X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Ftests%2Frecovery-small.sh;h=7ef48f428312e08d2d2bac1c0bbbf2442bf8f8b8;hp=8d6acd523ffd1d77080ce6421ba16109734b8070;hb=5d0201db65d366d8905a28d103a2a9b511c22ca7;hpb=cc29a264f74c06765e3b31551890f06a81fe6b03 diff --git a/lustre/tests/recovery-small.sh b/lustre/tests/recovery-small.sh index 8d6acd5..7ef48f4 100755 --- a/lustre/tests/recovery-small.sh +++ b/lustre/tests/recovery-small.sh @@ -2,18 +2,21 @@ set -e -# bug 5494 7288 5493 -ALWAYS_EXCEPT="24 27 52 $RECOVERY_SMALL_EXCEPT" +# bug 5494 5493 +ALWAYS_EXCEPT="24 52 $RECOVERY_SMALL_EXCEPT" PTLDEBUG=${PTLDEBUG:--1} LUSTRE=${LUSTRE:-`dirname $0`/..} . $LUSTRE/tests/test-framework.sh init_test_env $@ . ${CONFIG:=$LUSTRE/tests/cfg/$NAME.sh} +init_logging + +require_dsh_mds || exit 0 # also long tests: 19, 21a, 21e, 21f, 23, 27 -# 1 2.5 4 4 (min)" -[ "$SLOW" = "no" ] && EXCEPT_SLOW="17 26b 50 51 57" +# 1 2.5 2.5 4 4 (min)" +[ "$SLOW" = "no" ] && EXCEPT_SLOW="17 26a 26b 50 51 57" build_test_filter @@ -22,209 +25,239 @@ build_test_filter SETUP=${SETUP:-""} CLEANUP=${CLEANUP:-""} -cleanup_and_setup_lustre +check_and_setup_lustre + +assert_DIR rm -rf $DIR/[df][0-9]* test_1() { - drop_request "mcreate $MOUNT/1" || return 1 - drop_reint_reply "mcreate $MOUNT/2" || return 2 + drop_request "mcreate $DIR/f1" || return 1 + drop_reint_reply "mcreate $DIR/f2" || return 2 } run_test 1 "mcreate: drop req, drop rep" test_2() { - drop_request "tchmod 111 $MOUNT/2" || return 1 - drop_reint_reply "tchmod 666 $MOUNT/2" || return 2 + drop_request "tchmod 111 $DIR/f2" || return 1 + drop_reint_reply "tchmod 666 $DIR/f2" || return 2 } run_test 2 "chmod: drop req, drop rep" test_3() { - drop_request "statone $MOUNT/2" || return 1 - drop_reply "statone $MOUNT/2" || return 2 + drop_request "statone $DIR/f2" || return 1 + drop_reply "statone $DIR/f2" || return 2 } run_test 3 "stat: drop req, drop rep" +SAMPLE_NAME=f0.recovery-small.junk +SAMPLE_FILE=$TMP/$SAMPLE_NAME +# make this big, else test 9 doesn't wait for bulk -- bz 5595 +dd if=/dev/urandom of=$SAMPLE_FILE bs=1M count=4 + test_4() { - do_facet client "cp /etc/inittab $MOUNT/inittab" || return 1 - drop_request "cat $MOUNT/inittab > /dev/null" || return 2 - drop_reply "cat $MOUNT/inittab > /dev/null" || return 3 + do_facet client "cp $SAMPLE_FILE $DIR/$SAMPLE_NAME" || return 1 + drop_request "cat $DIR/$SAMPLE_NAME > /dev/null" || return 2 + drop_reply "cat $DIR/$SAMPLE_NAME > /dev/null" || return 3 } run_test 4 "open: drop req, drop rep" +RENAMED_AGAIN=$DIR/f0.renamed-again + test_5() { - drop_request "mv $MOUNT/inittab $MOUNT/renamed" || return 1 - drop_reint_reply "mv $MOUNT/renamed $MOUNT/renamed-again" || return 2 - do_facet client "checkstat -v $MOUNT/renamed-again" || return 3 + drop_request "mv $DIR/$SAMPLE_NAME $DIR/$tfile-renamed" || return 1 + drop_reint_reply "mv $DIR/$tfile-renamed $RENAMED_AGAIN" || return 2 + do_facet client "checkstat -v $RENAMED_AGAIN" || return 3 } run_test 5 "rename: drop req, drop rep" -[ ! -e $MOUNT/renamed-again ] && cp /etc/inittab $MOUNT/renamed-again +[ ! -e $RENAMED_AGAIN ] && cp $SAMPLE_FILE $RENAMED_AGAIN +LINK1=$DIR/f0.link1 +LINK2=$DIR/f0.link2 + test_6() { - drop_request "mlink $MOUNT/renamed-again $MOUNT/link1" || return 1 - drop_reint_reply "mlink $MOUNT/renamed-again $MOUNT/link2" || return 2 + drop_request "mlink $RENAMED_AGAIN $LINK1" || return 1 + drop_reint_reply "mlink $RENAMED_AGAIN $LINK2" || return 2 } run_test 6 "link: drop req, drop rep" -[ ! -e $MOUNT/link1 ] && mlink $MOUNT/renamed-again $MOUNT/link1 -[ ! -e $MOUNT/link2 ] && mlink $MOUNT/renamed-again $MOUNT/link2 +[ ! -e $LINK1 ] && mlink $RENAMED_AGAIN $LINK1 +[ ! -e $LINK2 ] && mlink $RENAMED_AGAIN $LINK2 test_7() { - drop_request "munlink $MOUNT/link1" || return 1 - drop_reint_reply "munlink $MOUNT/link2" || return 2 + drop_request "munlink $LINK1" || return 1 + drop_reint_reply "munlink $LINK2" || return 2 } run_test 7 "unlink: drop req, drop rep" #bug 1423 test_8() { - drop_reint_reply "touch $MOUNT/$tfile" || return 1 + drop_reint_reply "touch $DIR/$tfile" || return 1 } run_test 8 "touch: drop rep (bug 1423)" #bug 1420 test_9() { - pause_bulk "cp /etc/profile $MOUNT/$tfile" || return 1 - do_facet client "cp /etc/termcap $MOUNT/${tfile}.2" || return 2 + remote_ost_nodsh && skip "remote OST with nodsh" && return 0 + + pause_bulk "cp /etc/profile $DIR/$tfile" || return 1 + do_facet client "cp $SAMPLE_FILE $DIR/${tfile}.2" || return 2 do_facet client "sync" - do_facet client "rm $MOUNT/$tfile $MOUNT/${tfile}.2" || return 3 + do_facet client "rm $DIR/$tfile $DIR/${tfile}.2" || return 3 } run_test 9 "pause bulk on OST (bug 1420)" #bug 1521 test_10() { - do_facet client mcreate $MOUNT/$tfile || return 1 - drop_bl_callback "chmod 0777 $MOUNT/$tfile" || echo "evicted as expected" + do_facet client mcreate $DIR/$tfile || return 1 + drop_bl_callback "chmod 0777 $DIR/$tfile" || echo "evicted as expected" # wait for the mds to evict the client #echo "sleep $(($TIMEOUT*2))" #sleep $(($TIMEOUT*2)) - do_facet client touch $MOUNT/$tfile || echo "touch failed, evicted" - do_facet client checkstat -v -p 0777 $MOUNT/$tfile || return 3 - do_facet client "munlink $MOUNT/$tfile" + do_facet client touch $DIR/$tfile || echo "touch failed, evicted" + do_facet client checkstat -v -p 0777 $DIR/$tfile || return 3 + do_facet client "munlink $DIR/$tfile" } run_test 10 "finish request on server after client eviction (bug 1521)" #bug 2460 # wake up a thread waiting for completion after eviction test_11(){ - do_facet client multiop $MOUNT/$tfile Ow || return 1 - do_facet client multiop $MOUNT/$tfile or || return 2 + do_facet client multiop $DIR/$tfile Ow || return 1 + do_facet client multiop $DIR/$tfile or || return 2 cancel_lru_locks osc - do_facet client multiop $MOUNT/$tfile or || return 3 - drop_bl_callback multiop $MOUNT/$tfile Ow || echo "evicted as expected" + do_facet client multiop $DIR/$tfile or || return 3 + drop_bl_callback multiop $DIR/$tfile Ow || echo "evicted as expected" - do_facet client munlink $MOUNT/$tfile || return 4 + do_facet client munlink $DIR/$tfile || return 4 } run_test 11 "wake up a thread waiting for completion after eviction (b=2460)" #b=2494 test_12(){ - $LCTL mark multiop $MOUNT/$tfile OS_c - do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x115" + $LCTL mark multiop $DIR/$tfile OS_c + do_facet $SINGLEMDS "lctl set_param fail_loc=0x115" clear_failloc $SINGLEMDS $((TIMEOUT * 2)) & - multiop $MOUNT/$tfile OS_c & + multiop_bg_pause $DIR/$tfile OS_c || return 1 PID=$! #define OBD_FAIL_MDS_CLOSE_NET 0x115 - sleep 2 kill -USR1 $PID echo "waiting for multiop $PID" wait $PID || return 2 - do_facet client munlink $MOUNT/$tfile || return 3 + do_facet client munlink $DIR/$tfile || return 3 } run_test 12 "recover from timed out resend in ptlrpcd (b=2494)" # Bug 113, check that readdir lost recv timeout works. test_13() { - mkdir $MOUNT/readdir || return 1 - touch $MOUNT/readdir/newentry || return + mkdir -p $DIR/$tdir || return 1 + touch $DIR/$tdir/newentry || return # OBD_FAIL_MDS_READPAGE_NET|OBD_FAIL_ONCE - do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x80000104" - ls $MOUNT/readdir || return 3 - do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0" - rm -rf $MOUNT/readdir || return 4 + do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000104" + ls $DIR/$tdir || return 3 + do_facet $SINGLEMDS "lctl set_param fail_loc=0" + rm -rf $DIR/$tdir || return 4 } run_test 13 "mdc_readpage restart test (bug 1138)" # Bug 113, check that readdir lost send timeout works. test_14() { - mkdir $MOUNT/readdir - touch $MOUNT/readdir/newentry + mkdir -p $DIR/$tdir + touch $DIR/$tdir/newentry # OBD_FAIL_MDS_SENDPAGE|OBD_FAIL_ONCE - do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x80000106" - ls $MOUNT/readdir || return 1 - do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0" + do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000106" + ls $DIR/$tdir || return 1 + do_facet $SINGLEMDS "lctl set_param fail_loc=0" } run_test 14 "mdc_readpage resend test (bug 1138)" test_15() { - do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x80000128" + do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000128" touch $DIR/$tfile && return 1 return 0 } run_test 15 "failed open (-ENOMEM)" -READ_AHEAD=`cat $LPROC/llite/*/max_read_ahead_mb | head -n 1` +READ_AHEAD=`lctl get_param -n llite.*.max_read_ahead_mb | head -n 1` stop_read_ahead() { - for f in $LPROC/llite/*/max_read_ahead_mb; do - echo 0 > $f - done + lctl set_param -n llite.*.max_read_ahead_mb 0 } start_read_ahead() { - for f in $LPROC/llite/*/max_read_ahead_mb; do - echo $READ_AHEAD > $f - done + lctl set_param -n llite.*.max_read_ahead_mb $READ_AHEAD } test_16() { - do_facet client cp /etc/termcap $MOUNT + remote_ost_nodsh && skip "remote OST with nodsh" && return 0 + + do_facet client cp $SAMPLE_FILE $DIR sync stop_read_ahead #define OBD_FAIL_PTLRPC_BULK_PUT_NET 0x504 | OBD_FAIL_ONCE - do_facet ost1 "sysctl -w lustre.fail_loc=0x80000504" + do_facet ost1 "lctl set_param fail_loc=0x80000504" cancel_lru_locks osc # OST bulk will time out here, client resends - do_facet client "cmp /etc/termcap $MOUNT/termcap" || return 1 - do_facet ost1 sysctl -w lustre.fail_loc=0 + do_facet client "cmp $SAMPLE_FILE $DIR/${SAMPLE_FILE##*/}" || return 1 + do_facet ost1 lctl set_param fail_loc=0 # give recovery a chance to finish (shouldn't take long) sleep $TIMEOUT - do_facet client "cmp /etc/termcap $MOUNT/termcap" || return 2 + do_facet client "cmp $SAMPLE_FILE $DIR/${SAMPLE_FILE##*/}" || return 2 start_read_ahead } run_test 16 "timeout bulk put, don't evict client (2732)" test_17() { + local at_max_saved=0 + + remote_ost_nodsh && skip "remote OST with nodsh" && return 0 + + # With adaptive timeouts, bulk_get won't expire until adaptive_timeout_max + if at_is_enabled; then + at_max_saved=$(at_max_get ost1) + at_max_set $TIMEOUT ost1 + fi + # OBD_FAIL_PTLRPC_BULK_GET_NET 0x0503 | OBD_FAIL_ONCE # OST bulk will time out here, client retries - do_facet ost1 sysctl -w lustre.fail_loc=0x80000503 + do_facet ost1 lctl set_param fail_loc=0x80000503 # need to ensure we send an RPC - do_facet client cp /etc/termcap $DIR/$tfile + do_facet client cp $SAMPLE_FILE $DIR/$tfile sync - sleep $TIMEOUT - do_facet ost1 sysctl -w lustre.fail_loc=0 + # with AT, client will wait adaptive_max*factor+net_latency before + # expiring the req, hopefully timeout*2 is enough + sleep $(($TIMEOUT*2)) + + do_facet ost1 lctl set_param fail_loc=0 do_facet client "df $DIR" # expect cmp to succeed, client resent bulk - do_facet client "cmp /etc/termcap $DIR/$tfile" || return 3 + do_facet client "cmp $SAMPLE_FILE $DIR/$tfile" || return 3 do_facet client "rm $DIR/$tfile" || return 4 + [ $at_max_saved -ne 0 ] && at_max_set $at_max_saved ost1 return 0 } run_test 17 "timeout bulk get, don't evict client (2732)" test_18a() { - [ -z ${ost2_svc} ] && skip "needs 2 osts" && return 0 + [ -z ${ost2_svc} ] && skip_env "needs 2 osts" && return 0 - do_facet client mkdir -p $MOUNT/$tdir - f=$MOUNT/$tdir/$tfile + do_facet client mkdir -p $DIR/$tdir + f=$DIR/$tdir/$tfile cancel_lru_locks osc pgcache_empty || return 1 # 1 stripe on ost2 - lfs setstripe $f -s $((128 * 1024)) -i 1 -c 1 - - do_facet client cp /etc/termcap $f + $LFS setstripe -i 1 -c 1 $f + stripe_index=$($LFS getstripe -i $f) + if [ $stripe_index -ne 1 ]; then + $LFS getstripe $f + error "$f: stripe_index $stripe_index != 1" && return + fi + + do_facet client cp $SAMPLE_FILE $f sync - local osc2dev=`grep ${ost2_svc}-osc- $LPROC/devices | egrep -v 'MDT' | awk '{print $1}'` + local osc2dev=`lctl get_param -n devices | grep ${ost2_svc}-osc- | egrep -v 'MDT' | awk '{print $1}'` $LCTL --device $osc2dev deactivate || return 3 # my understanding is that there should be nothing in the page # cache after the client reconnects? @@ -237,18 +270,22 @@ test_18a() { run_test 18a "manual ost invalidate clears page cache immediately" test_18b() { - do_facet client mkdir -p $MOUNT/$tdir - f=$MOUNT/$tdir/$tfile - f2=$MOUNT/$tdir/${tfile}-2 + remote_ost_nodsh && skip "remote OST with nodsh" && return 0 + + do_facet client mkdir -p $DIR/$tdir + f=$DIR/$tdir/$tfile cancel_lru_locks osc pgcache_empty || return 1 - # shouldn't have to set stripe size of count==1 - lfs setstripe $f -s $((128 * 1024)) -i 0 -c 1 - lfs setstripe $f2 -s $((128 * 1024)) -i 0 -c 1 + $LFS setstripe -i 0 -c 1 $f + stripe_index=$($LFS getstripe -i $f) + if [ $stripe_index -ne 0 ]; then + $LFS getstripe $f + error "$f: stripe_index $stripe_index != 0" && return + fi - do_facet client cp /etc/termcap $f + do_facet client cp $SAMPLE_FILE $f sync ost_evict_client # allow recovery to complete @@ -257,13 +294,49 @@ test_18b() { # cache after the client reconnects? rc=0 pgcache_empty || rc=2 - rm -f $f $f2 + rm -f $f return $rc } run_test 18b "eviction and reconnect clears page cache (2766)" +test_18c() { + remote_ost_nodsh && skip "remote OST with nodsh" && return 0 + + do_facet client mkdir -p $DIR/$tdir + f=$DIR/$tdir/$tfile + + cancel_lru_locks osc + pgcache_empty || return 1 + + $LFS setstripe -i 0 -c 1 $f + stripe_index=$($LFS getstripe -i $f) + if [ $stripe_index -ne 0 ]; then + $LFS getstripe $f + error "$f: stripe_index $stripe_index != 0" && return + fi + + do_facet client cp $SAMPLE_FILE $f + sync + ost_evict_client + + # OBD_FAIL_OST_CONNECT_NET2 + # lost reply to connect request + do_facet ost1 lctl set_param fail_loc=0x80000225 + # force reconnect + sleep 1 + df $MOUNT > /dev/null 2>&1 + sleep 2 + # my understanding is that there should be nothing in the page + # cache after the client reconnects? + rc=0 + pgcache_empty || rc=2 + rm -f $f + return $rc +} +run_test 18c "Dropped connect reply after eviction handing (14755)" + test_19a() { - f=$MOUNT/$tfile + f=$DIR/$tfile do_facet client mcreate $f || return 1 drop_ldlm_cancel "chmod 0777 $f" || echo "evicted as expected" @@ -275,7 +348,7 @@ test_19a() { run_test 19a "test expired_lock_main on mds (2867)" test_19b() { - f=$MOUNT/$tfile + f=$DIR/$tfile do_facet client multiop $f Ow || return 1 do_facet client multiop $f or || return 2 @@ -289,13 +362,15 @@ test_19b() { run_test 19b "test expired_lock_main on ost (2867)" test_20a() { # bug 2983 - ldlm_handle_enqueue cleanup + remote_ost_nodsh && skip "remote OST with nodsh" && return 0 + mkdir -p $DIR/$tdir - multiop $DIR/$tdir/${tfile} O_wc & + $LFS setstripe -i 0 -c 1 $DIR/$tdir/${tfile} + multiop_bg_pause $DIR/$tdir/${tfile} O_wc || return 1 MULTI_PID=$! - sleep 1 cancel_lru_locks osc #define OBD_FAIL_LDLM_ENQUEUE_EXTENT_ERR 0x308 - do_facet ost1 sysctl -w lustre.fail_loc=0x80000308 + do_facet ost1 lctl set_param fail_loc=0x80000308 kill -USR1 $MULTI_PID wait $MULTI_PID rc=$? @@ -304,11 +379,13 @@ test_20a() { # bug 2983 - ldlm_handle_enqueue cleanup run_test 20a "ldlm_handle_enqueue error (should return error)" test_20b() { # bug 2986 - ldlm_handle_enqueue error during open + remote_ost_nodsh && skip "remote OST with nodsh" && return 0 + mkdir -p $DIR/$tdir - touch $DIR/$tdir/${tfile} + $LFS setstripe -i 0 -c 1 $DIR/$tdir/${tfile} cancel_lru_locks osc #define OBD_FAIL_LDLM_ENQUEUE_EXTENT_ERR 0x308 - do_facet ost1 sysctl -w lustre.fail_loc=0x80000308 + do_facet ost1 lctl set_param fail_loc=0x80000308 dd if=/etc/hosts of=$DIR/$tdir/$tfile && \ error "didn't fail open enqueue" || true } @@ -317,21 +394,21 @@ run_test 20b "ldlm_handle_enqueue error (should return error)" test_21a() { mkdir -p $DIR/$tdir-1 mkdir -p $DIR/$tdir-2 - multiop $DIR/$tdir-1/f O_c & + multiop_bg_pause $DIR/$tdir-1/f O_c || return 1 close_pid=$! - do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x80000129" + do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000129" multiop $DIR/$tdir-2/f Oc & open_pid=$! sleep 1 - do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0" + do_facet $SINGLEMDS "lctl set_param fail_loc=0" - do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x80000115" + do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000115" kill -USR1 $close_pid cancel_lru_locks mdc wait $close_pid || return 1 wait $open_pid || return 2 - do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0" + do_facet $SINGLEMDS "lctl set_param fail_loc=0" $CHECKSTAT -t file $DIR/$tdir-1/f || return 3 $CHECKSTAT -t file $DIR/$tdir-2/f || return 4 @@ -343,14 +420,14 @@ run_test 21a "drop close request while close and open are both in flight" test_21b() { mkdir -p $DIR/$tdir-1 mkdir -p $DIR/$tdir-2 - multiop $DIR/$tdir-1/f O_c & + multiop_bg_pause $DIR/$tdir-1/f O_c || return 1 close_pid=$! - do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x80000107" + do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000107" mcreate $DIR/$tdir-2/f & open_pid=$! sleep 1 - do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0" + do_facet $SINGLEMDS "lctl set_param fail_loc=0" kill -USR1 $close_pid cancel_lru_locks mdc @@ -366,22 +443,22 @@ run_test 21b "drop open request while close and open are both in flight" test_21c() { mkdir -p $DIR/$tdir-1 mkdir -p $DIR/$tdir-2 - multiop $DIR/$tdir-1/f O_c & + multiop_bg_pause $DIR/$tdir-1/f O_c || return 1 close_pid=$! - do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x80000107" + do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000107" mcreate $DIR/$tdir-2/f & open_pid=$! sleep 3 - do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0" + do_facet $SINGLEMDS "lctl set_param fail_loc=0" - do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x80000115" + do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000115" kill -USR1 $close_pid cancel_lru_locks mdc wait $close_pid || return 1 wait $open_pid || return 2 - do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0" + do_facet $SINGLEMDS "lctl set_param fail_loc=0" $CHECKSTAT -t file $DIR/$tdir-1/f || return 2 $CHECKSTAT -t file $DIR/$tdir-2/f || return 3 @@ -392,19 +469,19 @@ run_test 21c "drop both request while close and open are both in flight" test_21d() { mkdir -p $DIR/$tdir-1 mkdir -p $DIR/$tdir-2 - multiop $DIR/$tdir-1/f O_c & + multiop_bg_pause $DIR/$tdir-1/f O_c || return 1 pid=$! - do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x80000129" + do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000129" multiop $DIR/$tdir-2/f Oc & sleep 1 - do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0" + do_facet $SINGLEMDS "lctl set_param fail_loc=0" - do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x80000122" + do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000122" kill -USR1 $pid cancel_lru_locks mdc wait $pid || return 1 - do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0" + do_facet $SINGLEMDS "lctl set_param fail_loc=0" $CHECKSTAT -t file $DIR/$tdir-1/f || return 2 $CHECKSTAT -t file $DIR/$tdir-2/f || return 3 @@ -416,13 +493,13 @@ run_test 21d "drop close reply while close and open are both in flight" test_21e() { mkdir -p $DIR/$tdir-1 mkdir -p $DIR/$tdir-2 - multiop $DIR/$tdir-1/f O_c & + multiop_bg_pause $DIR/$tdir-1/f O_c || return 1 pid=$! - do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x80000119" + do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000119" touch $DIR/$tdir-2/f & sleep 1 - do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0" + do_facet $SINGLEMDS "lctl set_param fail_loc=0" kill -USR1 $pid cancel_lru_locks mdc @@ -438,19 +515,19 @@ run_test 21e "drop open reply while close and open are both in flight" test_21f() { mkdir -p $DIR/$tdir-1 mkdir -p $DIR/$tdir-2 - multiop $DIR/$tdir-1/f O_c & + multiop_bg_pause $DIR/$tdir-1/f O_c || return 1 pid=$! - do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x80000119" + do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000119" touch $DIR/$tdir-2/f & sleep 1 - do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0" + do_facet $SINGLEMDS "lctl set_param fail_loc=0" - do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x80000122" + do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000122" kill -USR1 $pid cancel_lru_locks mdc wait $pid || return 1 - do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0" + do_facet $SINGLEMDS "lctl set_param fail_loc=0" $CHECKSTAT -t file $DIR/$tdir-1/f || return 2 $CHECKSTAT -t file $DIR/$tdir-2/f || return 3 @@ -461,19 +538,19 @@ run_test 21f "drop both reply while close and open are both in flight" test_21g() { mkdir -p $DIR/$tdir-1 mkdir -p $DIR/$tdir-2 - multiop $DIR/$tdir-1/f O_c & + multiop_bg_pause $DIR/$tdir-1/f O_c || return 1 pid=$! - do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x80000119" + do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000119" touch $DIR/$tdir-2/f & sleep 1 - do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0" + do_facet $SINGLEMDS "lctl set_param fail_loc=0" - do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x80000115" + do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000115" kill -USR1 $pid cancel_lru_locks mdc wait $pid || return 1 - do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0" + do_facet $SINGLEMDS "lctl set_param fail_loc=0" $CHECKSTAT -t file $DIR/$tdir-1/f || return 2 $CHECKSTAT -t file $DIR/$tdir-2/f || return 3 @@ -484,20 +561,20 @@ run_test 21g "drop open reply and close request while close and open are both in test_21h() { mkdir -p $DIR/$tdir-1 mkdir -p $DIR/$tdir-2 - multiop $DIR/$tdir-1/f O_c & + multiop_bg_pause $DIR/$tdir-1/f O_c || return 1 pid=$! - do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x80000107" + do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000107" touch $DIR/$tdir-2/f & touch_pid=$! sleep 1 - do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0" + do_facet $SINGLEMDS "lctl set_param fail_loc=0" - do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x80000122" + do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000122" cancel_lru_locks mdc kill -USR1 $pid wait $pid || return 1 - do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0" + do_facet $SINGLEMDS "lctl set_param fail_loc=0" wait $touch_pid || return 2 @@ -512,7 +589,7 @@ test_22() { f1=$DIR/${tfile}-1 f2=$DIR/${tfile}-2 - do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x80000115" + do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000115" multiop $f2 Oc & close_pid=$! @@ -520,7 +597,7 @@ test_22() { multiop $f1 msu || return 1 cancel_lru_locks mdc - do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0" + do_facet $SINGLEMDS "lctl set_param fail_loc=0" wait $close_pid || return 2 rm -rf $f2 || return 4 @@ -528,7 +605,7 @@ test_22() { run_test 22 "drop close request and do mknod" test_23() { #b=4561 - multiop $DIR/$tfile O_c & + multiop_bg_pause $DIR/$tfile O_c || return 1 pid=$! # give a chance for open sleep 5 @@ -542,107 +619,156 @@ test_23() { #b=4561 } run_test 23 "client hang when close a file after mds crash" -test_24() { # bug 2248 - eviction fails writeback but app doesn't see it +test_24() { # bug 11710 details correct fsync() behavior + remote_ost_nodsh && skip "remote OST with nodsh" && return 0 + mkdir -p $DIR/$tdir + $LFS setstripe -i 0 -c 1 $DIR/$tdir cancel_lru_locks osc - multiop $DIR/$tdir/$tfile Owy_wyc & + multiop_bg_pause $DIR/$tdir/$tfile Owy_wyc || return 1 MULTI_PID=$! - usleep 500 ost_evict_client - usleep 500 kill -USR1 $MULTI_PID wait $MULTI_PID rc=$? - sysctl -w lustre.fail_loc=0x0 + lctl set_param fail_loc=0x0 client_reconnect [ $rc -eq 0 ] && error_ignore 5494 "multiop didn't fail fsync: rc $rc" || true } run_test 24 "fsync error (should return error)" -test_26() { # bug 5921 - evict dead exports by pinger +wait_client_evicted () { + local facet=$1 + local exports=$2 + local varsvc=${facet}_svc + + wait_update $(facet_active_host $facet) \ + "lctl get_param -n *.${!varsvc}.num_exports | cut -d' ' -f2" \ + $((exports - 1)) $3 +} + +test_26a() { # was test_26 bug 5921 - evict dead exports by pinger # this test can only run from a client on a separate node. - remote_ost || skip "local OST" && return - remote_mds || skip "local MDS" && return - OST_FILE=$LPROC/obdfilter/${ost1_svc}/num_exports - OST_EXP="`do_facet ost1 cat $OST_FILE`" - OST_NEXP1=`echo $OST_EXP | cut -d' ' -f2` - echo starting with $OST_NEXP1 OST exports + remote_ost || { skip "local OST" && return 0; } + remote_ost_nodsh && skip "remote OST with nodsh" && return 0 + remote_mds || { skip "local MDS" && return 0; } + + if [ $(facet_host mgs) = $(facet_host ost1) ]; then + skip "msg and ost1 are at the same node" + return 0 + fi + + check_timeout || return 1 + + local OST_NEXP=$(do_facet ost1 lctl get_param -n obdfilter.${ost1_svc}.num_exports | cut -d' ' -f2) + + echo starting with $OST_NEXP OST exports # OBD_FAIL_PTLRPC_DROP_RPC 0x505 - do_facet client sysctl -w lustre.fail_loc=0x505 - # evictor takes up to 2.25x to evict. But if there's a - # race to start the evictor from various obds, the loser - # might have to wait for the next ping. - echo Waiting for $(($TIMEOUT * 4)) secs - sleep $(($TIMEOUT * 4)) - OST_EXP="`do_facet ost1 cat $OST_FILE`" - OST_NEXP2=`echo $OST_EXP | cut -d' ' -f2` - echo ending with $OST_NEXP2 OST exports - do_facet client sysctl -w lustre.fail_loc=0x0 - [ $OST_NEXP1 -le $OST_NEXP2 ] && error "client not evicted" - return 0 + do_facet client lctl set_param fail_loc=0x505 + # evictor takes PING_EVICT_TIMEOUT + 3 * PING_INTERVAL to evict. + # But if there's a race to start the evictor from various obds, + # the loser might have to wait for the next ping. + + local rc=0 + wait_client_evicted ost1 $OST_NEXP $((TIMEOUT * 2 + TIMEOUT * 3 / 4)) + rc=$? + do_facet client lctl set_param fail_loc=0x0 + [ $rc -eq 0 ] || error "client not evicted from OST" } -run_test 26 "evict dead exports" +run_test 26a "evict dead exports" test_26b() { # bug 10140 - evict dead exports by pinger - client_df - zconf_mount `hostname` $MOUNT2 || error "Failed to mount $MOUNT2" - sleep 1 # wait connections being established - MDS_FILE=$LPROC/mdt/${mds1_svc}/num_exports - MDS_NEXP1="`do_facet $SINGLEMDS cat $MDS_FILE | cut -d' ' -f2`" - OST_FILE=$LPROC/obdfilter/${ost1_svc}/num_exports - OST_NEXP1="`do_facet ost1 cat $OST_FILE | cut -d' ' -f2`" - echo starting with $OST_NEXP1 OST and $MDS_NEXP1 MDS exports + remote_ost_nodsh && skip "remote OST with nodsh" && return 0 + + if [ $(facet_host mgs) = $(facet_host ost1) ]; then + skip "msg and ost1 are at the same node" + return 0 + fi + + check_timeout || return 1 + clients_up + zconf_mount `hostname` $MOUNT2 || + { error "Failed to mount $MOUNT2"; return 2; } + sleep 1 # wait connections being established + + local MDS_NEXP=$(do_facet $SINGLEMDS lctl get_param -n mdt.${mds1_svc}.num_exports | cut -d' ' -f2) + local OST_NEXP=$(do_facet ost1 lctl get_param -n obdfilter.${ost1_svc}.num_exports | cut -d' ' -f2) + + echo starting with $OST_NEXP OST and $MDS_NEXP MDS exports + zconf_umount `hostname` $MOUNT2 -f - # evictor takes up to 2.25x to evict. But if there's a - # race to start the evictor from various obds, the loser - # might have to wait for the next ping. - echo Waiting for $(($TIMEOUT * 4)) secs - sleep $(($TIMEOUT * 4)) - OST_NEXP2="`do_facet ost1 cat $OST_FILE | cut -d' ' -f2`" - MDS_NEXP2="`do_facet $SINGLEMDS cat $MDS_FILE | cut -d' ' -f2`" - echo ending with $OST_NEXP2 OST and $MDS_NEXP2 MDS exports - [ $OST_NEXP1 -le $OST_NEXP2 ] && error "client not evicted from OST" - [ $MDS_NEXP1 -le $MDS_NEXP2 ] && error "client not evicted from MDS" - return 0 + + # PING_INTERVAL max(obd_timeout / 4, 1U) + # PING_EVICT_TIMEOUT (PING_INTERVAL * 6) + + # evictor takes PING_EVICT_TIMEOUT + 3 * PING_INTERVAL to evict. + # But if there's a race to start the evictor from various obds, + # the loser might have to wait for the next ping. + # = 9 * PING_INTERVAL + PING_INTERVAL + # = 10 PING_INTERVAL = 10 obd_timeout / 4 = 2.5 obd_timeout + # let's wait $((TIMEOUT * 3)) # bug 19887 + local rc=0 + wait_client_evicted ost1 $OST_NEXP $((TIMEOUT * 3)) || \ + error "Client was not evicted by ost" rc=1 + wait_client_evicted $SINGLEMDS $MDS_NEXP $((TIMEOUT * 3)) || \ + error "Client was not evicted by mds" } run_test 26b "evict dead exports" test_27() { - remote_mds && { skip "remote MDS" && return 0; } mkdir -p $DIR/$tdir writemany -q -a $DIR/$tdir/$tfile 0 5 & CLIENT_PID=$! sleep 1 + local save_FAILURE_MODE=$FAILURE_MODE FAILURE_MODE="SOFT" facet_failover $SINGLEMDS #define OBD_FAIL_OSC_SHUTDOWN 0x407 - sysctl -w lustre.fail_loc=0x80000407 + do_facet $SINGLEMDS lctl set_param fail_loc=0x80000407 # need to wait for reconnect - echo -n waiting for fail_loc - while [ `sysctl -n lustre.fail_loc` -eq -2147482617 ]; do - sleep 1 - echo -n . - done + echo waiting for fail_loc + wait_update_facet $SINGLEMDS "lctl get_param -n fail_loc" "-2147482617" facet_failover $SINGLEMDS #no crashes allowed! kill -USR1 $CLIENT_PID wait $CLIENT_PID true + FAILURE_MODE=$save_FAILURE_MODE } run_test 27 "fail LOV while using OSC's" test_28() { # bug 6086 - error adding new clients - do_facet client mcreate $MOUNT/$tfile || return 1 - drop_bl_callback "chmod 0777 $MOUNT/$tfile" ||echo "evicted as expected" + do_facet client mcreate $DIR/$tfile || return 1 + drop_bl_callback "chmod 0777 $DIR/$tfile" ||echo "evicted as expected" #define OBD_FAIL_MDS_CLIENT_ADD 0x12f - do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x8000012f" + do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000012f" # fail once (evicted), reconnect fail (fail_loc), ok - df || (sleep 10; df) || (sleep 10; df) || error "reconnect failed" - rm -f $MOUNT/$tfile + client_up || (sleep 10; client_up) || (sleep 10; client_up) || error "reconnect failed" + rm -f $DIR/$tfile fail $SINGLEMDS # verify MDS last_rcvd can be loaded } run_test 28 "handle error adding new clients (bug 6086)" +test_29a() { # bug 22273 - error adding new clients + #define OBD_FAIL_TGT_CLIENT_ADD 0x711 + do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000711" + # fail abort so client will be new again + fail_abort $SINGLEMDS + client_up || error "reconnect failed" + return 0 +} +run_test 29a "error adding new clients doesn't cause LBUG (bug 22273)" + +test_29b() { # bug 22273 - error adding new clients + #define OBD_FAIL_TGT_CLIENT_ADD 0x711 + do_facet ost1 "lctl set_param fail_loc=0x80000711" + # fail abort so client will be new again + fail_abort ost1 + client_up || error "reconnect failed" + return 0 +} +run_test 29b "error adding new clients doesn't cause LBUG (bug 22273)" + test_50() { mkdir -p $DIR/$tdir # put a load of file creates/writes/deletes @@ -669,6 +795,9 @@ test_50() { run_test 50 "failover MDS under load" test_51() { + #define OBD_FAIL_MDS_SYNC_CAPA_SL 0x1310 + do_facet ost1 lctl set_param fail_loc=0x00001310 + mkdir -p $DIR/$tdir # put a load of file creates/writes/deletes writemany -q $DIR/$tdir/$tfile 0 5 & @@ -697,6 +826,7 @@ test_51() { run_test 51 "failover MDS during recovery" test_52_guts() { + do_facet client "mkdir -p $DIR/$tdir" do_facet client "writemany -q -a $DIR/$tdir/$tfile 300 5" & CLIENT_PID=$! echo writemany pid $CLIENT_PID @@ -714,6 +844,8 @@ test_52_guts() { } test_52() { + remote_ost_nodsh && skip "remote OST with nodsh" && return 0 + mkdir -p $DIR/$tdir test_52_guts rc=$? @@ -754,11 +886,12 @@ run_test 54 "back in time" # bug 11330 - liblustre application death during I/O locks up OST test_55() { - remote_ost && { skip "remote OST" && return 0; } + remote_ost_nodsh && skip "remote OST with nodsh" && return 0 mkdir -p $DIR/$tdir # first dd should be finished quickly + $LFS setstripe -c 1 -i 0 $DIR/$tdir/$tfile-1 dd if=/dev/zero of=$DIR/$tdir/$tfile-1 bs=32M count=4 & DDPID=$! count=0 @@ -773,8 +906,9 @@ test_55() { done echo "(dd_pid=$DDPID, time=$count)successful" - #define OBD_FAIL_OST_DROP_REQ 0x21d - do_facet ost sysctl -w lustre.fail_loc=0x0000021d + $LFS setstripe -c 1 -i 0 $DIR/$tdir/$tfile-2 + #define OBD_FAIL_OST_DROP_REQ 0x21d + do_facet ost1 lctl set_param fail_loc=0x0000021d # second dd will be never finished dd if=/dev/zero of=$DIR/$tdir/$tfile-2 bs=32M count=4 & DDPID=$! @@ -793,7 +927,7 @@ test_55() { echo "(dd_pid=$DDPID, time=$count)successful" #Recover fail_loc and dd will finish soon - do_facet ost sysctl -w lustre.fail_loc=0 + do_facet ost1 lctl set_param fail_loc=0 count=0 echo "step3: testing ......" while [ true ]; do @@ -813,19 +947,17 @@ run_test 55 "ost_brw_read/write drops timed-out read/write request" test_56() { # b=11277 #define OBD_FAIL_MDS_RESEND 0x136 touch $DIR/$tfile - do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x80000136" + do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000136" stat $DIR/$tfile - do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0" + do_facet $SINGLEMDS "lctl set_param fail_loc=0" rm -f $DIR/$tfile } run_test 56 "do not allow reconnect to busy exports" test_57_helper() { # no oscs means no client or mdt - while [ -e $LPROC/osc ]; do - for f in `find $LPROC -type f`; do - cat $f > /dev/null 2>&1 - done + while lctl get_param osc.*.* > /dev/null 2>&1; do + : # loop until proc file is removed done } @@ -834,12 +966,12 @@ test_57() { # bug 10866 pid=$! sleep 1 #define OBD_FAIL_LPROC_REMOVE 0xB00 - sysctl -w lustre.fail_loc=0x80000B00 + lctl set_param fail_loc=0x80000B00 zconf_umount `hostname` $DIR - sysctl -w lustre.fail_loc=0x80000B00 + lctl set_param fail_loc=0x80000B00 fail_abort $SINGLEMDS kill -9 $pid - sysctl -w lustre.fail_loc=0 + lctl set_param fail_loc=0 mount_client $DIR do_facet client "df $DIR" } @@ -847,14 +979,14 @@ run_test 57 "read procfs entries causes kernel crash" test_58() { # bug 11546 #define OBD_FAIL_MDC_ENQUEUE_PAUSE 0x801 - touch $MOUNT/$tfile - ls -la $MOUNT/$tfile - sysctl -w lustre.fail_loc=0x80000801 - cp $MOUNT/$tfile /dev/null & + touch $DIR/$tfile + ls -la $DIR/$tfile + lctl set_param fail_loc=0x80000801 + cp $DIR/$tfile /dev/null & pid=$! sleep 1 - sysctl -w lustre.fail_loc=0 - drop_bl_callback rm -f $MOUNT/$tfile + lctl set_param fail_loc=0 + drop_bl_callback rm -f $DIR/$tfile wait $pid do_facet client "df $DIR" } @@ -864,11 +996,11 @@ test_59() { # bug 10589 zconf_mount `hostname` $MOUNT2 || error "Failed to mount $MOUNT2" echo $DIR2 | grep -q $MOUNT2 || error "DIR2 is not set properly: $DIR2" #define OBD_FAIL_LDLM_CANCEL_EVICT_RACE 0x311 - sysctl -w lustre.fail_loc=0x311 + lctl set_param fail_loc=0x311 writes=$(LANG=C dd if=/dev/zero of=$DIR2/$tfile count=1 2>&1) [ $? = 0 ] || error "dd write failed" writes=$(echo $writes | awk -F '+' '/out/ {print $1}') - sysctl -w lustre.fail_loc=0 + lctl set_param fail_loc=0 sync zconf_umount `hostname` $MOUNT2 -f reads=$(LANG=C dd if=$DIR/$tfile of=/dev/null 2>&1) @@ -878,6 +1010,395 @@ test_59() { # bug 10589 } run_test 59 "Read cancel race on client eviction" -equals_msg `basename $0`: test complete, cleaning up +err17935 () { + # we assume that all md changes are in the MDT0 changelog + if [ $MDSCOUNT -gt 1 ]; then + error_ignore 17935 $* + else + error $* + fi +} + +test_60() { + MDT0=$($LCTL get_param -n mdc.*.mds_server_uuid | \ + awk '{gsub(/_UUID/,""); print $1}' | head -1) + + NUM_FILES=15000 + mkdir -p $DIR/$tdir + + # Register (and start) changelog + USER=$(do_facet $SINGLEMDS lctl --device $MDT0 changelog_register -n) + echo "Registered as $MDT0 changelog user $USER" + + # Generate a large number of changelog entries + createmany -o $DIR/$tdir/$tfile $NUM_FILES + sync + sleep 5 + + # Unlink files in the background + unlinkmany $DIR/$tdir/$tfile $NUM_FILES & + CLIENT_PID=$! + sleep 1 + + # Failover the MDS while unlinks are happening + facet_failover $SINGLEMDS + + # Wait for unlinkmany to finish + wait $CLIENT_PID + + # Check if all the create/unlink events were recorded + # in the changelog + $LFS changelog $MDT0 >> $DIR/$tdir/changelog + local cl_count=$(grep UNLNK $DIR/$tdir/changelog | wc -l) + echo "$cl_count unlinks in $MDT0 changelog" + + do_facet $SINGLEMDS lctl --device $MDT0 changelog_deregister $USER + USERS=$(( $(do_facet $SINGLEMDS lctl get_param -n \ + mdd.$MDT0.changelog_users | wc -l) - 2 )) + if [ $USERS -eq 0 ]; then + [ $cl_count -eq $NUM_FILES ] || \ + err17935 "Recorded ${cl_count} unlinks out of $NUM_FILES" + # Also make sure we can clear large changelogs + cl_count=$($LFS changelog $FSNAME | wc -l) + [ $cl_count -le 2 ] || \ + error "Changelog not empty: $cl_count entries" + else + # If there are other users, there may be other unlinks in the log + [ $cl_count -ge $NUM_FILES ] || \ + err17935 "Recorded ${cl_count} unlinks out of $NUM_FILES" + echo "$USERS other changelog users; can't verify clear" + fi +} +run_test 60 "Add Changelog entries during MDS failover" + +test_61() +{ + local mdtosc=$(get_mdtosc_proc_path $SINGLEMDS $FSNAME-OST0000) + mdtosc=${mdtosc/-MDT*/-MDT\*} + local cflags="osc.$mdtosc.connect_flags" + do_facet $SINGLEMDS "lctl get_param -n $cflags" |grep -q skip_orphan + [ $? -ne 0 ] && skip "don't have skip orphan feature" && return + + mkdir -p $DIR/$tdir || error "mkdir dir $DIR/$tdir failed" + # Set the default stripe of $DIR/$tdir to put the files to ost1 + $LFS setstripe -c 1 -i 0 $DIR/$tdir + + replay_barrier $SINGLEMDS + createmany -o $DIR/$tdir/$tfile-%d 10 + local oid=`do_facet ost1 "lctl get_param -n obdfilter.${ost1_svc}.last_id"` + + fail_abort $SINGLEMDS + + touch $DIR/$tdir/$tfile + local id=`$LFS getstripe $DIR/$tdir/$tfile | awk '$1 == 0 { print $2 }'` + [ $id -le $oid ] && error "the orphan objid was reused, failed" + + # Cleanup + rm -rf $DIR/$tdir +} +run_test 61 "Verify to not reuse orphan objects - bug 17025" + +check_cli_ir_state() +{ + local NODE=${1:-$HOSTNAME} + local st + st=$(do_node $NODE "lctl get_param mgc.*.ir_state | + awk '/imperative_recovery:/ { print \\\$2}'") + [ $st != ON -o $st != OFF ] || + error "Error state $st, must be ON or OFF" + echo -n $st +} + +check_target_ir_state() +{ + local target=${1} + local name=${target}_svc + local recovery_proc=obdfilter.${!name}.recovery_status + local st + + st=$(do_facet $target "lctl get_param -n $recovery_proc | + awk '/IR:/{ print \\\$2}'") + [ $st != ON -o $st != OFF ] || + error "Error state $st, must be ON or OFF" + echo -n $st +} + +set_ir_status() +{ + do_facet mgs lctl set_param -n mgs.MGS.live.$FSNAME="state=$1" +} + +get_ir_status() +{ + local state=$(do_facet mgs "lctl get_param -n mgs.MGS.live.$FSNAME | + awk '/state:/{ print \\\$2 }'") + echo -n ${state/,/} +} + +nidtbl_version_mgs() +{ + local ver=$(do_facet mgs "lctl get_param -n mgs.MGS.live.$FSNAME | + awk '/nidtbl_version:/{ print \\\$2 }'") + echo -n $ver +} + +# nidtbl_version_client [node] +nidtbl_version_client() +{ + local cli=$1 + local node=${2:-$HOSTNAME} + + if [ X$cli = Xclient ]; then + cli=$FSNAME-client + else + local obdtype=${cli/%[0-9]*/} + [ $obdtype != mds ] && error "wrong parameters $cli" + + node=$(facet_active_host $cli) + local t=${cli}_svc + cli=${!t} + fi + + local vers=$(do_node $node "lctl get_param -n mgc.*.ir_state" | + awk "/$cli/{print \$6}" |sort -u) + + # in case there are multiple mounts on the client node + local arr=($vers) + [ ${#arr[@]} -ne 1 ] && error "versions on client node mismatch" + echo -n $vers +} + +nidtbl_versions_match() +{ + [ $(nidtbl_version_mgs) -eq $(nidtbl_version_client ${1:-client}) ] +} + +target_instance_match() +{ + local srv=$1 + local obdtype + local cliname + + obdtype=${srv/%[0-9]*/} + case $obdtype in + mds) + obdname="mdt" + cliname="mdc" + ;; + ost) + obdname="obdfilter" + cliname="osc" + ;; + *) + error "invalid target type" $srv + return 1 + ;; + esac + + local target=${srv}_svc + local si=$(do_facet $srv lctl get_param -n $obdname.${!target}.instance) + local ci=$(lctl get_param -n $cliname.${!target}-${cliname}-*.import | \ + awk '/instance/{ print $2 }' |head -1) + + return $([ $si -eq $ci ]) +} + +test_100() +{ + do_facet mgs $LCTL list_param mgs.*.ir_timeout || + { skip "MGS without IR support"; return 0; } + + # disable IR + set_ir_status disabled + + local saved_FAILURE_MODE=$FAILURE_MODE + [ $(facet_host mgs) = $(facet_host ost1) ] && FAILURE_MODE="SOFT" + fail ost1 + + # valid check + nidtbl_versions_match && + error "version must differ due to IR disabled" + target_instance_match ost1 || error "instance mismatch" + + # restore env + set_ir_status full + FAILURE_MODE=$saved_FAILURE_MODE +} +run_test 100 "IR: Make sure normal recovery still works w/o IR" + +test_101() +{ + do_facet mgs $LCTL list_param mgs.*.ir_timeout || + { skip "MGS without IR support"; return 0; } + + set_ir_status full + + local OST1_IMP=$(get_osc_import_name client ost1) + + # disable pinger recovery + lctl set_param -n osc.$OST1_IMP.pinger_recov=0 + + fail ost1 + + target_instance_match ost1 || error "instance mismatch" + nidtbl_versions_match || error "version must match" + + lctl set_param -n osc.$OST1_IMP.pinger_recov=1 +} +run_test 101 "IR: Make sure IR works w/o normal recovery" + +test_102() +{ + do_facet mgs $LCTL list_param mgs.*.ir_timeout || + { skip "MGS without IR support"; return 0; } + + local clients=${CLIENTS:-$HOSTNAME} + local old_version + local new_version + local mgsdev=mgs + + set_ir_status full + + # let's have a new nidtbl version + fail ost1 + + # sleep for a while so that clients can see the failure of ost + # it must be MGC_TIMEOUT_MIN_SECONDS + MGC_TIMEOUT_RAND_CENTISEC. + # int mgc_request.c: + # define MGC_TIMEOUT_MIN_SECONDS 5 + # define MGC_TIMEOUT_RAND_CENTISEC 0x1ff /* ~500 * + local count=30 # 20 seconds at most + while [ $count -gt 0 ]; do + nidtbl_versions_match && break + sleep 1 + count=$((count-1)) + done + + nidtbl_versions_match || error "nidtbl mismatch" + + # get the version # + old_version=$(nidtbl_version_client client) + + zconf_umount_clients $clients $MOUNT || error "Cannot umount client" + + # restart mgs + combined_mgs_mds && mgsdev=mds1 + remount_facet $mgsdev + fail ost1 + + zconf_mount_clients $clients $MOUNT || error "Cannot mount client" + + # check new version + new_version=$(nidtbl_version_client client) + [ $new_version -lt $old_version ] && + error "nidtbl version wrong after mgs restarts" + return 0 +} +run_test 102 "IR: New client gets updated nidtbl after MGS restart" + +test_103() +{ + do_facet mgs $LCTL list_param mgs.*.ir_timeout || + { skip "MGS without IR support"; return 0; } + + combined_mgs_mds && skip "mgs and mds on the same target" && return 0 + + # workaround solution to generate config log on the mds + remount_facet mds1 + + stop mgs + stop mds1 + + # We need this test because mds is like a client in IR context. + start mds1 $MDSDEV1 || error "MDS should start w/o mgs" + + # start mgs and remount mds w/ ir + start mgs $MGSDEV + clients_up + + # remount client so that fsdb will be created on the MGS + umount_client $MOUNT || error "umount failed" + mount_client $MOUNT || error "mount failed" + + # sleep 30 seconds so the MDS has a chance to detect MGS restarting + local count=30 + while [ $count -gt 0 ]; do + [ $(nidtbl_version_client mds1) -ne 0 ] && break + sleep 1 + count=$((count-1)) + done + + # after a while, mds should be able to reconnect to mgs and fetch + # up-to-date nidtbl version + nidtbl_versions_match mds1 || error "mds nidtbl mismatch" + + # reset everything + set_ir_status full +} +run_test 103 "IR: MDS can start w/o MGS and get updated nidtbl later" + +test_104() +{ + do_facet mgs $LCTL list_param mgs.*.ir_timeout || + { skip "MGS without IR support"; return 0; } + + set_ir_status full + + stop ost1 + start ost1 $(ostdevname 1) "$OST_MOUNT_OPTS -onoir" || + error "OST1 cannot start" + clients_up + + local ir_state=$(check_target_ir_state ost1) + [ $ir_state = "OFF" ] || error "ir status on ost1 should be OFF" +} +run_test 104 "IR: ost can disable IR voluntarily" + +test_105() +{ + [ -z "$RCLIENTS" ] && skip "Needs multiple clients" && return 0 + do_facet mgs $LCTL list_param mgs.*.ir_timeout || + { skip "MGS without IR support"; return 0; } + + set_ir_status full + + # get one of the clients from client list + local rcli=$(echo $RCLIENTS |cut -d' ' -f 1) + + local old_MOUNTOPT=$MOUNTOPT + MOUNTOPT=${MOUNTOPT},noir + zconf_umount $rcli $MOUNT || error "umount failed" + zconf_mount $rcli $MOUNT || error "mount failed" + + # make sure lustre mount at $rcli disabling IR + local ir_state=$(check_cli_ir_state $rcli) + [ $ir_state = OFF ] || error "IR state must be OFF at $rcli" + + # make sure MGS's state is Partial + [ $(get_ir_status) = "partial" ] || error "MGS IR state must be partial" + + fail ost1 + # make sure IR on ost1 is OFF + local ir_state=$(check_target_ir_state ost1) + [ $ir_state = "OFF" ] || error "IR status on ost1 should be OFF" + + # restore it + MOUNTOPT=$old_MOUNTOPT + zconf_umount $rcli $MOUNT || error "umount failed" + zconf_mount $rcli $MOUNT || error "mount failed" + + # make sure MGS's state is full + [ $(get_ir_status) = "full" ] || error "MGS IR status must be full" + + fail ost1 + # make sure IR on ost1 is ON + local ir_state=$(check_target_ir_state ost1) + [ $ir_state = "ON" ] || error "IR status on ost1 should be OFF" + + return 0 +} +run_test 105 "IR: NON IR clients support" + +complete $(basename $0) $SECONDS check_and_cleanup_lustre -[ -f "$TESTSUITELOG" ] && cat $TESTSUITELOG || true +exit_status