在 2025/7/29 15:51, Nirjhar Roy (IBM) 写道:
When tested with blocksize/nodesize 64K on powerpc
with 64k pagesize on btrfs, then the test fails
with the folllowing error:
QA output created by 563
read/write
read is in range
-write is in range
+write has value of 8855552
+write is NOT in range 7969177.6 .. 8808038.4
I can reproduce the failure, although it's not 100% reliable, and indeed
with one tree block's size removed, it's back into the tolerance range.
write -> read/write
...
The slight increase in the amount of bytes that
are written is because of the increase in the
the nodesize(metadata) and hence it exceeds the tolerance limit slightly.
Fix this by increasing the write tolerance limit from 5% from 6%
for 64k blocksize btrfs.
Reported-by: Disha Goel <disgoel@xxxxxxxxxxxxx>
Signed-off-by: Nirjhar Roy (IBM) <nirjhar.roy.lists@xxxxxxxxx>
---
tests/generic/563 | 17 ++++++++++++++++-
1 file changed, 16 insertions(+), 1 deletion(-)
diff --git a/tests/generic/563 b/tests/generic/563
index 89a71aa4..efcac1ec 100755
--- a/tests/generic/563
+++ b/tests/generic/563
@@ -119,7 +119,22 @@ $XFS_IO_PROG -c "pread 0 $iosize" -c "pwrite -b $blksize 0 $iosize" -c fsync \
$SCRATCH_MNT/file >> $seqres.full 2>&1
switch_cg $cgdir
$XFS_IO_PROG -c fsync $SCRATCH_MNT/file
-check_cg $cgdir/$seq-cg $iosize $iosize 5% 5%
+blksz=`_get_block_size $SCRATCH_MNT`
+
+# On higher node sizes on btrfs, we observed slightly more
+# writes, due to increased metadata sizes.
+# Hence have a higher write tolerance for btrfs and when
+# node size is greater than 4k.
+if [[ "$FSTYP" == "btrfs" ]]; then
+ nodesz=$(_get_btrfs_node_size "$SCRATCH_DEV")
+ if [[ "$nodesz" -gt 4096 ]]; then
+ check_cg $cgdir/$seq-cg $iosize $iosize 5% 6%
+ else
+ check_cg $cgdir/$seq-cg $iosize $iosize 5% 5%
+ fi
+else
+ check_cg $cgdir/$seq-cg $iosize $iosize 5% 5%
+fi
Instead of the btrfs specific hack, I'd recommend to just enlarge iosize.
Double the iosize will easily make it to cover the tolerance of even
btrfs, but you still need a proper explanation of the change.
Thanks,
Qu
# Write from one cgroup then read and write from a second. Writes are charged to
# the first group and nothing to the second.
