LU-1365 tests: LARGEDIR and 3-level htree for local ldiskfs

Add conf-sanity.sh test for the LARGEDIR feature with Lustre + ldiskfs
to exceed both 2-level htree (with 1024-byte blocksize to reduce the
number of entries per htree level) and 2GB directory size (marked SLOW
since this takes a while to run). This is done using a smaller number
of hard-linked inodes (nlink_max = 65000), to avoid the overhead of
accessing and caching a large number of different inodes.

Running multiple threads on a single directory is serialized by the
VFS directory mutex, so the tests are run in parallel on all available
clients and mountpoints to speed up the test as much as possible.

Signed-off-by: Andreas Dilger <adilger@whamcloud.com>
Signed-off-by: Artem Blagodarenko <c17828@cray.com>
Change-Id: If52ef346a39a1c3236871edbb0327bf44fe1a392
Test-Parameters: trivial testlist=conf-sanity envdefinitions=SLOW=yes,FLAKEY=false
Reviewed-on: https://review.whamcloud.com/22009
Tested-by: Jenkins
Reviewed-by: James Nunez <jnunez@whamcloud.com>
Tested-by: Maloo <maloo@whamcloud.com>

diff --git a/lustre/tests/conf-sanity.sh b/lustre/tests/conf-sanity.sh
index 4bcb5f2..58cb879 100644 (file)
--- a/lustre/tests/conf-sanity.sh
+++ b/lustre/tests/conf-sanity.sh
@@ -74,8 +74,8 @@ init_logging
 require_dsh_mds || exit 0
 require_dsh_ost || exit 0
 
-#                                  8  22  40  (min)
-[ "$SLOW" = "no" ] && EXCEPT_SLOW="45 69 106"
+#                                  8  22  40  165  (min)
+[ "$SLOW" = "no" ] && EXCEPT_SLOW="45 69 106 111"
 
 assert_DIR
 
@@ -275,6 +275,10 @@ check_mount2() {
        echo "setup double mount lustre success"
 }
 
+generate_name() {
+       cat /dev/urandom | tr -dc 'a-zA-Z0-9' | fold -w $1 | head -n 1
+}
+
 build_test_filter
 
 if [ "$ONLY" == "setup" ]; then
@@ -8079,6 +8083,206 @@ test_109b()
 }
 run_test 109b "test lctl clear_conf one config"
 
+test_110()
+{
+       [[ $(facet_fstype $SINGLEMDS) != ldiskfs ]] &&
+               skip "Only applicable to ldiskfs-based MDTs"
+
+       do_facet $SINGLEMDS $DEBUGFS -w -R supported_features |grep large_dir ||
+               skip "large_dir option is not supported on MDS"
+       do_facet ost1 $DEBUGFS -w -R supported_features | grep large_dir ||
+               skip "large_dir option is not supported on OSS"
+
+       stopall # stop all targets before modifying the target counts
+       stack_trap "MDSCOUNT=$MDSCOUNT OSTCOUNT=$OSTCOUNT" EXIT
+       MDSCOUNT=1
+       OSTCOUNT=1
+
+       # ext4_dir_entry_2 struct size:264
+       # dx_root struct size:8
+       # dx_node struct size:8
+       # dx_entry struct size:8
+       # For 1024 bytes block size.
+       # First level directory entries: 126
+       # Second level directory entries: 127
+       # Entries in leaf: 3
+       # For 2 levels limit: 48006
+       # For 3 levels limit : 6096762
+       # Create 80000 files to safely exceed 2-level htree limit.
+       CONF_SANITY_110_LINKS=${CONF_SANITY_110_LINKS:-80000}
+
+       # can fit at most 3 filenames per 1KB leaf block, but each
+       # leaf/index block will only be 3/4 full before split at each level
+       (( MDSSIZE < CONF_SANITY_110_LINKS / 3 * 4/3 * 4/3 )) &&
+               CONF_SANITY_110_LINKS=$((MDSSIZE * 3 * 3/4 * 3/4))
+
+       local opts="$(mkfs_opts mds1 $(mdsdevname 1)) \
+                   --reformat $(mdsdevname 1) $(mdsvdevname 1)"
+       if [[ $opts != *mkfsoptions* ]]; then
+               opts+=" --mkfsoptions=\\\"-O large_dir -b 1024 -i 65536\\\""
+       else
+               opts="${opts//--mkfsoptions=\\\"/ \
+                       --mkfsoptions=\\\"-O large_dir -b 1024 -i 65536 }"
+       fi
+       echo "MDT params: $opts"
+       add mds1 $opts || error "add mds1 failed with new params"
+       start mds1 $(mdsdevname 1) $MDS_MOUNT_OPTS
+
+       opts="$(mkfs_opts ost1 $(ostdevname 1)) \
+               --reformat $(ostdevname 1) $(ostvdevname 1)"
+
+       if [[ $opts != *mkfsoptions* ]]; then
+               opts+=" --mkfsoptions=\\\"-O large_dir\\\" "
+       else
+               opts="${opts//--mkfsoptions=\\\"/ \
+                       --mkfsoptions=\\\"-O large_dir }"
+       fi
+       echo "OST params: $opts"
+       add ost1 $opts || error "add ost1 failed with new params"
+       start ost1 $(ostdevname 1) $OST_MOUNT_OPTS
+
+       MOUNT_2=yes mountcli || error "mount clients failed"
+
+       mkdir -v $DIR/$tdir || error "cannot create $DIR/$tdir"
+       local pids count=0 group=0
+
+       echo "creating $CONF_SANITY_110_LINKS in total"
+       while (( count < CONF_SANITY_110_LINKS )); do
+               local len=$((253 - $(wc -c <<<"$tfile-$group-40000-")))
+               local dir=DIR$((group % 2 + 1))
+               local target=${!dir}/$tdir/$tfile-$group
+               local long=$target-$(generate_name $len)-
+               local create=$((CONF_SANITY_110_LINKS - count))
+
+               (( create > 40000 )) && create=40000
+               touch $target || error "creating $target failed"
+               echo "creating $create hard links to $target"
+               createmany -l $target $long $create &
+               pids+=" $!"
+
+               count=$((count + create))
+               group=$((group + 1))
+       done
+       echo "waiting for PIDs$pids to complete"
+       wait $pids || error "createmany failed after $group groups"
+
+       cleanup
+
+       run_e2fsck $(facet_active_host mds1) $(mdsdevname 1) -n
+}
+run_test 110 "Adding large_dir with 3-level htree"
+
+test_111() {
+       [[ $(facet_fstype $SINGLEMDS) != ldiskfs ]] &&
+               skip "Only applicable to ldiskfs-based MDTs"
+
+       is_dm_flakey_dev $SINGLEMDS $(mdsdevname 1) &&
+               skip "This test can not be executed on flakey dev"
+
+       do_facet $SINGLEMDS $DEBUGFS -w -R supported_features |grep large_dir ||
+               skip "large_dir option is not supported on MDS"
+
+       do_facet ost1 $DEBUGFS -w -R supported_features | grep large_dir ||
+               skip "large_dir option is not supported on OSS"
+
+       # cleanup before changing target counts
+       cleanup
+       stack_trap "MDSSIZE=$MDSSIZE MDSCOUNT=$MDSCOUNT OSTCOUNT=$OSTCOUNT" EXIT
+       MDSCOUNT=1
+       OSTCOUNT=1
+       (( MDSSIZE < 2400000 )) && MDSSIZE=2400000 # need at least 2.4GB
+
+       local mdsdev=$(mdsdevname 1)
+
+       local opts="$(mkfs_opts mds1 $(mdsdevname 1)) \
+                   --reformat $(mdsdevname 1) $(mdsvdevname 1)"
+       if [[ $opts != *mkfsoptions* ]]; then
+               opts+=" --mkfsoptions=\\\"-O large_dir -i 1048576 \\\" "
+       else
+               opts="${opts//--mkfsoptions=\\\"/ \
+                       --mkfsoptions=\\\"-O large_dir -i 1048576 }"
+       fi
+       echo "MDT params: $opts"
+       __touch_device mds 1
+       add mds1 $opts || error "add mds1 failed with new params"
+       start mds1 $(mdsdevname 1) $MDS_MOUNT_OPTS
+
+       opts="$(mkfs_opts ost1 $(ostdevname 1)) \
+               --reformat $(ostdevname 1) $(ostvdevname 1)"
+       if [[ $opts != *mkfsoptions* ]]; then
+               opts+=" --mkfsoptions=\\\"-O large_dir \\\""
+       else
+               opts="${opts//--mkfsoptions=\\\"/ --mkfsoptions=\\\"-O large_dir }"
+       fi
+       echo "OST params: $opts"
+       __touch_device ost 1
+       add ost1 $opts || error "add ost1 failed with new params"
+       start ost1 $(ostdevname 1) $OST_MOUNT_OPTS
+
+       MOUNT_2=yes mountcli
+       mkdir $DIR/$tdir || error "cannot create $DIR/$tdir"
+       lfs df $DIR/$tdir
+       lfs df -i $DIR/$tdir
+
+       local group=0
+
+       local start=$SECONDS
+       local dirsize=0
+       local dirmax=$((2 << 30))
+       local needskip=0
+       local taken=0
+       local rate=0
+       local left=0
+       local num=0
+       while (( !needskip & dirsize < dirmax )); do
+               local pids=""
+
+               for cli in ${CLIENTS//,/ }; do
+                       local len=$((253 - $(wc -c <<<"$cli-$group-60000-")))
+                       local target=$cli-$group
+                       local long=$DIR/$tdir/$target-$(generate_name $len)-
+
+                       RPWD=$DIR/$tdir do_node $cli touch $target ||
+                               error "creating $target failed"
+                       echo "creating 60000 hardlinks to $target"
+                       RPWD=$DIR/$tdir do_node $cli createmany -l $target $long 60000 &
+                       pids+=" $!"
+
+                       group=$((group + 1))
+                       target=$cli-$group
+                       long=$DIR2/$tdir/$target-$(generate_name $len)-
+
+                       RPWD=$DIR2/$tdir do_node $cli touch $target ||
+                               error "creating $target failed"
+                       echo "creating 60000 hardlinks to $target"
+                       RPWD=$DIR2/$tdir do_node $cli createmany -l $target $long 60000 &
+                       pids+=" $!"
+
+                       group=$((group + 1))
+               done
+               echo "waiting for PIDs$pids to complete"
+               wait $pids || error "createmany failed after $group groups"
+               dirsize=$(stat -c %s $DIR/$tdir)
+               taken=$((SECONDS - start))
+               rate=$((dirsize / taken))
+               left=$(((dirmax - dirsize) / rate))
+               num=$((group * 60000))
+               echo "estimate ${left}s left after $num files / ${taken}s"
+               # if the estimated time remaining is too large (it may change
+               # over time as the create rate is not constant) then exit
+               # without declaring a failure.
+               (( left > 1200 )) && needskip=1
+       done
+
+       cleanup
+
+       (( $needskip )) && skip "ETA ${left}s after $num files / ${taken}s is too long"
+
+       run_e2fsck $(facet_active_host mds1) $(mdsdevname 1) -n
+}
+run_test 111 "Adding large_dir with over 2GB directory"
+
+
 cleanup_115()
 {
        trap 0