►
From YouTube: Community Meeting, February 28, 2022
Description
No description was provided for this meeting.
If this is YOUR meeting, an easy way to fix this is to add a description to your video, wherever mtngs.io found it (probably YouTube).
A
A
B
A
A
D
D
You
thank
you.
Thank
you
MJ,
so
we
have
two
things
on
the
agenda
today.
The
one
is
more
of
triage.
We
need
a
little
help.
We've
seen
some
strange
Behavior,
that's
the
second
one,
but
the
first
one,
we've
there's
been
a
new
release
of
kcp
and
and
kcp
Edge.
We've
noted
it
as
the
potential
to
be
a
good
first
issue
just
to
help
us
bomb
bump,
kcp
Edge
and
our
dependencies
up
to
the
new
kcp
binary
Mike.
Is
there
anything
you
wanted
to
add
there.
E
Not
obvious
that
I
need
to
if
anybody
has
any
questions,
go
ahead.
Yeah.
D
D
All
right,
if
you,
if
you're
interested,
please
sign
on
and
ask
us
to
assign
it
to
you
or
ask
us
if
you'd
like
to
work
with
us,
then
we'd
appreciate
that.
Thank
you.
D
We've
got
some
strange
Behavior
at
least
I
I.
See
it
as
strange
is
that
in
our
customizer
code
we
have
put
in
the
directives
for
the
compiler
the
we
want
this
to
be
a
namespace
controller.
So
having
that
we
removed
the
Gen
client
non-namespaced
directive,
and
when
we
issue
that
in
the
log
check
we
get
some,
we
get
an
error
that
I
don't
have
in
farming
now,
but
I
think
you
have
in
front
of
you.
There.
D
D
Yeah,
basically,
what
it's
saying
is
it's
the
customizer
namespace
Lister
is
is
missing,
is
is
Undeclared
and
so
I'm
wondering
if
anybody
out
there
has
had
this
same
type
of
issue
or
where
I
might
go
to
fix
it.
H
H
G
A
H
All
right,
so
there
is
a
PR
out,
like
you
have
to
re.
You
would
necessarily
like
today
or
like
whenever,
like
you
have
some
time
like
there's
some
rough
edges
but
we'll
show
today
is
actually
controller,
runtime
already
working
in
a
multi-cluster
scenario.
H
A
C
That's
just
a
just
a
small
tip
from
a
Mac
User.
If
you
go
to
display
preferences
and
just
you
know,
make
the
whole
DPI
setting
too
lower,
it
will
be
quite
nice
for
screen
sharing.
You
don't,
then
have
to
zoom
all
the
things
just
as
a
hint,
because
I'm
screen
sharing
a
lot
as
well.
H
Okay,
how
about
this
we'll
just
so
I'll
start
with.
H
Okay,
so
the
but
yeah,
basically
like
introduces
like
a
couple
of
things.
H
First,
that
is
like
what
is
a
logical
cluster
like
in
terms
of
like
kubernetes
in
general
and
basically
like
it
comes
down
to
the
logical
cluster
needs
to
give
controller
runtime
an
adapter
and
an
adapter
naming
a
CBD,
but
and
at
that,
like
I
wrote
like
at
this
very
simple
kind
adapter
this
morning,
which
basically
uses
the
kind
library
to
say
to
get
a
rest
config
like
when
you
give
it
a
name
or
to
list
clusters
or
to
watch
like
what
clusters
cycle.
H
We
have
and
basically
return
them
so
like
for
a
simple
list
like
you
just
have
like
kind
provided
list,
and
then
that
will
just
return
everything
that
like
starts
with
Fleet,
so
I
pre-created,
like
three
three
cluster
or
like
two
clusters
locally.
H
Wait,
that's
not
with
Fleet,
so
basically
test
one
and
three
test
two
and
then,
and
then
there's
like
a
watcher
as
well
that
basically
at
this
point
in
time
it
just
like
basically
lists
every
couple
of
seconds,
and
so
it's
like
do
I
have
any
changes,
do
I
have
any
new
clusters
or
I
have
deleted
any
clusters
so
because
what
we
want
to
see
is
like
a
cluster
come
and
go
and
controllers
the
controller
runtime
manager
reacting
to
it
and
then
so.
This
is
the
adapter
any
questions
on
this.
A
We
could
step
back
what's
the
final
goal
for
this,
because
from
this
one
that
looks
like
the.
H
Beat
I'll
yeah
I'll
get
into
it.
So
basically
like
right
now,
controller
runtime
supports
manager
only
against
a
single
cluster,
so
the
manager
has
a
single
config
and
then,
if
you
want
to
watch
objects
on
a
different
clusters
like
it's
up
to
you
to
actually
create
the
rest,
configs,
the
caches,
the
clients
and
so
on.
H
This
has
been
a
problem
forever
and
so
like
an
effort
that
I'm
trying
to
push
forward
is
say
that
we
want
to
do.
We
want
to
add,
like
a
native
support
for
Fleet
of
clusters
in
control
runtime.
What
would
that
look
like?
In
this
example?
We
have
a
very
simple
manager
that
runs
against
a
test
and
environment,
so
not
against
kind.
This
is
just
like
a
local
Cube
API
server
in
NCD.
H
This
could
be
kcp
or
what
could
be
anything
else
and
and
then
we
killed
the
manager
like
hey
I,
want
to
tell
you
how
to
find
logical
cluster
and
that's
where
the
adapter
comes
in,
and
the
adapter
right
now
just
works
against
client.
So
they
could.
There
will
be,
for
example,
a
kcp
adapter
in
the
future
that
could
work
against
like
workspaces.
H
And
then
the
other
I
guess
like
the
change
is
that
like
and
now
the
request
itself
as
it's
a
type,
has
a
cluster
in
it
so
like
it
will
come
with
a
logical
name
and
will
tell
you
this
request
came
from
this
cluster
they'll
act
upon
it.
Okay
question
so
far,.
E
H
So
that
would
be
up
to
the
adapter
to
say
for
a
given
rest
or
a
given
name
like
what
is
the
breast
conflict?
You
want
me
to
use-
and
this
is
a
very
deliberate
with
this
choice
of
like
using
address
config
instead
of
any
other
like
higher
level
struct,
because
you
know
for
kcp
specifically,
we
want
to
have
a
way
to
muck,
with
the
rest,
config
really
and
like
change
those
paths
to
like
use
the
workspaces
so
like
slash
cluster
slash
whatever
in
the
future.
H
This
is
not
shown
here
because
I
got
knocked
I
just
did
this
this
morning
before
this
meeting,
but
when
the
reconciled
request
comes
in
a
client
that
is
not
given
to
the
reconciler,
the
reconciler
had
must
go
through
the
manager
to
retrieve
a
cluster
and
then
retrieve
the
client.
So
the
client
will
get
cash
like
at
some
point
and
they
will
be
like
also
a
means
to
refresh
the
rest
can
take
in
the
future,
and
so
you
will
get
the
cluster
and
yeah.
H
The
cluster
has
like
also
like
a
bunch
of
other
things
in
it,
like
grass
mappers
cache
configs.
If
you
need
to
all
those
things,
but
for
click
mostly
properly
potential
just
need
a
client,
you
get
a
client
for
your
cluster
that
will
use
whatever
configuration
you
have
been
given.
H
H
So
this
is
just
a
way
to
make
it
very
generic,
so
that,
like
you,
don't
have
to
really
think
about
what
the
kcp
adapter
is
doing.
It
will
just
do
it
for
you
in
a
generic
fashion,
but
controller
runtime
does
not
for
a
company
to
Fork
at
that
point
to
work
with
kcp.
E
E
I'm
sorry
I
am
I,
could
understand
modifying
a
you
know,
a
server
URL
in
the
course
of
producing
a
rest,
config
but
you're
denying
that
so
I
don't
understand
how
the
akcp
adapter
here
could
produce
one
rest
config,
while
still
modifying
request
paths.
A
I
think
my
follow-up
question
might
answer
this
question
too.
So,
basically,
you
need
to
implement,
let's
say
my
AKs
writing
a
gas
adapter,
so
I
need
to
implement
a
better
interface
to
be
able
to
consume
aqs
clusters
and
each
and
every
time
I
want
to
interact
with
certain
cluster
in
azure.
H
So
that
is
correct,
but
in
there's
one
thing
that,
like
cluster,
the
rest
config
will
be
stored
locally
in
cash
like.
J
H
Like
in
memory
in
controller
runtime,
once
once,
you
have
retrieved
it,
okay.
A
G
F
H
So
you
mean
this
call
right
like
let's
get
yeah,
so
that's
a
good
question
so,
like
we'll,
we'll
have
to
dig
a
little
bit
into
into
how
the
manager
is
implemented,
but
at
a
high
level
get
cluster
goes
to
the
manager
and
say
it's
like
I.
Have
this
logical
name
I
need
a
cluster
out
of
it?
H
The
manager
will
ask
the
adapter
if
it
does
not
have
already
like
a
cluster
like
in
memory.
It
will
ask
adapter
I
need
a
rest
config
for
this
logical
name.
Do
you
have
one
and
if
it
has
one,
it
will
return
the
rest,
config
and
so.
F
That's
where
I
think
that's
what
was
I
think
that's
where
the
contusion
was.
Is
that
it's
not
just
one
rest,
config
mic,
it's
that
there's
a
rest,
config
per
logical
cluster
name
and
the
contract
with
the
adapter
is
I.
Give
you
a
logical
cluster
name.
You
give
me
back
a
rest.
Config
and,
however,
the
adapter
wants
to
implement
that
it
can.
H
Oh
I
meant
like
it
does
not
do
it
at
every
request
like
it
will
be
cached
at
some
point
and
yeah.
Then
then,
I
guess
like
there
is
also
like
the
we
have
I
haven't
covered
like
yet
in
controller
runtime,
like
the
difference
between
like
wild
cards,
requests
and
lesson
watch.
But
there
is
plans
to
support
those
as
well
and
I
can
link
the
hack
and
the.
If
folks
are
interested.
H
So
so
that
that
is
where
TCP
connection
for
workspace
I
would
say.
Yes
at
this
point
in
time,
that
is
correct.
H
F
E
Okay,
if
that
would
be
interesting,
I
have
a
separate
question:
I'll
just
put
my
hand
up
for
it.
Maybe
other
people
can
ask
other
questions.
First,.
E
Well:
okay,
if
doing
these
other
questions,
so
I
think
this
is
a
really
interesting
if
a
area
you're
addressing
I
think
it's
not
specific
to
controller
runtime
I,
think
that
you
know
this
idea
of
a
fleet
of
clusters
is,
you
know,
as
you've
noted
right,
it
appears
in
other
contexts
and
it
would
be
nice
to
have
and
an
interface
a
low
level
kind
of
Upstream
generally
agreed
upon
low-level
interface
to
it,
to
dealing
with
a
fleet
of
clusters
and
another
related
question
that
this
raises
actually
even
Beyond
a
fleet,
a
fleet,
it
implies
a
certain
Unity.
E
So
the
idea
is,
it
is
managed,
coherently,
consistently
and
to
identify
a
cluster.
You
only
need
like
a
name
and
it's
implicit.
What
Fleet
that's
part
of
you
know
in
general,
kubernetes
programming.
You
know
I've
already
been
writing
programs
that
deal
with
multiple
logical
clusters.
E
If
I
will,
if
you
will
or
API
servers
right
and
I
need
to
pass
a
kubeconfig
for
each
one
of
them,
and
you
know
in
general,
if
you're
doing
General
programming,
where
everything's
not
implicitly
in
one
cluster
once
in
one
cluster
or
even
implicitly,
in
the
same
Fleet,
you
know
there's
a
question
of
how
do
you
refer
to
something
that's
somewhere
else
and
that
gets
into
interfaces,
because,
if
you're
designing
apis
that
that
are
talking
about
things
that
are
not
all
in
the
same
cluster,
all
in
the
same
Fleet,
you
have
a
question
of
how
to
deal
with
references.
E
You
know
pretty
generally
I,
don't
know
if
there's
any
specific
answer
I'm
looking
for,
but
you
know
that
that's
that's
one
of
the
things
I'm
looking
at.
H
Yeah
and
I
think
that's
a
fair,
just
general,
like
the
the
general
concepts
of
like
how
do
I
refer
clusters
like
living
completely
somewhere
else
like
that
I
guess.
H
That's
not
like
something
that
we're
trying
to
answer
right
now,
like
we're,
trying
to
say:
okay,
like
I,
need
to
connect
to
multiple
clusters
to
a
flitter
clusters,
as
you,
as
you
said,
like
you
know
like
that,
that
could
be
usually
like
they're
managed
together
or
like
I,
can
get
them
together,
but
in
the
future
there
is
like
a
nothing
stopping
us
to,
for
example,
adding
more
adapters
to
a
manager
and
then,
like
the
question,
then,
is
like
how
do
we
make
sure
that,
like
I
can
reference
the
right
adapter
right,
like
I,
don't
want
to
go
to
an
to
like
kcp
for
a
concluster
and
vice
versa?
H
Right,
that's
gonna,
bring
even
more
I,
guess
complications,
but
we'll
figure
it
out
once
this
part,
I
guess
is
right.
E
Let
me
just
Point
again
to
some
again
future
Builds
on
this
I'm
interested
in
sharding
for
scale
and
in
yeah
yeah.
Actually,
you
know
I've
done
some
examples
of
this.
I
worked
a
simple
example
of
this
with
client-side
sharding
and
it
some
some
object.
Types
I
need
to
charge
for
scale
and
some
I.
E
Don't
if
the
you
know
if
the
coupe
API
server
code
could
deal
with,
for
example,
different
storage
for
different
types
of
objects,
or
maybe
this
adapter
layer
could
somehow
treat
different
kinds
of
objects
differently.
E
H
So
I
think,
like
we
have
a
couple
of
clusters:
I'm
just
going
to
run
not
this
my
example
and
we'll
walk
through
what
what's
happening
here
so
okay.
So
basically
the
manager
got
up
and
running
and
like
as
a
reminder,
the
managers
are
running
in
test
environment,
which
is
basically
a
cubic
server
and
NCD
running
locally.
It
that's
not
running
in
kind
and
then
we're
telling
the
manager
to
like
I
adapt
against
like
a
logical
clusters
like
a
running
in
kind.
H
And
so
as
you
see
here-
and
this
is
like
a
temporary
implementation
but
basically
like
it
will
the
manager
will
start
like
an
event
source
for
each
cluster
and
their
management.
So,
as
you
can
see
like
here
like
I,
say
and
I
guess,
maybe
I
forgot
to
mention
like
this
is
a
reconcile
that
only
watches
pods
to
watch
pods
across
every
namespace
and
so
like.
For
example,
here
I
see,
you
see
starting
an
Event
Source,
the
controller
kind
is
POD.
H
This
is
42
and
Fleet
One,
and
so
these
are
all
the
the
pause.
They're
gonna
be
under
management,
and
so
now,
if
I
go
to
test
you
cluster
and
I
do
like
keep
system
because.
H
G
H
Just
do
kind
net,
it's
gonna.
Add
a
label
here
called
complete
test.
One
see
this
that
was
edited
we'll
go
back
here.
You
can
see
that
the
correct
controller
got
and
cash
like
got
the
event,
and
then
we
detected
the
fleet
annotation.
Now,
what's
more
interesting
and
Lisa
for
me
is
like
okay,
what
happens
if
I
just
create
a
new
cluster
and
I
name?
It.
H
It'll
take
a
bit,
but
basically
the
general
idea
is
like
hey.
We
want
to
act
when
clustering
comes
and
go,
and
we
don't
want
to
leave
like
connection
hanging,
and
you
know,
what's
taking
so
long,
but
basically
like
the
the
manager
will
react
to
changes
because
the
adapter
will
inform
the
manager.
There
is
a
new
cluster
or
I
wanted
to
lead
a
cluster.
So
like
I.
G
H
Okay,
so
here
now
we
got
so
the
test
two
cluster
got
deleted
and
the
controller
got
the
shutdown
signal,
just
that
controller
so
like
for
any
other
controller,
that's
still
running
it
will
it
will
still
work,
and
so
this
is
still
going,
but
you
got
the
idea
like
and
then
the
all
the
workers
will
basically
be
shut
down.
H
H
G
Controllers
a
couple
in
theories.
H
So
as
of
today,
that
is
correct.
The
controller
is
now
tied
to
a
cluster,
but
the
control
implementation
in
control
around
time
is
like
very,
very
small.
The
reconciler
is
the
same.
It's
just
reference
like
multiple
times,
but
the
controller
and
the
watch
is
inside
a
controller.
H
They
will
be
basically
like
tied
to
a
cluster,
as
at
least
for
now
like
it
was
like
a
basically
like
the
simplest
way
to
ensure
cancellation
and
like
spinning
up
and
down
of
new
clusters,
so
the
in
the
future.
We
might
reuse
that.
F
And
we
can
talk
about
offline.
H
B
H
Will
not
actually
see
the
like.
This
reconcile
is
running
somewhere
else,
and
this
is
on
purpose
because,
like
we,
this
was
the
quickest
way
to
get
there
with
by
creating
new
controllers,
but
ultimately
like.
We
want
to
create
new
watches
in
a
controller.
But
that
is
like
a
larger
than
taking
like.
H
Which
might
be
a
good
embed
thing
right,
because
then
you
have
at
least
one
protein
per
cluster,
depending
on
like
how
many
you
want
to
spend
up
well,
but
then
that
also
Insurance
fairness
technically
so
versus,
like
sharing
it
like
it
will
just
be
like
Whoever.
Has
the
most
notifications
will
win?
A
E
Yeah,
okay,
so
yeah
I
mean
I
totally
expected
the
reconciler
to
be
shared.
That's
just
functional
I
was
surprised
at
a
controller
per,
but
it
sounds
like
you're
planning
on
getting
to
one
common
controller
right.
The
concern
is,
you
know
when
I
use
controller
runtime
I
I
write
my
main
in
a
way
that
explicitly
deals
with
the
One
controller,
so
I
I
have
to
write
my
main
differently
in
order
to
use
this
as
it
is
today,
right.
H
H
And
also
like,
we
want
to
make
sure
that
if
there
will
be
breaking
changes
there,
so,
for
example,
there
is
something
about
there's
an
ongoing
discussion
to
this
like.
Should
the
manager
have
the
like
a
cluster
associated
with
it
like
by
default,
because,
right
now
you
can
just
do
manager,
get
get
client
and
that
that
is
confusing
right.
It's
like,
because
what
client
am
I
getting
like
in
a
multi-cluster
scenario,
and
this
is
the
client
that's
associated,
which
is
test,
and
in
this
case,
but
for
new
users.
E
Course,
I
I
would
expect
to
have
to
opt
in
particularly
in
the
sort
of
architecture
where
I
have
to
say.
You
know
which
adapter
I
want
to
use.
So
clearly,
there's
nothing's
going
to
get
slipped
in
unawares,
but
I
would
expect
or
want
I
presume.
You
know
fairly
minimal
modifications,
but
yeah
I
mean
the
I
guess.
The
problem
is
the
existing
controller.
Runtime
is
kind
of
shot
through
with
the
Assumption
of
a
single
controller,
so
I
would
have
expected.
E
In
fact,
the
more
likely
approach
would
be
to
make
a
client
that
is,
you
might
call
multi-cluster
aware
right.
The
the
user
of
the
client
interface
I
can
get
one
client
object
and
when
I
make
a
request
on
it,
it
gets
steered
to
the
appropriate
cluster.
I
I
I
just
have
a
lot
of
question
to
that
one:
cash
shadow
Cross.
So
it's
a
different
curve
cells.
We
have
a
cashback
custom.
H
A
A
A
Cool
next
one
is
mine
and
I
will
try
to
do
it
very
fast,
because
it's
a
very
short
show
and
tell
so
I've
been
I've
been
playing
around
with
kcp
for
quite
a
while
now
and
in
particular,
I
was
working
on
the
opinionated
way
to
manage
self-managed
kcp
and
what
I'm
showing
now
is
deployed
in
the
public,
and
you
can
anybody
can
basically
poke
around
so
name.
Development
name
is
as
far
as
so
basically
what
it
allows-
and
you
have
to
trust
me
now
on
this
one.
A
A
A
So
it's
basically
Provisions
me
a
cluster
ads
configures,
my
coupon
today,
I'm
back
by
GDP
token
and
anything
I
interact
from
here.
I
can
do
it
as
a
normal
ACP
cooling,
tube
I
can
try
going
into
routes
and
to
three
dashes,
but
because
it's
basically
found
to
my
user
I
can
do
it.
I
only
have
access
to
the
ones
I
provision,
the
API.
A
In
the
background,
this
end
UPS
as
I
mentioned
in
there
in
the
common
tree
of
the
organizations.
So
I
can
basically,
as
a
user
I,
can
jump
in
provision
myself.
An
organization
in
a
future
potentially
add
other
users
to
that
organization,
and
either
we
can
use
my
workspace
as
I
shared
with
them
or
I,
basically
allow
them
to
provision
themselves,
it's
more
as
a
opiniated
way
how
you
can
use
kcp
to
create
something
here
in
this
case
it's
a
virtual
playground.
You
can
use
for
your
iot
controller
development,
etc,
etc.
A
K
Yes,
I
just
wanted
to
to
mention
that,
and
there
was
an
addition
on
the
TMC
side
that
was
just
matched
some
days
ago,
mainly
until
now,
since
only
deployments
were,
you
know,
sort
of
supported
at
least
their
semantics,
the
the
transformation
that
we
do
so
that
pods
created
Downstream
are,
you
know
pointing
back
to
kcp
and
also
the
transformation
that
we
do
on
the
Pod
specs,
so
that
pods
will
be
automatically
labeled
to
be
upsynced
back
to
kcp
as
well.
K
This
was
only
applied
on
deployments,
and
so,
if
someone
wanted
to
use
stateful
sets,
even
if
it's
not
completely
supported
semantic
wise,
you
know
typically
to
to
deploy
the
standard
workloads
using
a
postgres
database.
For
example,
it
was
just
not
working
so
I
mainly
added
a
matched
up
here.
That
applies
the
same
Transformations
on
all
the
prospect.
Cable,
if
we
can
say
so,
resources,
so
replicasets
and
stateful
sets
as
well.
K
I
didn't
do
anything
for
demon
sets,
because,
obviously,
since
it's
related
to
nodes
and
two
running
posts
that
are
related
to
to
nodes
things,
are,
you
know
a
bit
more
tricky
here,
but
at
least
that
should
make
anyone
able
to
start
using
and
playing
at
least
with
stateful
sets
or
replica
sets
in
the
future.
A
A
A
F
I
know
that
there
were
some
back
and
forth
in
slack
and
maybe
on
the
issue
in
this
one
around
some
of
the
meanings
of
I,
think
if
a
claim
was
accepted
or
not,
and
if
multiple
claims
were
accepted
or
not
so
I
think
we
can
put
this
in
the
backlog,
and
we
just
want
to
make
sure
that
we
take
some
time
to
review
the
proposal
and
provide
some
feedback
before
anybody
starts
working
on
it.
F
Up
I
mean
I'd
at
least
like
some
time
to
dedicate
to
thinking
through
the
proposal.
I
I
really
haven't
yet
so
you
can
put
the
status
assuming
you
have
permission
to
stick
it
in
backlog.
A
F
F
G
F
A
F
There's
a
little
bit,
that's
minimally
tricky
for
making
sure
that
the
entire
repo
compiles
correctly
with
the
right
relative
links,
since
that's
it,
because
package
client
is
now
going
to
become
its
own
go
module.
But
it's
the
same
thing
that
we
do
with
package
apis.
So
it
there
is
prior
art.
G
K
Teacher,
so
this
one
is
mainly
related
to
the
fact
that,
for
now
we
use
the
we
use
a
fixed
hard-coded
list
of
resources
for
adinking,
mainly
pods
and
endpoints,
and
even
we,
we
forbid
pods
to
be
synced
from
Upstream
to
Downstream
and
obviously
in
the
future.
We
would
at
least
need
to
have
a
way
to
customize
that,
and
you
know,
Define
a
sort
of
white
list
or
Blacklist.
We
have
to
discuss
that
of
resources
that
can
be
absinct
and
or
or
also
should
not
be
synced,
and
so
this
this.
K
Well,
fun
I
created
the
issue
first,
to
track
it
and
to
to
host
any
discussion
about
use
cases
we
already
have
in
mind
at
least
yes,
I
already
have
in
mind
some
ways
to
fix
this
and
to
answer
one
of
the
discussion
participants,
but
obviously
there
might
be
some
other
use
cases
as
the
one
Paulo
mentioned
which
might
drive.
You
know
the
design
in
in
other
directions
as
well.
So
I
mean
it's
it's
for
now.
It's
opened
the.
A
K
A
K
A
B
Yep
I
think
it's
mostly
about
kcp
making
sure
to
check
if
there
is
another
thing
to
get
before
draining
the
current
one,
because
if
there
are
workloads
running
on
the
same
Target
and
if
there
is
no
other
thing
Target
in
the
location,
I
think
the
workload
just
disappears.
We
don't
know
where
the
workload
is
so.
K
Yeah,
that's
mainly
about
I,
mean
deciding
whether
we
want
to
cancel
the
train.
If
there's
no
other,
you
know
place
to
to
to
send
yeah.
G
K
A
J
I
K
K
Not
true
for
now
I
assume,
we
would
have
to
think
about
adding
that
into
a
dedicated
epic.
Once
we
create
one.
A
F
F
Yeah
there's
enough
detail,
I
mean.
Basically,
we
may
want
to
have
a
more
of
a
meta
discussion
in
another
session
or
async
about
how
we're
using
the
GitHub
projects,
because
basically,
what
we're
doing
is
we're
reviewing
everything
that
we
haven't
reviewed
before
and
shoving
it
so
that
we
don't
have
to
review
it
again
and
eventually,
we'll
get
around
to
to
noticing
that
we,
you
know,
have
these
things
that
We've
triaged
once
and
maybe
we
should
fix
them.
So
we
may
want
to
refine
our
process.
F
This
one
I
think
is
one
that
we
probably
should
fix
sooner
rather
than
later,
so
I
would
be
inclined
to
put
it
in
next,
but
again
we're
not
really
using
the
projects.
The
way
that
we
probably
should
be.
A
C
Serious
I
would
like
to
look
into
that
just
to
learn
a
little
bit
more
I
like
the
session
ND
that
we
did
with
the
low
level
at
CD
debugging.
So
if,
if
there's
something
in
that
area
or
maybe
a
little
bit
of
higher
level,
I
think
this
is
a
nice
opportunity
to
learn
a
little
bit
more
about
the
internals
yeah.
F
This
one
probably
involves
our
HTTP
Handler
chain
and
filters
to
some
degree.
C
K
Yeah
this
one
probably
yeah,
okay,
not
sure
it's
it's
really
something
critical,
but
we
have
to
look
into
it.
It
was
assigned
to
Gian,
in
fact,
it's
probably
related
to
to
panning
work
already
in
an
issue
about,
in
any
case,
refactoring
the
option.
You
know
the
options
to
import
resources
from
the
physical
cluster.
C
Searches
I
believe
this
is
in
the
context
of
virtual
workspaces.
Isn't
it?
Yes,
there
is
a
yes
yes,
so
we
had
a
discussion
sounds
like
regarding
this,
and
this
is
not
a
bug.
I
mean
there
is
simply.
You
know
well
as
part
of
reverse
permission
claims.
What
I
did
is
to
build
an
authorizer
in
the
virtual
workspace
authorizer
chain
that
asserts
permission,
claims
the
non-reverse
part,
and
the
current
status
is
that
the
authorizer
that
I
backwarded
from
that
one
was
not
sufficient
to
solve
this
issue.
That's
the
tldr.
J
Next
double
check
about
admission:
did
we
say
the
stock,
which
already
has
information?
Maybe
it's
even
simpler,
I
think
it's
like.
We
discussed
that
quickly.
C
F
Yeah
I'll
add
that
comment
after
cool.
Basically,
what
happens
is
the
server
gets
a
panic.
F
Just
from
past
experience
and
other
projects,
where
we've
had
issues
that
are
add
metrics,
they
stay
open
forever,
because
nobody
knows
what
metrics
to
put.
K
In
yes,
so
so,
obviously
we
should
I
mean
there.
Just
should
be
more
description
here,
but
the
point
is
to
enable
metrics
in
the
Sinker
first
and
in
the
Sinclair.
You
have
a
number
of
controllers.
You
know
stack
Synchro
statue,
Sinker,
app
syncer
and
just
to
have
the
basic
normal
metrics
of
controllers
like
work,
your
size
and
new
stuff.
I
think
it
should
be
a
good
start.
K
K
Sure
I
mean,
and
as
soon
as
we
have
set
up
the
the
basic
Matrix
access,
you
know
endpoint
and
basic
Matrix
of
of
the
controllers.
Then
obviously
we
have
all
the
freedom
to
add.
You
know
more
feature
related
metrics
into
this.