►
From YouTube: Kubernetes kops office hours 20181109
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
We
can
proceed
with
the
agendas.
We
have
a
bunch
of
things
on
the
agenda,
but
I
would
say:
I
am
fans
of
some
of
the
companies
I
see
people
representing.
So
thank
you
for
everything
you
guys
do
and
all
right,
so
first
up
Ryan
an
update
on
the
111
beta,
1
12
alpha,
which
is
very
related
to
the
thing
I
added
immediately
on
the
agenda
next
up,
which
is
sincere,
apologies
for
the
delays
here.
This
is
entirely
my
fault.
A
A
Yes,
we've
also
had
an
issue
in
the
last
week
where
we
lost
quota
in
AWS,
so
dunno.
If
you
noticed
like
they,
see
all
the
kubernetes
PRS
in
order,
all
of
kubernetes
are
failing
because
of
cups.
Aws
failing
it's
not
it's
not
a
cops
issue
directly,
but
it
is
rather
a
quota
issue,
so
that
is
that
is
also
hops
up,
but
I
am
intending
to
make
up
lost
time.
A
The
yes,
once
we
cut
111
beta
I
think
we
can
cut
112
alpha
I
would
say
there
are
still
issues
in
kubernetes
111
which
are
being
found,
so
maybe
we're
not
too
far
behind
I
mean
we're
obviously
very
far
behind,
but
maybe
maybe
don't
run
it
in
fraud
just
yet
yeah
I'm,
just
thinking
of
particular
is
one
involving
node
addresses
going
missing.
There
is
a
PR
or
a
back
port
out
for
that.
A
So
if
you,
if
you
see
that
that
is
a
known
issue,
I,
don't
know
how
people
feel
about
that,
whether
we,
whether
people,
would
have
preferred
like
a
second
alpha
just
with
where
we
were
right
now
or
whether
people,
how
people
feel
hey.
A
Will
I
will
I
will
do
that
I
think
what's
next
on
the
agenda,
maybe
someone
can
talk
while
I
go.
Oh
really.
B
A
I
will
I
will
try
to
find
it
if
we
have
a
moment
and
we
can
talk
later.
Okay,
sorry,
thanks,
that's
good
mysticum
I
should
I
should
have
had
that
time.
I'm,
sorry,
okay,
but
yeah.
The
I
don't
know
how
people
feel
whether
we
would
have
preferred
another
alpha
with
just
whatever
we
had
got
into
date
or
whether
we
like
they
did
more
like
a
beta
being
more
ready,
but
yeah
I
know
we
I
know
we
are
behind
so
I
apologize.
C
A
That's
fair
and
we
can
try
to
it.
Maybe
we
could
try
to
do
a
scheduled
alpha
thing
make
a
schedule.
Alpha
works
for
me,
yeah
like
because
I
mean
alphas
are
not
necessarily
particularly
high
quality,
or
they
don't
have
the
same
quality
bar,
and
so
we
can
just
do
like
it
seems
like
we
could
do
alphas
on
a
timeline
right.
A
Next,
up
on
the
agenda,
Aaron
asked:
will
EDD
manager
be
recommended
for
production
deployments
in
cops
111
and
what
there's
anything
you
want
to
add?
There
I
think
that
the
question
is
reasonably
self-contained.
If
not,
the
answer
is
likely
not.
There
are
two,
so
the
good
news,
there's
great
news
here,
which
is
that
we
have
a
sub-project,
4
s,
CD
manager,
it's
going
to
merge
with
a
project
called
s,
CD
ADM.
A
That
project
has
been
approved
by
successor
life
cycle,
so
that
will
happen
and
we'll
be
spinning
that
up
we're
not
gonna
block
on
that
happening
on
the
sort
of
the
you
know
that
that
project
like
integrating
and
getting
it
all
spun
up,
but
that
that
will
be
so.
This
will
no
longer
be
a
cop,
specific
component
or
a
weird
like
and
ciliary
component.
It
will
be
a
it
is
a
sick,
bastard
lifecycle.
They
will
be
a
successor,
lifecycle
component,
which
is
which
is
great
new.
It's
a
signal
icicle
sub
project,
which
is
greatness.
A
The
we
do
have
two
issues,
at
least
with
three
issues
that
I
am
aware
of
which
block
it
becoming
recommended
for
production
deployments
in
cups.
111
the
working
backwards,
I
guess
TLS
certificates
for
there's
just
no
support
for
yet
I,
don't
think
that's
terribly
hard
to
add,
but
just
need
to
do
that.
There
is
an
issue
with
rolling
updates
in
h8
clusters,
so
the
issue
in
particular,
is
when
you
first
switch
I
think
from
EDD
from
protic.
You
bet
cities
or
the
legacy
HCV
management
to
ed
city
manager
in
an
HJ
cluster.
A
It
won't
or
the
odds
are
against
you
like
being
able
to
to
do
a
rolling
update
without
forcing
it.
We
were
basically
have
to
go
on
and
kick
some
of
the
ICD
manager
processes
to
have
that
work.
I
believe
it's
only
on
first
switch,
so
it
might
be
that
we
get
away
with
a
release.
Note
there
I,
don't
know
how
people
feel
about
that.
I
mean
I
need
to
like
figure
out
the
exact
scenario.
Is
there
the
the
more
complicated
issue
is
Calico
currently
talks
directly
to
ed
CD.
A
A
If,
if
categories
is
CRTs,
it
wouldn't
have
to
talk
to
Etsy
directly,
it
would
talk
to
the
API
server
in
the
API
server
talks
dead
city.
So
that's
that's.
A
better
model
for
us,
but
I
have
I
did
not
find
when
I
went
so
I
looked
at
this
problem
and
I
did
not
find
an
upgrade
path
from
where
we
are
today
to
that
I.
A
A
One
of
the
options
we
could
do
is
put
up
that
as
a
proposed
path
and
then
let
calico
be
motivated,
hopefully
to
to
provide
a
a
more
like
keep
you
on
calico
path,
but
honest
honestly
how
you've
looked
at
the
docks
and
there
were
no
that
didn't
seem
to
migration
path
from,
I
guess,
calico
to
with
etsy
D
to
calico
three.
We
see
our
DS.
D
A
A
Don't
know
whether,
like
yeah
psyllium,
is
a
little
odd
and
I,
don't
think
it's
it
isn't
using
the
it
isn't
using
NCD
for
the
same
purposes:
calico
it's
using
I
think
to
keep
track
of
like
their
their
mappings
of
like
and
what
they
call
it
but
identities
to
like
permissions.
So
it's
it'll
be
great
if
it
was
in
CRTs
but
I
think
that's
on
their
roadmap,
but
yeah
I
would
certainly
be
aware
that
it
could
be
tricky
in
future.
D
A
Absolutely
so
it's
it's
yeah
and
yes,
you
do
and
then
like.
If
psyllium
is
talking
to
it,
we
have
to
provide
the
those
certificates
to
psyllium,
and
we
have
to
you
know
at
some
stage
of
city
manager
is
gonna
get
better
at
or
but
call
me
I'd
see
the
ADM
now
so
I'll
use
its
new
name.
Etsy
da-dum
is
gonna
support
things
like
certificate
rotation
and
all
those
sorts
of
things
and
we're
gonna
have
to
integrate
clients
into
that
which
is
gonna,
be
less
than
fun.
So
yeah,
it's
it's
not
ideal.
A
One
of
the
interesting
options
might
be
to
run
a
the
sed
operator
or
an
EDD
operator
as
a
backing
store
for
psyllium.
If,
if
it
isn't
like,
if
we
can,
if
the
bootstrapping
still
works
as
it
were,
so
if
you
can
still
bring
up
psyllium
without
like
without
needing
silly
I,
miss
it
work
or
without
me
to
get
CD.
If
you
can
bootstrap
it,
then
everything
bootstrap,
the
underlying
sed.
A
Without
needing
psyllium
to
be
running
talking
to
Ed
CD,
then
you
could
run
a
psyllium
specific
at
CD,
which
which
would
be
separate
from
the
coven,
is
that
city?
Okay,
awesome.
Thank
you.
We
should
probably
talk
to
silly
on
that
yeah.
We
should
like
ass
silly
at
what
they
would
recommend,
because
it
is
it's
definitely
something
that
they
I'm
sure
they're,
aware
of
the
problem
and
you'll
be
good
to
chat
with
them
about
their
plans
and
how
they
would
move
customers
that
are
running
today.
I
just.
A
Yes,
I
would
definitely
I
think
I.
Think
I
serve
my
personally
to
have
another
look
at
the
ICG
operator
because
of
some
of
the
edge
cases
around
it.
We
can't
use
it
for
the
SED
ATM
use
case
because
it
relies
on
kubernetes.
So,
like
there's
a
circular
dependency
there,
the
last
time
I
looked
there
were
some
odd
edge
cases
around
like.
If
you
lose
quorum
it
would
delete
the
data,
so
I
hope
that's
been
fixed.
A
A
Next
on
the
agenda,
I
don't
have
anyone
else
wants
to
bring
up
anything,
but
next
on
the
agenda,
do
the
COPI,
Oh
Jesse
a.m.
eyes
have
a
changelog.
The
answer
is:
no,
they
don't
directly,
but
the
the
machinery
to
build
them
is
open-source
and
it
is
supposed
to
be,
or
it
is
turnkey
it's
turnkey
if
you
have
everything
working
so
in
other
words,
it
works
for
me
and
if
the
instructions
don't
work
for
you
then
like
please,
let
me
know
or
like
open
an
issue
there.
A
The
the
intention
is,
it
should
be
very
straightforward
to
build
your
own
images.
Should
you
choose
to
I?
Would
love
to
get
to
the
point
where
the
image
that
we
build
is
byte
for
byte,
identical
to
the
image
that
you
would
built.
We
are
a
long
way
away
from
that
just
generally
in
terms
of
the
state
of
the
world
like,
but
there's
some
great
stuff
happening
in
on
that
front.
But
yes,
so
today
the
best
option
is
to
well.
A
E
You
for
taking
my
question.
This
is
Daniel
there's.
This
is
just
chasing
a
kubernetes
issue
that
was
documented
on
the
actual
kubernetes
project.
Not
cops,
I'll,
probably
gather
a
little
bit
or
information
I,
don't
want
to
derail
the
meeting
and
chasing
down
an
issue,
but
well
we'll
try
to
gather
some
notes
and
maybe
get
in
the
next
office
meeting
with
those
ok.
A
E
So,
in
the
kubernetes
project
issue,
six,
four,
seven,
three,
six,
four,
seven,
four:
three
I'm!
Sorry:
it's
a
kernel
level,
integration
bug
with
kubernetes
and
we
are
actually
having
this
issue
in
our
production,
cops
environment
right
now,
and
so
we're
just
trying
to
see
if
upgrading
the
ami
would
help
that,
but
just
kind
of
a
shot
in
the
dark.
That's
why
we
are
attending
today,
I,
don't
it's
a
pretty
long
shot,
we'll.
A
See
there
is
a
so
you're
not
obligated
to
use
the
ami
right.
So
if
you
wanted
to,
you
could
try.
You
know
the
latest
a
bun
to
image
or
the
latest
red-hot
image
or
the
Debian
image,
if
any
or
the
latest
Amazon
system
Linux
image-
and
you
can
also,
if
you
create
a
second
instance
group
yeah,
you
are
able
to
like
specify
a
different
image
there.
A
So
you
could,
you
know,
have
one
or
two
nodes
since
your
firm
happens
there
and
you
can
use
tanks
and
or
you
can
use
label,
select
node
selectors
to
steer
workloads
or
contains
just
your
node
selectors
to
steer
workloads
to
it
and
taint
steer
worklets
away
from
it.
So
you
can
you
can
sort
of
introduce
it
that
way.
If
you,
if
you
would
like
to
try
yeah.
E
I
mean
our
issue:
is
it
only?
It
only
happens
in
our
production
environment,
our
test
environments,
don't
field
as
much
volume
or
traffic
and
like
basically,
we
can't
roll
our
kubernetes
clusters
with
the
new
version
of
X
application,
because
but
again
I
don't
want
to
derail
the
meeting
too
far,
but
we'll
probably
gather
more
information
and
maybe
join
the
next
office
hours
more
details
so
that
you
can
get
directly
to
the
issue.
That'll.
A
C
A
Great
thank
you
that
B,
that's
a
great
ring,
yeah
great
suggestion,
Mike
gr
Moni
asks
I
believe
there
was
a
mentioned
in
a
previous
office
hours
about
an
issue
with
fetching
protic.
You
too
often
can
we
get
a
recap.
Slash
reminder
on
that.
If
there's
anything
you
want
to
add
there,
no
no
just
I
I've
been
lurking
around
for
a
while
and
remember
hearing
a
conversation,
I
believe
it
was
poor.
Kid
was
being
fetched
and
we
were
hitting
some
limits
and
just
to
try
and
get
some
more
context
on
that.
A
F
G
A
Might
be
me
yes,
yes,
I've
been
looking
for
you.
Thank
you
so
much
for
coming
yeah.
Let
me
tell
you
the
so
this
is
there.
There
is
some
edge
case
where
I
guess,
a
small
subset
of
nodes
were
fetching
pretty
cute
very
often,
and
the
limit
we
were
hitting
was
financial.
It
was
costing
money
yep
thanks
for
AWS
like
helped
us
out
for
a
little
bit
long
term,
we're
gonna
move
to
so
they
gave
us
some
budget
to
like
ease
the
pain,
long
term
Thank,
You
dative
us.
A
For
that
long
term
we
are
going
to
get
our
artifacts
in
the
CNC.
F
is
like
gonna
host
our
artifacts,
and
it's
going
to
be
a
bigger
bucket
to
pay
for
these
whoopsies,
but
yeah.
We
added
some
like
back
office
to
later
versions
of
cups.
Yeah.
If
you
know
what
was
actually
the
underlying
issue,
that
would
be
super
helpful
yeah
for
sure.
So
what
I
can
do?
I
can
link
a
kurd
movie.
Ticket
I
have
open
it
the
hopefully
the
short
version
I
can
explain.
Is
we
had
clusters
with
live,
restore,
enabled
on
docker?
A
It
turns
out
that
live
restore.
If
you
restart
the
demon,
but
the
machine
is
still
alive,
docker
doesn't
restore
the
default
networks,
so
things
like
proto
cube
can't
start
because
they
want
the
host
network
and
there's
no
horse
network
on
the
box
and,
of
course,
we
restart
crota
cube
according
to
the
manifest
every
like
two
seconds
with
Apple
always
right.
So
it's
constantly
trying
to
grab
it,
and
it
is
slowly
every
two
seconds
that
launches
seven
Corp,
specific
containers,
the
can't
run,
and
then
it
does
it
again.
A
Two
seconds
later,
two
seconds
later
until
the
point
where
you
have
like
3,000
chrono,
cube
containers
sitting
on
this
machine,
and
so
we've
rolled
fleet-wide
with
ticking
live,
restore
off.
We
have
the
mobi
su
open,
I'll
link
it
in
the
agenda,
so
people
can
track
long
and
I'm
working
with
the
docker
folks
to
see
if
we
can
either
fix
that
or
for
future
versions.
Cuz.
It's
it's
a
neat
feature,
but
that's
a
horrible
bug
to
unfortunately
bring
to
the
table.
That
is,
that
is
wonderful
yeah.
A
If
you're
able
to
give
any
more
details
and
yeah
I
hadn't
thought
about
the
idea
that
pretty
cube
itself
would
be
restarting.
So
yeah
we're
yeah,
so
it
would
be
great
if
you
could
yeah
give
me
some
details
and
then
I
can
try
to
if
I
get
a
sort
of
speculative
like
fix,
but
yeah.
That
would
be
wonderful
to
get
more
deals
and
actually
the
real
fix
so
for
sure
cool.
Thank
you.
A
A
And
if
the
number
goes
way
down,
then
it
was
totally
odds.
Well,
thank
you
for
holding
up
I
appreciate
it.
Okay,
that's
great
news
that
made
my
day
alright.
Next
on
the
agenda,
Cooper
all
flying
cops
deployment.
I
think
we
talked
about
this
in
cigarettes.
Didn't
we
so
thank
you
for
coming
back
to
cop's
office
hours,
oh
yeah,.
F
Great
to
be
here,
I
was
I
just
wanted
to
know
if
there's
anymore,
if
you
had
any
time
to
think
about
it,
I
opened
up
an
issue
like
you
said
just
looking
for
any
kind
of
dependency
that
may
be
called
out.
So
after
we
talked
I
like
I
said:
I
got
all
the
terraform
scripts
pulled
everything
that
it
was
gonna
get
from
the
internet
put
it
in
my
little
nexus.
Repo
and
I'm,
deploying
all
of
this
from
the
gets
lab
CI.
So
I
made
a
custom.
F
Continuous
integration
pipeline
to
just
deploy
all
of
it
the
cops
and
then
deploy
all
of
my
custom
home
charts
of
stuff
on
top
of
it.
So
it's
just
building
on
a
cluster
and
then
putting
software
on
you
know
like
that,
makes
it
so
I
haven't
had
run
into
any
problems
yet,
but
I
haven't
nailed
to
fully
test
it.
I
just
want
to
know
like
I
was.
You
were
talking
about
the
network.
F
Plugins
I
switched
mine
from
the
default
to
calico
just
to
see
if
it
was
going
to
pull
something
different
from
the
internet,
but
instead
of
just
closed
a
big
archive
of
something
called
network.
Plugins
I,
don't
know
if
it
needs
anything
more
to
to
other
things,
and
do
you
recommend
a
different
overlay
network,
because
now
it's
the
time
for
me
anyway,.
A
So
I
think
the
I
think
that
Network
plugins
is
a
the
cni
Network
plugins,
which
is
a
binary
that
you
need
basically
for
any
of
the
non
built-in
network.
Plugins,
so
I
think
we
made
it.
So
if
you're
not
running
with
public
IPS,
you
basically
have
to
use
a
network,
plugin
I,
think
I,
don't
know.
If
anyone
here
is
currently
running
offline
or
there
was
some
functionality
that
actually
was
3/4
built
I
don't
have
any
ones
actually
using
that.
H
Elena
he's
the
only
thing
that
I
had
to
maybe
add
be
a
full
here.
It's
like
I,
using
a
tumulus
network
plugin
but
spot
in
cynics,
but
what
I
can't
say
is,
at
least
in
the
class
respect
I
can
specify
which
dark
I
want
to
use
for
the
CNI
for
that
killer,
networking
driver
so
I,
don't
know
if
you'll
have
that
option
with
Calico
I
haven't
tried
that,
but
you
could
specify
the
image
and
I
think
that,
with
respect
whichever
private
image
registry
few
needed
easily,
obviously
you're
gonna
want
to
test
and
investigate
I.
A
And
I
I
guess
we
got
so
close.
I,
obviously
have
been
a
little
I
think
the
next
one
eleven
has
to
take
priority,
but
I
would
love
to
actually
try
this
and
see
where
it
is
so.
I
will
dig
up
your
issue
Cooper
and
and
see
see
what
the
status
is,
because
it
would
be
fun
to
try
and
like
I.
Like
all
the
pieces
are
there
and
in
your
case,
does
Nexus
or
are
using
HTTP
does
Nexus
present
as
an
HTTP
proxy,
or
is
it
a
rewrite
of
image?
Url
image
names
I
just.
F
A
Yeah
I
know
I
know
someone
did
add
like
separately
from
the,
so
there
was
some
work
which
I
think
Chris
loved
and
someone
else
were
forget
was
working
on
to
basically
automate
what
you
were
working
on,
what
your
what
you
did,
but
they
don't
not,
although
through
easy
to
rewrite
artifact
locations
and
I
think
separately,
someone
else
implemented
the
ability
to
specify
HTTP
and
I
think
HTTP
proxies,
which
I
guess
might
like.
If
you
had,
if
you
had
a
proxy
that,
like
pulled
from
s3
or
whatever,
whatever
your
backing
store
was,
maybe
that
would
work.
A
H
Hi,
so
I
just
wanted
to
give
a
quick
update
so
there's
a
massive
PR
that
was
landed
recently
in
the
111
alpha
ads
finance
as
a
cloud
provider
in
cops,
and
so
essentially
so
spot
inst
is
a
company.
They
use
they're,
basically,
an
abstraction
for
spot
instances
on
AWS
or
I.
Forget
the
the
terminology,
but
like
just
non
reserved
instances
in
which
every
provider
you
and
so
are
non
on-demand
at
any
rate,
so
I'm
very
familiar
with,
at
least
for
the
AWS
side
of
things.
H
I
just
wanted
to
say
that
that
floor
request
has
landed
and
I'm
testing
it.
There
are
few
bugs
that
I
personally
have
brought
up
to
layer
in
the
original
author
that
purpose
the
ad
spotted
and
the
things
that
I've
concerned
myself.
What
they're
cared
about
have
been
addressed
by
leering
in
our
quests
that
have
been
open
against
cops,
ought
to
address
some
issues.
I
think
you
commented
on
at
least
one
is
ingesting.
Toys
I
saw
that
earlier
today.
So
that's
moving
right
along
I'm,
using
like
some
special
binary
that
I've
gotten
from
leering
directly.
D
H
I,
wouldn't
want
to
recommend
folks
do
too
much
like
anything
like
production
worthy
but
they're,
like
that's
still
in
alpha
feature
and
still
gated
with
a
future
fine,
but
I'm
testing
it
for
my
own
purposes.
We
as
a
company
beeswax,
we
work
very
closely
with
Escada.
It's
actually
doing
a
meet-up
next
week,
if
you're
in
New
York
there's
a
meet-up,
DevOps
and
drinks.
I'm
gonna
talk
a
little
bit
about
spot,
it's
in
beeswax
in
earth,
but
I
would
like
to
say
that
the
the
the
integration
works
it
works
well.
H
I
just
have
like
questions
I
think
mostly
can
I'm
new
to
cops
generally
also
I'm,
using
this
new
abstraction.
So
there's
like
a
few
things,
I'm
trying
to
figure
out
but
like
it
works,
it's
not
the
feature.
So
I
guess
that's
like
when
I
want
to
bring
up
and
point
to
I
had
a
question
about
I'm
trying
to
get
I'm
trying
to
like
make
cluster
upgrades.
Like
updates
and
upgrades
like
a
core
competency
like
something
we
do
routinely.
Maybe
we
have
like
summation
around
the
thing.
H
I'm
running
the
problem
problem
I'm
running
it
right
now,
I'm
trying
to
determine
so
whenever
I
just
do
a
dry
run
of
cops
update
cluster
and
you
know
don't
specify
yes
and
just
kind
of
want
to
see
like
what
has
changed.
My
cluster
always
needs
changes.
Some
of
that
was
because
of
like
SWAT
ants
in
the
spawn
with
sensory
integration
and
leerin
has
addressed
some
of
that.
H
But
I'm
trying
to
figure
out
like
the
problem
I'm
having
is
just
like
the
output,
is
very
verbose
and
I'm,
also
getting
like
a
lot
of
extra
information,
because
we
as
a
company
use
spot
inst
for
other
purposes
and
so
I'm,
getting
a
lot
of
extra
information
from
spa
tensed
about
those
extra
resources.
So
my
like
output
from
plus
update
cluster
is
a
bit
of
a
mess.
H
A
H
A
But,
as
you
say
there,
if
you're
running
the
integration
branch
or
their
their
integration
run
seems
unlikely
to
be
all
of
them.
I
don't
know
how.
If
we
can
I
get
you
if
you're
comfortable
that
you
can,
you
can
email
them
to
me
privately
or
put
them
on
back,
and
we
can
try
to
figure
out.
What's
going
on
well.
H
The
the
problem
in
having
is
like
I
guess
it's
like
if
I
I
should
be
use
a
different
account
like
as
finance
the
account
that
doesn't
have
other
respondents
has
elastic
groups,
their
equivalent
of
auto-scaling
groups
and
so
I'm.
My
logs
are
picking
up
all
of
the
auto
scan
groups
associated
with
our
account
sooner.
A
That's
actually
that's
that's
really
great
news,
because
that's
a
relatively
easy
fix
and
I
think
it's
the
same
fix
as
like
the
peer
they
looked
at
just
like
just
before
we
started.
So
if
they're,
the
way
that
cops
does
it
is
we
use
tags,
so
you
can
run
like
two
cops
clusters
in
a
single.
A
If
you
see
in
pure
a
negative,
eight
of
us
cops
I'm
sure
like
come
up
with
the
words
yeah,
but
and
and
they
don't
interfere
with
each
other
because
cuffs
filters
by
the
cluster
but
a
tag
and
the
tag
our
sponsor
the
cluster,
and
it
sounds
like
that-
isn't
working
force
but
inst,
and
we
that
is
a
relatively
easy
fix,
because
well,
if
spottin
supports
tags
on
elastic
groups,
then
it's
a
relatively
easy
fix.
Otherwise
I
think
we'll
need
to
come
up
with
some
some
way
to
partition
off
the
elastic
groups.
H
C
As
much,
it
almost
sounds
like
those
tags
like
when,
when
we
check
to
see
the
current
state
of
things,
those
tags
aren't
even
being
checked
at
all,
even
if
they
like,
because
I
feel
like
in
the
normal
condition,
they
would
be
checked
and
you
should
actually
get
zero
current
state
or
something
like
that.
So
there's
there's
like
yes,
it
sounds
to
me
just
like
dozens
that,
like
tag
related,
you
might
even
be
able
to
test
some
of
this
by
adding
a
tag
on
to
it.
H
A
This
problem,
like
happens
on
like
the
same
problem,
would
happen
on
AWS,
like
you
can
have
other
auto
scaling
groups
in
your
V
PC
cops
or
not,
and
we
don't
want
to
pick
them
up
in
cops.
We
want
to,
like
you
know,
ignore
the
things
that
cops
is
not
managing
so
but
yeah.
It's
it's
certainly
easy
to
get
wrong
and
yeah
I
think
yeah
yeah.
A
Definitely
definitely
yeah
I'm
glad
that
you're
running
without
yes,
because
who
knows
what
will
happen
if
it
takes
up
all
your
order
like
last
groups,
right,
I,
think,
awesome
and
yeah,
it's
great
to
see
and
I
will
I
will
catch
up
on
some
of
those
reviews,
but
yeah?
It
sounds
like
that
makes
sense,
given
the
review
I
did
this
morning
so
that
that's
great
Ryan
I
think
I
owe
you
some
reviews
on
cube
deploy.
It
looks
like
it's
that
true
I'm
bottom
now
sorry
do
I
owe
you
some
reviews
on
cube,
deploy.
G
A
A
A
G
A
G
A
Yeah,
it's
subtle,
so
it's
like
a
it's
a
race
condition
as
I
as
I
understand
it's
a
race
condition.
So
if
ec2
was
slow
to
respond,
I
guess
to
describe
instances
requests
it
would
maybe-
or
here
note
addresses
and
then
that
would
manifest
itself
as
you
not
being
able
to
exec
or
logs
or
proxy
to
the
pod
until
the
next
cycle.
So
it's
not
disaster,
but
we're
still
finding
stuff
in
111
was
I.
Think
the
message.
A
E
Justin
I
was
wondering
if
I
can
steal
a
little
bit
of
your
time
to
get
some
ammo
here,
so
we
currently
have
some
grumblings
within
our
team
management
to
potentially
that
switching
to
eks,
instead
of
going
on
cops-
and
I
was
wondering
if
you
could
help
provide,
you
know
a
good
reason
why
we
should
stick
with
cops.
You
know
cuz,
like
I'm,
trying
to
avoid
having
to
do
a
migration
and
getting
a
lot
of
heat
from
like
within
my
management
team
in
a
cold
world.
E
H
I'm
in
really
quickly
I,
don't
know
if
Justin
can
say
as
much
but
as
a
as
a
user
of
cops
and
as
these
are
things
on
services,
I'm,
not
just
an
e
KS
but
other
services.
I
can
say
that
one
reason
to
go
with
cops
great
living.
It
obviously
I,
don't
know
how
large
your
cluster
will
get,
but
potentially
using
ETS
and
having
a
I'm,
not
sure
it
there's
no
train
trail.
Transparency
and
innovation
is
like
great
limiting
for
different
API
calls
and
I'm,
not
sure
what
API
calls
here.
H
You
won't
need
to
make
any
on
your
eks
cluster,
but
if
you
I
I
have
experienced
in
my
six
months
is
working
at
a
startup
meeting
great
limited
on
some
different.
You
know
VF
services,
but
our
use
case
is
very
different
in
country
meet
up
in
DevOps
and
drinks.
Next,
Tuesday
I
check
out
whatever
you
kisses,
but
sales
pitches
side,
I'd,
say,
rate,
limiting
and
AWS
is
a
problem,
but
you
could
probably
some
of
that
is
mitigated
in
Cox.
If
you
have
your
own,
it's
like
HTTP
and
the
API
server.
H
C
Of
scale
you're
operating
in
open
to
we
have
looked
at
this
now
at
two
companies,
I'm
at
my
second
company
and
and
David
behind
me
recently
did
a
deep
dive
on
this
for
the
same
problem.
So
so
what
it
came
down
to
for
us
is
really
the
Masters.
The
control
plane
that
you
know
AWS
does
a
fantastic
job
of
managing
that
for
you.
Sometimes
you
don't
want
that
completely
managed,
like
you
know,
so
our
biggest
problem
is,
we
want
to
run
K
I.
Am
that
lets
you
manage
per
pod.
C
I
am
permissions
and
the
way
you
generally
deploy
that,
as
you
put
the
master,
the
server's
that
that
manage
you
know
the
actual
access.
You
run
that
on
your
master
notes
and
then
all
request
or
proxy
through
the
Masters
and
the
Masters.
Have
the
delegation
permissions
on
I?
Am
that
way
every
node
doesn't
need
that
permission
right.
You
can't
really
run
it
with
that
model.
If
you
don't
have
no
access
to
the
control
point.
So,
like
you
know,
in
accomplish
cluster
you'll
do
cube
CTL
get
pots.
C
I
Nothing
on
the
quote-unquote
master,
the
control
plane,
another
issues
as
of
right
now,
there's
a
way
to
actually
configure
the
control
plane.
So
if
you
want
to
say
we
were
doing
onec
stuff,
you
can
do
that
with
eks
I.
Don't
know
if
they're
gonna
have
the
ability
to
configure
that
kind
of
stuff
and
some
sort
of
dashboard
or
with
AWS
API
you
create,
they
did
as
of
right
now
and.