►
From YouTube: February 2021 OpenZFS Leadership Meeting
Description
At this month's meeting we discussed: release scheme and numbering; DirectIO PR; pool feature compatibility PR; force export PR.
meeting notes: https://docs.google.com/document/d/1w2jv2XVYFmBVvG1EGf-9A5HBVsjAYoLIFZAnWHhV-BM/edit
A
Hello:
everyone-
we
are
one
minute
past
the
hour,
so
we'll
get
started.
Welcome
to
the
february
2021
edition
of
the
opencfs
leadership
meeting,
I
see
a
few
folks
on
maybe
a
little
lighter
than
usual.
That's
all
right!
A
We
do
not
have
much
on
the
agenda
today,
so
lots
of
opportunities
for
discussion
and
q
a
so
I'll
start
with.
The
one
item
is
the
last
last
month's
meeting
we
talked
about
muhammad
from
seagate
was
talking
about
how
zfs
could
potentially
use
take
advantage
of
the
command
priority
interfaces,
and
there's
gonna
be
a
meeting
tomorrow
with
folks
from
seagate
to
to
talk
about
that.
So,
if
you're
interested
in
that
then
get
in
touch
with
muhammad
on
email
or
on
slack,
the
meeting
is
tomorrow.
A
It
looks
like
11
30..
Is
that
right,
yeah,
11,
30,
pacific
time.
A
B
Yeah,
I've
just
got
one
short
one.
I
don't
know
if
people
saw,
but
the
open,
cfs
2.02
was
tagged
yesterday
and
released.
So
I
pulled
in
some
fixes,
so
nothing
earth
shattering
in
there
just
some
bug
fixes
and
whatnot,
but
it's
available
cool.
C
B
Yeah
so
2o
pretty
much
has
everything
that
you
know.
The
only
big
thing
missing
from
at
the
moment
is:
let
me
go
back
so
right
now,
the
only
new
feature,
that's
in
master
that
isn't
in
2o
is
d-ray
right,
and
so
there
was
no
huge
rush
necessarily
to
get
2.1
out,
so
we
would
like
to
get
direct
iowa
in
it
as
well.
I
know
we'd
like
to
get
that
out
sooner
rather
than
later,
so
we're
working
to
get
that
work
wrapped
up,
but
it's
going
slower
than
I
would
have
liked.
B
So
I
don't
know
that
I
have
an
exact
an
exact
time
for
that
we
were
hoping
soon,
but
there
are
bugs
so
we're
working
on
the
bugs.
C
No
yeah,
those
bugs
slightly
worry
me,
I'm
just
thinking
about
timeline
in
context
of
freebies
13
release
is
going
to
be
branched
this
week
and
no
release.
Obviously
in
march,
but
right
now
we
appeared
in
state.
You
know
in
in
master
and
present
state
of
master
sounds
better
to
me
rather
than
state
of
after
drop
in
something
that's,
as
you
said,
not
exactly
stable,
so
yeah.
B
C
C
No
I'd
say
it's
like
when
it
when
it
was
the
merge
was
done.
There
was
kind
of
discussion,
that's
in
between
you
and
matt
pasey,
and
the
idea
to
have
2.1
release
it
before
timeline.
So
we
would
go
to
12.
2.1,
but
now
2.1
is
delayed
and
we
appear
in
master
middle
of
nowhere
between
2.0
to
0.1
and
how
then
to
follow
stable
branch
for
release
and
after
release
yeah.
C
B
Yeah,
we
don't
want
to
maintain
more
versions
than
we
have
to
right,
so
I
guess
we
should
figure
out
what
exactly
you
guys
need
and
what's
the
best
thing
to
stabilize
on,
like
I
say,
2.0
is
the
thing:
we've
been
probably
the
most
stable
thing
out
there
right.
It's
seen
the
most
testing
and
it's
only
just
missing
that
one.
So
if,
if
if
the
need
is
to
have
derailed,
then.
C
C
B
2.1
out
too,
but
it
has
not
it's
like
it's
been
delayed,
so
I
don't
have
an
exact
date
for
it.
A
B
C
Yeah,
I
I
have
no
problems
with
with
merging
the
things
that
can
be
emerged
that
that
just
what
to
do
is
already
with
the
code
that's
already
pushed
to
master
and
going
to
appear
in
this
particular
institution
already.
Shooting
is
already
branched,
just
not
13.0,
so
it's
going
to
get
too
stable,
we
would
have
to
revert
yeah.
A
If
we
wanted
to
release
a
2.1
based
on
what's
in
master,
now
would
and
then
release
a
2.2
with
with
direct
io
later.
What
would
the
impact
of
that
be?
Like
you?
Have
you
have
the
release?
You
know
the
release.
Work
would
be
more.
B
C
A
B
Yeah,
so
historically,
we've
done
just
one
branch
basically
and
kept
it
up,
which
is
fine,
but
the
branches
used
to
be
pretty
long
lived.
You
know
a
year
18
months,
something
like
that.
We
had
thought
about
keeping
that
you
know
doing
that
again
this
time,
but
I
don't
know,
I
don't
know
that
we
thought
about
it
that
much.
We
could
certainly
keep
doing
point
releases
on
that
branch.
A
It
seems
like
within
the
major
release,
you
know
the
two
release
that
it
should
just
be
one
branch,
and
you
know
if,
if
you
want
the
latest
stuff,
the
latest
then
you're
getting
the
latest
of
release
two
two
dot
whatever
the
biggest
thing
is
after
that.
B
I
think
part
of
the
problem
is
at
least
on
the
linux
side.
The
distributions
would
much
prefer
to
have
something
to
stabilize
on
and
get
bug
fixes
for
just
like
right.
That
would
be
the
two
release
right
yeah,
but
I
think
they
would
view
a
change
from
like
2.0
to
2.1.
That
adds
a
major
feature
as
being
a
bit
of
a
problem
right.
A
B
A
Think
that
we
should
stick
to
like
each
major
release
is
one
branch?
That's
the
stable,
like
that's
the
stable
stuff.
You
want
to
be
on
the
stable
release
you
get
on
to
the
you
know
you
get
onto
the
latest
release
and
you
know
we
maintain
like
two
major
releases
at
a
time.
Roughly,
you
know
less
one
or
two
major
releases
at
a
time
right
at
most
right,
like
maybe
we're
doing
backwards
to
0.8
and
we're
doing
backwards
to
two
to
two
right
now,
but
not
more
than
that.
A
B
B
They
would
like
that
to
exist
for
a
while
right
and
that
had
been
two
out
we,
I
had
been
targeting
2-0
to
get
point
releases
for
a
long
time,
just
with
little
back
ports
and
fixes
and
keep
that
branch
maintained.
That's
still,
honestly,
what
I'd
prefer
and
keeping
two
branches
isn't
terrible,
but
if,
if
d
raid
goes
in
2.1
and
direct
io
goes
in
2.2
right,
it
gets
a
little
show
me.
So
many
ranges.
C
B
C
Yeah
but
then
somebody
would
have
to
maintain
stable
mergers,
practically
maintain
custom,
stable
branch
duplicating
your
work.
That's
not
fun.
B
Yeah
exactly
well,
maybe
we
should
talk
about
it.
Like
maybe
offline,
I
don't
know,
maybe
being
covered
with
something
we
could
yeah.
I
don't
know
what
to
suggest.
C
B
A
I
mean
if
you
want
to
maintain
end
branches
like
it's.
It's
no,
you
know
I'm
not
doing
that
work.
So
if
you
want
to
do
that,
work
go
ahead,
but
I
to
me,
like
it
seems
like
a
better
use
of
our
collect
the
collective,
our
effort
to
maintain
fewer
branches,
fewer
branches
through
release
branches
and
to
say,
like
yeah.
You
know
either
we're
putting
a
major
release,
we're
putting
a
a
major
feature
into
a
into
an
existing
release
or
to
say
that
we're
making
a
new
release
a
new
major
release
for
that
new
major
feature.
C
Okay,
if
we,
if
we're
not
releasing
anything
now
and
just
holding,
taking,
depress
and
waiting
for
how
long
realistically
shall
we
wait
till
next
major
branch
we
if
it
will
be
with
direct
io
but
okay,
then
we
stable
directorio
will
be
another
18
months
or
we
have
something
in
more
realistic
time
frame
to
update
freebsd,
open
zfs
in
freebsd
13.1,
for
example,
just
in
next
miner,
just
to
restore
stable
state
within
reasonable
time
frame.
C
B
I
wouldn't
want
to
make
major
branches
very
frequently
myself.
We
had
talked
before
about
it
being
you
know,
18
months
or
12
months,
something
like
that
between
major
releases,
but
that
seems
like
an
awful
long
time
to
wait.
B
A
A
It
just
seems
to
me
like,
if
you're
I
mean,
if
you're
gonna
create
a
new,
if
you're
gonna
create
a
new
branch
and
you're
gonna
maintain
like
you're
gonna,
maintain
it
separately
as
like
there's
the
old
branch
and
then
there's
the
new
branch,
then,
what's
really
the
difference
between
calling
that
new
one
2.1
versus
2.
versus
3.0,
like
you
either
way
you
have
to
you
know
if
you're
saying
like
I
want
to
create
a
new
branch
and
maintain
and
we're
going
to
maintain
it
and
the
old
one.
B
B
Yeah
yeah
yeah.
I
guess
I'm
not
sure
how
to
answer
exactly
like
I'm
not
particularly
hung
up
on
the
number
of
it.
It's
just
how
many
branches.
To
maintain.
From
my
point
of
view,
I
would
prefer
to
maintain
fewer
right.
I
would
prefer
to
maintain
none.
If
people
want
to
help
step
up
and
maintain
a
branch,
we
can
get
a
maintainer
for
the
brain.
That
would
be
great.
B
So
so
the
plan
of
record,
like
you
said,
was
to
put
out
a
2-1
in
a
month
or
two
and
then
maintain
that
branch
and
the
2-0
branch.
It
sounds
like
that's
too
long
to
wait
for
freebsd.
C
B
Yeah,
so
I
mean
the
plan
of
record
for
2-1
was
to
get
direct.
I
o
in
as
soon
as
it
was
in
branch
it
and
hopefully
get
that
tag
release
out
in
q1.
So
you
know
sometime
in
the
next
two
months
at
least
branch
for
it.
I
don't
know
if
that
works
for
freebsd's
time
frame,
but
that
was
the
intention
and
I
was
going
to
maintain
a
2-0
and
a
2-1
branch.
C
Like
not
workable,
it's
already
in
public
repo,
if
somebody
already
used
it
durate
should
we
tell
sorry
guys
it's
not
known
there.
B
A
A
That
kind
of
locked
you
into
being
dependent
on
this
open,
zfs
release
happening
in
the
right
time
frame
for
you.
What's
the
plan?
Is
the
plan
going
forward
that
freebies
d
will
continue
to
like
use
open,
dfs
master
and
then
and
then
it
will
line
this
up
again
and
have
this
risk
again
or
is
it
going
to
switch
over
to
to
release
to
open
zfs
releases
in
general.
C
My
personal
opinion
would
be
great
to
have
main
one
on
main
respectively
and
get
stable
on
stable
branches,
but
that
would
require
at
least
some
kind
of
synchronization,
so
that
opencfs
would
be
branched
sometime
not
too
far
before
the
freebies,
stable
branched.
C
C
B
C
So
again,
three
busy
major
branches
are
forked
on
quite
a
schedule
and
long
far
in
advance,
and
this
was
now
and
next
will
be
in
about
18
months.
So
it's
kind
of
schedule
more
or
less
strict.
It's
not
feature
based
it's
more
like
more
time
based
so
by
the
time
things
happen
next
time
we
should
bet,
prepare
better.
Do
it
and
coordinate.
B
Yeah,
it
sounds
like
the
the
time
frame
flipping
here
really
caused
more
trouble
than
I
was
expecting
not
getting
this
stuff
done.
You
know
quite
release
out.
C
C
A
Brian,
are
you
for
the
future
releases
like
let's
say
after
this
after
we
release,
whatever
release
has
direct
io
in
it?
Are
you
thinking
that
minor
releases
like
2.2,
2.3
2.4
would
would
also
be
branches
that
we
would
maintain
like
in
parallel
with
with
previous,
you
know,
2.02.1,
or
would
we
not
have
those
generally,
and
we
would
just
have
a
two
one,
one
two
one,
two
two
and
three.
B
I
don't
know
I'm
I'm
open
to
suggestions.
I
was
of
the
I
was
thinking
that
we
could
make.
You
know
you
got
one
2.2
2.3
every.
B
I
don't
know
three
four
or
five
months
whatever
it
is
six
months,
something
like
that
from
something
a
little
more
frequent,
and
it
was
an
open
question
about
how
those
branches
would
all
get
maintained
right
if
we
just
maintain
the
last
one
or
we
do
something
like
you
know,
maintain
the
last
one
and
then
like
say
2.0.
We
just
maintain
for
a
longer
time
kind
of
like
an
lps
sort
of
thing.
A
And
I
think
also
so
that
folks
can
understand
based
on
the
number
like
what
does
that
mean
like
it
would
be
nice?
A
I
think
it'd
be
nice
to
be
able
to
look
at
the
number
and
then
understand
like
okay.
This
is
part
of
the
whatever
you
know
like
2
2.2
is
part
of
the
two
release
branch
right
and
so
like
within
two,
it's
all
just
accumulating
versus
being
like.
Oh,
you
know,
this
is
2.6
well
in
2.
in
2.4
we
made
that
a
maintained
branch.
So
it's
like
you
know,
four
five
and
six
are
are
together
and
then,
like
you
know
one
and
in
one
two
three
are
together.
You
know
what
I
mean
like.
A
A
I
don't
know
if
I
get
the
numbers
right,
but
it
was,
it
seems
like
it
would
be
nice
if
the
version
numbers
told
us
what
like
what's
effectively
a
major
release.
You
know
what
what's
its
own
release
train.
B
What
would
we
consider
a
major
release?
I
guess
would
be
a
question
then
like
what
denotes
something
as
a
major
release,
as
opposed
to
a
minor
release
so
like
with
the
current
version
numbering
we've
pretty
much
adopted
semantic
versioning.
So,
within
those
release
branches,
we've
cut
we're
pretty
much
guaranteeing
like
no
api
changes
for
that
kind
of
thing.
No
library
changes
so
we're
keeping
those
changes
out.
B
Well
like,
for
example,
in
all
the
the
two
branches
now
2.0
2.0.1.2
and
0.3.
There
have
been
no
abi
teams
and
they're
just
saying.
B
A
I
mean
I
thought
that
the
point
of
it
was
that
you're
you're
you're,
taking
less
risk
right
like
it's
changing
more
slowly
and
things
get
things
are
getting
cooked
on
the
main
branch
before
being
back
ported.
B
But
I
think
like
freebsd,
if
they
were
to
track
a
release
branch
too,
my
understanding
is
they
need
a
stable,
abi
too
right
if
they
won't
change
it.
That's
kind
of
expectation.
C
B
C
The
more
important,
even
not
library
interfaces,
but
that
too,
but
much
more
often
happens
between
kernel
user
space.
So
practically
you
can
update
your
kernel
and
still
be
able
to
boot
and
then
update
world
and
separate
step,
so
at
least
that
has
to
work
if
not
well.
Of
course,
you
can
just
have
applique
previous
applications
in
jail
with
previous
applications
and
libraries
and
all
also
supposed
to
work
with
newer
kernel
at
least
one
way.
C
The
compatibility
should
be
that's
why
on
fb,
we
have
shims
for
all
historical
versions
of
zfs
or
at
least
maintaining
main
functionality
so,
but
it
would
be
much
easier
rather
than
maintain
those
to
have
stability
in
upstream
versions.
Of
course,.
B
Yeah
I'd
say
we
want
that
kind
of
compatibility
between
major
versions
too.
Whatever
we
define
a
major
version
to
be
that
should
assuming
there's
no
incompatible
features.
A
A
A
Time,
okay,
so
it
seems
like
we
have
something
to
do
about
what
you
know,
what
we
really
mean
by
a
major
version
and
like
is
2.1
a
major
version,
and
if
so,
why,
like?
What
does
that?
Tell
us
about
the
numbers
that
we're
going
to
use
in
the
future.
C
Yeah,
I
think
the
main
we
should
do
is
write
some
kind
of
timeline,
some
plan.
What
versions
will
be
released
when
of
where
they
be
forked
from
and
how
long
they
are
going
to
be
supported?
Something
like
that
we
got
to
that
last
year
or
two,
because
it
was
quite
random
before,
but
now
it's
pretty
stable.
All
releases
are
announced
at
few
months
in
advance
from
branching
and
then
holds
up
regularly
updated
every
few
weeks.
C
C
A
I
don't
think
that
we
can
be
well.
I
don't
think
that
we
want
to
be
quite
as
strict
as
the
calendar
strictly
calendar-based
freebsd
I
mean
open.
Zfs
is
a
smaller
and
a
younger
project,
but
having
having
the
plan
written
down
somewhere,
including
like
branching
and
what's
gonna,
be
maintained
for
how
long
and
whatnot
would
be
would
definitely
be
good.
B
A
B
I
think
that's
something
we
should
have
to
decide
if
that
constitutes
a
major
release
or
or
what
or
what
kind
of
support
guarantees
we
want
to
offer.
For
that
all
right,
I
think,
for
the
200
release.
We
definitely
want
to
maintain
that
for
a
year
two
years,
whatever
it
is,
whatever
makes
sense
for
that
and
then
maybe
the
newer
two
releases
don't
get
sport
for
quite
as
long.
A
All
right
yeah,
it
sounds
like
I
guess
I
would
like
to
be
able
to
understand
like
assuming
that
like.
If
we
release
this,
if
we
do
this
release,
that
has
new
features.
A
Is
that
a
major
release
or
not,
and
then
how
do
we
make
sure
that
the
version
numbers
going
forward
from
there
makes
sense,
given
the
decision
that
we
make
for
the
version
number
of
the
next
of
of
d-raid
right.
A
All
right
cool!
So
since
I
guess
this
isn't
going
to
happen,
probably
before
the
next
meeting,
why
don't
we
can
we
give
brian?
Can
we
give
you
the
action
item
to
write
up
something
and
and
maybe
send
it
out
before
the
next
meeting,
and
then
we
can
talk
about
it
at
the
next
meeting.
B
Yeah,
I
can
do
something
like
that.
I
may
have
to
pull
somebody
in
to
get
someone
else's
thoughts,
but
if
I
put
something.
A
Together
cool,
oh
I
I
did
have
one
other
question
for
you,
brian,
getting
getting
back
to
the
actual
the
the
non-release
work
for
for
direct
ido
who's
who's
working
on
that
now
is
that
I
know
ryan
atkinson
was
doing
a
lot
of
work
on
that.
Is
he
still
working
on
that?
Are
you
working
on
that?
Is
anybody
else.
B
Yeah
brian
atkinson
is
still
working
on
that.
I'm
working
on
that
with
him,
so
we're
working
on
that
closely
iterating
to
get
it
all
shaped.
So
hopefully
we
should
get
an
updated
forward
class.
I
would
hope
this
week.
Okay,.
A
So
you
guys
are
still
you're
not
looking
for
credit
reviewers
right.
B
Now
soon
soon,
so
I
need
to
make
another
pass
of
the
code
review.
I
mean
it's
functional,
it
works.
It's
just.
You
know
polish
at
this
point,
but
yeah
soon
for
code,
reviewers,
all
right
cool,
that's
great!
B
B
I
don't
know
off
hand
what
the
full
request
number
is
for
that,
but
there's
an
implementation
there
that
lets.
You
basically
set
a
property
that
lets
you
limit,
which
features
get
automatically
enabled
when
you
do
a
pool
upgrade
and
which
ones
get
complained
about
when
you
do
a
vpool
status,
so
you
can
make
sure
that
you
lock
in
a
particular
compatibility
level,
be
great.
If
people
can
look
at
that
and
give
feedback
on
it,
there's
also
an
open
pull
request
for
the
forced
export
work.
We
talked
about
a
few
months
back
too.
A
Cool
yeah,
I
looked
at
the
at
the
feature,
compatibility
thing
and
that
looked
looked
really
good.
It
looked
like
it
was
close
to
being
done
and
I
haven't
looked
at
the
forced
export
stuff.
Yet.
B
I
think
the
forced
export
stuff
could
benefit
from
some
people
looking
at
it
just
to
make
sure
they're
happy
with
the
design
decisions
that
are
being
made
there,
because
it
touches
a
lot
of
things
like
the
transaction
handling
how
to
unwind
things.
It
all
looks
pretty
reasonable,
but
it
touches
a
lot
of
that
stuff.
So.
A
A
A
We
will
see
folks
in
four
weeks,
which
is
I
had
the
calendar
up.
I.
C
A
It's
the
second
again
2nd
of
march,
not
september,
contrary
to
my
tweet,
which
I
don't
know
where
september
came
from.
But
september
is
what
came
out
of
my
fingers
when
I
typed
that
yeah.
So
the
next
meeting
will
be
at
the
same
time
as
this
one
o'clock
pacific
on
march,
2nd
and
again,
if
you're
interested
in
the
interfaces
with
disk
drives
there'll
be
that
meeting
tomorrow
tomorrow
morning,
pacific
time,
all
right
thanks,
everyone.