►
From YouTube: May 2023 OpenZFS Leadership Meeting
Description
Agenda: Conference dates; ZIL locking
full notes: https://docs.google.com/document/d/1w2jv2XVYFmBVvG1EGf-9A5HBVsjAYoLIFZAnWHhV-BM/edit#
A
All
right:
well,
let's
get
started.
It's
welcome
to
the
May
2023
open,
ZFS
leadership,
meeting
not
too
much
on
the
agenda
today.
So
we'll
have
some
time
for
open-ended
discussion.
A
First
thing
I
want
to
mention.
Is
that
we're
starting
to
plan
next
year's
open,
ZFS,
developer
Summit
and
the
dates
that
we're
looking
at
are
October
16th
and
17th?
We
haven't
locked
those
in
yet
with
the
the
venue,
but
that's
what
we're
looking
at.
So
if
anybody
sees
conflicts
with
that
October,
16th
and
17th,
then
let
me
know.
A
A
B
A
All
right
next
item
on
the
agenda
looks
like
Zill,
locking
improvements.
C
I
made
the
patch
few
weeks
ago
that
should
dramatically
improve
performance.
At
least
it
does
on
my
tests
when
multiple
threats
are
trying
to
write
synchronously
to
the
same
data
set.
The
most
prominent
example
I
have
is
VMware
of
emotion
to
Isis
Target
when
it
uses
plenty
of
64
kilobyte
rights
same
time
from
multiple
threads
and
without
with
code
as
it
is
now,
it
creates
so
much
slow
contentions
that
practically
throttle
speed
is
about
1.8
gigabyte
per
second
also
minus
there.
C
The
problem
is
that
we're
doing
memory
copy
under
the
lock
and
that
by
definition
can
scale
only
one
CPU
at
the
time
copies
the
data
into
zeal,
so
I
created
PR
reference
to
that
ticket.
I
would
like
people
to
take
a
look.
The
idea
is
to
move
memory,
copy
and
zero
issue
out
of
the
lock,
so
the
only
thing
that
stays
under
the
lock
is
just
allocation
of
chain
of
blocks
linking
them
together.
C
So
it
still
creates
some
small
contentions,
the
science
practically
Global
lock
in
that
case,
but
it's
so
dramatically
smaller
that
getting
3.1
gigabyte
per
second
after
that
I
suspect,
may
not
no
longer
bottleneck
I.
Think
it's
bottleneck
on
VMware
side.
C
I,
see
one
rough
edge
around
the
fact
that
if
I'm
postponing
of
those
data
corporator
I
can't
there
are
cases
when
that
data
copy,
not
copy
but
right
of
indirect
block
for
indirect
linking
into
the
zoo
may
require
a
transaction
open
or
even
wait
for
transaction,
commit
that
I
can't
do
after
I
already
allocated
chain
of
blocks.
Otherwise
I'll
get
that
lock.
So
I
work
around
that
in
my
patch,
but
I
can't
say
that
I'm
exactly
happy
about
that
exception.
C
It's
just
makes
a
case
it
also
in
addition
to
that,
it
makes
case
of
testing
Zeal
a
bit,
not
so
strict
because
it
behaves
different
in
this
case.
So
pull
freezing
Works
has
to
work
differently
so
that
small
chunk
of
that
patches
they
can
plug
like.
So
if
anybody
has
better
ideas,
I
would
be
happy
to
hear
otherwise
it
works.
It
passes,
CI
so
and
general
review
and
commit
would
be
great.
Yes,
considering
how
big
is
Improvement,
unless
somebody
has
but
some
ideas
how
to
make
it
even
better.
A
And
I
see
that
prakash
took
a
look.
I
can
bug
him
as
well
remind
him
to
take
a
another
look.
Where
is
I'm
just
scanning
through
the
code?
Where
is
the
this
tricky
part
that
you
mentioned?
Look.
C
A
C
A
C
Yeah
I
had
to
do
exception
and
commit
transactions
earlier.
I
had
to
wait
for
commit
earlier
just
prematurely
to
avoid
waiting
later,
but
I
would
be
happy
to
not
do
that,
but
I
haven't
found
a
way
to
do
that.
So
actually
in
case
we
direct
block
I
can't
believe
it's
reasonable,
as
there
is
probably
no
other
way
but
I'm
not
exactly
sure,
with
full
freezing.
C
It's
maybe
I
didn't
dug
deeper,
not
deep
enough,
but
that
particular
line
of
2069
where
I
had
to
no
I
kept
existing
freeze,
I
guess
the
execution
or
txg
weight
scene
could
call
in
that
case,
that
makes
practice
I've
told
tests
different
from
production
environments
that
I'm
not
exactly
happy
about.
If
I
could
not
do
that
or
move
it
early
so
that
I
would
have
to
commit
early
and
wait
for
txg
sync
would
be
great
I.
A
Mean,
thankfully,
that
proofreezing
stuff
is
only
for
testing
so.
A
We're
not
testing
quite
exactly
the
same
thing
when
the
fusing
test
yeah
got
it
all
right.
C
A
A
Great
any
any
questions
about
that.
A
Right
other
items
of
discussion.
C
I
guess
everybody
is
waiting
for
2.2
release
that
was
result
Brian
here
last
night,
I
heard
was
waiting
for
force
that
commit
or
forcible
expert,
but
we
are,
but
he
lost
I
heard
he
was
going
to
look
on
what
the
status
and
whether
I
haven't
checked
myself.
C
It's
just
like
we
are
stretching
it
from
the
new
year
we
already
kind
of
Fairly
in
turn
us
in
a
state
where
we
would
like
new
ZFS
for
both
FreeBSD
and
Linux
versions.
Yeah.
We
are
in
Alpha
stages
and
boss
and
now
would
be
a
good
time.
A
Yeah
I
think
it's
been
quite
a
while
so
I
think
definitely
time
for
another
release,
so
yeah,
hopefully
they're
working
on
figuring,
sorting
it
out.
I
talked
to
Alan
a
little
bit
ago
and
it
sounded
like
he
thought
it
was
almost
done.
So
that's
good
and
then
I
saw
the
Mac.
A
Os
integration
seems
to
be
coming
along
looking
forward
to
seeing
that,
but
I
don't
know
that
we
have
any
I,
don't
have
any
more
info
on
that
specifically
I
assume
that'll
be
for
after
2.2
that'll
be
for,
like
maybe
we'll
call
that
3.0
or
something
we'll
see.
A
Okay,
we
may
have
a
very
short
meeting,
maybe
the
the
earlier
time
didn't
work
for
as
many
folks
this
time,
but
the
next
meeting
will
be
at
the
later
1
pm
time.
Four
weeks
from
today,
which
will
be
June
13th,
it
looks
like
sorry.
C
B
Many
of
the
previously
folks
are
obviously
recovering
from
BSD
can
which
just
wrapped
up,
and
of
course
there
was
much
discussion
of
open
ZFS
and
a
lunchtime
buff
and
MAV
can
confirm
and
I'd
like
to
propose
something:
mildly
non-technical,
so
I've
been
hosting
beehive
and
jail
calls
and
they
ping-pong
between
developers,
call
and
a
production.
Users
call
and
I
sense
that
there
might
be
interest
in
either
organized
through
you
and
Company
or
others.
A
a
periodic
call
discussing
just
production
user
topics.
B
A
A
So
if,
if
you
kind
of
know
who
those
people
are
I
would
encourage
you
to
organize
that
and
then
you
know
and
then
get
the
people
that
are
on
this
call
to
you
know,
go
attend
that
that
other
call
and
I'm
happy
to
coordinate,
like
you
know,
modifying
the
times
of
this
call
or
the
schedule
to
kind
of
make
it
easier
for
folks.
You
know,
maybe
we
do
like
whatever
like
every
other
month
or
we
do
every
every
third
month.
A
We
don't
have
this
meeting
and
instead
we
have
this
like
a
production
users,
oriented
meeting
I,
would
kind
of
look
to
someone
else
to
organize
that
and
like
send
out
the
word
and
get
people
to
attend
and
get
the
agenda
and
that
kind
of
stuff,
but
I'm
happy
to
attend
that
meeting
and
to
encourage
developers
to
attend
that
as
well.
Okay,.
B
So
fortunately,
oh
fortunately,
you
are
all
easy
to
find,
and
this
call
is
infrequent
enough
that
I
don't
think
it
would
conflict.
It's
more
of
a
response
to
Summits
at
conferences
where
this,
the
this
well
in
the
annual
conference
here
that
time
scale
is
a
bit
long
for
a
certain
topics.
So
this
is
suiting
this
really
well,
but
users,
just
sharing
their
discoveries
can
be
I,
don't
know
at
a
different
pace,
so
I
will
maybe
propose
some
times
that
are
not
guaranteed
to
conflict.
With
many
of
you
go
from
there.
A
A
Okay,
then
we'll
have
a
short
meeting
and
I'll
see
you
all
in
four
weeks,
thanks
and
thanks
for
that
great
suggestion.
Michael
look
forward
to
seeing
that
that
meeting
put
together
and
and
seeing
what
users
have
to
share
about
their
experiences.
My
pleasure,
thank
you.
So
much
for
your
work,
see
you.