►
From YouTube: Ceph RGW Refactoring Meeting 2023-05-24
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/contribute
What is Ceph: https://ceph.io/en/discover/
A
All
right
welcome
everybody
first
on
the
agenda
is
more
of
an
announcement
that
the
brief
release
candidate
is
coming
soon
sounds
like
we'll,
probably
try
to
cut
it
next
week,
so
I
want
to
make
sure
that
all
of
our
bug
fixes
that
have
merged
to
master
are
at
least
staged
for
Reef,
so
that
we
can
get
them
into
the
RC
for
testing.
A
A
And
so
maybe
maybe
we
can
schedule
a
meeting
with
Mark
and
see
how
to
get
those
rolling.
B
Yeah
I
think
we
will.
B
A
All
right
great
yeah,
it
looks
like
there
are
still
a
couple
multi-site
related
years,
open
against
Reef,
so
we'll
want
to
make
sure
that
those
get
in.
A
C
Yeah
I
have
two
fixes
to
the
Kafka
Library
that
are
causing
crashes
and
I
would
really
like
to
get
them
in
I
mean
those
are
open.
Prs,
so
I
need
them
to
be
merged
to
to
main
first
and
then
I'm
gonna
try
to
backboard
on
tourists.
C
So
the
the
the
the
crashes
are
not
found
in
our
testing,
like
in
tautology,
they're
actually
found
separately
by
an
upstream
user
and
by
Downstream
QE,
so
Downstream
key
verify
that
the
the
crashes
I
mean
the
fixes
are
fixing
the
crashes,
so
I
think
they're,
good
I'm,
Gonna,
Run
them
through
tautology
just
to
be
to
make
sure
I'm
haven't
broken
anything
but
other
than
that.
I
need
the
review
and
I
think
that
can
go
in.
C
C
Other
one
is,
there
was
one
place
when
I
was
missing
a
lock,
the
other
one
actually
is
a
potential
race
condition.
It
wasn't
the
thing
that
it
was
causing
the
the
crash,
but
I
mean
I.
Think
the
Edward
also
got
approval
format,
so
the
main
thing
to
and
the
other
one
is
like
one
line
of
adding
a
mutex
where
it
was
missing,
but
the
important
one
is
the
getting
rid
of
the
shared
pointer.
C
This
was
actually
someone
approaching
me.
They
have
a
a
project
where
they
need
they're
using
S3
storage
and
they
want
to
have
like
a
self-contained
solution
right,
it's
an
operator
in
in
kubernetes
or
openshift,
and
they
want
to
have
a
self-contained
solution
where
the
actual
storage
is
provided
is
part
of
the
solution.
C
So
I
was
proposing
to
use
a
rook,
but
Rook
actually
installed
the
full,
the
full
cluster,
and
this
is
a
little
heavy
weight
for
what
they
need.
So
I
was
wondering:
if
there
are
any
I
mean
they
have
to
be
work
on
The
Rook
side
to
enable
the
rgw
Standalone
node.
But
are
there
any
reasons
or
problems
that
would
prevent
us
from
running
DB
store
there?
So
and
also
do
we
have
like
documentation
or
explanation
of
how
exactly
to
set
up
DB
store,
because
the
The
Rook
operator
will
have
to
do
all
this
setup.
D
So
it
depends
on
what
they
want
to
do.
Dbstore
does
not
yet
support
everything.
C
D
C
Use
case
again,
I
I
know
it
doesn't
support
everything,
but
the
use
case
is
that
they
have
an
application
that
wants
to
use
an
S,
S3
storage
and
they
went
away
for
like
a
way
to
demo
or
a
way
to
install
that,
like
on
a
single
node,
kubernetes,
cluster
and-
and
just
you
know,
get
the
thing
running.
C
So,
for
example,
if
they
installed
Rook
with
Seth,
then
they
need
an
extra
disk
or
you
know
they
have
all
kind
of
other
requirements
and
need
more
CPU
more
memory,
and
they
want
something
simple:
they
can
just
Spin
and
show
that
it's
working
right.
E
Right
exactly
we'll
have
to
test
it.
Okay,
one-on-one
right
I
mean
what's
interesting
to
me.
What
would
be
valuable
to
me
is
the
framework,
the
framework
for
deploying
it
it
could
be.
You
know,
maybe
we
should
Mark
it
as
experimental,
and
maybe
we
still
need
to
wait
a
few
months
to
get
there
or
whatever.
If
we
get
some,
if
we
could
agree
on
some
point
where
we
say
it's
MVP
or
something.
D
E
That's
fair,
that's
that's!
Actually!
That's
pretty.
You
know
the
point.
I
don't
know
if
you
mentioned
it,
I
mean
they're
they're.
Currently
using
many
of
these
guys
right
I
mean,
and
so
that's
a
pretty
complete
system,
and
then
we
don't
we're
now
we're
now
we're
not
we're.
Not
we're
not
pretending
right
now
that
we
have
that
right.
Well,
so.
C
I'm,
what
I'm
trying
to
say
is
that
if
you
want,
if
you
want
to
demo
something
something,
if
you
want
to
demo
your
application,
you're
not
going
to
do
multi-side
failovers
you're
not
going
to
do
any
of
that.
This
is
not
part
of
any
demo
right.
So
it
doesn't
matter
that
we
don't
support
that
right.
D
D
E
D
See
so
I
don't
know
as
far
as
using
dbstore.
So
there
is
a
readme
in
the
DB
store
driver
directory.
Okay,.
C
D
It's
intended
for
running
dbstore
with
V
start,
and
so
it's
part
of
the
information
on
how
to
run
DB
stories,
in
fact,
in
v-start
itself,
but
between
that
readme
and
V
start
there
is
the
complete
information
you
need
and
it's
not
complicated,
there's
like
a
command
line
option
and
a
few
config
options
that
need
to
be
set,
and
that's
it.
Okay,.
C
D
Have
a
few
outstanding
changes
to
that
readme
that
I've
been
sort
of
building
up
as
I've
been
using
it
using
dbstore
to
make
it
clearer
for
new
people.
D
I
guess
I
can
push
that
if
that,
if
that
would
help,
but
it's
it's
not
complicated.
It's
really
dead,
simple.
If
you
use
vstart,
but
if
you
want
to
you
know,
use
it
outside
of
vstart
which
people
have
started
asking
to
do,
then
it's
going
to
need
a
little
bit
more
documentation,
but
it's
not.
C
Complicated
so
in
Rook,
of
course,
there's
no
V
start.
So
in
Rook,
like
you,
you
just
code
in
golang
all
the
steps.
That's
that's
what
the
Rook
operator
does
the
screen
the
steps.
So,
but
if
the
steps
are
clear
in
in
bash
and
some
read
me
then
and
I
guess
it
shouldn't
be
too
difficult
to
I
mean
running
like
admin
commands
or
setting
config
variables.
C
The
all
the
mechanics
exist
for
that,
of
course
in
group,
because
they
do
that
all
the
time
right,
the
only
thing
that
might
be
an
issue
there
are
like
installation,
dependencies
and
stuff
like
that.
So.
D
The
like
I
said:
there's
like
up
a
couple
of
config
options,
things
like
where
the
DB
is
stored
and
then
a
command
line
option
to
set
the
the
store
to
be
DB
store,
and
that's
really
it.
The
only
caveat
is
right
now,
I
do
not
believe
you
can
run
rgw
admin
commands
at
the
same
time
as
an
rgw
is
as
the
rgw
is
running,
because
I
don't
think
we
allow
multiple
access
to
our
sqlite
database.
F
This
is
a
conceptual
question:
where
does
the?
Where
does
the
DB
actually
live?.
D
C
Okay,
that's
great
so
I
kind
of
in
the
I
pasted
the
the
the
feature
request
that
I
opened
for
look
in
the
the
agenda
and
if
you
have
stuff
to
add
there
or
comments
or
something
then
feel
free
and
I'll
I'll.
Take
that
to
the
discussion
with
the
Rook
team
and
see
because
I
mean
it
will
require
some
work
on
their
end
because
for
for
example,
now
they
they
have
I
mean
the
installation
of
the
osds
and
all
that
stuff
is,
is
mandatory.
C
It's
not
optional
and
they'll
have
to
figure
out
how
to
do
that.
So
there's
some
some
work
to
do
on
The,
Rook
side,
but
I
want
to
make
sure
that,
on
on
our
end,
it's
clear
how
to
actually
set
up
the
lgw.
D
D
F
D
F
C
Yeah
I,
don't
think
sqlite
allows
that
yeah
yeah.
A
Yeah
but
I
mean
and
doesn't
Rook
do
like
Auto
scaling
to
add
more
rgw's,
and
that
assumes
that
they're
serving
the
same.
Oh
data.
E
I
mean
I
think
what
the
benefit
the
big
thing
about.
This
is
because
of
the
kind
of
a
framework
to
deploy
the
standalone
it'll.
Be
that
way.
I
should
also
be
thinking
about.
You
know
how
that
are
different.
How
Standalone
might
evolve
right?
We
might
want
to
do
DB
store,
because
there's
now,
when
we
have
podx
files,
we
might
want
to
support
that
making
that
available
on
the
same
framework.
D
F
D
A
Cool
well
I
think
this
is
a
great
idea.
Hey
just
want
to
make
sure
that
we
Market
it
as
experimental,
though
I
think
it'll
be
a
while
before
we're
ready
to
fully
support
a
DB
store
and
promise
backward
compatibility.
A
So
just
want
to
make
sure
that
users
aren't
putting
a
bunch
of
important
data
in
it
that
that
we
can't
support
long
term.
A
I
I
was
just
pointing
out
that
I
don't
think
we
really
have
a
strategy
for
making
Backward
Compatible
changes
to
the
DB
layer.
Currently.
A
And
so
I,
don't
think
that
we
can
really
promise
to
support.
This
is
anything
but
experimental
and
temporary
yeah.
B
E
A
C
E
B
Hi
yeah
I
added
that
so
I
think
there
was
another
fee
that
is
still
another
fee
where
we
would
want
to
have
notifications,
synced
and
eval
says
that.
Okay,
let
me
just
share
the
tracker
here.
B
That
we
need
system
objects
to
be
replicated
for
this
to
work.
So
I
just
wanted
to
understand
how
exactly
is
bucket
notification,
how
how
exactly
bucket
notifications
work
with
system
objects
and
what
needs
to
be
done
for
it
to
be
supported.
C
So
bug
notification,
they
store
all
the
information
about
the
topics
and
the
notifications
in
system
objects.
So
a
topic
is
where
to
send
notifications
to
like
the
Kafka
broker
or
mqp
broker,
or
something
like
that.
So
the
information,
the
the
IP
address,
board
or
kind
of
configuration
and
other
things
like
that
are
stored
in
a
system
object
holding
this
information.
So
this
is
the
topic
and
the
notification.
This
is
like
the
the
filter
and
the
type
of
events
you
want
to
notify
on
and
to
each
topic
to
send
and
from
which
bucket.
C
This
is
another
system
objects.
I
mean
it
is
related
to
the
bucket,
but
only
in
the
sense
that
it
has
the
bucket
name
it's
in
its
name,
but
other
than
that.
It's
just
a
system
object
where
we
store
this
information
that
has
the
topic
to
buckets.
This
is
what
we
call
the
I
think
we
also
call
that
the
topic,
but
it's
also
a
system,
object
that
we
store
and
all
these
objects
that
are
not
synced
between
between
site
between
zones.
So,
if
I'm
defining
identification
on
one
zone,
they
won't
be
effective
in
another
Zone.
B
Okay,
yeah
but
I,
don't
I,
don't
really
think
we
really
ever
discussed
about
supporting
system
objects.
Casey
do
you
know?
Was
there
any
plan.
A
A
This
bucket
notification,
stuff
might
be,
might
live
somewhere
else,
but
that's
probably
okay,
but
to
support
metadata
sync.
We
would
have
to
implement
all
of
the
metadata
interfaces
that
are
a
bit
of
a
mess
lately,
but
the
last
example
implementation
was
the
role
stuff
that
Prisa
did
for
STS.
B
Okay,
yeah
I'll
take
a
look
on
how
we
can
add
this
thanks.
Casey
thanks
anyway
sure.