[PATCH] xfs/291: fix spurious ENOSPC errors

[PATCH] xfs/291: fix spurious ENOSPC errors

[Dave Chinner]

From: Dave Chinner <dchinner@redhat.com>

xfs/291 tries to fill the filesystem almost full, so if the log size
changes with mkfs defaults then it's free space calculations are not
longer valid and so it throws lots of ENOSPC errors during a run.
This is not fatal for this test, but it does increase the runtime of
it and fill the 291.full file with unnecessary errors.

The number of frag files it creates is also too many for a 512 byte
inode filesystem (by about 900) so reduce the number of inodes
initially created so the test works ofr 512 byte inodes. With 512
byte inodes, the free space histogram looks like this after the frag
phase:

   from      to extents  blocks    pct
      1       1   10730   10730 100.00

And for 256 byte inodes:

   from      to extents  blocks    pct
      1       1   12388   12388 100.00

So these changes do not affect the intended operation of the test.

Signed-off-by: Dave Chinner <dchinner@redhat.com>
---
 tests/xfs/291 | 9 ++++++---
 1 file changed, 6 insertions(+), 3 deletions(-)

diff --git a/tests/xfs/291 b/tests/xfs/291
index d5a9b08..0f3f051 100755
--- a/tests/xfs/291
+++ b/tests/xfs/291
@@ -46,7 +46,7 @@ _supported_os IRIX Linux
 # real QA test starts here
 rm -f $seqres.full
 _require_scratch
-_scratch_mkfs_xfs -n size=16k -d size=128m >> $seqres.full 2>&1
+_scratch_mkfs_xfs -n size=16k -l size=5m -d size=128m >> $seqres.full 2>&1
 _scratch_mount
 
 # First we cause very badly fragmented freespace, then
@@ -54,13 +54,16 @@ _scratch_mount
 # should come out very fragmented as a result
 
 # Step 1: Cause badly fragmented free space
+# 
+# XXX: this needs to be factored to match what generic/204 does to support
+# different inode sizes without ENOSPC
 mkdir $SCRATCH_MNT/fragdir
-for I in `seq 0 27200`; do
+for I in `seq 0 26200`; do
         (echo data > $SCRATCH_MNT/fragdir/f$I) >> $seqres.full 2>&1
 done
 sync
 
-for I in `seq 0 2 27200`; do
+for I in `seq 0 2 26200`; do
         rm -f $SCRATCH_MNT/fragdir/f$I >> $seqres.full 2>&1
 done
 sync
-- 
1.8.4.rc3

_______________________________________________
xfs mailing list
xfs@oss.sgi.com
http://oss.sgi.com/mailman/listinfo/xfs

Re: [PATCH] xfs/291: fix spurious ENOSPC errors

[Eric Sandeen]

On 1/22/14, 12:46 AM, Dave Chinner wrote:
> From: Dave Chinner <dchinner@redhat.com>
> 
> xfs/291 tries to fill the filesystem almost full, so if the log size
> changes with mkfs defaults then it's free space calculations are not
> longer valid and so it throws lots of ENOSPC errors during a run.
> This is not fatal for this test, but it does increase the runtime of
> it and fill the 291.full file with unnecessary errors.

Hm so you've set the log size to 5m (-l size=5m) but how does that
interact with any "-l size=XXX" set in MKFS_OPTS ? 

(It's certainly no worse than what I had in my original test; I just
wonder if we should ignore MKFS_OPTS altogether in this test, and
completely manually specify all options?)

((but I guess that would restrict this test too much, wouldn't it))

Fine for now, I suppose:

Reviewed-by: Eric Sandeen <sandeen@redhat.com>

-Eric

> The number of frag files it creates is also too many for a 512 byte
> inode filesystem (by about 900) so reduce the number of inodes
> initially created so the test works ofr 512 byte inodes. With 512
> byte inodes, the free space histogram looks like this after the frag
> phase:
> 
>    from      to extents  blocks    pct
>       1       1   10730   10730 100.00
> 
> And for 256 byte inodes:
> 
>    from      to extents  blocks    pct
>       1       1   12388   12388 100.00
> 
> So these changes do not affect the intended operation of the test.
> 
> Signed-off-by: Dave Chinner <dchinner@redhat.com>
> ---
>  tests/xfs/291 | 9 ++++++---
>  1 file changed, 6 insertions(+), 3 deletions(-)
> 
> diff --git a/tests/xfs/291 b/tests/xfs/291
> index d5a9b08..0f3f051 100755
> --- a/tests/xfs/291
> +++ b/tests/xfs/291
> @@ -46,7 +46,7 @@ _supported_os IRIX Linux
>  # real QA test starts here
>  rm -f $seqres.full
>  _require_scratch
> -_scratch_mkfs_xfs -n size=16k -d size=128m >> $seqres.full 2>&1
> +_scratch_mkfs_xfs -n size=16k -l size=5m -d size=128m >> $seqres.full 2>&1
>  _scratch_mount
>  
>  # First we cause very badly fragmented freespace, then
> @@ -54,13 +54,16 @@ _scratch_mount
>  # should come out very fragmented as a result
>  
>  # Step 1: Cause badly fragmented free space
> +# 
> +# XXX: this needs to be factored to match what generic/204 does to support
> +# different inode sizes without ENOSPC
>  mkdir $SCRATCH_MNT/fragdir
> -for I in `seq 0 27200`; do
> +for I in `seq 0 26200`; do
>          (echo data > $SCRATCH_MNT/fragdir/f$I) >> $seqres.full 2>&1
>  done
>  sync
>  
> -for I in `seq 0 2 27200`; do
> +for I in `seq 0 2 26200`; do
>          rm -f $SCRATCH_MNT/fragdir/f$I >> $seqres.full 2>&1
>  done
>  sync
> 

_______________________________________________
xfs mailing list
xfs@oss.sgi.com
http://oss.sgi.com/mailman/listinfo/xfs

Re: [PATCH] xfs/291: fix spurious ENOSPC errors

[Dave Chinner]

On Wed, Jan 22, 2014 at 03:52:48PM -0600, Eric Sandeen wrote:
> On 1/22/14, 12:46 AM, Dave Chinner wrote:
> > From: Dave Chinner <dchinner@redhat.com>
> > 
> > xfs/291 tries to fill the filesystem almost full, so if the log size
> > changes with mkfs defaults then it's free space calculations are not
> > longer valid and so it throws lots of ENOSPC errors during a run.
> > This is not fatal for this test, but it does increase the runtime of
> > it and fill the 291.full file with unnecessary errors.
> 
> Hm so you've set the log size to 5m (-l size=5m) but how does that
> interact with any "-l size=XXX" set in MKFS_OPTS ? 

see _scratch_mkfs_xfs:

        # a mkfs failure may be caused by conflicts between
        # $MKFS_OPTIONS and $extra_mkfs_options
        if [ $mkfs_status -ne 0 -a ! -z "$extra_mkfs_options" ]; then
                (
                echo -n "** mkfs failed with extra mkfs options "
                echo "added to \"$MKFS_OPTIONS\" by test $seq **"
                echo -n "** attempting to mkfs using only test $seq "
                echo "options: $extra_mkfs_options **"
                ) >> $seqres.full

                # running mkfs again. overwrite previous mkfs output files
                _scratch_mkfs_xfs_opts $extra_mkfs_options \
                        2>$tmp_dir.mkfserr 1>$tmp_dir.mkfsstd
                local mkfs_status=$?
        fi

So, it will drop the MKFS_OPTIONS and just use the test specific
options if there is a conflict/failure.

> (It's certainly no worse than what I had in my original test; I just
> wonder if we should ignore MKFS_OPTS altogether in this test, and
> completely manually specify all options?)

That's effectively what it already does automatically.... :)

Cheers,

Dave.
-- 
Dave Chinner
david@fromorbit.com

_______________________________________________
xfs mailing list
xfs@oss.sgi.com
http://oss.sgi.com/mailman/listinfo/xfs