►
From YouTube: Hackathon presentations - 2018 OpenZFS Developer Summit
A
A
So
the
way
this
is
gonna
work
is
everyone
here
who
did
something
today
that
includes
you
know,
having
a
discussion
and
in
the
webpage
writing
code.
Even
if
your
code
doesn't
compile
come
up
here,
you
can
come
up
if
you
worked
with
a
bunch
of
people,
come
up
in
a
group.
We're
sending
send
one
delegate
up
to
tell
us
what
you
work
on.
If
you,
if
you
do,
have
something
that
you
want
a
demo,
you
want
to
show
us
your
webpage
or
accessory
code.
B
A
C
A
A
F
A
A
A
First
I'd
like
to
share
ITIL
'ti
that
I
wrote
because
between
ZFS
send
em
ZFS
received,
there's
a
lot,
there's
often
just
sort
of
a
pipeline
of
weak
little
software
to
try
to
get
yours.
Your
screen
across
to
another
prompt
will
win
or
whatever
so
I
wrote
that
TCP
standard
input
multiplexer
that
scatters
anything
that
receives
a
non-standard
in
across
multiple
TCP
streams,
terminates
it
on
the
other
end
and
guarantees
the
order.
A
So
if
any
one
of
the
intermediate
connections
gets
broken,
it
will
be
started
or,
and
it
is
also
able
to
paralyze
a
traffic
monsoon,
multiple
l1,
l2
or
l3
networks.
So
if
you
have
like
40
gig
InfiniBand
10
gig
e
at
a
one
gig
Fiverr
it
will,
it
will
actually
use
all
of
those
connections
to
the
investor
on
occasion
screams
and
for
the
purposes
of
a
hackathon
I.
A
B
A
B
A
A
A
C
Worry
I
worked
with
three
things
on
three
things:
I'm,
not
gonna
present
any
of
them.
I
tell
you
the
first
thing
certain
when
I
worked
on
I've
had
a
newcomer
trapped
with
a
hackathon
which
was
yesterday,
and
then
we
worked
on
more
today
and
had
some
really
good
results
from
that
and
that's
been
running
and
I'm
sure
they're
happy
there.
So
everyone
go
check
out
our
new
comer
wiki
page
and
don't
really
like
send
us
resources
or
suggestions
for
things.
Things.
H
C
Do
to
make
it
even
more
on
you
concurrently
second
thing:
I
did
was
I
talked
with
Karen
and
Michael
Brian
and
Alan,
and
some
other
folks,
the
batplane
open
source
community
development.
We
talked
about
ask
people
for
money
and
what
kind
of
events
Beliveau.
We
also
talked
about
the
process
for
working
on
in
creating
community-wide
conduct
penalties.
I
C
D
H
Because
this
is
the
best
logo
in
the
world,
so
first
of
all,
I
want
to
thank
Julian.
He
actually
submitted
a
patch
to
help
suppress
some
unneeded
and
kind
of
crafty
errors
that
were
popping
up
in
the
winter
when
we
were
checking
the
code
because
musi
wall
is
actually
technically
you're
in
an
ash,
not
POSIX
Sh,
but
I
mean
every
limitation
of
a
shelf
supports
ash.
So
so
not
really
a
problem,
but
it
was
getting
annoying
and
people
kept
opening
issue.
So
Julian
I
do
want
to
thank
you
for
that.
H
H
D
D
G
I
G
So
it's
it's
the
one
where
one
of
the
incarnations
of
it
is.
If
you
have
review
receiving
a
sparsely
populated
data
set
so
with
a
bunch
of
sparse
files,
you
will
it'll
be
super
slow
incremental
overseas
because
it
activates
the
right
throttle
when
it
shouldn't,
because
it
doesn't,
the
right
file
doesn't
take
pulls
into
account,
but
I
just
pushed
a
pull
request
to
ZFS
and
Linux
to
fix
that
and
my
testing
so
far
four
times
faster
than
the
buggy
version.
D
G
A
F
A
A
Realize
that
we
do
have
the
proc
TX
G's
in
ZFS,
on
when
I
said,
gives
you
some
information
so
I,
just
really.
It
was
an
observability
script
that
I
tried
to
like
just
pretty
up
the
format
a
little
bit
for
what
I
wanted
it
to
do.
So
I
have
show
you
what
the
old
format
looks
like
it's
not
quite
wide
enough.
F
A
A
K
G
A
Generation
of
open,
ZFS
leadership,
so
we
talked
about
a
bunch
of
the
changes
that
are
kind
of
coming
down.
The
pike
in
terms
of
basically
the
center
of
mass
of
opens,
the
best
development
it
is
shifting.
Slash,
has
shifted
from
limos
to
ZFS
on
Linux
and
how
we're
going
to
deal
with
that.
So
the
first
thing
that
we're
gonna
do
deal
with.
It
is
to
schedule
a
meeting
another
meeting,
so
we're
gonna
set
up
monthly
meetings.
A
Basically,
we
want
to
assign
responsibility
for
getting
those
features
into
all
the
other
platforms.
So,
for
example,
like
encryption
is
indeed
on
Linux,
it's
not
yet
in
the
most
and/or
FreeBSD.
So
what
you
know
who
do
we?
Who
do
we
need
to
call
pay,
bribe,
etc
to
get
the
job
done
and
and
same
for
all
the
other
futures
going
into
the
future?
So
that
was
me
and
like
a
third
of
the
other
people
over
there?
Okay.
H
K
K
K
K
Thank
you
for
addressing
leadership
issues
because
it's
so
easy
to
just
geek
out,
and
you
know
let
hope
that
someone
else
will
jump
in
and
do
it.
But
that's
not
always
gonna
happen
on
its
own.
Just
play
say:
please
wait
and
let
me
know
if
anything
shows
up
there.
Otherwise
I'm
a
big
screen
I
can
just
try
to
hold
it
up.
K
Screw
it
up
a
big
screen:
I'll
use
that
okay,
so
I
I
love
you
all
I,
don't
like
tribes,
so
I
pick
and
choose
and
I
see
open,
ZFS
and
I.
Think
it's
one
of
the
greatest
open-source
technologies
ever
produced
by
the
human
race
and
I've
been
using
ZFS
on
on
three
nests
for
quite
some
time.
Helping
people
get
off
cliffs
and
all
that
get.
F
K
K
K
Simple
things
like,
let's
just
DD
m
to
Meg's
and
do
it
again
and
again
and
again
and
import/export
basically
at
the
moment,
just
kind
of
touch
it
gently,
because
there
are
points
where,
if
you
hit
the
value
of
the
RP
just
panics,
it's
just
it
does
stuff
like
that.
But
it's
a
thing
and
with
git
bash
you
can
have
like
a
bash
shell
and
use
POSIX
scripts
and
embrace.
G
B
B
G
G
A
L
A
Great
so
in
ZFS
on
a
loom
Osiris,
we
have
this
really
cool
in
sis
event,
infrastructure
which
is
really
damned
handy.
If
you
just
want
to
say,
hey
here's
an
event,
something
happened,
somebody
else
deal
with
it
and
then
somebody
else
can
subscribe
to
them
and
you
get
them
and
it's
great,
and
so,
when
the
port
to
the
ZFS
on
Linux.
A
Event
system,
and
so
those
began,
as
ed
said,
is
the
ZFS
event
daemon
that
plugs
into
this
thing.
So
we
use
one
source
code
for
open,
ZFS
and
then
the
SIS
events
then
get
translated,
may
show
up
in
Z&Z.
Then
you
can
write
scripts
that
subscribe
to
these
things
that
do
something
with
it,
which
is
very
cool
so
back
in
the
day
with
since
events,
I
wrote
this
little
thing
that
would
take
that
events
or
is
this
events
and
then
throw
them
in
simplex
DB
with
some
information
and
I.
This
is
it.
A
This
is,
you
know
an
enormous
amount
of
hard
work.
I
did
actually
just
finish
to
here,
but
more
importantly,
I
think
is
what
you
get
out
of
it
and
what
you
get
out
of
it
on
your
graphs
like
this.
Is
these
events
so
at
some
point
in
time
something
happened
and
then
over
here
you
see
here
and
then
diving
or
diving
is
the
name
of
the
pool
of
a
diving
pool.
F
A
Scripts
started
and
there's
not
much
data
there
and
it
finished
in
just
a
few
seconds,
but
that
way
I
can
really
thin
what's
happening
in
the
system
to
events
you
know
we
have
an
error
or
something
into
epic
cat
occurs
and
your
latency
goes
to
hell.
We
can
show
that
in
the
same
dashboard,
so
that's.
F
A
D
B
M
M
This
has
been
kind
of
a
commonly
requested
thing
amongst
many
ZFS
on
linux
users,
particularly
those
that
are
using
like
network
network
style
backends
that
talk
to
cloud
services
s3,
whatever
a
lot
of
companies
have
their
own
proprietary
thing
and
it's
amazing
how
often
they
do
stop
working
yeah,
I
go
request
gets
sent
and
for
whatever
reason,
the
completion
never
gets
back.
So
that's
the
problem
that
we're
trying
to
deal
with
here.
M
Well,
the
real
problem
is
that,
when
this
happens,
generally
speaking,
the
only
recourse
to
recover
is
to
reboot
the
system
and
system
has
additional
pools
on
it.
Well,
while
you
got
Valentine,
the
system
has
to
report
everything.
So
this
was
a
research
project
today
that
I
just
started
I'm
on
my
way
here
to
see
this
is
even
possible
and
it's
something
I've
been
thinking
about
for
quite
a
while
a
lot
of
possible
different
approaches,
but
I
actually
kind
of
got
it
working
today
here,
a
hackathon,
so
the
approaches
I
took
is
I
leveraged
facility.
M
It's
the
kind
of
leverage
is
some
of
the
logic
and
what's
called
the
continue
mode
which
was
developed
some
time
ago
to
deal
with
more
transient
type
of
errors,
we're
a
kiss-ass
fabric
or
something
that
drop
a
completion
and
again
the
whole
thing
goes
to
lunch
because,
generally
speaking,
the
default
mode
is
just
to
wait
and
wait.
Wait,
wait!
M
H
M
Of
leveraging
that
infrastructure
a
little
bit
so
my
plan
to
make
this
work
was
to,
as
I
said,
build
something
into
that
infrastructure.
So
what
I've
got
right
now
that
I
by
the
demo
is
actually
a
test
that
I
am
running
my
test
machine
in
the
window
and
it's
a
VM
guest
and
it
has
a
pool
on
it
that
that
we're
going
to
look
at
and
I've
added
some
some
testing
framework
to
the
kernel
right
now
to
help
me
learn
about
whether
this
is
possible.
M
I
did
start
integrating
it
into
the
zpool
export
command.
With
a
new
argument
at
the
moment,
I'm
not
actually
sure
what
to
call
the
future
abandoning
a
pool
or
boarding
of
pool
discarding
a
pool,
something
like
that.
Whatever
in
any
case,
there
are
some
issues
of
ordering
of
operations
that
I
ran
into
trying
to
integrate
it
out
to
you
and
Busey
cool
and
that
so
when
I
demo
here
is
actually
going
to
be
using
my
testing
framework,
which
I'm
going
to
briefly
describe
in
order
to
test
this
feature.
M
I
have
built
a
I've
added,
a
Linux
of
tunable
parameter
into
which
you
can
write
a
string.
That's
the
name
of
Aveda
yeah
that
will
cause
the
reader
of
gifts,
driver
to
not
request
a
completion
event
for
the
next
IO
to
that
disc,
and
therefore
this
is
a
request
to
completion
event.
We
don't
get
a
completion
of
that
essentially
simulating
this
condition.
M
F
M
B
M
To
drop
a
completion,
those
parameter
up,
there
is
also
part
of
my
experimentation,
as
I
wanted
to
figure
out
what
what
error
we
could
plug
into
the
cio
de
temps
mollify,
the
rest
of
the
pipeline
and
txg
layer
and
everything
else.
So
right
now,
I'm
playing
with
PIL,
which
is
there,
are
no
five.
Ultimately
I
don't
expect.
This
is
something
that
would
be
a
tunable.
M
This
is
here
again
strictly
as
part
of
the
as
part
of
the
chess
platform,
so
that
little
bottom
statement
right
there
says
we're
just
going
to
drop
a
completion
on
SS
da
four
and
that's
going
to
cause
the
whole.
Damn
thing
hang
up
that
window
right.
There
is
on
the
Left
window
without
a
console
of
the
video,
so
we'll
be
able
to
see
the
console
messages.
I
added
one
console
message
to
be
able
to
tell
that
the
dead
man
has
actually
triggered
the
feature.
M
So
what
the
feature
is
is
I
added
a
new
flag
to
the
spa
structure
called
and
I'm
right
now,
I'm,
calling
it
spa,
abort
so
to
spot
flight
is
set
that
handles
sort
of
the
other
end
of
this.
The
CIO
Deadman
is
handling
the
lower
end
and
making
sure
that
the
aisle
pipeline
doesn't
stall,
and
the
goal
of
that
is
to
right
now
kill
the
system.
That's
a
pool
is
screwed.
It
needs
to
be
abandoned,
yeah.
M
M
M
I've
got
this
working
and
and
then
went
and
started,
adding
all
the
infrastructure
to
the
I
octal
and
the
equal
command
and
everything
else.
So
there's
actually
already
methods
you
can
use,
but,
as
I
said,
there's
sort
of
these
ordering
of
problem
there's
a
fluttering
problem
in
that
the
unmounting
has
to
be
done
at
just
the
right
time,
etc,
etc,
and
I
also
need
to
rely
on
the
lazy
on
mount
feature
which
is
not
directly
supported
right
now
by
the
live
GFS
library.
Actually
it
is,
but
it
actually
has
to
be
yours,
so
fewer
complications.
M
M
So
the
demo
is
going
to
be
this
script
right
here,
which
is
going
to
destroy
it,
create
a
dataset
and
then
just
are
syncing
the
whole
system
over
to
it.
So
there's
a
lot
of
stuff
happening,
so
I'm
going
to
clear
that
up
right
now,
yeah
it's
running
now
the
data
set
to
mixes
so
now
we're
we're
we're
copying
stuff
and
we
can
see
it's
writing
stuff
out
to
it.
Okay,
now
I'm
gonna
break
it.
M
There
we
go
kill
my
arson
commands
which,
prior
to
the
future
half
the
time
you
couldn't
even
kill
those
arson
commands
because
they've
got
files
open
on
the
pool,
yeah
I
think
we'll
find
right
now
there.
Now
you
can
just
try
doing
this
again,
the
other
things
don't
even
running
anymore,
so
they
got
killed
and
we
can
see
that
the
one
Flay
got
set.
So
now
the
spa
is
in
abandon
mode.
Okay.
M
This
would
normally
be
something
that
would
be
triggered
by
right
now:
I'm
using
zpool
export
capital
a
but
as
I
said
that
doesn't
work
because
I've
not
finished
the
whole
up
or
user
interface.
Part
of
this.
So
now
we
have
to
manually
do
some
of
the
things
that
seed
blocks.
What
we'll
do
so?
We
need
to
unmount
the
file
systems.
I've
got
here
that
my
hand
right
now,
as
you
know,
and
seat
pool
export,
normally
unknown
file
systems.
It
gets
the
name
of
the
file
system,
sorts
in
the
motherÃs
order
and
it
unmask
em.
F
M
M
M
F
F
Because
you
guys
said
you
want
to
participate
on
the
channel
and
I've
moved
up
to
my
colleagues
at
Intel.
You
get
the
test
code,
that's
missing
from
as
in
fully
classic
and
they
agreed
to
get
it
to
us.
So
we
go
to
naturally
and
then
I
took
an
attempt
at
be
basing
delays,
master
and
down
to
four
files
as
well
collisions,
and
it's
mostly
the
sleep,
checkpoint
and
bus
terminal.
Our
friends,
my
friends.
M
M
D
E
So
essentially,
the
idea
here
is
going
to
be
able
to
set
some
be
their
specific
properties
and
in
the
first
iteration
is
only
going
to
apply
to
a
top-level,
be
there
so
properties
which
we
are
intending
to
sell
are
things
like
all
the
difference
you
develop
video
to
a
location
settings
then
we're
going
to
be
able
to
and
not
accept
some
be
there.
You
don't
want
to
allocate
any
data
to
we're
going
to
also
be
able
to
get
some
do
their
properties,
such
as,
like
some
stats
like
how
much,
how
much
do
locate
to
it?
E
G
E
We
figured
out
the
most
the
hardest
ones,
those
questions
and
we
haven't
created
the
open,
CSS
slack
channel.
Yet
I
think
that
would
be
a
good
idea
to
do.
I
mean
we
took
a
picture
of
our
scribbles
on
the
board.
I
think
Alan.
You
have
volunteered
to
make
this
thing's
dick,
like
yeah
yeah
I
have
a
bit
of
a
first
I've
already.
A
D
A
A
D
J
So
it's
it
makes
sense
to
me
at
least
that
it's
kind
of
a
different
use
case,
because
what
the
Zil,
it's
just
there's
one
piece
of
thing:
they're
ready
that
as
fast
as
the
skin,
rather
than
we
have
all
this
data,
so
I
don't
have
an
answer,
but
I
wanted
to
find
out.
Is
that
true?
And
so
would
it
be
worth
it
to
investigate
sort
of
a
lightweight
Zil
pipeline
I
thought
some
of
this,
because
it's
just
maybe
highlights
still
how
how
the
zio
pipeline
works.
J
So
the
first
thing
here
is
is
in
the
Zil
I
put
this
fly
in
just
so.
When
my
code
blows
up,
I
can
still
don't
get
mechanically.
So
what
I
want
to
engage
the
code
here,
I
think
the
you
man,
you
can
see.
I've
just
added
this
new
function,
zio
and
tenth
log
right,
which
is
right
down
here
and
all
that
does
is
really.
It
comes
down
to
instincts.
Instantiates
Wow
passes
in
a
different
fight
line,
so
that'll
be
used
later.
So
this
is
a
new
pipeline
that
was
pretty
easy
to
create.
J
G
J
It
down
into
what
the
stages
were,
and
so
what
I
came
up
with
here
is
I
have
an
async
issue,
which
it
just
made
sense.
That
was
the
first
stage
issued
that
async
and
then
I
just
thought.
I
could
do
it
in
two
stages,
because
there's
one
stall
like
George
described
as
I,
went
through
all
the
stages.
There
was
only
one
they
returned
at
nology
IO
at
the
end
of
each
stage.
If
you're
just
going
to
go
to
the
next
one,
it
just
returns
back
to
see.
J
J
The
nice
thing
about
him
was
then
I
could
put
a
breakpoint
in
here
and
then
just
kind
of
step
through
and
see
what
that
space
did
in
that
context,
and
then,
hopefully
that
would
help
me
figure
out
what
is
the
common
case
so
I'm
just
kind
of
working
through
those
and
didn't
get
all
the
way
through
them.
I
did
if
I'm
understanding
things
right.
J
It
seems
like
most
of
those
stages
and
right
start,
don't
do
much
for
the
basic
common
case
for
the
Zil
anyway,
so
I
got
rid
of
a
bunch
of
them
I'm
just
but
I'm
not
going
to
copy
down
that
hasn't
been
all
tested.
So
maybe
there's
something
there
that
would
help
you
in
in
understanding
out
of
the
pipeline.
F
L
Okay,
first
of
all,
I
said
before
that
they
come
from
us
Ilona,
but
I
work
in
either
line.
So
one
of
my
colleagues
tell
told
me,
oh
you
are
not
you
didn't
all.
You
didn't
tell
that
you
work
with
an
amazing
deeming
in
either
landing
good,
so
yeah
we
working
the
amazing
idle,
and
so
that's
what
they
could
say
that
and
that
form
is
called
idle
and
so
I
remember
this
time.
L
Okay,
so
what
is
something
very
simple?
The
the
thing
is
that
sometimes,
when
they
execute
from
command
line,
Zenda
V
or
set
poor
comments,
they
look
so
you
get
this
doubt
you
get
there
or
when
you
have
a
problem
or
is
Z
is
killed
and
somebody
remove
several
drives
at
some
point.
Some
point
you
get,
I,
suspend
it
and
you
can
get
like
non-responsive
commands,
even
if
you
try
to
kill
them.
L
That's
especially
bad
if
you
have
crumbs
that
do
something
based
on
scrapping
from
the
police
tables
or
whatever.
So
what
I
did
is
something
very
simple
which
executes
the
command
with
a
timeout
and
if
not,
it
kills
it
so
to
show
that
very,
very
simple
I
will
less
from
my
folder
without
a
mode
of
five
seconds
that
comes
instantly
and
in
the
church
line.
I
have
them
the
error
error
level
of
this
comment
so
weeks
ago,
because
everything
work
if
I
put
a
random
directly
and
doesn't
exist.
L
It
brings
our
level
2
from
LS
and
and
then,
and
the
error
message
has
has
been
captured
and
they're
all
put
understand
the
error
of
standard
error.
So
are
we
okay,
okay,
okay,
so
so
the
idea
about
that
is
that
you
can
learn
the
or
whatever
you
can
say
that
along
time
out,
I
will
show
I
show
I
will
show
it.
We
with
this.
We
sleep
for
10
seconds
and
then
we
do
LS.
L
L
You
can
call
this
script
from
other
programs
or
whatever
or
from
your
groans
instead
of
calling
the
DB
or
set
full
status
or
whatever
you
call
this
script
with
the
timeout,
for
example,
10
seconds,
and
then
the
comment
and
you
get
better
level
also
when
you
are
on
the
heavy,
lower,
rebuilding
coping,
etc,
simple
command
like
the
first
list
or
whatever
can
take
a
lot
of
seconds,
then
you
can
having
that
you
wait
for
20
seconds
or
whatever,
and
you
can
have
like
a
a
child
of
several
comments
in
there,
so
that
it
is
very
simple.
G
G
I'm
Tom,
so
I
just
basically
did
a
few
things
where
I
had
a
lot
of
really
helpful
discussions
with
quite
a
few
people
as
part
of
the
period
discussions
and
then
afterwards
that
George
and
I
worked
on
a
book
related
to
art,
where
we
found
out
that
maybe
hatch
locks
are
not
quite
as
safe
as
we
thought
they
were,
and
so
we're
gonna
be
doing
some
work
to
try
to
figure
out.
Well,
we've
done
about
that.
I
Yeah
hi
I'm,
Prasanna
and
I
worked
on
trying
to
get
some
visualizations
for
the
gate
state.
So
Seraphim
has
added
on
new
HDD
probes
like
static
box
in
the
pin,
the
port
for
the
change
of
transaction
states.
So
it's
put
in
like
open
to
queries
for
syncing
to
thing,
as
you
can
see
here,
so
we're
trying
to
also
utilize.
You
drank
a
couple
of
Visio
visualizations.
One
was
with
the
state
maps.
So
let
me
be
variable
to
just
this
is
a
very
small
sample,
so
we
were
able
to
see
kind
of
transitions.
F
I
This
one-
and
you
can
see
like
a
see
numbers
here
and
you
can
zoom
in
and
do
more
things,
so
that
was
that
was
mainly
one
part
still
trying
to
get
our
heads
around
the
state
map
and
the
other
thing
was
team,
had
an
idea
to
connect
farm
use
like
a
ganja
using
Python.
So
we
are
trying
to
get
this,
but
this
is
not
the
actual
one.
I
I
I
B
B
B
D
F
H
Right,
so
this
has
nothing
to
do
with
the
hackathon.
Today,
I
did
just
post
a
brief
blurb
in
and
slack
if
anybody
saw
that.
What
pretty
sure
none
of
you
know
is
that
for
the
last
three
years
I
have
been
running
a
Linux
distribution
called
sanity,
Linux
originally
was
called
pure
Linux,
and
everyone
gave
me
crack
than
the
name
just
fill
it
with
you
and
and
imlah.
So
now,.
H
H
There's
a
lot
of
extra
crap
that
goes
into
this
I'm
one
person
who
is
not
very
much
of
a
program
perfect
and
running
it
just
roses
a
little
beyond
my
skills,
admittedly
use
it
was
a
very
ambitious
project
but
I
when
I
started
it.
It
is
something
I
want
to
bring
back
to
life
and
anyone
would
like
to
contribute
and
the
you
know
one
off,
maybe
lend
some
help,
maybe
help
me
get
it
a
little
better.
H
A
And
in
the
open,
ZFS
main
open,
DFS
channel
I
posted
a
link
to
a
Google
Form.
Please
submit
one
vote.
This
is
open
for
folks
in
the
stream
as
well.
If
you've
been
watching,
I
I
know
how
many
of
you
there
are.
So
please
do
not
abuse
it.
This
is
the
first
time
that
we've
allowed
voting
from
non
local
participants.