►
From YouTube: July 2019 OpenZFS Leadership Meeting
Description
We discussed the Developer Summit conference; ZoL + SIMD; waiting for TRIM to complete; RAIDZ removal and expansion; and the feature compatibility matrix.
Meeting notes: https://docs.google.com/document/d/1w2jv2XVYFmBVvG1EGf-9A5HBVsjAYoLIFZAnWHhV-BM/edit?ts=5d125dd0#
A
A
A
I
would
love
to
hear
from
every
one
of
you
about
what
cool
stuff
you're
building
into
ZFS
or
building
with
CFS,
or
you
know
any
of
your
other
thoughts
related
to
is
the
efest
in
the
community
in
a
presentation
at
the
conference,
so
send
me
an
email
with
your
presentation
proposal
you
have
until
August
19th.
So
that's
a
little
bit
less
than
one
month
to
send
me
your
what
you
want
to
present
on
at
the
conference
for
those
who
haven't
been
before
the
the
conference
is
two
days.
A
The
first
day
is
presentations
to
the
whole
group
in
in
a
kind
of
auditorium
room
and
then
the
second
day
is
probably
will
have
some
shorter,
more
informal
presentations
and
discussions
in
the
morning
and
and
then
a
lot
of
a
hackathon.
Will
you
have
a
chance
to
work
with
folks
that
you
may
only
know
from
this
call
on
some
some
little
enhancement
or
thing?
A
That's
bugging
you
about
CFS
and
the
goal
is
kind
of
have
something
to
show
for
it
does
it
have
to
be
done,
and
this
is
not
a
work
on
a
long
time
back
on?
It's
just.
You
know
a
few
hours
that
will
have
to
throw
something
together.
Maybe
the
concept
or
maybe
it's
just
researching
something
and
coming
up
with
a
more
concrete
proposal
or
design,
and
the
conference
will
be
at
the
same
location
as
last
year.
B
A
A
C
D
D
A
B
D
A
A
B
A
I
think
we
hold
people
last
year
and
at
least
most
of
the
folks
didn't
didn't,
have
a
problem
with
that.
But
yeah
there's
some
discount
tickets
available
and
if
it
is
a
financial
burden,
then
reach
out
to
us,
and
we
can
also
okay,
see
ya.
The
alan
was
mentioning
that
the
freebsd
foundation
may
have
travel
grants
as
well
and.
A
A
F
A
I
mean
did
this:
this
kind
of
stuff
has
been
supported
for
10
years
in
the
hardware
like
there's,
probably
ten
at
least
ten
year-old
CPUs
that
have
support
for
some
of
these
instructions
and
the
later
ones
have
more.
You
know
more
different
cancer
instructions
and
the
GOL
code,
basically
like
it
tests
all
the
different
ways
that
you
can
do
stuff.
It
has
new
limitations
for
each
of
the
kind
of
subsets
of
instructions
that
are
available,
and
then
it
tests
up
see
which
ones
are
the
fastest.
Okay,.
C
C
F
Specifically,
the
ability,
the
presence
of
the
instructions,
is
in
the
CPU
ID
thing
and
it
determines
at
runtime
whether
it
can
use
any
of
these
sets
of
instructions
and
which
particular
ones.
There
are
multiple
versions
of
the
vector
instructions
and
whether
or
not
it
can
use
the
AES
instructions.
We
say.
F
Right
correct
the
the
issue
was
that
for
reasons
I
won't
get
into
Linux
ended
up
removing
the
interface
ability
to
have
the
kernel
saved
a
floating
point
register
set,
which
is
a
necessary
portion
of
this.
As
a
result,
Brian
had
to
go
and
put
the
code
to
do
that
back
in
the
cradle,
and
this
is
dumb,
but
it
does
work.
I
will.
A
Yeah
yeah
I
think
this
has
been
in
gol
for
quite
some
time
several
years,
but
as
far
as
I
know,
it's
not
available
and
I
need
the
other
operating
systems.
I
imagine
you
know
it's
mostly
a
matter
of
kind
of
doing
what
Brian
did
recently,
which
is
you
know,
figuring
how
to
save
and
restore
the
registers
correctly.
On
on
on
your
specific
operating
system,.
C
C
G
H
A
B
A
You,
the
kind
of
on
demand,
trim
feature
where
you
do
default
written.
You
can
track
the
status
of
how
much
is
doing,
but
if
you
have
Auto
trim
on
you,
there's
no
way
to
measure
how
many
unmapped
are
queued.
That
haven't
been
done
yet,
and
we
have
a
client
that
has
a
pullback
by
a
slow,
remote
device
and
they
want
to
be
able
to
tell
once
all
the
they're
done
before
they
export
the
pool.
And
so
I
was
thinking
something
like
the
the
freedom
property.
A
We
have
a
sink
destroy,
be
able
to
tell
how
much
of
how
big
the
range
tree
of
things
you
want
to
unmask
is,
though,
or
something
to
that
effect
yeah.
How
can
you
remind
me
how
exactly
we
keep
track
of
like
how
that
batching
works
of
the
continuous
trimming
like
I?
Remember,
there's
something
where
it
takes
like
32,
TX
Gees
worth,
and
then
it
like
issues,
all
those
and
then
it's
bashing
up.
The
next
set
of
32
is:
is
that
I
pretty
right
how
it
works,
because
I
was
mostly
familiar
with
the
freebsd
one?
C
A
By
the
wayside,
yeah
but
I
know
there
is
a
delay,
partly
so
that
if
we're
just
going
to
overwrite
the
LBA,
we
don't
bother
trimming
you
yeah
and
so
on.
But
as
I
recall,
that
was
pretty.
It
was
like
some
kind
of
fixed
number
of
ticks
geez,
so
it
may
be
interesting
to
have
a
functionality.
That's
basically
like
wait.
Please
wait
for
all
the
queued
up
trimming
work
to
be
completed,
which
would
wait
for
like
that
number
of
keys,
or
something
like
that.
A
I
mean
like
under
the
full
sync
man,
but
for
like
example,
people
so
like.
Maybe
it's
like
a
flag
dizzy,
full
sync
or
something
like
that.
You
know
use
evil,
sync
trim
and
it
it
waits
for
all
the
trim
used
to
be
done.
I
think
that
might
be
I
think
that
makes
all
that
use
case
a
little
bit
better
because
the
because
it's
like
a
continuous
activity
of
you
know
more
stuff
being
queued
up
to
free
and
then
like
something
freed
like
that
value
is
gonna,
just
be
jumping
all
over
the
place.
I
A
That
is,
this
is
more
around
waiting
for
the
user
initiated
things,
so
you
can
say
like
wait
for
scrub
to
complete
wait
for
device
removal
to
complete
wait
for
initialization
to
complete.
It
does
also
have
wait
for
wait
for
the
freeing
property
to
go
to
zero
and
I.
Remember.
There
was
some
kind
of
interesting
design
traces
around
how
exactly
that
worked.
I.
A
A
C
C
A
I
A
I
A
C
I
will
try
to
reproduce
it
with
the
high
school
about
south
of
petabyte.
It's
very
much
I
have
a
system
where
I
can
try
to
reproduce
it,
but
it's
time
to
time
it
can
work
correctly.
But
after
some
crashes
it
has
on
import
operation
and
I
have
to
use
some
recovery
operation
in
the
import
command,
and
only
after
that
who
can
be
a
curation.
A
C
A
A
A
C
C
A
Too
easy
yeah,
yeah
I,
don't
really
have
any
progress
reports
since
the
since
that
preview
that
I
posted
a
couple
months
ago,
I
was
hoping
that
would
get
some
some
takers
who
would
test
it
out.
I
think
I
need
to
ping
ping.
Some
folks
who
expressed
interest
earlier
but
I
haven't
followed
through
yet.
A
So
yeah
we
need
some
testing
I'm,
also
hoping
to
get
some
more
time
personally
to
spend
on
continuing
to
get
it
all
working.
But
right
now
it's
definitely
like
it's
used.
It's
the
functionality.
Is
there
Don
dis
forma
is
going
to
change,
so
you
know
you
can't
use
it
in
production
by
any
means,
but
like
basically,
all
the
functionality
is
there
and
there's
a
few
race
conditions
that
need
to
be
closed
and
it
needs
to
be
taught
to
him
taught
about
if
there
are
device
failures
during
the
during
the
extension.
C
A
Yeah,
you
can
you,
you
should
I
mean
I,
haven't
tested
that
nothing
in
the
design
that
that
it
should
just
work
like
he
should
be
able
to
have
multiple
raids,
eat
up
all
the
devices
and
add
one
disk
to
one
of
them
and
then
add
one
disk
to
another
in
one
of
them
and
then
at
the
second
is
to
the
first
one
or
whatever
like
it.
It
doesn't
really
care
about
there
being
other
top
level
vetoes
and.
G
G
A
A
A
But
if
you
actually
want
to
preserve
the
redundancy,
then
it
can't
really
be
done
in
the
same
manner
as
expansion,
because,
like
the
expansion
is
just
looking
at
the
allocated
space
and
like
rearranging
in
the
disks,
it
isn't,
it
isn't
rewriting
any
existing
blocks
or
apparently
it's
moving
the
parity
and
the
existing
blocks
around,
but
it
isn't
like
recomputing
parody.
But
if
you
wanted
to
shrink
it,
then
you
would
have
to
you
know
if
you
have
like
four
disks.
A
If
you
have
a
five
disks
and
one
of
them
is
parity,
and
now
you
would
have
four
disks
and
one
the
new
parity
like
now.
This
one
has
to
be
the
parity
of
diesel
of
three.
Not
this
would
be
the
parity
of
these
other
four
and
there's
really
no
way
around
that.
So
it
would
be
much
more
complicated
and
you
know
okay.
C
A
You
know
for
data
and
one
parity
and
then,
if
I
move
that
to
some
other
raid-z,
that's
also
really
one.
That's
also
five
wide.
If
I
have
one
like
one
logical
block
that
goes
across
these
five
and
I,
take
the
the
first
three
sectors
and
I
put
them
on
the
first
three
sectors
here,
but
then
I
take
the
second
two
sectors
and
put
them
on
those
same
discs.
A
A
A
F
A
A
A
C
A
C
A
C
D
J
A
A
K
K
A
And
I
think
this.
This
call
this
time
is
too
early
for
Jurgen
to
join
us,
so
I
won't
ask
him
for
him.
I
just
want
to
check
it
check
in
with
him
separately
on
OS,
X
or
Windows
stuff,
but
it
looks
like
he's,
definitely
updated
a
few
of
these
and
he
has
like
the
trim
of
the
new
waves
already.
You
know
a
sex
so
he's
continuing
to
pored
over
stuff
from
ZFS
on
Linux.
I
A
I
think
the
the
idea,
this
especially,
was
to
track
kind
of
work.
That's
in
progress,
but
rather
than
like
feature
requests
or
you
know,
feature
ideas,
so
I'm
not
really
that
I
don't
really
know
what
whether
that
applies
to
this
Apple
work
for
for
Linux
or
not.
Do
you
know
what
the
status
of
that
is.
I
All
I
know
that
there's
a
PR
out
there
that
people
have
been
asking
for
code
review
on
and
yeah
I'm
I,
don't
think
it's
it's
alway
there
yet
to
be
put
in
through,
but
yeah
just
with
how
we
have
the
other
people
starting
to
base
off
at
ZFS.
I.
Don't
know
if
that's
gonna
be
an
issue
with
NFS
not
being
fully
there
yeah.
A
That
your
specific
question
it
sounds
like
there
is
this
is
a
work
in
progress.
Is
the
pull
request
that
you're
talking
about
the
and
interest
before
I
call
support
yet
yeah
yeah
I
mean
I,
let's
go
ahead
and
add
it
then
I
think
any
like
significant.
You
know
know
where
these
work
we
should
add
in
here.
We
have
others
kind
of
somewhat
things.
I
That
would
be
my
understanding.
I
mean
I,
I
kind
of
feel
that
was
like
a
base
tenant
feature
of
ZFS
from
from
the
get-go
and
it
just
never
had
the
proper
OS
backing
to
really
get
implemented.
Lon
Linux
until
at
some
point
you're
within
the
last
couple
years
and
you're,
taking
advantage
of
that
time
to
do
that
now,
yeah.
A
I
mean
I
guess
then
we
could
argue
that
that
should
be
in
the
earlier
category.
Then,
since
you
know
it
is
present
on
the
most,
but
maybe
I'll
read
some
more
about
how
exactly
it
works
to
see
if
I
want
to
move
it
up,
there
yeah
I,
think
tracking
that
and
any
other
work
in
progress
that
folks
think
should
be
on
here.
We
we
can
definitely
go
ahead
and
add
it
generally
we're
talking
about
features
which
I
think
this
definitely
is
one.
C
A
G
A
F
A
J
F
F
C
F
A
A
There
does
ed
standard
stuff,
I,
finally
got
a
new
PR
for
it
opened
up
a
couple
weeks
ago,
cool.
This
is
basically
reporting
the
latest
version
from
FreeBSD
over
to
0l
and
could
use
a
little
help.
It
passes
all
the
Linux
build
tests,
but
it
fails
to
build
in
all
of
the
like
test,
builds
I
penned
slightly
confusing.
So
well
so
wait
what
test.
A
So
let
me
just
rephrase
what
you
said,
which
is
it
compiles,
but
it
is,
but
the
tests
do
not
pass.
No,
no!
No.
It
compiles
in
all
the
compiled
tests,
but
it
doesn't
compile
it
any
of
the
test.
Builds
Oh,
interesting,
yeah
whoa,
that's
backwards
all
right!
Well,
maybe
it
doesn't,
but
all
of
the
build
tests
incentivize
past,
but
all
of
the
tests.
I
E
A
Basically,
yes,
okay,
cool
and
in
terms
of
the
implementation?
What
what's
left
dude
besides
fix
the
bugs
that
are
yeah,
basically
much
of
them.
Getting
all
the
tests
to
pass
I
think
it's
finally
ready.
Okay,
so
I'm
gonna
go
ahead
and
remove
the
working
progress
then
doesn't
sound.
Like
duh
I
mean
the
only
work
that
your
that
you
know
you
have
to
do
is
fix
the
test
failures.
A
I
One
I
comforted
on
about
because
the
compression
level
is
sort
of
the
separate
property
making
sure
that
that
doesn't
end
up
going
sideways
with.
If
you
have
inheritance
and
you
use
set
of
s,
rename
to
move
it
to
a
different
parent
or
something
that
we
are.
You
know
with
as
a
standard
compression
level
set
when
compress
equals.
I
was
at
four
or
something
yeah.