Re: [PATCH v5 0/9] fstests: add tests for btrfs' raid-stripe-tree feature

[Johannes Thumshirn <Johannes.Thumshirn@wdc.com>]

On 08.12.23 02:19, Anand Jain wrote:
> 
> 
> On 12/7/23 17:41, Filipe Manana wrote:
>> On Thu, Dec 7, 2023 at 9:03 AM Johannes Thumshirn
>> <johannes.thumshirn@wdc.com> wrote:
>>>
>>> Add tests for btrfs' raid-stripe-tree feature. All of these test work by
>>> writing a specific pattern to a newly created filesystem and afterwards
>>> using `btrfs inspect-internal -t raid-stripe $SCRATCH_DEV_POOL` to verify
>>> the placement and the layout of the metadata.
>>>
>>> The md5sum of each file will be compared as well after a re-mount of the
>>> filesystem.
>>>
>>> ---
>>> Changes in v5:
>>> - add _require_btrfs_free_space_tree helper and use in tests
>>> - Link to v4: https://lore.kernel.org/r/20231206-btrfs-raid-v4-0-578284dd3a70@wdc.com
>>>
>>> Changes in v4:
>>> - add _require_btrfs_no_compress to all tests
>>> - add _require_btrfs_no_nodatacow helper and add to btrfs/308
>>> - add _require_btrfs_feature "free_space_tree" to all tests
>>> - Link to v3: https://lore.kernel.org/r/20231205-btrfs-raid-v3-0-0e857a5439a2@wdc.com
>>>
>>> Changes in v3:
>>> - added 'raid-stripe-tree' to mkfs options, as only zoned raid gets it
>>>     automatically
>>> - Rename test cases as btrfs/302 and btrfs/303 already exist upstream
>>> - Link to v2: https://lore.kernel.org/r/20231205-btrfs-raid-v2-0-25f80eea345b@wdc.com
>>>
>>> Changes in v2:
>>> - Re-ordered series so the newly introduced group is added before the
>>>     tests
>>> - Changes Filipe requested to the tests.
>>> - Link to v1: https://lore.kernel.org/r/20231204-btrfs-raid-v1-0-b254eb1bcff8@wdc.com
>>>
>>> ---
>>> Johannes Thumshirn (9):
>>>         fstests: doc: add new raid-stripe-tree group
>>>         common: add filter for btrfs raid-stripe dump
>>>         common: add _require_btrfs_no_nodatacow helper
>>>         common: add _require_btrfs_free_space_tree
>>>         btrfs: add fstest for stripe-tree metadata with 4k write
>>>         btrfs: add fstest for 8k write spanning two stripes on raid-stripe-tree
>>>         btrfs: add fstest for writing to a file at an offset with RST
>>>         btrfs: add fstests to write 128k to a RST filesystem
>>>         btrfs: add fstest for overwriting a file partially with RST
>>>
>>>    common/btrfs        |  17 +++++++++
>>>    common/filter.btrfs |  14 +++++++
>>>    doc/group-names.txt |   1 +
>>>    tests/btrfs/304     |  56 +++++++++++++++++++++++++++
>>>    tests/btrfs/304.out |  58 ++++++++++++++++++++++++++++
>>>    tests/btrfs/305     |  61 ++++++++++++++++++++++++++++++
>>>    tests/btrfs/305.out |  82 ++++++++++++++++++++++++++++++++++++++++
>>>    tests/btrfs/306     |  59 +++++++++++++++++++++++++++++
>>>    tests/btrfs/306.out |  75 +++++++++++++++++++++++++++++++++++++
>>>    tests/btrfs/307     |  56 +++++++++++++++++++++++++++
>>>    tests/btrfs/307.out |  65 ++++++++++++++++++++++++++++++++
>>>    tests/btrfs/308     |  60 +++++++++++++++++++++++++++++
>>>    tests/btrfs/308.out | 106 ++++++++++++++++++++++++++++++++++++++++++++++++++++
>>>    13 files changed, 710 insertions(+)
>>> ---
>>> base-commit: baca8a2b5cb6e798ce3a07e79a081031370c6cb8
>>
>> Btw this base commit does not exist in the official fstests repo.
>> That commit is from the staging branch at https://github.com/kdave/xfstests
>>
>> A "git am" will fail because the official fstests repo doesn't have
>> _require_btrfs_no_block_group_tree() at common/btrfs,
>> so it needs to be manually adjusted when applying the 3rd patch.
>>
>> I tried the tests and they look good, so:
>>
>> Reviewed-by: Filipe Manana <fdmanana@suse.com>
>>
>> One question I missed before. Test 304 for example does a 4K write and
>> expects in the golden output to get a 4K raid stripe item.
>> What happens on a machine with 64K page size? There the default sector
>> size is 64K, will the write result in a 64K raid stripe item or will
>> it be 4K? In the former case, it will make the test fail.
>>
> 
> Testing on a 64K pagesize. Will run it. Apologies for intermittent
> responses; OOO until December 21.

Thanks Anand!

I don't have a 64k page size system to test, but I _think_ Filipe is 
right, that will fail. I think we should skip these tests on non 4k sectors.