Pool allocation classes misplacing small file blocks

Due to an off-by-one condition in spa_preferred_class() we are picking
the "normal" allocation class instead of the "special" one for file
blocks with size equal to the special_small_blocks property value.

This change fix the small code issue, update the ZFS Test Suite and the
zfs(8) man page.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Don Brady <don.brady@delphix.com>
Signed-off-by: loli10K <ezomori.nozomu@gmail.com>
Closes #8351
Closes #8361
This commit is contained in:
loli10K 2019-02-08 21:32:12 +01:00 committed by Matthew Ahrens
parent 0902c4577f
commit 4417096956
3 changed files with 47 additions and 4 deletions

View File

@ -1521,9 +1521,11 @@ This feature must be enabled to be used
.Pc . .Pc .
.It Sy special_small_blocks Ns = Ns Em size .It Sy special_small_blocks Ns = Ns Em size
This value represents the threshold block size for including small file This value represents the threshold block size for including small file
blocks into the special allocation class. Valid values are zero or a blocks into the special allocation class. Blocks smaller than or equal to this
power of two from 512B up to 128K. The default size is 0 which means no value will be assigned to the special allocation class while greater blocks
small file blocks will be allocated in the special class. will be assigned to the regular class. Valid values are zero or a power of two
from 512B up to 128K. The default size is 0 which means no small file blocks
will be allocated in the special class.
.Pp .Pp
Before setting this property, a special class vdev must be added to the Before setting this property, a special class vdev must be added to the
pool. See pool. See

View File

@ -1851,7 +1851,7 @@ spa_preferred_class(spa_t *spa, uint64_t size, dmu_object_type_t objtype,
* zfs_special_class_metadata_reserve_pct exclusively for metadata. * zfs_special_class_metadata_reserve_pct exclusively for metadata.
*/ */
if (DMU_OT_IS_FILE(objtype) && if (DMU_OT_IS_FILE(objtype) &&
has_special_class && size < special_smallblk) { has_special_class && size <= special_smallblk) {
metaslab_class_t *special = spa_special_class(spa); metaslab_class_t *special = spa_special_class(spa);
uint64_t alloc = metaslab_class_get_alloc(special); uint64_t alloc = metaslab_class_get_alloc(special);
uint64_t space = metaslab_class_get_space(special); uint64_t space = metaslab_class_get_space(special);

View File

@ -24,6 +24,39 @@
verify_runnable "global" verify_runnable "global"
#
# Verify the file identified by the input <inode> is written on a special vdev
# According to the pool layout used in this test vdev_id 3 and 4 are special
# XXX: move this function to libtest.shlib once we get "Vdev Properties"
#
function file_in_special_vdev # <dataset> <inode>
{
typeset dataset="$1"
typeset inum="$2"
zdb -dddddd $dataset $inum | awk '{
# find DVAs from string "offset level dva" only for L0 (data) blocks
if (match($0,"L0 [0-9]+")) {
dvas[0]=$3
dvas[1]=$4
dvas[2]=$5
for (i = 0; i < 3; ++i) {
if (match(dvas[i],"([^:]+):.*")) {
dva = substr(dvas[i], RSTART, RLENGTH);
# parse DVA from string "vdev:offset:asize"
if (split(dva,arr,":") != 3) {
print "Error parsing DVA: <" dva ">";
exit 1;
}
# verify vdev is "special"
if (arr[1] < 3) {
exit 1;
}
}
}
}}'
}
claim="Removing a special device from a pool succeeds." claim="Removing a special device from a pool succeeds."
log_assert $claim log_assert $claim
@ -53,6 +86,13 @@ done
log_must sync_pool $TESTPOOL log_must sync_pool $TESTPOOL
log_must zpool list -v $TESTPOOL log_must zpool list -v $TESTPOOL
# Verify the files were written in the special class vdevs
for i in 1 2 3 4; do
dataset="$TESTPOOL/$TESTFS"
inum="$(stat -c '%i' /$TESTPOOL/$TESTFS/testfile.$i)"
log_must file_in_special_vdev $dataset $inum
done
# #
# remove a special allocation vdev and force a remapping # remove a special allocation vdev and force a remapping
# N.B. The 'zfs remap' command has been disabled and may be removed. # N.B. The 'zfs remap' command has been disabled and may be removed.
@ -67,6 +107,7 @@ log_must sync_pool $TESTPOOL
sleep 1 sleep 1
log_must zdb -bbcc $TESTPOOL log_must zdb -bbcc $TESTPOOL
log_must zpool list -v $TESTPOOL
log_must zpool destroy -f "$TESTPOOL" log_must zpool destroy -f "$TESTPOOL"
log_pass $claim log_pass $claim