►
Description
Kubernetes Storage Special-Interest-Group (SIG) Object Bucket API Standup Meeting - 11 January 2021
Meeting Notes/Agenda: -
Find out more about the Storage SIG here: https://github.com/kubernetes/community/tree/master/sig-storage
A
All
right
so
jeff,
okay,
so,
let's
see
who
do
we
all
have
here?
Perfect,
okay,
nicholas
is
not
here,
but
that's
okay.
A
A
So
last
week
we
were
talking
about
finalizers
and
also
we
touched
upon
documentation
a
little
bit.
I've
been
keeping
an
eye
on
the
common
chat
and
jeff.
Am
I
right
in
understanding
that
you're
looking
at
the
documentation
a
little
bit.
B
Yeah,
I
did
a
little
look
at
the
weekend,
but
but
trainees
going
to
the
last
I've
heard
from
him,
he'll
push
it
into
the
official
repo
and
then
I'll
just
create
a
pr
against
the
spec.md
file.
A
Okay
sounds
good
so
which
part
of
the
so
can
someone
tell
me
which
repo
is
this
pull
request
on
said.
C
A
C
B
A
Understood
understood,
so
let
me
take
a
look
at
this,
so
what
do
we
need
to
do
before
we
push
it
to
in
upstream.
C
In
making
it
super
correct
right,
it
is
like
in
the
iterations
we'll
correct
all
the
issues.
B
Right
and
it
would
be
good
to
get
other
review
comments
on
it.
The
the
whole
decision
last
week
was
that
the
kep
is
static.
It's
not
going
to
be
updated,
it's
merged
and
done,
but
we
need
a
way
to
document
cozy,
yeah
and
well.
This
is
how
we're
gonna
do
it
and
there'll
be
prs,
can
be
written
against
the
document
to
keep
it
current.
A
Okay,
so
how
about
we
do
this
swinging
yeah?
Let's
start
by
making
a
pull
request
directly
top
stream.
I
think
that
sounds
good.
I
will
not
define
so
how
about
we
just
get
the
architecture
right,
because
I
think
it
might
be
easier
to
review
the
pr
in
small
chunks,
okay
and
yeah
a.
C
Yeah,
this
is
yeah
this
you
know
it's
either
we
can
embed
images
or
you
know
either
we
can
do
similar
to
what
we
are
doing
here
right,
which
is
yeah.
A
A
Yeah
yeah,
this
can
be
cool.
I
was
thinking
more
like
sidecar,
but
overall
architecture,
this.
This
is
co
right
and
then
maybe
like
in
our
picture.
We
have
like
a
more
detail.
You
will
have
three
components
where
the
sidecar
talks
to
the
controller
and
yeah
and
then
maybe,
as
a
third
picture,
we
can
define
the
resources.
A
A
Yeah,
so
I
think
architecture
wise
see
generally
architectures,
you
show
two
things.
One
is:
what
are
all
the
components?
Two
is
how
they
interact.
So
if
you
can
show
just
that,
that'll
be
that'll.
Be
enough.
As
someone
who
comes
in
new
say
they
don't
understand
how
cozy
works,
they
should
be
able
to
get
a
decent
sense
of
it.
Looking
at
the
architecture,
diagnosis
sounds.
A
So
right
now
we're
just
going
over
some
of
the
things
hey
right
now
we're
just
going
over
some
of
the
work
that
has
been
done
in
the
last
week.
A
I
know
you
haven't,
I
don't
know
if
you've
been
able
to
join
after
you
know
after
the
new
year
started
in
case
in
case
you
weren't
here
last
week
last
week
we
talked
a
little
bit
about
how
our
finalizers
are
set
up
and
also
we
talked
about
documentation
being
needed
because
more
and
more
people
from
the
community
are
interested
and
are
asking
questions
or
making
feature
requests,
and
it's
high
time
that
we
have
some
documentation
in
place.
So
they
understand
where
we
are
at
so
okay
stages.
A
Is
here
hi,
just
hey,
sir
hey,
so
I
wanted
to
bring
to
your
attention
some
of
the
changes
we
made
to
the
customized
template
while,
while
trying
to
you
know
integrate
everything,
okay,
so
kubernetes,
I
think
I
made
this
pull
request
to.
Let
me
see
is
this
csi.
A
Okay,
it
was
approved
okay,
so
we
were
trying
to
get
this
working
digits
and
the
current
way
we've
set
up.
The
customized
templates
for
the
anti-cozy
project
is
to
have
a
single
manifest
that
will
deploy
all
of
the
different
components
in
in
central
repo,
correct
right.
A
So
one
of
the
issues
we
faced
with
that
was
the
controller
report.
One
of
the
issues
we
faced
with
that
was.
It
was
not
very
clear
how
to
reference,
how
to
reference
the
different
crds
or
files
from
other
repositories
right
here
such
that
we
always
get
the
latest
you're
looking
at
base.
A
I
think
you
need
to
go
into
full
stack,
yeah,
full
stacks,
no,
actually
even
in
base
there's
an
issue,
you
see
the
crds
don't
get
deployed
when
you
put
it
like
that
they
work
if
you're
doing
customized
build,
but
customize
is
a
part
of
cubectl
and
the
cubectl
version
of
customers.
Even
the
latest
does
not
do
that.
Does
not
work
with
this.
A
A
So
so
we
were
experimenting
with
different
mechanisms,
even
for
our
integration.
We
were
trying
to
see
what's
the
best
way
to
do
it
and
see
the
intention
of
doing
it.
The
previous
way
was,
if
I
have
a
place
to
write
so
I'll,
just
write
here,
just
just
to
show
you
what
we
were
thinking
before
earlier.
The
idea
was,
they
should
just
be
able
to
do
cube,
ctl,
create
dash
k
and
then
deploy
slash,
full
stack.
A
A
For
the
time
being,
we
came
up
with
a
solution.
That's
simpler
and
it'll
help
us
move
forward,
so
what
we
did
instead
of
instead
of
trying
to
have
a
full
stack,
you
know
folder
inside
inside
a
central
repo.
A
It
seemed
to
be
easier
to
just
have
a
customized
template
per
repository,
so
what
we
did
customization
yaml
in
the
root
of
the
repository
and
that
just
points
to
some
resources
under
the
resource
directory
and
now
what
people
can
do
is
to
deploy
this
specific
repository.
They
just
do
cube,
ctl,
create
hk
github.com
or
wait
six
dot.
A
Okay,
it
is
the
I
o
slash,
cozy,
cs,
adapter
and
so
on.
So
if
I
wanted
to
deploy
a
different
one,
you
know
say
the
controller.
I
would
just
do
that.
So
in
order
to
deploy
the
full
stack,
I
would
have
to
issue
four
cube.
Sql
commands
instead
of
one,
but
that
is
an
okay
compromise,
given
that
it's
simpler
for
us
to
build
and
we
can
move
forward.
Okay,
so
is,
is
that
what
you
want
to
do
for
so
I
have
a
pr
open
for
site
car
right
or
for
privilege
right.
D
A
I
I
can
go
ahead
and
refactor
that
similar
to
what
you
have
done
for
the
adapter.
That
would
be
good.
That's
that's
what
I
was
hoping
to
do.
Yes,
okay,
perfect,
all
right
and
then
what
do
you
want
to
do
for
controller
then?
Like
so,
do
we
want
to
update
that
as
well?
Yeah,
yeah?
Okay,
we
can
take
out
that
deploy
and
we
can
just
follow
the
same
structure
there
as
well.
Okay,
so
we
are
not
doing
full
stack
anywhere.
Basically,
it's
not
good
stacking
here
yeah
at
this
point.
A
So
these,
when
do
the
crds
get
applied
like
when
you
apply
the
controller
when
you
create
the
controller
changes
or
well,
that
could
also
be
a
customized
template,
so
cozy
api,
okay.
So,
okay,
so
we
want
to
add
customers
to
api
repo
as
well.
Now,
yes,
so
the
crds
get
deployed.
A
Okay,
make
sense,
yeah,
okay,
going
back
to
this,
so
so
I'm
going
to
put
this
in
in
progress
yeah.
So
this
one
that
can
be
merged
directly.
Okay,.
C
A
A
A
Doing
this
one
we
are
not
doing
so,
we
should
basically
say
we
decided
not
to
have
a
single
customize
that
will
deploy
right,
yeah.
Okay,
I'm
just
going
to
move
this
archivist.
That's
right!
Review
are
back
for
controller.
I
believe
you've
done
this.
C
Already
right
yeah,
we
have
done
that
too,
and
we
can
close
them.
Actually,
when
we
do
the
prs
we
should,
we
should
either
do
fixes
with
the
bug
number
so
that
gets
automatically
closed
or
we
close
so
right
right.
We
I
I
can
clean
that
up.
Okay,.
A
All
right,
so
what
about
this
thing?
I
think
this
is
also
done.
A
Okay,
so
so
they
just
you're
clear
right
what
needs
to
be
done?
Next?
Yes,
yes,
okay,
okay,
so
streamy
you're,
going
to
do
the
documentation
and
jeff
is
going
to
follow
up
with
that
and
the
next
thing
so
rob.
What
are
you
working
on
right
now.
D
So
in
doing
the
integration
stuff
right,
we
found
that
bug
with
a
definition
of
context,
and
then
I
was
working
on
so
I
fixed
I
fixed
that
we,
you
know
you
created
a
temporary
to
move
forward,
but
I
fixed
that
and
in
a
pr
I
haven't
pushed
up
and
I
was
working
on
re-importing
the
logging
such
that
we
can
control
the
logging
level
and
ran
into
some
issues
with
that
and
didn't
do
haven't
been
able
to
do
much
else.
Apart
from
the
the
work
we
were
doing
to
try
to
get
integration
going.
C
Yeah
yeah,
so
probably
we
should
open
issue
because
some
some
of
the
projects
are
using
glock.
D
Yeah,
okay,
it's
though
it's
got
a
way,
complicated
setup
and
just
for
some
reason
I
don't
know
if
there's
something
unique
and
what
I'm
doing
in
sidecar
or
not,
but
just
doing
an
initialization
is
throwing
things
into
a
loop.
So
I'm
I
need
to
dig
into
that.
A
Got
it
so
rob?
Can
you
take
that
up?
Can
you
make
sure
I'm
gonna
work
on
that
this
week?
Okay,
so
there
are
two
other
issues
assigned
to
you.
So
have
you
already
finished
them
or
what's
the
what's
the
status
of
these,
which
are
there
two
issues?
Add
removal
finalizes
on
bucket.
A
Okay
sounds
good,
okay
and
on
bucket
access,
this
will
also
be
assigned
to
you
right.
A
Well,
once
you've
finished
this
task,
yeah
I'll
leave
you
assigned
for
now
and
yeah
depending
on
who's
screen.
We
can.
D
A
Okay,
so
unassigned
you,
then
so
it's
it's
open
for
they,
just
if
he
is
done
with
this
current
task.
A
Absolutely
okay,
so
the
next
thing
is
chris:
are
you
here.
A
I
wish
so
have
you
started
working
on
finalizer
for
buckets.
E
Yeah
I
started
working
on
that
a
little
bit
over
the
weekend.
Just
trying
to
I
just
figure
out
like
finalizers
more,
I
haven't
worked
too
much
with
them
before
so.
A
Okay,
so
yeah,
if
you
want
to
go
over
them,
you
know
together
I'm
happy
to
get
on
a
call,
maybe
yeah
we
can
talk
more,
but
if
you're
confident
or
you
know,
if
you,
if
you
want
to
take
it
up
yourself-
that's
fine
too
yeah
I'll
leave
it
up
to
you.
E
Yeah
I'll
ping,
you,
if
I
have
any
questions.
A
Sure
that
sounds
good
all
right,
so
the
next
thing
I
want
to
bring
up
is
like
I
was
saying
there
are
more
people
from
the
community
interested
in
contributing
and
also
in
trying
this
out.
Last
week
we
got
some
interest
from
the
rook
dot
io
community.
Today
this
morning
I
think
there
was
a
new
issue
added
by
this
person
from
india.
A
For
usage
and
performance,
prometheus
metrics,
so
things
like
usage
capacity
capacity
might
be
applicable
in
some
cases.
A
Okay,
so
I
mean
he's
asked
for
performance,
for
instance
latency
trooper
niops.
These
don't
really
apply
to
especially
iops.
A
These
are
all
functions
that
are
that
that
can
only
be
measured
from
the
client
that
is
making
the
request
to
actually
all
three
of
them.
They
can
only
be
measured
by
the
client.
That's
making
the
request
to
the
back
end,
because
the
back
end
does
not
differentiate
one
client
from
another
and
depending
on
where
the
client
is.
These
numbers
are
going
to
change.
A
So
I
don't
think
it
makes
sense
to
do
performance
this
way
in
terms
of
latency
throughput
ios
per
bucket.
However,
we
could
have
some
performance
metrics
from
the
bucket
something
like
average.
You
know
I
08
or
write
speed
per
mb
or
something
like
that.
F
F
I
mean
the
server
knows
how
much
data
it's
served
over
how
much
time
but
but
throughput
it
depends
on
the
latency
right.
No,
I
mean,
like
you
know,
I
see
what
you
mean:
you're
pushing
a
megabyte
per
second,
you
know
that
you've
sent
a
megabyte
per
second
out
the
front
door
received
it
in
the
front
in
the
front
door.
Similarly,
with
iops,
you
know
how
many
requests
you've
serviced.
Of
course,
you
don't
know
if
the
other
guy
ever
saw
the
responses
you
sent,
but
that's
not
your
problem.
A
F
So
let's
talk
about
these
three
things:
there
is
a
concept
of
server-side
latency,
so
from
the
time
that
the
response
shows
up
at
your
front
door
at
the
time
the
response
leaves
if,
if
that's
more
than
zero
seconds,
that's
server
induced
latency.
A
The
thing
is
so
talking
about
performance
with
object,
storage,
the
the
the
bottleneck
or
the
slowest
part
of
this
entire
setup
is
going
to
be.
The
wire
is
going
to
be
the
network,
it's
not
going
to
be
the
amount
of
time
it
takes
to
retrieve
the
data.
It
depends
on
your
architecture,
regardless
of
the
architecture.
The
the
network
is
going
to
be
your
slowest
part,
not.
A
F
A
The
so
what
I've
seen
from
the
industry
is,
I
mean
we
have
to
do
this
right.
We
have
to
give
them
metrics
that
are
that
are
usable
by
them
and
and
to
start
with,
we
have
to
give
them
metrics
that
mean
the
most
like.
Yes,
this
metric
might
be
useful.
However,
I
think
there's
a
higher
priority
in
terms
of
like
from
the
client
side,
throughput
and
latency
and
iops
make
way
more
sense
for
for
a
customer
or
someone
who's
using
cozy
compared
to
how
the
servers
behave.
F
F
Product
but
but
it's
not
but
you're,
right
like
once
once
we
once.
This
is
the
most
important
thing
to
work
on
yeah.
We
should
get
it
right
and
client-side
numbers
are
valuable.
I
was
just
only.
I
was
just
pointing
out
that
there
are
numbers
you
can
generate
on
the
server
and
they're
not
worthless
they're,
just
maybe
not
the
best
numbers
you
can
get.
I
agree
and
and
furthermore,
like
we
don't
have
a
footprint
on
the
client
side,
we're
not
in
a
position
to
give
you
anything
other
than
what
the
server
knows.
A
You're
right
so
so,
let's
create
a
new
milestone
for
this.
So
extremely
I
don't
remember
how
to
create
a
new
milestone.
Can
we
put
this
as
just
v1?
For
now
I
mean
we
haven't
even
crossed
alpha.
A
A
Yeah,
okay,
so
just
as
a
you
know,
I
had
some
few
a
few
thoughts
on
this.
I
think,
as
a
first
step
in
order
to
implement
something
like
this,
we'll
need
to
introduce
a
new
grpc
api.
Again,
we
shouldn't
worry
about
this
too
much,
but
just
as
a
just
as
an
early
idea
of
how
we
can
implement
this
we're
going
to
need
a
new
grpc
api
that
will
get
exposed
from
the
provisional
sidecar.
A
A
Okay,
all
right,
so
we're
almost
out
of
time.
I
think
I
have
a
good
sense
of
how
we
are
moving
forward,
so
I
wanted
to
give
one
more
update,
so
we
are
almost
well
we're
actually
pretty
much
ready
to
show
a
demo
on
thursday,
so
we've
got
all
the
three
components
working
together.
A
We
I've
seen
the
basic
workflows
and
they
work.
We
just
have
to
put
together
a
demo.
There
are.
There
are
a
few
things
that
are
remaining
before
we
can
do
the
demo,
which
is
getting
these
three
pull
requests
in
yeah
this
one
we
we're
fixing
pages
is
working
on
it.
A
We'll
only
do
the
demo
if
these
four
things
are
in
place,
because
that's
the
that's
the
expectation
we
set
for
the
demo
saying
that
you
know
when,
when
we
do
make
the
demo
as
much
as
we
show
will
actually
be
usable
in
in
as
close
to
a
production-ready
manner,
as
we
can
at
this
point.
A
So
yeah
look
forward
to
the
demo
on
thursday.
If,
if
anything
holds
us
back
or
if
something
comes
up
there,
any
blockers
we'll
bring
it
up
on
thursday.
A
The
other
thing
we'll
bring
up
on
thursday
is
how
we're
setting
up
the
finalizers
across
the
three
components,
and
you
know
how
they
get
created,
and
it
is
that
thing
so
you
know
stuff
like
that.
So
that's
it
from
me.
Does
anyone
have
any
questions.