►
From YouTube: April 23, 2019 OpenZFS Leadership Meeting
Description
We discussed DRAID Meetup; ZoL 0.8; ZoL porting; enabling compression by default; and updating LZ4.
Agenda and meeting notes: https://docs.google.com/document/d/1w2jv2XVYFmBVvG1EGf-9A5HBVsjAYoLIFZAnWHhV-BM
A
B
The
current
meeting.
Date
is
May
3rd.
So
that's
what
two
weeks,
or
so
it's
the
day
after
docker
con
in
San
Francisco.
So
some
people
who
are
traveling
for
that
or
you
know,
could
be
useful
or
convenient
and
and
the
whole
idea
is
to
let's
get
down.
You
know,
flesh
out
any
remaining
design
decisions
that
need
to
be
done
and
then
identify
the
work
that
remains
to
bring
it
into
the
fold
and
make
it
happen.
B
So
if
you're
interested
by
all
means
reach
out
to
myself
or
anyone
in
the
group,
we
have
a
dear
aide
slack
channel
or
locate
dear
aide,
locate
us
and
RSVP,
because
we'll
need
to
have
a
list
for
people
to
get
into
the
building.
But
I
appreciate
it
can
show
up
that's
interested
in
the
project
and
let's
get
some
flesh
on
the
bones.
B
A
A
B
A
I
mean
we
can
do
zoom
I.
Think.
As
you
mentioned
it's,
it
can
be
difficult
to
coordinate
that
like
a
discussion
with
what
remote
and
local
folks,
so
you
might
need
primarily
like
read-only
for
remote
folks,
but
you
know
in
easily
make
that
available
and
the
audio
in
that
room.
I
think
works
pretty
well
as
long
as
there's
only
one
discussion
going
on,
but.
C
D
A
A
You
Richard,
it
would
be
great
if
you
could
put
together
like
an
agenda
agenda
doc
beforehand
and
then
I
think
that
might
be
a
great
way
for
remote
folks
to
participate
as
well
like
in
advance
of
the
meeting.
You
know
by
adding
comments
or
questions
to
the
agenda.
Doc
about
you
know,
designing
points
that
they
are
interested
in
and
then
we
can
cover
it.
You
know
during.
A
E
I'm
here
could
everybody
hear
me:
yeah
awesome,
so
yeah,
just
a
quick
update
for
folks.
So
last
week
we
tagged
the
final
receipt
release
candidate,
there's
an
o
8o
r
c
4
that
was
released
so
that
feature
complete
and
has
been
soaking
for
about
a
week.
Now
it's
seen
a
lot
of
testing
before
that.
Obviously,
since
everything
has
been
available,
but
this
is
what
we're
encouraging
people
to
test
as
much
as
possible
and
kick
the
tires
on.
E
E
Zooming,
nothing
new
comes
in,
that's
major,
which
we
haven't
seen
yet,
hopefully
just
a
couple
of
weeks,
because
it
has
been
soaking
for
quite
a
long
time
now.
We
do
have
a
lot
of
people
who
were
on
the
master
branch
and
have
run
previous,
our
C's,
so
we're
just
trying
to
address
the
last
remaining
significant
issues.
People
have
reported
all
right
so
like
a
day
a
week,
a
month,
a
couple
of
weeks,
so
probably
two
week,
three
weeks.
If
we
find
something
right,
the
work
we're
trying
to
wrap
it
up.
E
A
Cool,
that's
that's
great!
Actually,
why
don't?
We
stick
on
your
other
agenda
item
I
just
ordering
I
do
Brian
you're
asking
if
there
is
if
there
are
folks
here
who
have
been
working
on
porting
ZL
commits
to
other
platforms.
If
you
have
any
feedback,
maybe
he
library
on
what
kind
of
feedback
you're
looking
for.
E
Yeah
I
was
just
kind
of
curious
what
other,
what
problems
people
have
run
into
porting
features
from
Linux
to
the
other
platforms
Orange.
How
that
process
is
going
particularly
I'm.
Looking
for
feedback
on
things,
we
can
be
doing
better.
Make
that
process
easier
or
better
ways
to
engage
with
people
when
we're
making
design
decisions
or
things
we
shouldn't
be
doing
or
problems.
You
found
importing
that
kind
of
thing.
C
C
Remember,
I
to
some
definition:
what
and
redefine
it
only
not
specific
room.
Illumise
specific.
Probably
we
have
to
identify
a
list
of
definitions
and
provided
pure
platform
with
some
redefinition
or
something
else'
also
function,
naming
convention
definition
with
ZFS
prefix
or
something
else
because
per
platform
we
can
redefine
some
function.
Implementation
visit
on
+
on
specific.
E
So
we've
tried
to
keep
as
consistent
as
possible
with
the
illumos
stuff,
but
since
most
of
us
are
Kalu,
most
developers
I'm
sure
we've
diverged
in
areas
so
yeah
I'm
all
for
documenting
those
places
where
we've
diverged
be
nice
to
have
a
list
of
that
kind
of
stuff.
I
know
we've
added
or
extended
some
interfaces
in
that
way,
I
think
getting
all
that
kind.
C
C
It
was
some
issue
in
some
VFS
where,
where
we
are
different
between
illumise
and
linux,
and
probably
try
to
define
some
specific
of
function,
what
we
can
implement
your
platform
will
be
better,
but
I
have
a
question
how
we
can
report
about
this
issue,
probably
for
some
portability
issue
or
some
design
issue
encoding.
How
can
I
report
about
this
issue.
E
Well,
probably,
the
issue
tracker
is
the
best
spot.
To
start
I
mean
we
could
look
at
setting
up
a
wiki
page
or
some
things
may
be
on
the
open,
ZFS
Wiki
to
like
document
the
differences
but
I
an
issues,
an
easy
spots.
To
start
it's
none
of
the
FS
on
life
issue
tracker
and
that
we
can
figure
out
where
to
go
from
there.
Yes,.
C
A
Mean
I
think
it
could,
although
the
the
trick
with
the
wiki
is
that
you
know
it's
very
easy
for
that
to
get
out
of
day
or
not
be
maintained,
I
think
it's
a
lot
easier
for
the
folks
who
are
reporting
stuff
to,
like
you
know,
just
shoot
an
email
or
file
an
issue
and
with
the
current
state
or
current
issues
that
they're
running
into
so
you
know,
maybe
the
developer
at
open,
ZFS
Alan.
This
might
be
like
any
easier
for
more,
like
good
first
step
for
reporting
those
things.
F
E
I
suppose
another
option
would
be
we
could
have
a
porting
document
in
the
source
tree
or
something
like
that
where
we,
you
know,
document
the
differences
and
then
people
could
open
changes
against
that.
My
concern,
the
mailing
list,
is
that
we'll
just
end
up
with
a
lot
of
you
know:
emails
that
won't
get
collated
anywhere,
yeah.
A
I
guess
it
depends
on
if
we're
talking
about
things
that
are
like
differences,
that
need
to
be
documented
for
posterity
versus.
Like
you
know,
hey,
it
looks
like
you
guys
are
using
this
macro.
Could
you
please
use
this
other
function
instead
or
whatever
sure
you
know
what
I
mean,
which
is
more
like
time
in
in
context
sensitive
and
can
be
right.
E
Yeah
I
think
there's
probably
two
classes
of
things
here,
like
one
changes
that
we
can
fix
right,
we
just
used
the
wrong
function
or
macro
or
whatever.
We
just
need
to
apply
the
fix
to
get
in
sync
and
then
probably
a
whole
nother
group,
where
we
just
have
to
be
fundamentally
different
in
this
way
right.
Just
because
it's
the
way
the
kernel
works
or
whatever
we
need
to
document
those
differences.
A
Yeah,
it's
definitely
getting
like
Jerry
getting
your
notes.
You
know,
feel
free
to
just
shoot
us
out
on
the
mailing
list
and
then
maybe,
when
we
look
at
the
specifics,
we
can
figure
out
which
of
those
are
things
that
can
be
resolved
versus
things
that
just
need
to
be
documented.
I
mean
for
the
things
that
need
to
be
documented,
like
I'm,
fine
with
me
on
the
wiki
or
in
the
repo
I
mean
the
wiki
might
be
like
more
of
the
correct
place
of
doing
it,
because
it
is
issues
that
affect
all
the
platforms.
A
But
in
the
case
where
there's
like,
where
it's
like,
when
you're
pulling
stuff
into
Lumos
from
ZL,
we
need
to
be
aware
of
these.
Differences
like
it
might
make
sense
to
put
those
into
like
a
document
in
the
illumos,
repo
or
other
documentation.
Just
because,
like
I,
feel,
like
I,
am
usually
looking
at
the
source.
More
than
I
am
at
that
at
the
wiki.
So
it
might
be
easier
for
people
to
find
that
in
sources,
obviously
they're
already
working
with
the
source
if
they
doing
40.
C
Okay,
thanks
and
I
think
we
can
try
to
start
with
the
email
list.
Probably
try
to
ask
for
some
proposals
and,
as
a
final
decision,
try
to
put
information
to
the
wiki
or
some
solid
place.
Is
it
okay,
yeah
I
mean
last
question
to
my
list:
first,
probably
try
to
find
a
better
place
where
we
can
share
this
info
between
platform
yeah.
A
A
Cool
helpful
to
update
other
folks
in
in
modern,
open,
ZFS
community
by
just
a
name
like
a
link
or
a
copy
of
the
email
that
you
have
there
to
the
FreeBSD.
This
course
that'd
crossbow
us
that
over
to
the
developer,
open
ZFS
list,
just
gently
sweet
idea
where
ya
folks,
out
on
the
West
End
Linux,
that
aren't
necessarily
on
the
call.
G
G
E
G
I
I
And
then
deltax
has
a
whole
bunch
of
changes
that
are
in
the
pipe
to
go
into
master
as
soon
as
the
audit
is
cut,
and
so
my
plan
is
to
rebase
to
master
as
soon
as
the
Delft
exchanges
have
gone
and
because
I'd
like
to
have
a
little
bit
of
crescent
period
and
master
and
so
on,
but
it's
not
really
fair
to
have
any
reasonably
tested
changes.
Waiting
on
you
know
the
the
previously
support
discussion,
so
I
think
I
mean
I.
I
A
A
A
A
You
know
not
expert
users.
The
I
think
that,
in
my
opinion,
the
downside
is
that
it,
the
current
at
least
the
current
implementation,
even
with
lz4,
does
have
some
cpu
costs,
and
maybe
a
more
minor
concern
is
that
it
does
change
the
space
usage
in
a
way
that
might
be
surprising
and
it
can
introduce,
like
you
know,
running
out
of
space
in
ways
that
you
didn't
otherwise
expect.
A
Like
you
know,
let's
say
you
create
a
giant
Oracle
database
file
and
that
writes
out
the
whole
file.
It
touches
every
block
of
it.
Now
it's
consuming
that
space
and
then,
but
if
it
were
compressed
to
compress
that
each
block
would
compress
sound,
really
small,
but
your
if
you
were
relying
on
the
fact
that
they
okay
will
I
initialize
this
Oracle
file,
now
I'll
always
be
able
to
write
to
it.
The
spaces
like
effectively
been
reserved,
then
that
kind
of
idea
would
no
longer
work.
A
C
As
some
comments,
for
example,
old
spark
machine
do
not
support
compression,
enable
it
on
tool
from
open
boot
program
for
boot
system,
but
we
can
enable
compression
on
data
set.
What
would
we
are
using
with
the
US,
and
we
are
using,
do
not
enabled
compression
on
our
pool
and
enable
compression
and
the
special
data
set,
for
example,
where
we
host
the
primary
files
and
something
else
so
I
do
not
enable
compression.
A
C
C
A
A
C
Yes,
and
because
open
boot
spark
question
need
to
do
not
enable
compression
on
powerful,
where
you
have
to
really
have
good
information.
I
Sorry
back
to
the
issue
we
discussed
several
times
about
deep,
portable
and
defaults
and
how
boot
pools
need
to
be
different
period
because
it's
going
to
be
dependent
on
whatever
method,
the
Machine
boots
and
so
making
something
default
for
data
pools
is
one
thing,
but
whatever
your
only
use
to
create
a
boot
pool
or
a
pool
that
you're
gonna
boot
from
is
gonna
have
to
be
always.
Gonna
have
to
be
system
dependent
right.
K
A
You'd
argue,
like
you
know,
if
you're
creating
a
pool,
you
really
have
to
know
what
you're
doing
and
there
or
like
whatever
we
do
with
compression.
You
know
you
can
learn
to
set
Z
to
do
z
blow,
create
no
compression
equals
off
if
you
need
to
or
whatever
yeah-
and
you
know
really
the
question
kind
of
boils
down
to
for
like
people
that
are
because
those
pools
people
have
typing
is
equal
to
Korea
at
the
command
line
to
create
their
rules
generally.
A
Getting
if,
like
you
know,
installed
from
some
booth
order
or
they're
like
following
some
crazy
script,
recipe
on
the
web
with
like
a
thousand
instructions
and
adding
one
more
is
no
big
deal.
So
really
it's
like
people
that
will
create,
but
no
arguments
to
create
their
data
pool.
You
know,
should
it
be
compressed
or
not
I.
H
H
G
H
K
H
You
know
what
the
one
drawback
that's
pretty
consistent
with
turning
compression
on
is
it
does
oftentimes
impact
rate
speed
a
lot
of
times.
You
will
take
a
right
speed
drop
simply
because
the
system
will
you
know.
Writing
is
the
most
taxing
thing
to
begin
with
and
if
you
have
a
CPU
issue
at
all,
turning
on
the
compression
will
its
first
exposure
will
be
your
right.
Speed
drops,
but
the
vast
majority
of
people
don't
care
about
that
yeah
they
go.
That's.
I
A
A
Doesn't
help,
isn't
saving
you
space
ads
or
CPU,
and
you
you
are
like
you
know,
running
low
on
CPU,
because
you're
doing
you
know
raid
z3
calculations
and
whatever
other
for
those
folks
like
are
we
are
we
okay
with
you
know
in
the
next
release,
or
you
know
to
lease
now
or
whatever
telling
them.
You
know,
yeah,
I'm,
gonna
guess,
look
bu
you
now
you're
gonna
have
to
do
you
know.
Xiv
will
create
Oh
compression.
It
was
lost
well.
A
G
H
Answer:
you're
not
comparing
my
answer
to
that
would
be
compression
short
circuits,
and
so,
if
you're,
actually
writing
incompressible
data,
there
isn't
a
huge
CPU
overhead
to
that,
because
you
know
the
looks
at
the
first
part,
the
blog
and
although
I
can't
compress
that
gives
up
on
it.
You
know
the
the
when
compression
actually
has
noticeable
CPU
overhead
is
when
it
can
compress
well.
I
F
G
I
I
also
think
that
the
people
who
are
using
ZFS
based
systems
for
video
editing-
oh
they're,
not
the
ones
who
are
doing
the
zpool
creed
or
ZFS
great
commands
themselves-
are
going
through
some
UI
and
so
I
think
it's
up
to
their
administrators
or
to
the
system
integrators
or
creating
it.
Whatever
embedded
systems
to
be
able
to
at
least
say
this
data
set
is
has
compression
on
the
compression
is
not
working
well,
you
might
get
better
performance
by
turning
it
off
I.
H
H
G
The
grated
case
is
not
so
bad
because
as
soon
as
you
that,
if
has
set
compensable,
is
off,
you
get
a
lot
of
new
rights.
Don't
and
don't
the
decompress
isn't
what's
slowing
anything
down
right.
So
it's
it's
not
like
some
other
features
where
you
have
to
do
a
bunch
of
work
to
try
to
make
it
stop
doing
whatever
it
was
doing
yes
undoing.
Dee-Do
is
a
far
different
yeah.
H
G
A
A
I
guess
both
like
create
the
full
requests,
but
also
figure
out
like
what
what
the
rollout
looks
like
you
know
what
the
timeline
looks
like
what
the
notification
looks
like
you
know.
Do
we
do
this
in
this
release
or
the
next
release,
or
you
know
how
long
notice
fury
do
we
have
and
answering
those
kinds
of
questions
I
think
are
still
up
in
the
air,
but
I
think
you
know,
probably
somebody
should
just
come
out.
Come
come
up
with
the
proposal
and
then,
what's
in
my
ear,
shakes
from
there.
J
J
J
A
A
A
Very
well
yes,
if
United
States,
you
probably
have
that
Monday
off
cool.
So
the
last
item
that
I
had
was
there
could
have
conduct
folks,
probably
saw
either
the
email
or
the
pull
requests,
or
the
slack
message
about
the
code
of
conduct
that
we're
working
on
just
wanted
to
give
folks
an
additional
opportunity
to
give
us
any
feedback.
I'm
on
this
call.
A
Well,
we're
also
you
know
we're
also
open
to,
and
we
have
received
some
feedback
so
far
that
you
know
from
folks
who
you
know
supported
the
proposal
in
general,
but
had
either
you
know
small,
specific
concerns
or
just
suggestions
of
how
to
read.
You
know
reword
things.
So
if
you
have
that
kind
of
feedback
as
well
we'd
like
to
hear
it
feel
free
to
email,
us
or
let
us
know
now,.
G
A
G
Enough
like
that,
this
there's
an
upright
end
up
mattering
that
much
because,
if
you
overwrite
with
the
same
day,
debt
the
first
time,
the
knock
right
won't
work
and
it'll
update
to
the
new
Texan.
But
then,
when
you
confess
that
they
did
but
the
third
time
it
will
match.
Yeah
I
mean
there's
only
one
extra
right,
but
yeah
I
think
having
the
switch
to
go
back.
There's
on
C's
yeah.
A
I
mean
the
issue
is
that
if
you're,
actually
we
rely
on
this,
it
means
that
you
could
roughly
double
the
storage
requirements.
You
know
say,
might
go
great
like
in
the
next.
You
know
release
of
our
software.
Maybe
you
might
like
it
may
end
up
using
twice
your
storage.
It's
not
a
great
story
right,
I'm.
A
Know
you
have
all
these
snapshots,
and
so
you
you'd
and
obstructions
overwrite,
the
original
Breeden
and
the
snapshots
and
the
new
version
is
the
newer
snapshots.
We've
got
definitely
issue:
it's
the
space
savings,
not
you
and
IO
savings
that
are,
it
was
important
to
us
only.
We
could
rewrite
the
block
point
as
somehow.
A
H
G
Not
that
anyone
is
aware
of
maybe
we
would
maybe.
A
A
C
G
Yes,
that
was
definitely
you
know.
As
matt
said.
The
first
plan
would
be
just
to
get
the
decompression
performance
improvements
so
that
we
don't
change
the
on
disk
version
at
all
and
then
evaluate
whether
it's
worth
doing
the
compression
management's
as
well
and
probably
having
a
switch
to
be
able
to
switch
back
to
the
old
one,
possibly
off
by
default
at
first.
So
that
test
will
also.
C
Additional
question:
if
we
have
updated
before-
and
we
have
very
large
all
the
data-
have
you
tasted
big
data
with
the
compress
the
compression,
for
example,
you
have
a
several
terabytes
of
all
the
data
with
the
core
compression
and
try
to
decompress
it
and
take
a
look.
You
have
know
something
broken
for
region
also,
if
you
can
use
a
different
block
size
of
the
bed,
you.
G
G
H
D
H
42
is
a
different
compression
method
at
first
and
then
switch
the
decompressor
over
and
then
I.
Don't
know
that
you'd
really
even
need
to
switch.
The
compressor
over
people
could
make
that
switch
manually.
That
way
systems
could
still
read
and
write
using
both
algorithms
if
they
needed
to
be
lower
risk
at
any
rate,
yeah.
E
A
G
And
that
actually
reminds
me
of
the
other
issue.
I
was
going
to
ask
about
so
for
some
settings
like
compress
and
checksum.
We
have
a
value
of
on
that's
a
default,
but
over
time
that
default
has
changed
right,
who
buys
equals
bond
moved
from
LGBT
to
I
was
like
4
and
checks
come
on
move
from
selector
to
deflector
for
your
arm,
there's
not
a
good
way
for
the
user
to
tell
which
one
is
using
when
it's
set
to
arm
like.
If
you
do
is
Oedipus
get
compress.
It
just
says
on.