Re: [LSF/MM TOPIC] FS, MM, and stable trees

Re: [LSF/MM TOPIC] FS, MM, and stable trees

[Luis Chamberlain]

On Thu, Mar 10, 2022 at 01:51:22PM -0500, Josef Bacik wrote:
> [root@fedora-rawhide ~]# cat /xfstests-dev/local.config
> [btrfs_normal_freespacetree]
> [btrfs_compress_freespacetree]
> [btrfs_normal]
> [btrfs_compression]
> [kdave]
> [btrfs_normal_noholes]
> [btrfs_compress_noholes]
> [btrfs_noholes_freespacetree]

+ linux-btrfs and zone folks.

I simplified these as follows, please let me know if the names are
alright. I think we may be able to come up with something more
clever than btrfs_dave. The raid56/noraid56 exist just for the
defaults of the distro/btrfs-progs. The expunge list is used
to determine if something is raid56 or not sadly given we
have no groups for putting tests into the raid56 group. The idea
is some distros don't support raid56 so that is the goal with
the noraid56 config.

The name needs to be: $FS_$FANCY_SINGLE_SPACED_NAME

Each guest spawned will have that same hostname. And likewise
the expunges are collected for each guest hostname. The hostname
is used to pick the expunge directory so to ensure it reflects
the baseline.

You may want to look at this expunge file:

https://github.com/mcgrof/kdevops/blob/master/workflows/fstests/expunges/opensuse-leap/15.3/btrfs/unassigned/btrfs_noraid56.txt\x02

[default]
TEST_DEV=@FSTESTSTESTDEV@
TEST_DIR=@FSTESTSDIR@
SCRATCH_DEV_POOL="@FSTESTSSCRATCHDEVPOOL@"

SCRATCH_MNT=@FSTESTSSCRATCHMNT@
RESULT_BASE=$PWD/results/$HOST/$(uname -r)

[btrfs_raid56]
MKFS_OPTIONS='-f'
FSTYP=btrfs

[btrfs_noraid56]
MKFS_OPTIONS='-f'
FSTYP=btrfs

[btrfs_normalfreespacetree]
LOGWRITES_DEV=@FSTESTSLOGWRITESDEV@
MKFS_OPTIONS="-K -f -O ^no-holes"
MOUNT_OPTIONS="-o space_cache=v2"
FSTYP=btrfs

[btrfs_compressfreespacetree]
MOUNT_OPTIONS="-o compress=zlib,space_cache=v2"
MKFS_OPTIONS="-K -f -O ^no-holes"

[btrfs_normal]
LOGWRITES_DEV=@FSTESTSLOGWRITESDEV@
MKFS_OPTIONS="-K -O ^no-holes -R ^free-space-tree"
MOUNT_OPTIONS="-o discard=async"

[btrfs_compression]
MOUNT_OPTIONS="-o compress=zstd,discard=async"
MKFS_OPTIONS="-K -O ^no-holes -R ^free-space-tree"

[btrfs_kdave]
MKFS_OPTIONS="-K -O no-holes -R ^free-space-tree"
MOUNT_OPTIONS="-o discard,space_cache=v2"

[btrfs_normalnoholes]
LOGWRITES_DEV=@FSTESTSLOGWRITESDEV@
MKFS_OPTIONS="-K -O no-holes -f -R ^free-space-tree"

[btrfs_compressnoholes]
MKFS_OPTIONS="-K -O no-holes -f -R ^free-space-tree"
MOUNT_OPTIONS="-o compress=lzo"

[btrfs_noholesfreespacetree]
MKFS_OPTIONS="-K -O no-holes -f"
MOUNT_OPTIONS="-o space_cache=v2"

I see nothing for NVMe ZNS.. so how about 

[btrfs_zns]
MKFS_OPTIONS="-f -d single -m single"
TEST_DEV=@FSTESTSTESTZNSDEV@
SCRATCH_DEV_POOL="@FSTESTSSCRATCHDEVZNSPOOL@"

[btrfs_simple]
TEST_DEV=@FSTESTSTESTSDEV@
MKFS_OPTIONS="-f -d single -m single"
SCRATCH_DEV_POOL="@FSTESTSSCRATCHDEVPOOL@"

The idea being btrfs_simple will not use zns drives behind the scenes
but btrfs_zns will.

  Luis

btrfs profiles to test was: (Re: [LSF/MM TOPIC] FS, MM, and stable trees)

[Luis Chamberlain]

On Fri, Mar 11, 2022 at 06:07:21PM -0800, Luis Chamberlain wrote:
> On Thu, Mar 10, 2022 at 01:51:22PM -0500, Josef Bacik wrote:
> > [root@fedora-rawhide ~]# cat /xfstests-dev/local.config
> > [btrfs_normal_freespacetree]
> > [btrfs_compress_freespacetree]
> > [btrfs_normal]
> > [btrfs_compression]
> > [kdave]
> > [btrfs_normal_noholes]
> > [btrfs_compress_noholes]
> > [btrfs_noholes_freespacetree]
> 
> + linux-btrfs and zone folks.
> 
> The name needs to be: $FS_$FANCY_SINGLE_SPACED_NAME

Actually using_underscores_is_fine for the hostnames so we can keep
your original except kdave :) and that just gets mapped to btrfs_kdave
for now until you guys figure out what to call it.

Likewise it would be useful if someone goees through these and gives me
hints as to the kernel revision that supports such config, so that if
testing on stable for instance or an older kernel, then the kconfig
option for them does not appear.

> I see nothing for NVMe ZNS.. so how about 
> 
> [btrfs_zns]
> MKFS_OPTIONS="-f -d single -m single"
> TEST_DEV=@FSTESTSTESTZNSDEV@
> SCRATCH_DEV_POOL="@FSTESTSSCRATCHDEVZNSPOOL@"
> 
> [btrfs_simple]
> TEST_DEV=@FSTESTSTESTSDEV@
> MKFS_OPTIONS="-f -d single -m single"
> SCRATCH_DEV_POOL="@FSTESTSSCRATCHDEVPOOL@"
> 
> The idea being btrfs_simple will not use zns drives behind the scenes
> but btrfs_zns will.

I went with:

[btrfs_simple]
[btrfs_simple_zns]

If there are other ZNS profiles we can use / should test please let me know.

Thanks,

  Luis

Re: btrfs profiles to test was: (Re: [LSF/MM TOPIC] FS, MM, and stable trees)

[Josef Bacik]

On Mon, Mar 14, 2022 at 03:45:28PM -0700, Luis Chamberlain wrote:
> On Fri, Mar 11, 2022 at 06:07:21PM -0800, Luis Chamberlain wrote:
> > On Thu, Mar 10, 2022 at 01:51:22PM -0500, Josef Bacik wrote:
> > > [root@fedora-rawhide ~]# cat /xfstests-dev/local.config
> > > [btrfs_normal_freespacetree]
> > > [btrfs_compress_freespacetree]
> > > [btrfs_normal]
> > > [btrfs_compression]
> > > [kdave]
> > > [btrfs_normal_noholes]
> > > [btrfs_compress_noholes]
> > > [btrfs_noholes_freespacetree]
> > 
> > + linux-btrfs and zone folks.
> > 
> > The name needs to be: $FS_$FANCY_SINGLE_SPACED_NAME
> 
> Actually using_underscores_is_fine for the hostnames so we can keep
> your original except kdave :) and that just gets mapped to btrfs_kdave
> for now until you guys figure out what to call it.
> 

Lol you didn't need to save the name, I just threw that in there because Sterba
wanted me to test something specific for some patch and I just never deleted it.

> Likewise it would be useful if someone goees through these and gives me
> hints as to the kernel revision that supports such config, so that if
> testing on stable for instance or an older kernel, then the kconfig
> option for them does not appear.
> 

I'm cloning this stuff and doing it now, I got fed up trying to find the
performance difference between virtme and libvirt.  If your shit gives me the
right performance and makes it so I don't have to think then I'll be happy
enough to use it.  Thanks,

Josef

Re: btrfs profiles to test was: (Re: [LSF/MM TOPIC] FS, MM, and stable trees)

[Luis Chamberlain]

On Tue, Mar 15, 2022 at 10:23:53AM -0400, Josef Bacik wrote:
> On Mon, Mar 14, 2022 at 03:45:28PM -0700, Luis Chamberlain wrote:
> > On Fri, Mar 11, 2022 at 06:07:21PM -0800, Luis Chamberlain wrote:
> > > On Thu, Mar 10, 2022 at 01:51:22PM -0500, Josef Bacik wrote:
> > > > [root@fedora-rawhide ~]# cat /xfstests-dev/local.config
> > > > [btrfs_normal_freespacetree]
> > > > [btrfs_compress_freespacetree]
> > > > [btrfs_normal]
> > > > [btrfs_compression]
> > > > [kdave]
> > > > [btrfs_normal_noholes]
> > > > [btrfs_compress_noholes]
> > > > [btrfs_noholes_freespacetree]
> > > 
> > > + linux-btrfs and zone folks.
> > > 
> > > The name needs to be: $FS_$FANCY_SINGLE_SPACED_NAME
> > 
> > Actually using_underscores_is_fine for the hostnames so we can keep
> > your original except kdave :) and that just gets mapped to btrfs_kdave
> > for now until you guys figure out what to call it.
> > 
> 
> Lol you didn't need to save the name, I just threw that in there because Sterba
> wanted me to test something specific for some patch and I just never deleted it.

Heh, I figured, I just didn't know what the hell to name that.

> > Likewise it would be useful if someone goees through these and gives me
> > hints as to the kernel revision that supports such config, so that if
> > testing on stable for instance or an older kernel, then the kconfig
> > option for them does not appear.
> > 
> 
> I'm cloning this stuff and doing it now, I got fed up trying to find the
> performance difference between virtme and libvirt.  If your shit gives me the
> right performance and makes it so I don't have to think then I'll be happy
> enough to use it.  Thanks,

Groovy.

  Luis