/$tfile && error "rename must fail" check_pdo_conflict $PID1 && { wait $PID1; error "rename isn't blocked"; } - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 42e "pdirops: mkdir and rename (tgt) ==============" @@ -1702,7 +1732,7 @@ test_42f() { sleep 1 mv $DIR2/$tfile $DIR2/$tfile-2 || error "rename must succeed" check_pdo_conflict $PID1 && { wait $PID1; error "rename isn't blocked"; } - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 42f "pdirops: mkdir and rename (src) ==============" @@ -1715,7 +1745,7 @@ test_42g() { sleep 1 stat $DIR2/$tfile > /dev/null || error "stat must succeed" check_pdo_conflict $PID1 && { wait $PID1; error "getattr isn't blocked"; } - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 42g "pdirops: mkdir vs getattr ==============" @@ -1728,7 +1758,7 @@ test_42h() { sleep 1 ls -lia $DIR2/ > /dev/null check_pdo_conflict $PID1 && { wait $PID1; error "readdir isn't blocked"; } - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 42h "pdirops: mkdir vs readdir ==============" @@ -1743,7 +1773,7 @@ test_43a() { sleep 1 mkdir $DIR2/$tfile || error "mkdir must succeed" check_pdo_conflict $PID1 && { wait $PID1; error "mkdir isn't blocked"; } - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 43a "pdirops: unlink vs mkdir ==============" @@ -1757,7 +1787,7 @@ test_43b() { sleep 1 $MULTIOP $DIR2/$tfile oO_CREAT:O_EXCL:c || error "create must succeed" check_pdo_conflict $PID1 && { wait $PID1; error "create isn't blocked"; } - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 43b "pdirops: unlink vs create ==============" @@ -1772,7 +1802,7 @@ test_43c() { sleep 1 link $DIR2/$tfile-2 $DIR2/$tfile || error "link must succeed" check_pdo_conflict $PID1 && { wait $PID1; error "link isn't blocked"; } - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 43c "pdirops: unlink vs link ==============" @@ -1786,7 +1816,7 @@ test_43d() { sleep 1 rm $DIR2/$tfile && error "unlink must fail" check_pdo_conflict $PID1 && { wait $PID1; error "unlink isn't blocked"; } - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 43d "pdirops: unlink vs unlink ==============" @@ -1801,7 +1831,7 @@ test_43e() { sleep 1 mv -u $DIR2/$tfile-2 $DIR2/$tfile || error "rename must succeed" check_pdo_conflict $PID1 && { wait $PID1; error "rename isn't blocked"; } - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 43e "pdirops: unlink and rename (tgt) ==============" @@ -1815,7 +1845,7 @@ test_43f() { sleep 1 mv $DIR2/$tfile $DIR2/$tfile-2 && error "rename must fail" check_pdo_conflict $PID1 && { wait $PID1; error "rename isn't blocked"; } - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 43f "pdirops: unlink and rename (src) ==============" @@ -1829,7 +1859,7 @@ test_43g() { sleep 1 stat $DIR2/$tfile > /dev/null && error "stat must fail" check_pdo_conflict $PID1 && { wait $PID1; error "getattr isn't blocked"; } - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 43g "pdirops: unlink vs getattr ==============" @@ -1843,7 +1873,7 @@ test_43h() { sleep 1 ls -lia $DIR2/ > /dev/null check_pdo_conflict $PID1 && { wait $PID1; error "readdir isn't blocked"; } - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 43h "pdirops: unlink vs readdir ==============" @@ -1859,7 +1889,7 @@ test_43i() { $LFS mkdir -i 1 $DIR2/$tfile || error "remote mkdir must succeed" check_pdo_conflict $PID1 && { wait $PID1; error "remote mkdir isn't blocked"; } - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 43i "pdirops: unlink vs remote mkdir" @@ -1874,7 +1904,7 @@ test_44a() { sleep 1 mkdir $DIR2/$tfile && error "mkdir must fail" check_pdo_conflict $PID1 && { wait $PID1; error "mkdir isn't blocked"; } - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 44a "pdirops: rename tgt vs mkdir ==============" @@ -1888,7 +1918,7 @@ test_44b() { sleep 1 $MULTIOP $DIR2/$tfile oO_CREAT:O_EXCL:c && error "create must fail" check_pdo_conflict $PID1 && { wait $PID1; error "create isn't blocked"; } - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 44b "pdirops: rename tgt vs create ==============" @@ -1903,7 +1933,7 @@ test_44c() { sleep 1 link $DIR2/$tfile-3 $DIR2/$tfile && error "link must fail" check_pdo_conflict $PID1 && { wait $PID1; error "link isn't blocked"; } - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 44c "pdirops: rename tgt vs link ==============" @@ -1917,7 +1947,7 @@ test_44d() { sleep 1 rm $DIR2/$tfile || error "unlink must succeed" check_pdo_conflict $PID1 && { wait $PID1; error "unlink isn't blocked"; } - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 44d "pdirops: rename tgt vs unlink ==============" @@ -1933,7 +1963,7 @@ test_44e() { sleep 1 mv $DIR2/$tfile-3 $DIR2/$tfile || error "rename must succeed" check_pdo_conflict $PID1 && { wait $PID1; error "rename isn't blocked"; } - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 44e "pdirops: rename tgt and rename (tgt) ==============" @@ -1948,7 +1978,7 @@ test_44f() { sleep 1 mv $DIR2/$tfile $DIR2/$tfile-3 || error "rename must succeed" check_pdo_conflict $PID1 && { wait $PID1; error "rename isn't blocked"; } - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 44f "pdirops: rename tgt and rename (src) ==============" @@ -1962,7 +1992,7 @@ test_44g() { sleep 1 stat $DIR2/$tfile > /dev/null || error "stat must succeed" check_pdo_conflict $PID1 && { wait $PID1; error "getattr isn't blocked"; } - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 44g "pdirops: rename tgt vs getattr ==============" @@ -1976,7 +2006,7 @@ test_44h() { sleep 1 ls -lia $DIR2/ > /dev/null check_pdo_conflict $PID1 && { wait $PID1; error "readdir isn't blocked"; } - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 44h "pdirops: rename tgt vs readdir ==============" @@ -1993,7 +2023,7 @@ test_44i() { $LFS mkdir -i 1 $DIR2/$tfile && error "remote mkdir must fail" check_pdo_conflict $PID1 && { wait $PID1; error "remote mkdir isn't blocked"; } - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 44i "pdirops: rename tgt vs remote mkdir" @@ -2008,7 +2038,7 @@ test_45a() { sleep 1 mkdir $DIR2/$tfile || error "mkdir must succeed" check_pdo_conflict $PID1 && { wait $PID1; error "mkdir isn't blocked"; } - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 45a "pdirops: rename src vs mkdir ==============" @@ -2022,7 +2052,7 @@ test_45b() { sleep 1 $MULTIOP $DIR2/$tfile oO_CREAT:O_EXCL:c || error "create must succeed" check_pdo_conflict $PID1 && { wait $PID1; error "create isn't blocked"; } - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 45b "pdirops: rename src vs create ==============" @@ -2037,7 +2067,7 @@ test_45c() { sleep 1 link $DIR2/$tfile-3 $DIR2/$tfile || error "link must succeed" check_pdo_conflict $PID1 && { wait $PID1; error "link isn't blocked"; } - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 45c "pdirops: rename src vs link ==============" @@ -2051,7 +2081,7 @@ test_45d() { sleep 1 rm $DIR2/$tfile && error "unlink must fail" check_pdo_conflict $PID1 && { wait $PID1; error "unlink isn't blocked"; } - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 45d "pdirops: rename src vs unlink ==============" @@ -2066,7 +2096,7 @@ test_45e() { sleep 1 mv $DIR2/$tfile-3 $DIR2/$tfile || error "rename must succeed" check_pdo_conflict $PID1 && { wait $PID1; error "rename isn't blocked"; } - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 45e "pdirops: rename src and rename (tgt) ==============" @@ -2080,7 +2110,7 @@ test_45f() { sleep 1 mv $DIR2/$tfile $DIR2/$tfile-3 && error "rename must fail" check_pdo_conflict $PID1 && { wait $PID1; error "rename isn't blocked"; } - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 45f "pdirops: rename src and rename (src) ==============" @@ -2094,7 +2124,7 @@ test_45g() { sleep 1 stat $DIR2/$tfile > /dev/null && error "stat must fail" check_pdo_conflict $PID1 && { wait $PID1; error "getattr isn't blocked"; } - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 45g "pdirops: rename src vs getattr ==============" @@ -2108,7 +2138,7 @@ test_45h() { sleep 1 ls -lia $DIR2/ > /dev/null check_pdo_conflict $PID1 && { wait $PID1; error "readdir isn't blocked"; } - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 45h "pdirops: unlink vs readdir ==============" @@ -2124,7 +2154,7 @@ test_45i() { $LFS mkdir -i 1 $DIR2/$tfile || error "create remote dir must succeed" check_pdo_conflict $PID1 && { wait $PID1; error "create remote dir isn't blocked"; } - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 45i "pdirops: rename src vs remote mkdir" @@ -2139,7 +2169,7 @@ test_46a() { sleep 1 mkdir $DIR2/$tfile && error "mkdir must fail" check_pdo_conflict $PID1 && { wait $PID1; error "mkdir isn't blocked"; } - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 46a "pdirops: link vs mkdir ==============" @@ -2153,7 +2183,7 @@ test_46b() { sleep 1 $MULTIOP $DIR2/$tfile oO_CREAT:O_EXCL:c && error "create must fail" check_pdo_conflict $PID1 && { wait $PID1; error "create isn't blocked"; } - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 46b "pdirops: link vs create ==============" @@ -2167,7 +2197,7 @@ test_46c() { sleep 1 link $DIR2/$tfile $DIR2/$tfile && error "link must fail" check_pdo_conflict $PID1 && { wait $PID1; error "link isn't blocked"; } - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 46c "pdirops: link vs link ==============" @@ -2181,7 +2211,7 @@ test_46d() { sleep 1 rm $DIR2/$tfile || error "unlink must succeed" check_pdo_conflict $PID1 && { wait $PID1; error "unlink isn't blocked"; } - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 46d "pdirops: link vs unlink ==============" @@ -2196,7 +2226,7 @@ test_46e() { sleep 1 mv $DIR2/$tfile-3 $DIR2/$tfile || error "rename must succeed" check_pdo_conflict $PID1 && { wait $PID1; error "rename isn't blocked"; } - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 46e "pdirops: link and rename (tgt) ==============" @@ -2211,7 +2241,7 @@ test_46f() { sleep 1 mv $DIR2/$tfile $DIR2/$tfile-3 || error "rename must succeed" check_pdo_conflict $PID1 && { wait $PID1; error "rename isn't blocked"; } - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 46f "pdirops: link and rename (src) ==============" @@ -2225,7 +2255,7 @@ test_46g() { sleep 1 stat $DIR2/$tfile > /dev/null || error "stat must succeed" check_pdo_conflict $PID1 && { wait $PID1; error "getattr isn't blocked"; } - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 46g "pdirops: link vs getattr ==============" @@ -2240,7 +2270,7 @@ test_46h() { ls -lia $DIR2/ > /dev/null check_pdo_conflict $PID1 && { wait $PID1; error "readdir isn't blocked"; } - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 46h "pdirops: link vs readdir ==============" @@ -2256,7 +2286,7 @@ test_46i() { $LFS mkdir -i 1 $DIR2/$tfile && error "remote mkdir must fail" check_pdo_conflict $PID1 && { wait $PID1; error "remote mkdir isn't blocked"; } - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 46i "pdirops: link vs remote mkdir" @@ -2271,7 +2301,7 @@ test_47a() { sleep 1 mkdir $DIR2/$tfile && error "mkdir must fail" check_pdo_conflict $PID1 && { wait $PID1; error "mkdir isn't blocked"; } - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 47a "pdirops: remote mkdir vs mkdir" @@ -2286,7 +2316,7 @@ test_47b() { multiop $DIR2/$tfile oO_CREAT:O_EXCL:c && error "create must fail" check_pdo_conflict $PID1 && { wait $PID1; error "create isn't blocked"; } - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 47b "pdirops: remote mkdir vs create" @@ -2301,7 +2331,7 @@ test_47c() { sleep 1 link $DIR2/$tfile-2 $DIR2/$tfile && error "link must fail" check_pdo_conflict $PID1 && { wait $PID1; error "link isn't blocked"; } - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 47c "pdirops: remote mkdir vs link" @@ -2316,7 +2346,7 @@ test_47d() { rmdir $DIR2/$tfile || error "unlink must succeed" check_pdo_conflict $PID1 && { wait $PID1; error "unlink isn't blocked"; } - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 47d "pdirops: remote mkdir vs unlink" @@ -2332,7 +2362,7 @@ test_47e() { mv -T $DIR2/$tfile-2 $DIR2/$tfile && error "rename must fail" check_pdo_conflict $PID1 && { wait $PID1; error "rename isn't blocked"; } - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 47e "pdirops: remote mkdir and rename (tgt)" @@ -2347,7 +2377,7 @@ test_47f() { mv $DIR2/$tfile $DIR2/$tfile-2 || error "rename must succeed" check_pdo_conflict $PID1 && { wait $PID1; error "rename isn't blocked"; } - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 47f "pdirops: remote mkdir and rename (src)" @@ -2362,7 +2392,7 @@ test_47g() { stat $DIR2/$tfile > /dev/null || error "stat must succeed" check_pdo_conflict $PID1 && { wait $PID1; error "getattr isn't blocked"; } - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 47g "pdirops: remote mkdir vs getattr" @@ -2384,13 +2414,13 @@ test_51a() { local filesize local origfile=/etc/hosts - filesize=`stat -c %s $origfile` + filesize=$(stat -c %s $origfile) # create an empty file - $MCREATE $DIR1/$tfile + $MCREATE $DIR1/$tfile || error "can't create $DIR1/$tfile" # cache layout lock on both mount point - stat $DIR1/$tfile > /dev/null - stat $DIR2/$tfile > /dev/null + stat $DIR1/$tfile > /dev/null || error "stat $DIR1/$tfile failed" + stat $DIR2/$tfile > /dev/null || error "stat $DIR2/$tfile failed" # open and sleep 2 seconds then read $MULTIOP $DIR2/$tfile o_2r${filesize}c & @@ -2398,12 +2428,13 @@ test_51a() { sleep 1 # create the layout of testing file - dd if=$origfile of=$DIR1/$tfile conv=notrunc > /dev/null + dd if=$origfile of=$DIR1/$tfile conv=notrunc > /dev/null || + error "dd $DIR1/$tfile failed" # MULTIOP proc should be able to read enough bytes and exit sleep 2 - kill -0 $pid && error "multiop is still there" - cmp $origfile $DIR2/$tfile || error "$MCREATE and $DIR2/$tfile differs" + kill -0 $pid 2> /dev/null && error "multiop is still there" + cmp $origfile $DIR2/$tfile || error "$origfile and $DIR2/$tfile differs" rm -f $DIR1/$tfile } @@ -2416,7 +2447,7 @@ test_51b() { local tmpfile=`mktemp` # create an empty file - $MCREATE $DIR1/$tfile + $MCREATE $DIR1/$tfile || error "mcreate $DIR1/$tfile failed" # delay glimpse so that layout has changed when glimpse finish #define OBD_FAIL_GLIMPSE_DELAY 0x1404 @@ -2426,10 +2457,11 @@ test_51b() { sleep 1 # create layout of testing file - dd if=/dev/zero of=$DIR1/$tfile bs=1k count=1 conv=notrunc > /dev/null + dd if=/dev/zero of=$DIR1/$tfile bs=1k count=1 conv=notrunc >/dev/null || + error "dd $DIR1/$tfile failed" wait $pid - local fsize=`cat $tmpfile` + local fsize=$(cat $tmpfile) [ x$fsize = x1024 ] || error "file size is $fsize, should be 1024" @@ -2441,11 +2473,12 @@ test_51c() { [ $OSTCOUNT -ge 2 ] || { skip "need at least 2 osts"; return; } # set default layout to have 1 stripe - mkdir -p $DIR1/$tdir + mkdir $DIR1/$tdir $LFS setstripe -c 1 $DIR1/$tdir # create a file with empty layout - $MCREATE $DIR1/$tdir/$tfile + $MCREATE $DIR1/$tdir/$tfile || + error "$MCREATE $DIR1/$tdir/$tfile failed" #define OBD_FAIL_MDS_LL_BLOCK 0x172 do_facet $SINGLEMDS $LCTL set_param fail_loc=0x172 @@ -2458,8 +2491,14 @@ test_51c() { # write something to the file, it should be blocked on fetching layout dd if=/dev/zero of=$DIR2/$tdir/$tfile bs=1k count=1 conv=notrunc - local cnt=$($LFS getstripe -c $DIR2/$tdir/$tfile) - [ $cnt -eq $OSTCOUNT ] || error "have $cnt stripes, expected $OSTCOUNT" + local stripecnt=$($LFS getstripe -c $DIR2/$tdir/$tfile) + wait $pid + + # lod_qos.c::min_stripe_count() allows setstripe with a default stripe + # count to succeed with only 3/4 of the number of stripes (rounded up), + # so creating striped files does not fail if an OST is offline or full + [ $stripecnt -ge $((OSTCOUNT - $OSTCOUNT / 4)) ] || + error "layout wrong: getstripe -c $stripecnt < $OSTCOUNT * 3/4" rm -fr $DIR1/$tdir } @@ -2496,7 +2535,7 @@ test_54_part1() echo "==> rename vs getattr vs setxattr should not deadlock" mkdir -p $DIR/d1/d2/d3 || error "(1) mkdir failed" - do_facet mds $LCTL set_param fail_loc=$1 + do_facet mds1 $LCTL set_param fail_loc=$1 mv -T $DIR/d1/d2/d3 $DIR/d1/d3 & PID1=$! @@ -2518,7 +2557,7 @@ test_54_part2() { echo "==> rename vs getattr vs open vs getattr should not deadlock" mkdir -p $DIR/d1/d2/d3 || error "(1) mkdir failed" - do_facet mds $LCTL set_param fail_loc=$1 + do_facet mds1 $LCTL set_param fail_loc=$1 mv -T $DIR/d1/d2/d3 $DIR/d1/d3 & PID1=$! @@ -2563,7 +2602,7 @@ test_55a() { mkdir -p $DIR/d1/d2 $DIR/d3 || error "(1) mkdir failed" #define OBD_FAIL_MDS_RENAME4 0x156 - do_facet mds $LCTL set_param fail_loc=0x80000156 + do_facet mds1 $LCTL set_param fail_loc=0x80000156 mv -T $DIR/d1/d2 $DIR/d3/d2 & PID1=$! @@ -2581,7 +2620,7 @@ test_55b() mkdir -p $DIR/d1/d2 $DIR/d3 || error "(1) mkdir failed" #define OBD_FAIL_MDS_RENAME4 0x156 - do_facet mds $LCTL set_param fail_loc=0x80000156 + do_facet mds1 $LCTL set_param fail_loc=0x80000156 mv -T $DIR/d1/d2 $DIR/d3/d2 & PID1=$! @@ -2599,7 +2638,7 @@ test_55c() mkdir -p $DIR/d1/d2 $DIR/d3 || error "(1) mkdir failed" #define OBD_FAIL_MDS_RENAME4 0x156 - do_facet mds $LCTL set_param fail_loc=0x156 + do_facet mds1 $LCTL set_param fail_loc=0x156 mv -T $DIR/d1/d2 $DIR/d3/d2 & PID1=$! @@ -2627,7 +2666,7 @@ test_55d() touch $DIR/f1 #define OBD_FAIL_MDS_RENAME3 0x155 - do_facet mds $LCTL set_param fail_loc=0x155 + do_facet mds1 $LCTL set_param fail_loc=0x155 mv $DIR/f1 $DIR/$tdir & PID1=$! sleep 2 @@ -2731,7 +2770,7 @@ test_70b() { # LU-2781 } run_test 70b "remove files after calling rm_entry" -test_71() { +test_71a() { local server_version=$(lustre_version_code $SINGLEMDS) [[ $server_version -lt $(version_code 2.1.6) ]] && @@ -2740,7 +2779,8 @@ test_71() { # Patch not applied to 2.2 and 2.3 branches [[ $server_version -ge $(version_code 2.2.0) ]] && [[ $server_version -lt $(version_code 2.4.0) ]] && - skip "Need MDS version at least 2.4.0" && return + skip "Need MDS version earlier than 2.2.0 or at least 2.4.0" && + return checkfiemap --test || { skip "checkfiemap not runnable: $?" && return; } @@ -2772,7 +2812,34 @@ test_71() { echo $can2 [ $can3 -eq $can4 ] || error $((can2-can1)) "cancel RPC occured." } -run_test 71 "correct file map just after write operation is finished" +run_test 71a "correct file map just after write operation is finished" + +test_71b() { + local server_version=$(lustre_version_code $SINGLEMDS) + + [[ $server_version -lt $(version_code 2.1.6) ]] && + skip "Need MDS version at least 2.1.6" && return + + # Patch not applied to 2.2 and 2.3 branches + [[ $server_version -ge $(version_code 2.2.0) ]] && + [[ $server_version -lt $(version_code 2.4.0) ]] && + skip "Need MDS version earlier than 2.2.0 or at least 2.4.0" && + return + [[ $OSTCOUNT -ge 2 ]] || { skip "need at least 2 osts"; return; } + + checkfiemap --test || + { skip "error $?: checkfiemap failed" && return; } + + mkdir -p $DIR1/$tdir + + $LFS setstripe -c -1 $DIR1/$tdir || error "setstripe failed" + dd if=/dev/urandom of=$DIR1/$tdir/$tfile bs=40K count=1 + [ "$(facet_fstype ost$(($($GETSTRIPE -i $DIR1/$tdir/$tfile) + 1)))" = \ + "zfs" ] && + skip "ORI-366/LU-1941: FIEMAP unimplemented on ZFS" && return 0 + checkfiemap $DIR1/$tdir/$tfile 40960 || error "checkfiemap failed" +} +run_test 71b "check fiemap support for stripecount > 1" test_72() { local p="$TMP/sanityN-$TESTNAME.parameters" @@ -2817,14 +2884,16 @@ test_73() { error "setfattr1 failed" getfattr -n user.attr1 $DIR2/$tfile || error "getfattr1 failed" getfattr -n user.attr1 $DIR1/$tfile || error "getfattr2 failed" - clear_llite_stats + clear_stats llite.*.stats # PR lock should be cached by now on both clients getfattr -n user.attr1 $DIR1/$tfile || error "getfattr3 failed" # 2 hits for getfattr(0)+getfattr(size) - [ $(calc_llite_stats getxattr_hits) -eq 2 ] || error "not cached in $DIR1" + [ $(calc_stats llite.*.stats getxattr_hits) -eq 2 ] || + error "not cached in $DIR1" getfattr -n user.attr1 $DIR2/$tfile || error "getfattr4 failed" # 4 hits for more getfattr(0)+getfattr(size) - [ $(calc_llite_stats getxattr_hits) -eq 4 ] || error "not cached in $DIR2" + [ $(calc_stats llite.*.stats getxattr_hits) -eq 4 ] || + error "not cached in $DIR2" rm -f $DIR2/$tfile restore_lustre_params < $p @@ -2869,7 +2938,7 @@ test_76() { #LU-946 skip "Need MDS version at least 2.5.53" && return remote_mds_nodsh && skip "remote MDS with nodsh" && return - local fcount=2048 + local fcount=$((MDSCOUNT * 256)) declare -a fd_list declare -a fid_list @@ -2885,20 +2954,28 @@ test_76() { #LU-946 # drop all open locks and close any cached "open" files on the client cancel_lru_locks mdc - echo -n "open files " - ulimit -n 8096 + local open_fids_cmd="$LCTL get_param -n mdt.*.exports.'$nid'.open_files" + local fid_list=($(do_nodes $(comma_list $(mdts_nodes)) $open_fids_cmd)) + local already=${#fid_list[@]} + for (( i = 0; i < $already; i++ )) ; do + log "already open[$i]: $($LFS fid2path $DIR2 ${fid_list[i]})" + done + + echo -n "opening files: " + ulimit -n $((fcount + 50)) for ((i = 0; i < $fcount; i++)); do touch $DIR/$tdir/f_$i - local fd=$(free_fd) - local cmd="exec $fd<$DIR/$tdir/f_$i" - eval $cmd + local fd=$(free_fd ${fd_list[i]}) + local open_cmd="exec $fd<$DIR/$tdir/f_$i" + eval $open_cmd + fd_list[i]=$fd - echo -n "." + + (( $i % 32 == 0 )) && echo -n "." done echo - local get_open_fids="$LCTL get_param -n mdt.*.exports.'$nid'.open_files" - local fid_list=($(do_nodes $(comma_list $(mdts_nodes)) $get_open_fids)) + fid_list=($(do_nodes $(comma_list $(mdts_nodes)) $open_fids_cmd)) # Possible errors in openfiles FID list. # 1. Missing FIDs. Check 1 @@ -2908,19 +2985,28 @@ test_76() { #LU-946 # 5. Valid FID, points to some other file. Check 3 # Check 1 - [ ${#fid_list[@]} -ne $fcount ] && - error "${#fid_list[@]} != $fcount open files" - - for (( i = 0; i < $fcount; i++ )) ; do - cmd="exec ${fd_list[i]} /dev/null 2>&1 + do_nodes $CLIENTS $myRUNAS \ + dd if=/dev/zero of="$dir/nrs_r_$HOSTNAME" bs=1M count=$n || + error "dd at 0 on client failed (1)" for ((i = 0; i < $n; i++)); do - do_nodes $CLIENTS $myRUNAS dd if=/dev/zero\ - of="$dir/nrs_w_$HOSTNAME" bs=1M seek=$i count=1\ - > /dev/null 2>&1 & + do_nodes $CLIENTS $myRUNAS dd if=/dev/zero \ + of="$dir/nrs_w_$HOSTNAME" bs=1M seek=$i count=1 || + error "dd at ${i}MB on client failed (2)" & local pids_w[$i]=$! done do_nodes $CLIENTS sync; cancel_lru_locks osc for ((i = 0; i < $n; i++)); do - do_nodes $CLIENTS $myRUNAS dd if="$dir/nrs_w_$HOSTNAME"\ - of=/dev/zero bs=1M seek=$i count=1 > /dev/null 2>&1 & + do_nodes $CLIENTS $myRUNAS dd if="$dir/nrs_w_$HOSTNAME" \ + of=/dev/zero bs=1M seek=$i count=1 > /dev/null || + error "dd at ${i}MB on client failed (3)" & local pids_r[$i]=$! done cancel_lru_locks osc @@ -2967,26 +3055,30 @@ nrs_write_read() { } test_77a() { #LU-3266 - do_facet $SINGLEMDS lctl set_param ost.OSS.*.nrs_policies="fifo" + oss=$(comma_list $(osts_nodes)) + do_nodes $oss lctl set_param ost.OSS.*.nrs_policies="fifo" nrs_write_read return 0 } run_test 77a "check FIFO NRS policy" - test_77b() { #LU-3266 - do_facet $SINGLEMDS lctl set_param ost.OSS.*.nrs_policies="crrn" - do_facet $SINGLEMDS lctl set_param ost.OSS.*.nrs_crrn_quantum=1 + oss=$(comma_list $(osts_nodes)) + + do_nodes $oss lctl set_param ost.OSS.*.nrs_policies="crrn" \ + ost.OSS.*.nrs_crrn_quantum=1 echo "policy: crr-n, crrn_quantum 1" nrs_write_read - do_facet $SINGLEMDS lctl set_param ost.OSS.*.nrs_crrn_quantum=64 + do_nodes $oss lctl set_param ost.OSS.*.nrs_crrn_quantum=64 echo "policy: crr-n, crrn_quantum 64" nrs_write_read + # cleanup + do_nodes $oss lctl set_param ost.OSS.ost_io.nrs_policies="fifo" return 0 } run_test 77b "check CRR-N NRS policy" @@ -2994,44 +3086,29 @@ run_test 77b "check CRR-N NRS policy" orr_trr() { local policy=$1 - for i in $(seq 1 $OSTCOUNT) - do - do_facet ost"$i" lctl set_param \ - ost.OSS.ost_io.nrs_policies=$policy - do_facet ost"$i" lctl set_param \ - ost.OSS.*.nrs_"$policy"_quantum=1 - do_facet ost"$i" lctl set_param \ - ost.OSS.*.nrs_"$policy"_offset_type="physical" - do_facet ost"$i" lctl set_param \ - ost.OSS.*.nrs_"$policy"_supported="reads" - done + oss=$(comma_list $(osts_nodes)) - echo "policy: $policy, ${policy}_quantum 1, ${policy}_offset_type \ - physical, ${policy}_supported reads" + do_nodes $oss lctl set_param ost.OSS.ost_io.nrs_policies=$policy \ + ost.OSS.*.nrs_"$policy"_quantum=1 \ + ost.OSS.*.nrs_"$policy"_offset_type="physical" \ + ost.OSS.*.nrs_"$policy"_supported="reads" + + echo "policy: $policy, ${policy}_quantum 1, ${policy}_offset_type physical, ${policy}_supported reads" nrs_write_read - for i in $(seq 1 $OSTCOUNT) - do - do_facet ost"$i" lctl set_param \ - ost.OSS.*.nrs_${policy}_supported="writes" - do_facet ost"$i" lctl set_param \ - ost.OSS.*.nrs_${policy}_quantum=64 - done - echo "policy: $policy, ${policy}_quantum 64, \ - ${policy}_offset_type physical, ${policy}_supported writes" + do_nodes $oss lctl set_param ost.OSS.*.nrs_${policy}_supported="writes" \ + ost.OSS.*.nrs_${policy}_quantum=64 + + echo "policy: $policy, ${policy}_quantum 64, ${policy}_offset_type physical, ${policy}_supported writes" nrs_write_read - for i in $(seq 1 $OSTCOUNT) - do - do_facet ost"$i" lctl set_param \ - ost.OSS.*.nrs_${policy}_supported="reads_and_writes" - do_facet ost"$i" lctl set_param \ - ost.OSS.*.nrs_${policy}_offset_type="logical" - done - echo "policy: $policy, ${policy}_quantum 64, \ - ${policy}_offset_type logical, ${policy}_supported reads_and_writes" + do_nodes $oss lctl set_param ost.OSS.*.nrs_${policy}_supported="reads_and_writes" \ + ost.OSS.*.nrs_${policy}_offset_type="logical" + echo "policy: $policy, ${policy}_quantum 64, ${policy}_offset_type logical, ${policy}_supported reads_and_writes" nrs_write_read + # cleanup + do_nodes $oss lctl set_param ost.OSS.ost_io.nrs_policies="fifo" return 0 } @@ -3053,33 +3130,75 @@ tbf_rule_operate() shift 1 do_facet $facet lctl set_param \ - ost.OSS.ost_io.nrs_tbf_rule="$@" + ost.OSS.ost_io.nrs_tbf_rule="$*" [ $? -ne 0 ] && - error "failed to operate on TBF rules" + error "failed to run operate '$*' on TBF rules" +} + +tbf_verify() { + local dir=$DIR/$tdir + local client1=${CLIENT1:-$(hostname)} + local myRUNAS="$3" + + mkdir $dir || error "mkdir $dir failed" + $LFS setstripe -c 1 $dir || error "setstripe to $dir failed" + chmod 777 $dir + + echo "Limited write rate: $1, read rate: $2" + echo "Verify the write rate is under TBF control" + local rate=$(do_node $client1 $myRUNAS dd if=/dev/zero of=$dir/tbf \ + bs=1M count=100 oflag=direct 2>&1 | awk '/bytes/ {print $8}') + echo "Write speed is $rate" + + # verify the write rate does not exceed 110% of TBF limited rate + [ $(bc <<< "$rate < 1.1 * $1") -eq 1 ] || + error_ignore LU-9140 "The write rate ($rate) exceeds 110% of preset rate ($1)" + + cancel_lru_locks osc + + echo "Verify the read rate is under TBF control" + rate=$(do_node $client1 $myRUNAS dd if=$dir/tbf of=/dev/null \ + bs=1M count=100 iflag=direct 2>&1 | awk '/bytes/ {print $8}') + echo "Read speed is $rate" + + # verify the read rate does not exceed 110% of TBF limited rate + [ $(bc <<< "$rate < 1.1 * $2") -eq 1 ] || + error_ignore LU-9140 "The read rate ($rate) exceeds 110% of preset rate ($2)" + + cancel_lru_locks osc + rm -rf $dir || error "rm -rf $dir failed" } test_77e() { - for i in $(seq 1 $OSTCOUNT) - do - do_facet ost"$i" lctl set_param \ - ost.OSS.ost_io.nrs_policies="tbf\ nid" - [ $? -ne 0 ] && - error "failed to set TBF policy" - done + local server_version=$(lustre_version_code ost1) + [[ $server_version -ge $(version_code 2.7.58) ]] || + { skip "Need server version newer than 2.7.57"; return 0; } + + oss=$(comma_list $(osts_nodes)) + + do_nodes $oss lctl set_param ost.OSS.ost_io.nrs_policies="tbf\ nid" + [ $? -ne 0 ] && error "failed to set TBF policy" + + local idis + local rateis + if [ $(lustre_version_code ost1) -ge $(version_code 2.8.54) ]; then + idis="nid=" + rateis="rate=" + fi # Only operate rules on ost1 since OSTs might run on the same OSS # Add some rules - tbf_rule_operate ost1 "start\ localhost\ {0@lo}\ 1000" + tbf_rule_operate ost1 "start\ localhost\ ${idis}{0@lo}\ ${rateis}1000" local address=$(comma_list "$(host_nids_address $CLIENTS $NETTYPE)") local client_nids=$(nids_list $address "\\") - tbf_rule_operate ost1 "start\ clients\ {$client_nids}\ 100" - tbf_rule_operate ost1 "start\ others\ {*.*.*.*@$NETTYPE}\ 50" + tbf_rule_operate ost1 "start\ clients\ ${idis}{$client_nids}\ ${rateis}100" + tbf_rule_operate ost1 "start\ others\ ${idis}{*.*.*.*@$NETTYPE}\ ${rateis}50" nrs_write_read # Change the rules - tbf_rule_operate ost1 "change\ localhost\ 1001" - tbf_rule_operate ost1 "change\ clients\ 101" - tbf_rule_operate ost1 "change\ others\ 51" + tbf_rule_operate ost1 "change\ localhost\ ${rateis}1001" + tbf_rule_operate ost1 "change\ clients\ ${rateis}101" + tbf_rule_operate ost1 "change\ others\ ${rateis}51" nrs_write_read # Stop the rules @@ -3089,19 +3208,20 @@ test_77e() { nrs_write_read # Cleanup the TBF policy - for i in $(seq 1 $OSTCOUNT) - do - do_facet ost"$i" lctl set_param \ - ost.OSS.ost_io.nrs_policies="fifo" - [ $? -ne 0 ] && - error "failed to set policy back to fifo" - done + do_nodes $oss lctl set_param ost.OSS.ost_io.nrs_policies="fifo" + [ $? -ne 0 ] && error "failed to set policy back to fifo" nrs_write_read return 0 } run_test 77e "check TBF NID nrs policy" test_77f() { + local server_version=$(lustre_version_code ost1) + [[ $server_version -ge $(version_code 2.7.58) ]] || + { skip "Need server version newer than 2.7.57"; return 0; } + + oss=$(comma_list $(osts_nodes)) + # Configure jobid_var local saved_jobid_var=$($LCTL get_param -n jobid_var) if [ $saved_jobid_var != procname_uid ]; then @@ -3110,25 +3230,27 @@ test_77f() { "$FSNAME.sys.jobid_var" procname_uid fi - for i in $(seq 1 $OSTCOUNT) - do - do_facet ost"$i" lctl set_param \ - ost.OSS.ost_io.nrs_policies="tbf\ jobid" - [ $? -ne 0 ] && - error "failed to set TBF policy" - done + do_nodes $oss lctl set_param ost.OSS.ost_io.nrs_policies="tbf\ jobid" + [ $? -ne 0 ] && error "failed to set TBF policy" + + local idis + local rateis + if [ $(lustre_version_code ost1) -ge $(version_code 2.8.54) ]; then + idis="jobid=" + rateis="rate=" + fi # Only operate rules on ost1 since OSTs might run on the same OSS # Add some rules - tbf_rule_operate ost1 "start\ runas\ {iozone.$RUNAS_ID\ dd.$RUNAS_ID\ tiotest.$RUNAS_ID}\ 1000" - tbf_rule_operate ost1 "start\ iozone_runas\ {iozone.$RUNAS_ID}\ 100" - tbf_rule_operate ost1 "start\ dd_runas\ {dd.$RUNAS_ID}\ 50" + tbf_rule_operate ost1 "start\ runas\ ${idis}{iozone.$RUNAS_ID\ dd.$RUNAS_ID\ tiotest.$RUNAS_ID}\ ${rateis}1000" + tbf_rule_operate ost1 "start\ iozone_runas\ ${idis}{iozone.$RUNAS_ID}\ ${rateis}100" + tbf_rule_operate ost1 "start\ dd_runas\ ${idis}{dd.$RUNAS_ID}\ ${rateis}50" nrs_write_read "$RUNAS" # Change the rules - tbf_rule_operate ost1 "change\ runas\ 1001" - tbf_rule_operate ost1 "change\ iozone_runas\ 101" - tbf_rule_operate ost1 "change\ dd_runas\ 51" + tbf_rule_operate ost1 "change\ runas\ ${rateis}1001" + tbf_rule_operate ost1 "change\ iozone_runas\ ${rateis}101" + tbf_rule_operate ost1 "change\ dd_runas\ ${rateis}51" nrs_write_read "$RUNAS" # Stop the rules @@ -3138,13 +3260,8 @@ test_77f() { nrs_write_read "$RUNAS" # Cleanup the TBF policy - for i in $(seq 1 $OSTCOUNT) - do - do_facet ost"$i" lctl set_param \ - ost.OSS.ost_io.nrs_policies="fifo" - [ $? -ne 0 ] && - error "failed to set policy back to fifo" - done + do_nodes $oss lctl set_param ost.OSS.ost_io.nrs_policies="fifo" + [ $? -ne 0 ] && error "failed to set policy back to fifo" nrs_write_read "$RUNAS" local current_jobid_var=$($LCTL get_param -n jobid_var) @@ -3158,13 +3275,94 @@ test_77f() { run_test 77f "check TBF JobID nrs policy" test_77g() { - for i in $(seq 1 $OSTCOUNT) - do - do_facet ost"$i" lctl set_param \ - ost.OSS.ost_io.nrs_policies="tbf\ nid" - [ $? -ne 0 ] && - error "failed to set TBF policy" + local server_version=$(lustre_version_code ost1) + [[ $server_version -ge $(version_code 2.7.58) ]] || + { skip "Need server version newer than 2.7.57"; return 0; } + + oss=$(comma_list $(osts_nodes)) + + do_nodes $oss lctl set_param ost.OSS.ost_io.nrs_policies="tbf\ nid" + [ $? -ne 0 ] && error "failed to set TBF policy" + + do_nodes $oss lctl set_param ost.OSS.ost_io.nrs_policies="tbf\ jobid" + [ $? -ne 0 ] && error "failed to set TBF policy" + + local idis + local rateis + if [ $(lustre_version_code ost1) -ge $(version_code 2.8.54) ]; then + idis="jobid=" + rateis="rate=" + fi + + # Add a rule that only valid for Jobid TBF. If direct change between + # TBF types is not supported, this operation will fail. + tbf_rule_operate ost1 "start\ dd_runas\ ${idis}{dd.$RUNAS_ID}\ ${rateis}50" + + # Cleanup the TBF policy + do_nodes $oss lctl set_param ost.OSS.ost_io.nrs_policies="fifo" + [ $? -ne 0 ] && error "failed to set policy back to fifo" + return 0 +} +run_test 77g "Change TBF type directly" + +test_77h() { + [ $(lustre_version_code ost1) -ge $(version_code 2.8.55) ] || + { skip "Need OST version at least 2.8.55"; return 0; } + + local old_policy=$(do_facet ost1 \ + lctl get_param ost.OSS.ost_io.nrs_policies) + local new_policy + + do_facet ost1 lctl set_param \ + ost.OSS.ost_io.nrs_policies="abc" + [ $? -eq 0 ] && error "should return error" + + do_facet ost1 lctl set_param \ + ost.OSS.ost_io.nrs_policies="tbf\ abc" + [ $? -eq 0 ] && error "should return error" + + do_facet ost1 lctl set_param \ + ost.OSS.ost_io.nrs_policies="tbf\ reg\ abc" + [ $? -eq 0 ] && error "should return error" + + do_facet ost1 lctl set_param \ + ost.OSS.ost_io.nrs_policies="tbf\ abc\ efg" + [ $? -eq 0 ] && error "should return error" + + new_policy=$(do_facet ost1 lctl get_param ost.OSS.ost_io.nrs_policies) + [ $? -eq 0 ] || error "shouldn't LBUG" + + [ "$old_policy" = "$new_policy" ] || error "NRS policy should be same" + + return 0 +} +run_test 77h "Wrong policy name should report error, not LBUG" + +tbf_rule_check() +{ + local facet=$1 + local expected=$2 + local error_message=$3 + local rule_number=0 + for rule in $expected; do + rule_number=$((rule_number + 1)) done + local stop_line=$(($rule_number + 3)) + local awk_command="awk 'NR >= 4 && NR <= $stop_line {print \$1}'" + + local output=$(do_facet $facet lctl get_param \ + ost.OSS.ost_io.nrs_tbf_rule | + eval $awk_command | + tr "\n" " " | + sed 's/[ ]*$//') + if [ "$output" != "$expected" ]; then + error "$error_message, expected '$expected', got '$output'" + fi +} + +test_77i() { + [ $(lustre_version_code ost1) -ge $(version_code 2.8.55) ] || + { skip "Need OST version at least 2.8.55"; return 0; } for i in $(seq 1 $OSTCOUNT) do @@ -3174,52 +3372,233 @@ test_77g() { error "failed to set TBF policy" done - # Add a rule that only valid for Jobid TBF. If direct change between - # TBF types is not supported, this operation will fail. - tbf_rule_operate ost1 "start\ dd_runas\ {dd.$RUNAS_ID}\ 50" + tbf_rule_check ost1 "default" "error before inserting any rule" + + tbf_rule_operate ost1 "start\ before\ jobid={jobid}\ rate=1000" + tbf_rule_check ost1 "before default" \ + "error when inserting rule 'before'" + + tbf_rule_operate ost1 "start\ after\ jobid={jobid}\ rate=1000\ rank=default" + tbf_rule_check ost1 "before after default" \ + "error when inserting rule 'after'" + + tbf_rule_operate ost1 "start\ target\ jobid={jobid}\ rate=1000\ rank=after" + tbf_rule_check ost1 "before target after default" \ + "error when inserting rule 'target'" + + echo "Move before itself" + tbf_rule_operate ost1 "change\ target\ rank=target" + tbf_rule_check ost1 "before target after default" \ + "error when moving before itself" + + echo "Move to higher rank" + tbf_rule_operate ost1 "change\ target\ rank=before" + tbf_rule_check ost1 "target before after default" \ + "error when moving to higher rank" + + echo "Move to lower rank" + tbf_rule_operate ost1 "change\ target\ rank=after" + tbf_rule_check ost1 "before target after default" \ + "error when moving to lower rank" + + echo "Move before default" + tbf_rule_operate ost1 "change\ target\ rank=default" + tbf_rule_check ost1 "before after target default" \ + error "error when moving before default" # Cleanup the TBF policy - for i in $(seq 1 $OSTCOUNT) - do - do_facet ost"$i" lctl set_param \ + do_nodes $(comma_list $(osts_nodes)) \ + $LCTL set_param ost.OSS.ost_io.nrs_policies=fifo + return 0 +} +run_test 77i "Change rank of TBF rule" + +test_77j() { + local idis + local rateis + if [ $(lustre_version_code ost1) -ge $(version_code 2.8.60) ]; then + idis="opcode=" + rateis="rate=" + fi + + do_nodes $(comma_list $(osts_nodes)) \ + lctl set_param jobid_var=procname_uid \ + ost.OSS.ost_io.nrs_policies="tbf\ opcode" \ + ost.OSS.ost_io.nrs_tbf_rule="start\ ost_r\ ${idis}{ost_read}\ ${rateis}5" \ + ost.OSS.ost_io.nrs_tbf_rule="start\ ost_w\ ${idis}{ost_write}\ ${rateis}20" + [ $? -ne 0 ] && error "failed to set TBF OPCode policy" + + nrs_write_read + tbf_verify 20 5 + + do_nodes $(comma_list $(osts_nodes)) \ + lctl set_param ost.OSS.ost_io.nrs_tbf_rule="stop\ ost_r" \ + ost.OSS.ost_io.nrs_tbf_rule="stop\ ost_w" \ ost.OSS.ost_io.nrs_policies="fifo" - [ $? -ne 0 ] && - error "failed to set policy back to fifo" - done + + # sleep 3 seconds to wait the tbf policy stop completely, + # or the next test case is possible get -EAGAIN when + # setting the tbf policy + sleep 3 +} +run_test 77j "check TBF-OPCode NRS policy" + +test_77k() { + [[ $(lustre_version_code ost1) -ge $(version_code 2.9.53) ]] || + { skip "Need OST version at least 2.9.53"; return 0; } + + do_nodes $(comma_list $(osts_nodes)) \ + lctl set_param ost.OSS.ost_io.nrs_policies="tbf" \ + ost.OSS.ost_io.nrs_tbf_rule="start\ ext_w\ jobid={dd.$RUNAS_ID}\&opcode={ost_write}\ rate=20" \ + ost.OSS.ost_io.nrs_tbf_rule="start\ ext_r\ jobid={dd.$RUNAS_ID}\&opcode={ost_read}\ rate=10" + + nrs_write_read "$RUNAS" + tbf_verify 20 10 "$RUNAS" + + local address=$(comma_list "$(host_nids_address $CLIENTS $NETTYPE)") + local client_nids=$(nids_list $address "\\") + do_nodes $(comma_list $(osts_nodes)) \ + lctl set_param ost.OSS.ost_io.nrs_tbf_rule="stop\ ext_w" \ + ost.OSS.ost_io.nrs_tbf_rule="stop\ ext_r" \ + ost.OSS.ost_io.nrs_tbf_rule="start\ ext_w\ nid={0@lo\ $client_nids}\&opcode={ost_write}\ rate=20" \ + ost.OSS.ost_io.nrs_tbf_rule="start\ ext_r\ nid={0@lo\ $client_nids}\&opcode={ost_read}\ rate=10" + + nrs_write_read + tbf_verify 20 10 + + do_nodes $(comma_list $(osts_nodes)) \ + lctl set_param ost.OSS.ost_io.nrs_tbf_rule="stop\ ext_w" \ + ost.OSS.ost_io.nrs_tbf_rule="stop\ ext_r" \ + ost.OSS.ost_io.nrs_tbf_rule="start\ ext\ nid={0@lo\ $client_nids}\&jobid={dd.$RUNAS_ID}\ rate=20" + + nrs_write_read "$RUNAS" + tbf_verify 20 20 "$RUNAS" + + do_nodes $(comma_list $(osts_nodes)) \ + lctl set_param ost.OSS.ost_io.nrs_tbf_rule="stop\ ext" \ + ost.OSS.ost_io.nrs_tbf_rule="start\ ext_a\ jobid={dd.$RUNAS_ID},opcode={ost_write}\ rate=20" \ + ost.OSS.ost_io.nrs_tbf_rule="start\ ext_b\ jobid={dd.$RUNAS_ID},opcode={ost_read}\ rate=10" + + nrs_write_read "$RUNAS" + # with parameter "RUNAS", it will match the latest rule + # "ext_b" first, so the limited write rate is 10. + tbf_verify 10 10 "$RUNAS" + tbf_verify 20 10 + + do_nodes $(comma_list $(osts_nodes)) \ + lctl set_param ost.OSS.ost_io.nrs_tbf_rule="stop\ ext_a" \ + ost.OSS.ost_io.nrs_tbf_rule="stop\ ext_b" \ + ost.OSS.ost_io.nrs_policies="fifo" + + sleep 3 +} +run_test 77k "check the extended TBF policy with NID/JobID/OPCode expression" + +test_77l() { + if [ $(lustre_version_code ost1) -lt $(version_code 2.9.54) ]; then + skip "Need OST version at least 2.9.54" + return 0 + fi + + local dir=$DIR/$tdir + + mkdir $dir || error "mkdir $dir failed" + $LFS setstripe -c $OSTCOUNT $dir || error "setstripe to $dir failed" + chmod 777 $dir + + local nodes=$(comma_list $(osts_nodes)) + do_nodes $nodes lctl set_param ost.OSS.ost_io.nrs_policies=delay \ + ost.OSS.ost_io.nrs_delay_min=4 \ + ost.OSS.ost_io.nrs_delay_max=4 \ + ost.OSS.ost_io.nrs_delay_pct=100 + [ $? -ne 0 ] && error "Failed to set delay policy" + + local start=$SECONDS + do_nodes "${SINGLECLIENT:-$HOSTNAME}" "$RUNAS" \ + dd if=/dev/zero of="$dir/nrs_delay_$HOSTNAME" bs=1M count=1 \ + oflag=direct conv=fdatasync || + { do_nodes $nodes lctl set_param ost.OSS.ost_io.nrs_policies="fifo"; + error "dd on client failed (1)"; } + local elapsed=$((SECONDS - start)) + + # NRS delay doesn't do sub-second timing, so a request enqueued at + # 0.9 seconds can be dequeued at 4.0 + [ $elapsed -lt 3 ] && + { do_nodes $nodes lctl set_param ost.OSS.ost_io.nrs_policies="fifo"; + error "Single 1M write should take at least 3 seconds"; } + + start=$SECONDS + do_nodes "${SINGLECLIENT:-$HOSTNAME}" "$RUNAS" \ + dd if=/dev/zero of="$dir/nrs_delay_$HOSTNAME" bs=1M count=10 \ + oflag=direct conv=fdatasync || + { do_nodes $nodes lctl set_param ost.OSS.ost_io.nrs_policies="fifo"; + error "dd on client failed (2)"; } + elapsed=$((SECONDS - start)) + + [ $elapsed -lt 30 ] && + { do_nodes $nodes lctl set_param ost.OSS.ost_io.nrs_policies="fifo"; + error "Ten 1M writes should take at least 30 seconds"; } + + do_nodes $nodes lctl set_param ost.OSS.ost_io.nrs_policies="fifo" + [ $? -ne 0 ] && error "failed to set policy back to fifo" + return 0 } -run_test 77g "Change TBF type directly" +run_test 77l "check NRS Delay slows write RPC processing" test_78() { #LU-6673 + local server_version=$(lustre_version_code ost1) + [[ $server_version -ge $(version_code 2.7.58) ]] || + { skip "Need server version newer than 2.7.57"; return 0; } + local rc - for i in $(seq 1 $OSTCOUNT) - do - do_facet ost"$i" lctl set_param \ - ost.OSS.ost_io.nrs_policies="orr" & - do_facet ost"$i" lctl set_param \ - ost.OSS.*.nrs_orr_quantum=1 - rc=$? - # Valid return codes are: - # 0: Tuning succeeded - # ENODEV: Policy is still stopped - # EAGAIN: Policy is being initialized - [ $rc -eq 0 -o $rc -eq 19 -o $rc -eq 11 ] || - error "Expected set_param to return 0|ENODEV|EAGAIN" - done + oss=$(comma_list $(osts_nodes)) + do_nodes $oss lctl set_param ost.OSS.ost_io.nrs_policies="orr" & + do_nodes $oss lctl set_param ost.OSS.*.nrs_orr_quantum=1 + rc=$? + # Valid return codes are: + # 0: Tuning succeeded + # ENODEV: Policy is still stopped + # EAGAIN: Policy is being initialized + [ $rc -eq 0 -o $rc -eq 19 -o $rc -eq 11 ] || + error "Expected set_param to return 0|ENODEV|EAGAIN" # Cleanup the ORR policy - for i in $(seq 1 $OSTCOUNT) - do - do_facet ost"$i" lctl set_param \ - ost.OSS.ost_io.nrs_policies="fifo" - [ $? -ne 0 ] && - error "failed to set policy back to fifo" - done + do_nodes $oss lctl set_param ost.OSS.ost_io.nrs_policies="fifo" + [ $? -ne 0 ] && error "failed to set policy back to fifo" return 0 } run_test 78 "Enable policy and specify tunings right away" +test_79() { + remote_mds_nodsh && skip "remote MDS with nodsh" && return + test_mkdir -p $DIR/$tdir + + # Prevent interference from layout intent RPCs due to + # asynchronous writeback. These will be tested in 130c below. + do_nodes ${CLIENTS:-$HOSTNAME} sync + + setfattr -n trusted.name1 -v value1 $DIR/$tdir || + error "setfattr -n trusted.name1=value1 $DIR/$tdir failed" + +#define OBD_FAIL_MDS_INTENT_DELAY 0x160 + local mdtidx=$($LFS getstripe -M $DIR/$tdir) + local facet=mds$((mdtidx + 1)) + stat $DIR/$tdir + set_nodes_failloc $(facet_active_host $facet) 0x80000160 + getfattr -n trusted.name1 $DIR/$tdir 2> /dev/null & + local pid=$! + sleep 2 + +#define OBD_FAIL_MDS_GETXATTR_PACK 0x131 + set_nodes_failloc $(facet_active_host $facet) 0x80000131 + + wait $pid + return 0 +} +run_test 79 "xattr: intent error" + test_80a() { [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return local MDTIDX=1 @@ -3399,7 +3778,7 @@ test_81() { touch d0/0 || error "create 0 failed" mv d0/0 d1/0 || error "rename d0/0 d1/0 failed" stat d0/0 && error "stat mv filed succeed" - mv $DIR2/$tdir/d1/0 $DIR2/$tdir/d0/0 || "rename d1/0 d0/0 failed" + mv $DIR2/$tdir/d1/0 $DIR2/$tdir/d0/0 || error "rename d1/0 d0/0 failed" stat d0/0 || error "stat failed" local t=$(ls -ai $DIR1/$tdir/d0 | sort -u | wc -l) @@ -3568,47 +3947,49 @@ test_92() { run_test 92 "create remote directory under orphan directory" test_93() { - dd if=/dev/zero of=$DIR2/$tfile bs=4k count=2 conv=fsync - - local before=$(date +%s) - local evict - - $LCTL mark write -#define OBD_FAIL_LDLM_PAUSE_CANCEL 0x312 - $LCTL set_param fail_val=5 fail_loc=0x80000312 - dd if=/dev/zero of=$DIR/$tfile conv=notrunc oflag=append bs=4k count=1 & - local pid=$! - sleep 2 - -#define OBD_FAIL_LDLM_PAUSE_CANCEL_LOCAL 0x329 - $LCTL set_param fail_val=6 fail_loc=0x80000329 - $LCTL mark kill $pid - kill -ALRM $pid - - dd if=/dev/zero of=$DIR2/$tfile conv=notrunc oflag=append bs=4k count=1 - - wait $pid - dd if=/dev/zero of=$DIR/$tfile bs=4k count=1 conv=fsync - - evict=$(do_facet client $LCTL get_param \ - osc.$FSNAME-OST*-osc-*/state | \ - awk -F"[ [,]" '/EVICTED ]$/ { if (t<$5) {t=$5;} } END { print t }') - - [ -z "$evict" ] || [[ $evict -le $before ]] || - (do_facet client $LCTL get_param \ - osc.$FSNAME-OST*-osc-*/state; - error "eviction happened: $evict before:$before") -} -run_test 93 "signal vs CP callback race" + local rc1=0 + local rc2=0 + local old_rr + + mkdir -p $DIR1/$tfile-1/ + mkdir -p $DIR2/$tfile-2/ + local old_rr=$(do_facet $SINGLEMDS lctl get_param -n \ + 'lod.lustre-MDT*/qos_threshold_rr' | sed -e 's/%//') + do_facet $SINGLEMDS lctl set_param -n \ + 'lod.lustre-MDT*/qos_threshold_rr' 100 + #define OBD_FAIL_MDS_LOV_CREATE_RACE 0x163 + do_facet $SINGLEMDS "lctl set_param fail_loc=0x00000163" + + $SETSTRIPE -c -1 $DIR1/$tfile-1/file1 & + local PID1=$! + sleep 1 + $SETSTRIPE -c -1 $DIR2/$tfile-2/file2 & + local PID2=$! + wait $PID2 + wait $PID1 + do_facet $SINGLEMDS "lctl set_param fail_loc=0x0" + do_facet $SINGLEMDS "lctl set_param -n \ + 'lod.lustre-MDT*/qos_threshold_rr' $old_rr" + + $GETSTRIPE $DIR1/$tfile-1/file1 + rc1=$($GETSTRIPE -q $DIR1/$tfile-1/file1 | + awk '{if (/[0-9]/) print $1 }' | sort | uniq -d | wc -l) + $GETSTRIPE $DIR2/$tfile-2/file2 + rc2=$($GETSTRIPE -q $DIR2/$tfile-2/file2 | + awk '{if (/[0-9]/) print $1 }' | sort | uniq -d | wc -l) + echo "rc1=$rc1 and rc2=$rc2 " + [ $rc1 -eq 0 ] && [ $rc2 -eq 0 ] || + error "object allocate on same ost detected" +} +run_test 93 "alloc_rr should not allocate on same ost" log "cleanup: ======================================================" # kill and wait in each test only guarentee script finish, but command in script # like 'rm' 'chmod' may still be running, wait for all commands to finish # otherwise umount below will fail -wait_update $HOSTNAME "fuser -m $MOUNT2" "" || true - -[ "$(mount | grep $MOUNT2)" ] && umount $MOUNT2 +[ "$(mount | grep $MOUNT2)" ] && wait_update $HOSTNAME "fuser -m $MOUNT2" "" || + true complete $SECONDS rm -f $SAMPLE_FILE