►
From YouTube: Ceph RGW Refactoring Meeting 2023-01-08
Description
Join us every Wednesday for the Ceph RGW Refactoring meeting: https://ceph.io/en/community/meetups
Ceph website: https://ceph.io
Ceph blog: https://ceph.io/en/news/blog/
Contribute to Ceph: https://ceph.io/en/developers/contrib...
What is Ceph: https://ceph.io/en/discover/
B
I'm
curious
I
was
just
looking
for
a
restaurant
to
create
a
work
plan
for
the
different
parts
of
the
current
team,
refactoring
workflow
refactoring,
and
you
and
I
had
a
conversations
about
this
earlier
today
and
and
you
you
made
a
point
that
it's
it's
your
sense
of
the
plan.
B
Your
plan
is
that
is
that
this
I
know
often
I'd
often
imagine
that
we
would
have
one
set
of
protein
flows,
somehow
spanning
multicide
and
and
and
the
request
interface
and
that's
possible,
and
that's
not
how
you
see
it
going
forward
and-
and
you
explain
that
there's
we
were
kind
of,
they
were
probably
getting
everything
we
want
out
of
out
of
out
of
the
current
system
with
optional
yields,
or
you
know,
asio
integrated
yield
to
get
Cooperative.
B
You
know
full
Cooperative
scheduling
and
on
about
a
small
thread
pool,
which
is
what
I
think
we
really
want
out
of
it.
So,
but,
but
we
haven't,
you
mentioned
that
we
haven't
done
all
the
things
we
need
to
do
there.
It
works
also
proceeding
to
get
on
the
other
side.
The
other
path
with
six
foot
plus
20
Co
routine.
So
I'd
like
to
get
us
to
to
fill
out
a
sort
of
a
work,
breakdown,
structure,
kind
of
a
list
of
all
the
backlog
items
and
other
and
other
broken.
How
they're?
B
How
they're
sequenced
to
do
these
two
things
and
it
seems
like
if
you
haven't,
got
everything
we
need
for
the
for
the
request.
Workflow,
that's
that's
a
gap.
We
should
be
working
on
this
year.
C
A
Yeah
agreed
I
think
Red
House
GW
admin
generally
should
be
the
only
thing
calling
that.
C
I
think
you
can
do
something
similar
to
what
I've
been
doing
in
the
various
neorados
tests
to
even
get
rid
of.
In
those
cases.
B
A
Right
well,
I
I
I
could
definitely
see
both
life
cycle
and
GC
being
Rewritten
in
terms
of
async,
although
I'm
not
sure,
there's
a
huge
benefit
in
doing
that.
B
A
B
B
C
B
Well,
we
would
always
want
well
I
I
personally
think
this
is
that
this
is
a
that
this
is
a
secondary
Point
like
five
seconds,
and
it
doesn't
have
so
many
threads
that
they're.
But
it's
a
problem
so
you'd
replace
it
with
something,
but
we
can
make
it
work,
but
I
think
but
I
recommend
they
focus
on
not
taking
away
the
ability
to
do
the
Ferrari
admin
and
life
cycle
to
do
what
they
do,
but
but
make
sure
we
get
to
something
efficient
for
the
for
living
in
your
face.
A
E
C
The
one
thing
there
is
that
I
kind
of
worry
that
that
would
that
we
would
basically
have
to
essentially
Force
every
zipper
back
end
to
to
to
actually
support
asynchrony
I'm,
not
opposed
to
doing
that.
I'd
love
to
force
every
zipper
back
into
actually
supporting
synchrony,
but
I
think
we
actually
would
have
to
support
it.
Force
every
zipper
back
in
to
actually
do
that.
A
Yes
or
I
mean
what
I
always
have
to
run
more
threads
for
back
ends
that
don't
have
async.
Oh.
C
B
B
A
Then
I
think
it's
okay
for
other
backends
to
decide
on
their
own
and
convert
things
to
async
on
their
own
time.
Frames.
B
That
sounds
more
straightforward
than
I
thought.
It
was
in
terms
of
that
main
interface,
or
you
know
that
now
the
Roger
goes
what
well
I
guess.
Where
were
we
in
the
process?
We
did
a
lot
of
work
to
to
optionally,
unify
things
and
and
I
guess
I
guess,
there's
work
to
be
done
in
zipper.
Is
that
is
that
fair?
A
Ali,
what
do
you
remember
from
from
this
passes?
We
were,
we
picked
at
least
some
requests
and
tried
to
get
rid
of
all
of
The
Blocking
weights,
but
they're
still
still
several
in
there.
F
Yeah
I
I,
don't
think
I
made
it
past,
create
bucket
or
I
might
have
gotten
create
bucket
done,
but
then
I
didn't
move
on.
To
put
I
mean
this
is
something
that
I
can
take
up
or
maybe
me
and
somebody
else
could
take
up
because
it's
just
it's
just
throwing
them
all
through
there
and
then
knocking
out
the
no
yields
before
the
the
before
the
blocking
on
libretos
calls
right.
B
F
Yeah,
what
if
I,
go
through
and
do
an
op
and
then
shoot
him
or
I?
Guess
really
back
with
the
instructions
maybe
next
week,
and
then
we
can
go
from
there.
It
should
be
pretty
straightforward.
F
I
think
there
was
just
something
I
needed
to
do
to
print
the
back
Trace
out
of
all
the
functions
that
hit
at
a
certain
point
in
GDB,
where
the
blocking
liberators
call
happens
for
when
for
when
that's
hit,
I
I
yeah
I'm.
Trying
to
do
my
best
to
remember
but
yeah
that
mean
I
yeah
I'd
be
happy
to
work
on
it
with
calpash
a.
C
If
I
can
make
a
suggestion,
it
might
be
simpler
just
to
go
file
by
file
and
say
you
know
if
we
know
the
only
places
that
the
null
yields
should
exist,
at
least
for
now
are
in
rgw
admin
and
yeah
things
like
life
cycle,
then
we
could
go
through
and
fight
that
you
could
just
go
through
file
by
file
and
just
knock
them
out
on
that.
You
know,
knock
them
out
on
a
file
and
all
its
agencies.
That
sort
of
thing.
F
I
think
op
IOP
is
the
better
approach,
but
yeah
I
can.
A
Well,
yeah
I
think,
in
my
view,
like
put
option,
get
up
are
going
to
be
give
us
the
most
benefit
so
in
terms
of
prioritization,
maybe
start
with
crucial
apps
like
that,
but
then
eventually
going
file
by
file
and
knocking
out
everything
else,
I
think
is
the
only
way
to
to
be
sure
that
it's
complete
yeah.
B
So
that's
that's
like
a
good
plan,
so
we're
coming
back
quickly
to
zip
to
the
zipper
interface.
Are
there
gaps
there
or
not
or
not?
B
F
Mean
I
would
guess
the
only
gabson
zipper
would
be
some
of
the
calls
in
zipper.
Just
don't
take
optional
yields,
I'm,
not
sure,
if
that's
the
case
but
I'm
gonna
guess
the
only
zip
related
changes
would
just
be
adding
optional
yields
to
certain
function
calls
if
they're,
if
they
all
don't
already
have
them.
A
A
I
mean
so
I.
Don't
think
that
there's
anything
special
that
we
need
to
do
for
zipper
like.
B
E
E
So
that
would
be
where
I
would
start.
Yeah
I
think
I
agree
with
that
with
all
those
no
yields
and
Trace
them
back
up
to
see
where
they
come
in
and
fix
the
API
to
pass
them
in
if
it
doesn't
already.
B
Well,
like
the
thought
process
I
have
is
we're
about
to
start
a
new,
a
new
zipper
driver
for
the
that's
good.
That's
going
to
be
operating,
that's
going
to
that's
going
to
be
operating
on
apparently
on
the
posix
interface,
which
also
got
a
bunch
of
up
calls
going
from
gpfs.
B
We
get
the
implementation
of
that
driver
to
be
as
a
synchronous
as
we
can
from
go
so
that
we
don't,
for
example,
become
thread
starved
if
we,
if
we
want
to
run
that
system
with
a
small
requested
pool.
C
A
B
A
B
C
C
G
B
A
B
Okay,
that
that
covers
everything
I
wanted
to
cover
I,
feel
I
feel,
like
I
feel,
like
that,
the
other
other
hypothesis
for
taking
up
so
much
how
much
time,
thanks
for
helping
that.
F
Yeah
I
know
I'll
I'll
try
to
report
back
with
an
OP
gone
through
by
the
next
meeting.
A
And
on
the
d4n
side,
the
redis
client
I
think
our
integration
is
currently
synchronous,
but
I
saw
on
the
Boost
mailing
list
that
somebody's
proposing
a
seo-based,
redis
client
library
to
boost.
So,
if
that's
accepted,
then
we
have
a
good
option
for
making
that
stuff.
Async.
Also.
B
F
Would
we
want
to
make
that
go
after
samara's
well,
after
samar's
initial
PR's
merged.
B
B
A
Yeah,
it
does
look
high
quality,
not
sure
how
much
of
a
community
it
has
or
whether
they
would
keep
up
with
updates
in
redis
or
new
versions,
but
being
included
in
boost
makes
that
a
lot
more
likely.
B
A
All
right
so
C,
plus
plus
20
Co
routines
for
multi-site,
are
still
in
progress.
A
Adam
has
done
some
groundwork
for
neorados
to
support
those
and
I.
Think
Uval
is
doing
some
prototyping
for
metadata.
Sync,
myself,
I've
been
looking
at
how
to
do
the
yield
spawn
window
thing
that
our
current
core
routines
do
for
bounded
concurrency,
that's
a
little
tricky
with
C
plus
plus
20
co-routines,
or
with
azio
at
least,
and
we
need
something
that
kind
of
tracks
the
core
routines
we
spawn
and
wait
for
them
to
complete
before
spawning
more
so
I've
been
working
on
a
class
that
does
that
so
atom
and
you've.
A
E
A
A
B
It's
me:
if
we've
actually
kind
of
benefited
I
mean
we
should
probably
try
not
I
mean
we
should
try
not
to
I,
mean
I,
don't
like
Gary
I
guess,
but
we're
not
We're
Not
Gonna.
We
need
it
until
I
think
either.
Maybe
it
can
be
a
post
post,
a
backboard
if
that's
easy,
to
do
whatever
it
makes
sense.
So
I
guess
I.
E
A
D
E
E
There
are
the
AIO
stuff
is
all
gone,
so
basically
we're
left
with
stuff
that
is
rgwsi
Zone
and
rgwsi
sysops,
which
is
used
by
Zone
and
rgw
period.
Those
three
things
are
the
ones
that
are
left
and
the
syslop
stuff
is
only
still
required
because
it's
used
by
the
zone
and
period
stuff.
D
D
E
Right
right,
yeah,
my
reference
yeah.
No,
so
yeah,
that's
the
problem,
it's
loadable
modules
doesn't
build.
We
need
the
Zone
group
stuff,
but
but
you
have
all
we
have,
but
we
have
all
the
pieces.
No,
we
don't
The
Zone
group
work
is
not
done.
That
was
well
I'm,
sorry,
I,
misunderstood
and,
and
so
my
my
question
was:
is
the
Zone
group
stuff
close
or
should
I
start
working
on
it
or
is
Casey
want
to
finish
it
himself?.
F
B
A
E
B
B
D
Yeah,
that's
what
we
want
to
do
and
that's
what
we
we
want.
Chaos
and
what's
the
other
one,
so
those
guys
have
can
start
working
with
the
framework
correctly
and.
B
A
I
would
I
would
encourage
Caleb
to
kind
of
split
out
anything
in
that
PR.
That's
that
could
be
merged
without
going
all
the
way
to
separating
the
libraries
so
that
we
can
stop
carrying
all
of
it
in
a
future
branch.
A
E
Yeah,
that
seems
like
a
reasonable
Way
Forward
is
to
start
merging
the
the
file
split
up
changes
without
the
actual
module,
the
hard
module
changes
in
cmake.
A
Well,
the
the
pr
has
tons
of
changes
to
source
files,
and
it
seems
like
all
of
those
could
probably
emerge
without
the
cmake
part,
that
splits
libraries,
that
that
is
what
it.
E
D
E
D
E
A
B
I,
don't
think
so
offhand.
We
definitely
do
want
to
merge
that
we
can
we'll
see.
First,
isn't
ready
at
least
at
least
well
I.
Don't
know
if
that's
true
data
sync
is
is
potentially
ready,
I
guess,
except
it
hasn't
been
evaluated,
except
by
by
sorry
or
and
I.
Don't
know.
If
it's
critical
that
it'd
be
in
before
Reef
I
mean
we
can
always
backboard
it.
E
B
B
B
G
A
I
mean
if
it,
if
it
merges
before
code
freeze,
then
it
can
make
grief.
I
wasn't
I
wasn't
planning
to
have
it
ready
for
a
reef.
B
What
different
notification
back
you
know
or
different
bidding
integration?
That's
just
that
was
soloing
effects
two
weeks
ago,
so
it
seems
like
it
seems,
I
mean
at
least
in
summers
some
respects.
But
this
is
just
what
happens?
There's
I,
don't
think,
there's
any
risk
of
thick
rod.
I
mean
it's
gonna,
be
a
thing
to
be
taken
over
by
somebody
an
extended.
Regardless
of
what
happens.
If
you
know
I,
think
or
I,
don't
remember
how
long
or
spending
on
being
around
that
that
was
like
three
more
weeks:
February.
E
B
That
is
branches
and
which
one
right
then.
B
F
A
B
A
Yeah
so
I
mean
I'll
I'll
make
a
a
concerted
effort
to
you
know,
follow
up
with
review
on.
A
D
G
Foreign
flight
stuff
merge
for
Reef,
at
least
the
you
know
the
early
version
of
it
I'm
down
to
one
last
comment
of
yours:
Casey,
which
I
need
to
do
a
quick
test
on
to
see
whether
I
can
replace
a
string
of
you
with
a
string
but
other
than
that
I
think
it's
I
was
able
to
address
all
your
issues.
G
Should
be
safe
because
yeah
it
really
doesn't
yeah,
it
doesn't
interfere
with
any
existing
processes
other
than
startup.
You
know,
there's
a
new
front-end
installed
and
a
new
Port
open,
but
but
only
if
you
turn
it
on.
A
Okay,
it
also
changes
how
we
build
Arrow,
but
is
that
only
if
you
have
it
on.
G
It
changes
well
I
mean
the
fact
that
the
cmake
files
are
modified
or
does
it
change?
How
maybe
it
does
maybe
I
did
turn
on
a
flag
to
to
add
in
some
new
you're
you're
right.
It
does
actually
add
in
fold
in
some
new
libraries
on
the
Arrow
side
through
some
Flags.
That's
correct.
B
A
Yeah
I
I,
don't
think
it
I,
don't
think
that
should
really
affect
anything
for
reef
at
all
of
the
of
the
distros
we're
building
for
are
still
building
from
sub
module.
So
it's
not
like
we're
adding
new
dependencies
to
flight
packages.
B
A
Self
context
or
sep
config
has
support
for
experimental
features,
but
I'm,
not
a
big
fan
of
it.
I
think.
A
B
Let
Eric
know
how
he
did
Mark
this.
You
know
so
it's
like,
so
that
everybody
knows
that
that's
the
case.
A
G
Okay,
yeah
they're,
definitely
some
configurables,
so
I
I
will
make
sure
those
are
marked
in
there.
A
A
Do
you
mean
the
fifo
branch?
Yes,
it
might
not
actually
need
a
rebase.
The
the
javafix
was
just
from
cleaning
up
packages
on
the
builders
cool.