►
From YouTube: Closing by Matt Ahrens
Description
No description was provided for this meeting.
If this is YOUR meeting, an easy way to fix this is to add a description to your video, wherever mtngs.io found it (probably YouTube).
A
So
that's
not
right.
Let
me
just
put
up
the
slide,
so
we
have
a
couple
of
options
of
activities.
Now.
A
dinner
is
not
going
to
be
ready
for
another
about
15
minutes,
so
we
can
take
a
break
and
get
out
of
this
hot
room
or
if
you
guys
are
interested,
we
can
do
some
Q&A
about
maybe
the
old
times.
We
can
get
a
mark
to
come
up
here
and
you
know
we
can
just
take
whatever
random
questions
folks
have
about.
You
know
back
in
the
day
of
ZFS
or
future
stuff.
A
What
do
we
want
to
see
in
ZFS
I'm
also
told
that
a
special
guest
may
be
arriving
in
like
15
minutes
or
so
who
may
be
able
to
join
us
all
right
folks,
too
hot
to
do
that
in
here.
You
want
to
do
folks,
have
questions.
If
there's
no
questions,
oh
no
mark
wasn't
feeling
well
all
right.
Well,
sorry
mark,
then
we
have
special
Q&A
time
with
me
and
let's
say
any
any
other
gfs
developers
who
want
to
come
up
here
and
field
questions
about
George.
Why
don't
you
come
up
here.
A
A
So
you're
asking
about
like
a
library
to
do
kind
of
programmatically
control,
ZFS
yeah.
So
in
terms
of
the
administrative
actions
you
can
do
that
with
libs
with
libs
EFS.
You
know
the
ZFS
command
it
pretty
much
just
calls
in
to
lib
gfs,
but
that's
an
unstable
interface
and
we've
changed
it
a
bunch
of
your
time
and
it's
kind
of
hard
to
use
programmatically,
because
the
error
handling
is
it
tends
to
do
a
lot
of
things
with
one
function.
Call
like
you.
A
Do
one
function
call
to
lib
ZFS
and
it
does
like
whole
bunch
of
stuff
and
it
might
fail
somewhere
in
the
middle,
and
you
get
some
error
codes.
So
we
created
lib
ZFS
core,
which
is
like
pretty
much
a
thin
wrapper
layer
around
the
cocktails
to
the
colonel
and
the
ideas
that
that
will
become
a
stable
interface
with
defined
air
semantics.
A
B
C
C
A
I
guess
for
the
video.
Basically,
the
the
comment
was
that
there's
a
bunch
of
other
system
level,
debugging
stuff
that
is
applicable
on
Lumos,
like
kmm
flags
and
you
know,
probably
make
sense
to
look
for
similar
kinds
of
things
like
I
know
on
freebsd.
They
have
this
witness
lock
order
checking.
So
that
would
be
a
good
example
of
things
that
we
would
want
to
turn
on
like
when
running
the
test.
Suite
I.
A
B
I
had
a
back
in
the
day
story
that
tied
into
some
of
the
stuff
that
mark
was
discussing,
so
he
mentioned
that
the
kind
of
the
initial
hello
world
for
ZFS
was
going
to
be
this
test,
where
they
do
a
big
bring
over
of
all
of
the
date
command,
compile
it
and
like
make
sure
it
works,
so
have
people
here
played
around
with
something
called
Zille
tests?
Are
you
familiar
with
something
called
Zille
test?
What.
B
B
A
C
A
So
I
think
that
gets
a
little
bit
to
what
Mark
was
talking
about
earlier
in
his
talk
about
like
the
VFS
locking
and
the
idea
that
we
didn't
want
to
replicate
that
in
ZFS,
so
like
direct
I,
owe
at
least
in
in
solaris
lecture
Lumos,
the
pointing
of
direct
I
always
like
to
disable
this
file
level,
locking
so
that
you
could
have
it
like
a
database
doing
concurrent
rights
to
the
same
thought
to
one
file
all
at
the
same
time.
So
in
ZFS
we
didn't
want
to
support
direct
I.
A
Oh,
we
wanted
to
make
it
like
not
necessary
to
have
that
hack,
because
we
saw
a
direct
I/o
is
essentially
like
a
hack
because,
like
the
locking
was
support.
So
you
know,
Mark
didn't
talk
about
this,
but
one
of
the
things
that
he
implemented
was
the
range
locking
in
the
Zil.
So
when
you
are
doing
a
right
system
call
you,
you
can
get
the
POSIX
semantics
of
no
overlapping
rights,
but
with
locking
just
the
range
that's
being
rings.
A
So
you
can
still
have
a
whole
bunch
of
Rights
covering
different
ranges
of
the
file
going
on
concurrently.
So
we
don't
have
any
plans
to
add
direct
I/o
support.
Hopefully
there's
no
need
to
do
that
at
least
on
the
most.
If
there's,
if
there's
like
Linux
specific
stuff,
then
maybe
we
should
talk
offline.
A
It's
a
little
bit
harder,
because
you
know
we're
normally
doing
like
check
something
and
all
this
other
stuff
and
putting
the
IO
off
until
sinking
context,
because
we're
copy-on-write,
so
that
would
probably
be
much
more
difficult,
but
doing
something
like
that
for
the
read
path.
I
think
totally
makes
sense.
I
think.