| Message ID | 20241023103930.432190-1-hch@lst.de (mailing list archive) |
|---|---|
| State | Deferred, archived |
| Headers | show |
| Series | xfs: remove the post-EOF prealloc tests from the auto and quick groups | expand |
On Wed, Oct 23, 2024 at 12:39:30PM +0200, Christoph Hellwig wrote: > These fail for various non-default configs like DAX, alwayscow and > small block sizes. Shouldn't we selectively _notrun these tests for configurations where speculative/delayed allocations don't work? I had started on a helper to try to detect the situations where the tests cannot ever pass, but never quite finished it: diff --git a/common/xfs b/common/xfs index 557017c716e32c..5cb2c102e2c04f 100644 --- a/common/xfs +++ b/common/xfs @@ -2238,3 +2238,34 @@ _scratch_xfs_scrubbed() { $XFS_SCRUBBED_PROG "${scrubbed_args[@]}" "$@" $SCRATCH_MNT } + +# Will this filesystem create speculative post-EOF preallocations for a file? +_require_speculative_prealloc() +{ + local file="$1" + local tries + local overage + + # Now that we have background garbage collection processes that can be + # triggered by low space/quota conditions, it's possible that we won't + # succeed in creating a speculative preallocation on the first try. + for ((tries = 0; tries < 5; tries++)); do + rm -f $file + + # a few file extending open-write-close cycles should be enough + # to trigger the fs to retain preallocation. write 256k in 32k + # intervals to be sure + for i in $(seq 0 32768 262144); do + $XFS_IO_PROG -f -c "pwrite $i 32k" $file >> $seqres.full + + # Do we have more blocks allocated than what we've + # written so far? + overage="$(stat -c '%b * %B - %s' $file | bc)" + test "$overage" -gt 0 && return 0 + done + done + + _notrun "Warning: No speculative preallocation for $file after " \ + "$tries iterations." \ + "Check use of the allocsize= mount option." +} --D > Signed-off-by: Christoph Hellwig <hch@lst.de> > --- > tests/xfs/629 | 2 +- > tests/xfs/630 | 2 +- > tests/xfs/631 | 2 +- > tests/xfs/632 | 2 +- > 4 files changed, 4 insertions(+), 4 deletions(-) > > diff --git a/tests/xfs/629 b/tests/xfs/629 > index 58beedc03a8b..e2f5af085b5f 100755 > --- a/tests/xfs/629 > +++ b/tests/xfs/629 > @@ -8,7 +8,7 @@ > # > > . ./common/preamble > -_begin_fstest auto quick prealloc rw > +_begin_fstest prealloc rw > > . ./common/filter > > diff --git a/tests/xfs/630 b/tests/xfs/630 > index 939d8a4ac37f..df7ca60111d6 100755 > --- a/tests/xfs/630 > +++ b/tests/xfs/630 > @@ -8,7 +8,7 @@ > # > > . ./common/preamble > -_begin_fstest auto quick prealloc rw > +_begin_fstest prealloc rw > > . ./common/filter > > diff --git a/tests/xfs/631 b/tests/xfs/631 > index 55a74297918a..1e50bc033f7c 100755 > --- a/tests/xfs/631 > +++ b/tests/xfs/631 > @@ -8,7 +8,7 @@ > # > > . ./common/preamble > -_begin_fstest auto quick prealloc rw > +_begin_fstest prealloc rw > > . ./common/filter > > diff --git a/tests/xfs/632 b/tests/xfs/632 > index 61041d45a706..3b1c61fdc129 100755 > --- a/tests/xfs/632 > +++ b/tests/xfs/632 > @@ -9,7 +9,7 @@ > # > > . ./common/preamble > -_begin_fstest auto prealloc rw > +_begin_fstest prealloc rw > > . ./common/filter > > -- > 2.45.2 > >
On Wed, Oct 23, 2024 at 10:23:51AM -0700, Darrick J. Wong wrote: > On Wed, Oct 23, 2024 at 12:39:30PM +0200, Christoph Hellwig wrote: > > These fail for various non-default configs like DAX, alwayscow and > > small block sizes. > > Shouldn't we selectively _notrun these tests for configurations where > speculative/delayed allocations don't work? Perhaps turning off speculative delalloc by adding the mount option "-o allocsize=<1 fsb>" to these tests would result in them always have the same behaviour? -Dave.
On Thu, Oct 24, 2024 at 08:40:42AM +1100, Dave Chinner wrote: > On Wed, Oct 23, 2024 at 10:23:51AM -0700, Darrick J. Wong wrote: > > On Wed, Oct 23, 2024 at 12:39:30PM +0200, Christoph Hellwig wrote: > > > These fail for various non-default configs like DAX, alwayscow and > > > small block sizes. > > > > Shouldn't we selectively _notrun these tests for configurations where > > speculative/delayed allocations don't work? > > Perhaps turning off speculative delalloc by adding the mount > option "-o allocsize=<1 fsb>" to these tests would result in them > always have the same behaviour? But it also kinda defeats what they are trying to test? At least that's my understanding, but your wrote them, though..
On Wed, Oct 23, 2024 at 10:23:51AM -0700, Darrick J. Wong wrote: > On Wed, Oct 23, 2024 at 12:39:30PM +0200, Christoph Hellwig wrote: > > These fail for various non-default configs like DAX, alwayscow and > > small block sizes. > > Shouldn't we selectively _notrun these tests for configurations where > speculative/delayed allocations don't work? > > I had started on a helper to try to detect the situations where the > tests cannot ever pass, but never quite finished it: If you get that to fully work it seems like a good idea, yes.
On Wed, Oct 23, 2024 at 10:23:51AM -0700, Darrick J. Wong wrote: > On Wed, Oct 23, 2024 at 12:39:30PM +0200, Christoph Hellwig wrote: > > These fail for various non-default configs like DAX, alwayscow and > > small block sizes. > > Shouldn't we selectively _notrun these tests for configurations where > speculative/delayed allocations don't work? > > I had started on a helper to try to detect the situations where the > tests cannot ever pass, but never quite finished it: > > diff --git a/common/xfs b/common/xfs > index 557017c716e32c..5cb2c102e2c04f 100644 > --- a/common/xfs > +++ b/common/xfs > @@ -2238,3 +2238,34 @@ _scratch_xfs_scrubbed() { > > $XFS_SCRUBBED_PROG "${scrubbed_args[@]}" "$@" $SCRATCH_MNT > } > + > +# Will this filesystem create speculative post-EOF preallocations for a file? > +_require_speculative_prealloc() > +{ > + local file="$1" > + local tries > + local overage > + > + # Now that we have background garbage collection processes that can be > + # triggered by low space/quota conditions, it's possible that we won't > + # succeed in creating a speculative preallocation on the first try. > + for ((tries = 0; tries < 5; tries++)); do > + rm -f $file > + > + # a few file extending open-write-close cycles should be enough > + # to trigger the fs to retain preallocation. write 256k in 32k > + # intervals to be sure > + for i in $(seq 0 32768 262144); do > + $XFS_IO_PROG -f -c "pwrite $i 32k" $file >> $seqres.full > + > + # Do we have more blocks allocated than what we've > + # written so far? > + overage="$(stat -c '%b * %B - %s' $file | bc)" > + test "$overage" -gt 0 && return 0 > + done > + done > + > + _notrun "Warning: No speculative preallocation for $file after " \ > + "$tries iterations." \ > + "Check use of the allocsize= mount option." > +} Before we remove these cases from auto group, if above function can help these test cases to be stable passed as expected. I'm glad to consider it at first :) Thanks, Zorro > > --D > > > Signed-off-by: Christoph Hellwig <hch@lst.de> > > --- > > tests/xfs/629 | 2 +- > > tests/xfs/630 | 2 +- > > tests/xfs/631 | 2 +- > > tests/xfs/632 | 2 +- > > 4 files changed, 4 insertions(+), 4 deletions(-) > > > > diff --git a/tests/xfs/629 b/tests/xfs/629 > > index 58beedc03a8b..e2f5af085b5f 100755 > > --- a/tests/xfs/629 > > +++ b/tests/xfs/629 > > @@ -8,7 +8,7 @@ > > # > > > > . ./common/preamble > > -_begin_fstest auto quick prealloc rw > > +_begin_fstest prealloc rw > > > > . ./common/filter > > > > diff --git a/tests/xfs/630 b/tests/xfs/630 > > index 939d8a4ac37f..df7ca60111d6 100755 > > --- a/tests/xfs/630 > > +++ b/tests/xfs/630 > > @@ -8,7 +8,7 @@ > > # > > > > . ./common/preamble > > -_begin_fstest auto quick prealloc rw > > +_begin_fstest prealloc rw > > > > . ./common/filter > > > > diff --git a/tests/xfs/631 b/tests/xfs/631 > > index 55a74297918a..1e50bc033f7c 100755 > > --- a/tests/xfs/631 > > +++ b/tests/xfs/631 > > @@ -8,7 +8,7 @@ > > # > > > > . ./common/preamble > > -_begin_fstest auto quick prealloc rw > > +_begin_fstest prealloc rw > > > > . ./common/filter > > > > diff --git a/tests/xfs/632 b/tests/xfs/632 > > index 61041d45a706..3b1c61fdc129 100755 > > --- a/tests/xfs/632 > > +++ b/tests/xfs/632 > > @@ -9,7 +9,7 @@ > > # > > > > . ./common/preamble > > -_begin_fstest auto prealloc rw > > +_begin_fstest prealloc rw > > > > . ./common/filter > > > > -- > > 2.45.2 > > > > >
diff --git a/tests/xfs/629 b/tests/xfs/629 index 58beedc03a8b..e2f5af085b5f 100755 --- a/tests/xfs/629 +++ b/tests/xfs/629 @@ -8,7 +8,7 @@ # . ./common/preamble -_begin_fstest auto quick prealloc rw +_begin_fstest prealloc rw . ./common/filter diff --git a/tests/xfs/630 b/tests/xfs/630 index 939d8a4ac37f..df7ca60111d6 100755 --- a/tests/xfs/630 +++ b/tests/xfs/630 @@ -8,7 +8,7 @@ # . ./common/preamble -_begin_fstest auto quick prealloc rw +_begin_fstest prealloc rw . ./common/filter diff --git a/tests/xfs/631 b/tests/xfs/631 index 55a74297918a..1e50bc033f7c 100755 --- a/tests/xfs/631 +++ b/tests/xfs/631 @@ -8,7 +8,7 @@ # . ./common/preamble -_begin_fstest auto quick prealloc rw +_begin_fstest prealloc rw . ./common/filter diff --git a/tests/xfs/632 b/tests/xfs/632 index 61041d45a706..3b1c61fdc129 100755 --- a/tests/xfs/632 +++ b/tests/xfs/632 @@ -9,7 +9,7 @@ # . ./common/preamble -_begin_fstest auto prealloc rw +_begin_fstest prealloc rw . ./common/filter
These fail for various non-default configs like DAX, alwayscow and small block sizes. Signed-off-by: Christoph Hellwig <hch@lst.de> --- tests/xfs/629 | 2 +- tests/xfs/630 | 2 +- tests/xfs/631 | 2 +- tests/xfs/632 | 2 +- 4 files changed, 4 insertions(+), 4 deletions(-)