Thread (13 messages) 13 messages, 2 authors, 2016-02-29

Re: [PATCH v2 RESEND 2/5] fstests: btrfs: Add basic test for btrfs in-band de-duplication

From: Dave Chinner <david@fromorbit.com>
Date: 2016-02-28 22:26:32
Also in: fstests

On Wed, Feb 24, 2016 at 04:06:33PM +0800, Qu Wenruo wrote:
quoted hunk ↗ jump to hunk
Add basic test for btrfs in-band de-duplication, including:
1) Enable
2) Re-enable
3) On disk extents are refering to same bytenr
4) Disable

Signed-off-by: Qu Wenruo <redacted>
---
 common/defrag       |   8 ++++
 tests/btrfs/200     | 125 ++++++++++++++++++++++++++++++++++++++++++++++++++++
 tests/btrfs/200.out |  19 ++++++++
 tests/btrfs/group   |   1 +
 4 files changed, 153 insertions(+)
 create mode 100755 tests/btrfs/200
 create mode 100644 tests/btrfs/200.out
diff --git a/common/defrag b/common/defrag
index 942593e..34cc822 100644
--- a/common/defrag
+++ b/common/defrag
@@ -47,6 +47,14 @@ _extent_count()
 	$XFS_IO_PROG -c "fiemap" $1 | tail -n +2 | grep -v hole | wc -l| $AWK_PROG '{print $1}'
 }
 
+_uniq_extent_count()
+{
+	file=$1
+	$XFS_IO_PROG -c "fiemap" $file >> $seqres.full 2>&1
+	$XFS_IO_PROG -c "fiemap" $file | tail -n +2 | grep -v hole |\
+		$AWK_PROG '{print $3}' | sort | uniq | wc -l
+}
This needs comments ot explain how it is different to _extent_count.
Also should probably be named _extent_count_unique()
quoted hunk ↗ jump to hunk
+
 	min=$1
diff --git a/tests/btrfs/200 b/tests/btrfs/200
new file mode 100755
index 0000000..f2ff542
--- /dev/null
+++ b/tests/btrfs/200
@@ -0,0 +1,125 @@
+#! /bin/bash
+# FS QA Test 200
+#
+# Basic btrfs inband dedup test, including:
+# 1) Enable
+# 2) Uniq file extent number
        Unique.
+# 3) Re-enable
+# 4) Disable
I don't understand what 2-4 are describing. As a test summary,
"Basic btrfs inband dedup test" is sufficient.
+_supported_fs btrfs
+_supported_os Linux
+_require_scratch
+_require_btrfs_subcommand dedup
+_require_btrfs_fs_feature dedup
+_require_btrfs_mkfs_feature dedup
+
+# File size is twice the maximum file extent of btrfs
+# So even fallbacked to non-dedup, it will have at least 2 extents
+file_size=$(( 256 * 1024 * 1024 ))
Used for xfs_io, so "file_size=256m" is all that is needed here.
+_scratch_mkfs "-O dedup" >> $seqres.full 2>&1
+_scratch_mount
+
+do_dedup_test()
+{
+	backend=$1
+	dedup_bs=$2
+
+	_run_btrfs_util_prog dedup enable -s $backend -b $dedup_bs $SCRATCH_MNT
+	$XFS_IO_PROG -f -c "pwrite -b $dedup_bs 0 $dedup_bs" \
+		$SCRATCH_MNT/initial_block | _filter_xfs_io
+
+	# sync to ensure dedup hash is added into dedup pool
+	sync
xfs_io -fs  or xfs_io ... -c "fsync" ... ?
+	$XFS_IO_PROG -f -c "pwrite -b $dedup_bs 0 $file_size" \
+		$SCRATCH_MNT/real_file | _filter_xfs_io
+	# sync again to ensure data are all written to disk and
+	# we can get stable extent map
+	sync
Again, why now just do a sync write or fsync from the xfs?
+
+	# Test if real_file is de-duplicated
+	nr_uniq_extents=$(_uniq_extent_count $SCRATCH_MNT/real_file)
+	nr_total_extents=$(_extent_count $SCRATCH_MNT/real_file)
+
+	echo "uniq/total: $nr_uniq_extents/$nr_total_extents" >> $seqres.full
+	# Allow a small amount of dedup miss, as commit interval or
+	# memory pressure may break a dedup_bs block and cause
+	# smalll extent which won't go through dedup routine
+	if [ $nr_uniq_extents -ge $(( $nr_total_extents * 5 / 100 )) ]; then
+		echo "Too high dedup failure rate"
+	fi
_within_tolerance
+
+	# Also check the md5sum to ensure data is not corrupted
+	md5=$(_md5_checksum $SCRATCH_MNT/real_file)
+	if [ $md5 != $init_md5 ]; then
+		echo "File after in-band de-duplication is corrupted"
+	fi
Nope. Just echo the md5sum to the golden output file.

+}
+
+# Create the initial file and calculate its checksum without dedup
+$XFS_IO_PROG -f -c "pwrite 0 $file_size" $SCRATCH_MNT/csum_file | \
+	_filter_xfs_io
+init_md5=$(_md5_checksum $SCRATCH_MNT/csum_file)
+echo "md5 of the initial file is $init_md5" >> $seqres.full
Just echo the md5sum to the golden output file.

Cheers,

Dave.
-- 
Dave Chinner
david@fromorbit.com
Keyboard shortcuts
hback out one level
jnext message in thread
kprevious message in thread
ldrill in
Escclose help / fold thread tree
?toggle this help