►
From YouTube: May 2021 OpenZFS Leadership Meeting
Description
At this month's meeting we discussed: new maintainers; vdev properties; userland zfs; FreeBSD summit; OSX port
meeting notes: https://docs.google.com/document/d/1w2jv2XVYFmBVvG1EGf-9A5HBVsjAYoLIFZAnWHhV-BM/edit#
A
I
sent
an
email
about
this.
You
may
have
seen
it
but,
as
you
guys
probably
know,
brian
bellandorf
has
been
doing
like
almost
all
of
the
work
of
maintaining
the
the
prs
and
issues
in
opengfs
for
the
past.
I
don't
know
close
to
a
decade
now
and
as
the
project's
grown
we
have
you
know
more
and
more
volume
of
contributions
which
is
great
and
but
it's
it's
quite
a
workload
for
one
person
to
do
so.
We've
recruited
a
few
new
maintainers.
A
Hopefully,
brian
will
be
able
to
remain
in
kind
of
a
catch-all,
consulting
role
and-
and
I
will
probably
continue
on
like
helping
out
other
folks
but
not
owning
areas.
Myself
and
tony
hutter
will
continue
managing
the
back
ports
and
doing
the
back
ports
yeah,
as
well
as
the
zed,
the
event
daemon
cool.
A
So
thanks
to
our
three
new
volunteers,
hopefully
you
will
find
it
rewarding
to
contribute
to
the
community
in
this
way
and
thanks
for
your
efforts,
so
you
probably
saw
a
few
of
those
integrations
happen.
I
think
just
yesterday
starting
but
look
out,
you
know.
Oh,
the
last
thing
I
should
mention
is
that
we're
using
the
assignee
of
each
pr
to
indicate
which
maintainer
is
responsible
for
it?
A
So
shortly
after
you
open
the
pr,
you
should
see
that
somebody
one
of
the
maintainers
was
assigned
to
it
and
they're
kind
of
your
point
of
contact
if
you're
like,
why
is
my
request
not
getting
integrated,
or
what
do
I
need
to
do
next?
How
do
I
find
people
to
review
this?
The
the
assigned
code
reviewer
there
in
github
is
a
person
that
you
should
ask
those
questions
of.
B
A
A
Yeah
concrete
thing
that
that
would
apply
to
today
would
be
like
a
shift.
For
example,
where
I
mean.
Maybe
this
is
not
a
good
idea,
but,
like
day
shift
is
specific
to
each
v
dev.
It
may
not
be
a
good
idea
to
have
different
ones
for
different.
You
know
normal
class
v
devs,
but
you
can
do
it
and
right
now
you
have
the
like
dash
o
a
shift
equals
whatever,
and
that
applies
to
at
least
all
the
normal
class
ones.
I
don't
know
if
it
applies
to
other
ones,.
A
C
D
A
A
little
tricky
in
terms
of
like
how
you
would
want
to
parse
the
command
line.
It
would
be
much
easier
if,
if,
like
you
had
to
add,
add
v
devs
one
at
a
time.
You
know
if
there
was
a
like
zip,
create
that
created
some
like
empty
v
dev
and
then
you
would
like
add
each
veto.
Sorry,
an
empty
pool
with
no
v
devs
and
then
afterwards,
like
add
one
video
at
a
time,
but
that's
not
how
it
works.
Today,
I
guess
we
could.
A
B
A
Especially
when
you
have
like
you
have
to
handle
both
like
properties
that
keep
that
are
set
on
a
top
level,
vita
versus
ones
that
are
on
leafs
yeah
and
you
might
have
both
or
neither
you
might
have
one
vita,
which
is
both
the
top
level
and
the
leaf.
So
that
seems
pretty
tricky.
So
I
mean
maybe
what.
E
A
E
B
E
B
Right
now,
most
of
the
properties
that
the
bdf
property
stuff
is
adding
are
read
only
or
whatever,
or
would
be
able
to
be
set
after
without
a
disadvantage.
But
at
some
point
you
could
see
it
being
different
like
even
I
don't
know
if
it
makes
sense,
but
controlling
the
number
of
meta,
slabs
or
the
size
of
the
meta
slabs
or
something
or
anything,
that's
create
time.
Only
like
that.
E
B
A
A
B
B
But
you
know
getting
just
the
the
pool
status
other
than
online
or
degraded
is,
is
one
of
the
properties
of
the
pool,
but
that
you
know
the
scan,
progress
or
just
the
there's
the
action
line.
If
something
is
wrong
and
a
bunch
of
other
ones,
there's
no
way
to
extract
those
other
than
you
know
pursing
the
human
readables
equal
status.
B
I
looked
at
doing
it,
but
a
lot
of
the
logic
for
it
all
lives
in
the
user
space
side,
which
is
a
little
weird
to
do
with
properties
like
I
think
for
some
of
them.
You
know
we
can
pass
the
enum
or
whatever
value
as
the
property,
and
then
libsetfs
can
convert
that
into
the
string,
but
for
some
of
them
it
it
looks
a
little
harder
to
do.
Just
all
the
text
comes
from
either
the
z-pool
command
or
lib
zfs,
not
from
the
other
side
of
zfs.
A
A
Yeah
I
mean
I
feel
like
action
especially
is
like
complicated
but
scan
and
those
related
ones,
I
think,
could
probably
be
made
machine
personable
pretty
straightforwardly.
I
mean
they
are.
Basically
all
those
numbers
are
just
coming
up
from
the
kernel
and,
like
I
wouldn't
bother
with
say
the
estimate,
but
just
which
is
calculated
in
user
land,
but
being
able
to
report
like
there
is
a
scan
in
progress
and
like
this
is
how
much
it's
done,
and
this
is
how
much
it
has
to
do.
G
G
A
B
B
A
And
to
not
worry
about
that
too
much,
and
then
you
know,
I
think
that
we
should
kind
of
separately
address
the
problem
of
like
having
some
user-friendly
easy-to-read
output.
That's,
like
hey,
tell
me
about
this
thing
and
which
you
know
might
include
some
properties,
but
not
all
of
them
like
we
could.
A
Come
up
with
some
aliases
like
that
or
you
know,
come
up
some
new
sub
command,
that's
like
zfs
info
data
set
and
then
it's
like
a
full
screen
text.
Display
of
you
know
various
stuff
about
it.
You
know
I'm
just
making
stuff
up,
but
you
know
like
I
would
solve
that
problem
separately.
Then
then,
making
like
get
all
be
pretty
because
it's
clearly
not
as
it
stands
today.
D
H
A
A
H
H
So
we
disabled
this
file
system,
jfs
file
system
and
exported
this
zfs
volume
through
the
ice
casing,
so
that
iscsi
part
we
have
returned
and
they
have
also
added
support
for
the
application
and
the
rebuilding.
So
with
that,
actually
we
can
run
this
multiple
jfs
process
in
user
space,
which
provides
the
replication
and
also
the
rebuilding
if
any
replica
gets
out
or
if
you
want
to
add
any
new
replica.
So
that
way,
it
adds
support
for
the
multi-tenancy
environment.
A
H
A
A
Yeah,
I
think
that
would
I
think
that
people
probably
have
a
opinion
on
what
on
the
perform.
They
probably
have
a
preconceived
notion
on
the
performance
of
user
land
file
systems
and
zfs
due
to
experience
with
fuse,
and
if,
if
we
can
help
to
dispel
that
I
mean
if
it
is
faster,
then
that
would
definitely
be
helpful
to
you
know:
inform
people
that
hey
this
is.
This
is
not
just
like
a
toy.
It
can
be
used
for
production
and
get
good
get
performance.
A
A
Aside
from
that,
I
guess
what's
the
status
like
are
you,
I
saw
you
open
the
blue
request,
it's
like
not
huge.
It's
only
about
a
little
more
than
a
thousand
lines
of
code.
Would
you
is
this
ready
for
review?
Is
this
like
the
whole
thing,
or
are
we
waiting
for
additional
changes
that
you're
planning
to
make.
H
H
F
H
A
Cool
and
it
you
mentioned
that,
there's
changes
required
to
the
useless
iscsi
server
have
have
you
engaged
with
whatever
the
community
is
that
maintains
that
to
see
you
know
if
they
would
accept
those
changes
upstream,
because
that's
kind
of
you
know
for
folks
who
are
not.
You
know
you
using
in
your
product.
It
would
be
nice
if
there
was
like
a
stan.
You
know
we
need
some
way
to
actually
use
this.
D
D
A
D
F
A
A
And
yeah
I
mean,
I
think
that
I
would
love
to
see
this
idea
of
userland
zfs
become
more
useful
for
more
use
cases,
I'm
also
interested
in
portability
of
the
userland
zfs.
Like
I
I'm
assuming
that
you're
just
running
this
on
z
on
linux,
it
would
be
really
cool
if
we
could,
if
we
could,
you
know
if
we're
only
using
like
posix
interfaces,
then
in
theory
it'll
work
on
any
unix
which
would
be
really
cool.
H
H
A
I
think
we
need
to
figure
out
like
it
sounds
like
you
know,
integrating
it
with
the
iscsi
server
nice,
because
the
server
is
not
something.
That's
you
know
part
of
our
code
base.
Obviously,
so
we
need
to
figure
out
like
what
that
is,
how
universal
it
is,
or
would
we
integrate
this
and
kind
of
be
like
if
you
want
to
use
this,
you
have
to
be
using
this
other.
A
B
Hacky
way,
I
think
one
thing
we
talked
about
a
couple
years
ago
me
you
and
paul
dodeck
was
being
able
to
do
send
receives
in
user
space
for
even
just
for
the
to
improve
the
send
receive
tests,
and
that
might
be
a
relatively
straightforward
thing
and
something
that
wouldn't
be
too
complicated
for
the
user
space
one
because
it's
mostly
just
you
know,
doing
the
zfs
bits
and
reading
or
writing
to
a
file.
Descriptor.
H
A
Yeah,
okay,
so,
like
you,
have
your
own
protocol
for
reading
and
writing
z
vols
that
the
iscsi
server
talks
to
right.
Yes,
yeah,
okay!
So
then
I
mean,
if
that's
going
to
be
integrated,
then
we
would
need
some
way
of
testing
that
either
with
a
dependency
on
this
external
isolde
server
or
with
you
know,
some
kind
of
test
programs.
That
would
talk
the
protocol
that
that
you've
invented.
H
A
Cool
yeah.
That
would
be
great,
I
mean
I
think,
that
it
probably
wouldn't
be
too
hard
to
actually
hook
that
up
with
the
there's
seraphim.
You
might
remember
what
the
name
of
this
thing
is,
but
the
you
know
on
linux,
there's
something
like.
Oh
it's
there's.
One
of
them
is
called
views,
buse,
that's
like
block
driver
and
user
space.
I
I
A
C
F
G
A
A
A
G
F
A
That
is
another
open,
zfs
platform
right
in
addition
to
linux
and
previoust,
I
think
we
would
want
to
see
kind
of
similar
at
least
a
plan
for
getting
to
similar
level
of
functionality,
including
like
support
for
tunables
and
observability,
like
for
example.
Well,
I
guess
things
like
a
lot
of
that.
Stuff
is
just
through
iactal.
So
as
long
as
you
implement
all
the
actuals,
then
things
like
xepo
xepo
iostat
use
the
iactals,
so
that
might
be
sufficient.
A
A
Is
there
is
a
very
rudimentary
way
to
set
tunables
in
in
user
land
that
I
think
zdb
or
z
test
has
like
a
dash
o
and
then
it
basically
like
sets.
It
uses
like
the
runtime
linker,
to
look
up
the
symbol
with
that
name
and
then
sets
it.
So
it's
like
very
janky,
but
it's
kind
of
analogous
to
what
we
have
on
had
or
had
on
solaris,
which
is
like
you
can
use
the
debugger
to
book
values
into
global
variables.
A
C
Yes,
I
mentioned
at
the
last
leadership
meeting
and
I've
been
trying
to
drum
up
some
interest
in
reviewing
this
since
then,
and
we
figured
out
the
problem
with
the
freebsd
test.
That
was
failing.
So
it's
ready
for
review
again,
so
I'm
just
trying
to
get
some
reviewers
and
maybe,
in
light
of
the
stuff
you
announced
in
the
beginning
of
this
call,
we
can
figure
out.
Who
should
be
this?
I
need
to
this
and
the
point
of
contact
that
I
should
be
bugging.
A
A
K
A
A
It's
online.
I
think
it's
on
zoom
and
it's
free
to
attend
and
I'll.
Be
I'm
going
to
be
giving
a
talk
on
friday
that
friday,
the
11th
about
raids,
the
expansion
hopefully
featuring
at
least
announcements
that
might
be
interesting
to
developers,
if
not
end,
users
of
redzig's
expansion,
hoping
to
have
the
final
pr
out
by
then.
D
B
A
B
It
I
don't
think
so.
There's
one
talk
on
thursday
about
a
kind
of
recorder
for
our
storage
subsystem,
cam,
that
you
know
not
really
zfs,
but
if
you're
into
storage
and
finding
out
what
happened
to
your
storage
or
whatever.
That
might
be
interesting,
and
then
the
continuous
integration,
brainstorming
and
planning
might
be
of
interest
to
some
people
figuring
out
how
to
improve
previously
integration
with
open
zfsci
and
so
on.
B
B
A
B
B
And
I
just
reached
out
to
mark,
but
if
anybody
we're
also
looking
to
finish
reviewing
and
land
the
forced
export
stuff,
I
think
we
have
the
last
failing
test
sorted.
Finally,
and
that
will
be
handy
if
encountered
a
couple
of
people
in
chat
over
the
last
couple
of
months,
who
would
have
benefited
from
being
able
to?
You
know,
force
export
the
the
pool
that
was
on
the
usb
stick,
that
the
cat
stole
or
whatever
things
of
that
nature.
A
A
B
A
A
L
L
A
L
A
L
L
L
J
L
A
B
B
A
B
I
think
the
right
thing
to
do
probably
is
just
instead
of
what
the
the
if
deaths
they
have
here
is
just
make
a
macro
that
decodes
it
properly
and
use
that
for
accessing
it.
Instead
of
the
struct
we
had
been
using
because
the
struct
is
the
wrong
way
around
on
the
opposite,
endian,
but
yeah,
I
think
we
should
be
able
to
just
make
it
always
be
the
way
it
is
now
and
not
like.
We
did
the
right
thing
and
wrote
it
to
the
disc
forced
to
be
in
big
endian.
It's
just
when
we
were.