►
From YouTube: 2018-06-05 Rook Community Meeting
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
So
I
think
we've
had
some
conversations
recently
about.
You
know
what
is
essential
for
the
milestone
to
be
completed
and
I.
Think
we've
made
some
good
progress
recently
on
some
of
those
issues.
I
think
that's
probably
about
time
now
to
clear
off
the
like
two
halves
on
the
project
board
that
we're
looking
at
now,
because
we
need
to
converge
and
get
0.8
release
out
sooner
than
later,
so
I
want
to
start
making.
C
D
D
That's
two
issues
here:
why
that's
because
of
the
multi
multi
backends,
API
change,
let's
make
it
hard
to
upgrade
in
my
opinion,
make
is
upgrading
is
Nostrum
to
this
review.
The
second
is
that
OSD
right
now
use
the
device
map
device
that
device
config.
Not
so,
if
you
have
a
point,
seven
code
in
the
system-
and
you
want
to
upgrade
from
27
to
28,
the
API
change
is
one
one
of
those
things
and
then
the
device.
D
How
do
you
map
the
devices
from
the
OSD
existing
as
this
to
the
ones
that
you
are
using
the
devices
configured
as
be
a
significant
challenge?
The
third
thing
is
the
OSD
and
a
model
I
use
in
deployments
versus
replica
assets
in
point
seven.
So
all
these
things
connected
makes
me
believe
that
we
probably
will
have
a
break
in
changes
in
28,
so
how
we're
gonna
deal
with
upgrade,
or
should
we
support,
upgrade
at
all
I.
B
Wonder
if
we
can
find
a
way
to
either
support
the
or
not
seven
running
in
some
different
mode
like
those
that
were
using
daemon
sets,
maybe
they
could
continue
using
a
daemon
set
in
some
way,
but
I
know
there's
been
a
lot
of
change
here,
so
it's
gonna
be
a
challenge
to
support
one
way
or
another.
I,
unfortunately,
haven't
had
a
chance
to
really
think
about
that
upgrade
path.
Yet.
A
So,
just
to
be
clear
here
does
its:
are
you
guys
saying
it
sounds
like
what
you're
saying
is
that
there
are
some
upgrade
obstacles
that
are
beyond
the
scope
of
just
what
you
need
to
be
concerned
with
specifically
about
the
OSD
swimming
in
their
own
pods?
There's
other
things
beyond
that
scope,
yeah
so
with
with
the
upgrade
so
in
the
work
that
was
done
to
support
multiple
storage
providers
and
types.
You
know
we
updated
the
upgrade
guide
and
you
know
going
through
that
results
in
a
functional.
A
D
Certain
things:
well,
why
does
the
US
still
continue
to
use
the
API?
We
want
alpha
1
anyone
upgrade
from
there.
It's
going
to
be
some
challenge.
The
second
thing
is
the
switch
you
forgot
replica
statue,
team
assets,
I
haven't
figured
out
how
that's
upgraded
their
work,
the
and
then
how
because
the
device
is
being
used
by
the
existing
or
a
species
into
config
Maps,
that's
going
to
be
the
third,
so.
A
D
A
Talk
about
the
first
one,
real
quick
I
mean
the
you
know
the
about
types.
You
know
the
brook
that
iov
1
alpha
1.
So
we
have
you
know
automatic
conversion
code
migration
code
by
the
operator
that
you
know
converts
all
those
types
and
updates
since
the
new
ones.
So
is
there
there's
you're
saying
that
there's
still
an
issue
outstanding.
D
A
Exactly
so
it
like,
if
you
were
so
for
a
a
user
scenario
here,
it's
has
a
0.7
cluster
deployed
the
steps
in
the
upgrade
guide
to
you
know
to
delete
the
operator
and
let
it
recreate
itself
that
when
the
operator
comes
up,
it
will
automatically
convert
all
the
view.
1
alpha,
1
types
to
you
know
view
and
alpha
2
or
the
sefirot
wrote
that
io
types
you
know
recreating
the
CR
DS
and
doing
all
that
migration
automatically
and
I've
tested
that.
B
C
G
D
When
I
we
baste
about
to
adopt
the
master
code
I'm
using
the
the
Yamaha
was
using,
was
not
quite
working,
so
I
just
updated,
my
llamo
with
the
staff
API
and
instead
of
the
one
API
and
then
I
started
working.
So
my
impression
was
the
master
code.
If
I
just
use
them
to
do,
one
half
of
an
API
is
possible.
I
should
try
this
again
to
confirm.
D
So
if
that's
okay,
if
the
API
is
not
a
big
issue,
so
we
can
move
on
to
the
next
one
that
config
map
the
device
config
method-
that's
in
my
opinion,
couldn't
be
some
challenge
figure
out.
What's
the
devices
used
by
the
existing
OS,
these
Justin
question
actually
should
be
my
grace,
the
existing
part
or
just
should
we
just
let
them
run
like
they
are
right
now
and
then,
if
request
new
clusters,
we
use
the
new
schemes.
A
A
E
Woman
is
that
I
think
I
think
it
actually
would
be
okay
to
say
that
OSDs
that
all
run
in
a
single
pod
remain
that
way
and
I
I,
don't
see
why
that's
a
problem
I
think
over
time.
You
know
it
converges,
but
it
doesn't
mean
that
I
mean
as
long
as
as
long
as
people
upgrade
to
0.8
and
things
continue
to
work.
It's
fine.
E
D
A
The
the
same
purse-
and
the
same
could
be
said
about
you
know
what
do
I
mean
was
talking
about
with
month,
like
that
Mons
changing
from
replica
sets
to
deployments.
The
same
could
be
said
for
that
too.
If
they're,
you
know
state
if
the
real
goal
here
is
maintaining
a
functional
cluster
that
you
know
conserve
data
requests,
so
you
know
it,
it
doesn't
mess
the
important
part.
Isn't
it
so
that
everything
is
running
on
the
latest?
A
You
know
formats
of
the
components
to
keep
the
cluster
running,
but
as
long
as
it's
a
functional
cluster,
that's
what
really
I
am
concerned
about,
but
that
does
have
the
severe
caveats
that
you
know.
I
do
not
want
to
necessarily
get
it
have
a
bunch
of
code
paths
in
the
code
base
where
we're
having
to
deal
with
all
these
legacy.
You
know
types
and
in
components
and
carry
that
burden
for
you
know
forever.
A
B
E
C
C
The
discs
or
the
tourists
are
already
prepared,
so
I
haven't
seen
any
change
to
the
partition
layout
that
is
needed
for
photos
DS.
The
only
thing
that
we
right
now
have
this
issue
that
the
partitions
are
not
created
any
more.
The
DB
and
twelve
partitions
are
still
as
far
as
I
understand,
at
least
allowing
the
person
to
in
some
way
tell
tops
Alpharetta
to
start
a
new
deployment
photos,
the
other
job
to
prepare
those
T's
or
more
or
less
see.
C
The
proprietor
of
ten
sees
that
the
old
STIs
are
already
existent
and
then
just
adds
them
to
the
nose.
Just
could
also
be
done
many
of
these
because
they're
well,
we
don't
have
follow
many
upwards
it.
So
it's
totally.
Okay,
if
the
user
has
to
manually,
do
some
stuff
I
think
we
should
only
go
for
keeping
old
code
in
a
codebase
unless
there
is
no
no
way
to
do.
C
A
D
A
E
E
B
A
We
have
a
an
agenda
item
for
that
coming
up,
so
we'll
get
into
that
discussion.
It
just
is
that
yeah,
that's
definitely
an
issue
we
want
to
talk
about,
but
so
let's
try
to
wrap
up.
0.8
discussion,
though,
are
there
other
outstanding
tickets
on
this
board
here
included
in
the
0.8
milestone
that
are
of
at
risk
or
ones
that
we
are
concerned
about.
B
E
A
C
C
A
C
A
C
H
A
Okay,
that
sounds
good,
Tony,
okay.
So,
let's
move
on
to
the
next
agenda
item
today
and
the
Alex.
That
is
exactly
what
I
believe
you
wanted
to
talk
about.
So
in
general,
we've
had
some
reliability
concerns
with
our
jenkins
continuous
integration
infrastructure.
Some
of
those
have
been
sort
of
addressed
or
made
somewhat
more
reliable,
but
there
is
the
biggest
new
and
and
blocking
issue
that
we
have
right
now.
Is
that
the
GCE
instances
we
have
to
run
tests
in
to
end
integration?
A
So
we
need
to
figure
out
why
the
GCE
nodes
cannot
come
up
online
and
start
running
tests
and
Ilya
I
was
looking
at
this
yesterday
and
I
was
helping
him,
but
I
don't
know
if
we
what's
exactly
the
issue
is
or
what
you
know.
An
estimation
for
a
resolution
is
so
there's
not
I
think
this
is
a
high
priority
and
we're
gonna
continue
working
on
this,
but
it
don't
necessarily
have
a
resolution.
A
E
A
C
B
A
Yeah,
so
Ilya
I'll
try
to
find
some
time
with
earlier
this
morning.
So
you
continue
looking
into
this,
but
we're
getting
to
a
state
now
where
disabling
GC,
just
to
unblock
ourselves
and
move
forward,
does
make
sense.
C
C
E
C
A
Okay,
I
think
we
had
a
discussion
about
upgrade
and
migration
in
the
context
of
you
know,
though
SD
pod
changes
and
then
just
a
little
bit
more
on
that
Travis,
you
updated
the
pull
request,
template
to
have
a
checklist
item
about
making
sure
that
each
pull
request
has
thought
through
upgrade
and
migration
impacts.
Correct.
Yes,.
A
Cool
so
I
liked
it
in
general,
as
a
community,
we
are
being
more
focused
and
mindful
of
upgrade
impacts
to
our
user
base,
because
with
each
release
and
with
each
you
know,
with
the
growth
of
our
community
and
more
users
on
the
current
code
base,
you
know
we
have
a
fairly
fair
amount
of
people
kind
of,
depending
on
their
clusters,
continuing
to
run
across
releases.
So
as
a
community
I
think
that
you
know,
we've
done.
A
Not
I'm,
probably
not
yet
Toni,
because
the
upgrade
the
the
support
we
have
for
upgrades
does
have
manual
steps.
There's
not
full
automated
upgrade,
that's
been
done
and
you
know
in
the
operators
so
having
an
integration
test
about
it.
You
know
we
have
to
also
incorporate
some
of
those
manual
steps,
so
it
might
be
very
kind
of
difficult
at
this
point
so
far.
Okay,
but
I
like
that
idea,
though
absolutely
that.
A
C
B
A
Yeah
well
in
yes,
that's
a
good
point
Alex
and
that
it
will
also
be
impacted
by
adding
more
storage
providers
as
well.
You
know
like
cement,
you
know
integration
tests
or
cockroach
TB
integration
tests
that
will
only
serve
to
further
increase
that
build
time
so
parallelizing.
Some
of
those
efforts
where
they
can
be
you
know
done
that
way
is,
is
definitely
a
smart
idea.
A
H
E
B
Yeah,
there's
definitely
been
interest,
I
think
a
few
people
don't
like
that
to
get
or
whatever,
but
I
think
the
right.
Now
the
documentation
we
have
as
far
as
creating
an
object,
store
and
consuming
it.
We
mentioned
here,
go
to
the
tool
box
and
run
run
this
SEF
command
to
go,
create
a
user,
and
then
it's
so
the
you
know
just
from
a
basic
walkthrough
perspective.
We
know
we
need
a
user
and
creating
us,
the
creating
a
user
with
the
CRD
would
make
that
flow.
B
A
lot
better
I
think
it's
not
clear
what
the
design
looks
like
for
the
CRD,
like
you
know,
if
it
generates
a
secret,
how
do
you?
How
do
you
know
that?
How
do
you
get
that
secret
back
from
the
user
we
just
created
how
to
consume
it
anyway,
so
I
think
there's
some
design
questions,
but
right
now
the
documentation
is
pretty
manual
as
far
as
creating
an
object,
store
user
write,
something
that.
H
H
So
we'd
have
to
tie
in
that
user
to
every
single
type
of
storage,
once
the
buckets
guarantee
goes
in
like
Tigh
users
to
the
buckets
ie
just
to
the
file
system
and
SEF,
and
that
that
becomes
a
I
mean
a
burden
really
I
don't
know
it
is.
Was
that
the
intention,
or
was
it
just
to
automate
the
creation
of
users
and
then
they
would
go
in
and
manually
deal
with
the
permissions
issues?
H
B
H
B
F
H
B
F
B
H
F
A
Yeah
and
the
way
I've
looked
at
this
so
far
is,
is
in
a
fairly
simple
context
that
Tony
I
in
general
I
see
value
in
automation
of
you
know,
steps
there
where
people
have
to
go
and
copy/paste
things.
If
it's
just
that
the
nice
think
that's
still
a
win,
because
right
now
you
have
to
go
to
the
tool
box,
then
you
have
to
run
some
commands
and
then
copy
this
text
and
paste
it
somewhere
else,
and
you
know
like
find
the
right
thing
to
put
it
in,
and
all
that
and
simple
automation
around
that.
A
A
H
As
far
as
this,
these
abstractions
guys
so
I'm
working
on
a
dock
for
bucket
abstractions
that
we
can
use
that
are
tied
to
a
set,
a
mini
object,
storage
and
there.
So
right
now,
it's
it's
super
early
phase
and
it's
we're
unable
to
tie
the
actual
bucket
to
the
application
lifecycle
right.
The
same
way
we
do
with
volumes.
So
the
what
people
have
to
do
is
manually,
create
a
bucket
and
then
go
off
and
create
the
thing
right
and
then
there's
this
kind
of
janky
naming
scheme.
H
That's
so
it
will
create
the
credentials
and
like
a
binding
with
a
certain
naming
scheme
based
on
what
you
name
the
bucket
and
that's
how
you
tie
it
to
your
application
propagate
the
credentials
in
but
moving
forward.
That's
not
really
something
we
should
keep
doing
so.
Some
of
the
Fond
around
that
was
going
to
six
storage
and
see
if
we
can
kind
of
broaden
the
abstractions
from
volume.
So
it's
not
just
volume.
Now
it
could
be
bucket
or
something
else
and
actually
tie
it
to
a
pod
lifecycle.
H
F
H
C
C
What
you
know
Bob
from
hajikko
yeah,
it's
the
secret
storage
and
stuff
like
that,
leads
to
the
second
part
there
with
the
users.
Have
you
thought
about
kind
of
having
a
generic
user
CLU,
which
kind
of
like
role
based
access,
like
Auerbach,
uses
findings
to
find
to
certain
pockets
to
certain
storms
or
even
other
stuff?
H
I
mean
I
can
look
into
that,
because
it
is
right
now
it's,
it
seems,
really
fragile.
This
is
not
fragile,
but
like
limiting
this
design.
Everything
as
we've
done
certain
weighting
relying
on
a
naming
scheme,
but
if
we
were
able
to
I
guess
deal
with
this
like
bucket
mounting
in
quotes
right
via
just
like
the
secrets
that
are
generated
and
propagating
them
into
the
pods.
That
would
be
really
nice
anyway.
H
H
E
C
H
And
it
once
the
not
having
to
modify
a
pod
spec
to
accommodate
for
this
and
then
the
other
is
you
know
if
we're
able
to
modify
the
pot
spec,
what
the
world
would
look
like
so
yeah
and
users
really
complicates
this.
That's
why
I
didn't
know
what
the
priority
was
with
having
an
object,
store
user
or
things
like
that,
because
I
would
change
its
design
quite
a
lot
because
then
you
have
multiple
sets
of
credentials
that
are
being
created,
and
then
you
got
new
configurations
and
permissions
on
the
buckets
things
like
that.
So.
F
A
F
F
H
F
F
E
F
H
H
H
A
E
H
H
E
A
B
B
A
C
C
It's
the
four
finale:
it's
letting
well
I
can't
fall
out
the
bottom
edge
on
I.
I,
probably
don't
see
a
downside
with
it
between
having
when
the
user
creates
a
new
issue
having
a
selection
between
vistas
and
backrub
for
distance,
the
feature,
requests
or
something
regular
and
then
getting
a
different
template.
B
C
C
C
A
G
A
G
G
Like
seeing
what
happens
when
nodes
fail
or
making
sure
that
we
like
tests,
that
rook
is
stable
on
an
upgrade
from
one
kubernetes
version
to
another
and
I
I
guess
I'm
also
trying
to
see
like
does
everyone
use,
mini
cube
for
development?
Does
this
every
you
know?
Do
people
spin
up
small
kubernetes
clusters.
A
Yeah,
that's
a
fantastic
question
Blaine,
so
there
so
Sebastian
Han
had
done
some
work
to
create
a
script
that
brings
up
I,
believe
a
multiple
node
cluster
using
I
think
I
think
it
uses
the
pert
on
Linux
machines.
So
there's
support
for
that.
I
personally
have
not
used
it
since
I
have
a
Mac
but
I
think
that
I
was
working
at
least
four
Sebastian's
workflows,
as
anybody
else
have
is.
B
Under
the
documentation,
it's
it's
a
link
to
a
multi,
no
development,
environment
yeah.
It's
definitely
a
great
question
now,
because
during
development
I'd
say,
yeah
mini
cube
is
the
common
thing,
but
to
really
test
it
yeah
integration
and
upgrade
and
multi
node
yeah.
You
need
more
than
mini
cube
to
really.
F
G
Right
I
guess
so
one
of
the
things
I
have
found
trying
to
do
that
on
my
local
machine
is
that
my
my
desk
becomes
a
bottleneck.
G
F
G
A
Yeah
I
personally
plain
haven't
used
in
OpenStack,
really
at
all
one
solution
that
we
were
Travis
and
I
had
used
in
the
past
when
we
were
still
at
quantum,
was
using
the
core
OS
vagrant
repo,
to
bring,
if
you
know,
multiple
virtual
machines
that
are
running
car
OS
instances
and
then
using
cube
ATM
to
deploy.
You
know,
multi
node
kubernetes
cluster
across
those
that
had
some
pretty
good
success
for
us
in
terms
of
being
able
to
run,
you
know
multiple
node
clusters
that
have
multiple
disks
and
that
we
I
do
not
recall
running
into
any.
A
A
I'd
be
interesting
interested
to
see
you
know
what
failures
you
had
already
run
into
because
you
know
in
general
having
a
multiple
node
solution
for
developers
and
their
workflow
is
pretty
important
to
me.
So
you
know,
if
you're
having
problems
with
that
and
that's
you
know
kind
of
blocking
things
that
you
want
to
progress
on
then
you
know
I
would
like
to
be
able
to
solve
those
or
you
don't
understand
what
those
issues
are.
Yeah.
G
Okay,
I
might
start
trying
to
to
focus
again
on
bringing
up
a
virtualized
cluster
I
mean
we
have
a
like.
Souza
has
a
the
our
cast
product,
which
is
like
a
an
operating
system
environment
that
already
has
kubernetes
up
and
running
I
I.
Don't
think
that,
like
there's
anything
particularly
different
about
that
that
should
be
making
the
environment
more
difficult.
G
Yeah,
it's
I
guess
it's
good
to
know
that
having
a
multi,
node
development
test
environment
is
something
that
is
deemed
important.
I've
been
a
little
cautious
to
like
jump
totally
on
mini
cube,
just
because
it
seems
like
it's
a
somewhat
of
a
degraded
environment
compared
to
what
will
exist
in
the
real
world.
Yeah.
A
That's
an
absolutely
valid
point:
Blaine
in
your
IDE
I've,
seen
mini
cube,
as
you
know,
at
the
quickest
possible
way
to
you
know,
have
a
kubernetes
cluster
that
can
test
something
on.
You
know
at
the
sacrifice
of
you,
know
more
realistic,
real-world
scenarios,
but
yeah
so
losing
losing
that
ability
to
test
in
more
realistic
scenarios
is,
is
definitely
the
drawback.
There's.
B
C
If
we
look
into
the
community
and
everyone
to
see
what
would
be
a
feasible
solution
to
you,
yeah
kind
of
have
an
environment
that
I
push
my
pull
request.
Rockets
build
and
the
only
thing
I
have
to
do
is
connect
you
to
commit
to
some
kubernetes
cluster,
which
maybe
even
start
it
up
with
requests
or
something
because,
in
my
case,.
C
C
A
All
right
yeah
in
general,
if
anyone
neither
has
say
I
blame
you
can
feel
free
to
share
some
of
the
issues
you're
running
into
on
a
multi
node
test
environment
and
then
in
general.
Anyone
who
has
ideas
about
you
know
good
ways
to
do.
Testing
from
you
know
into
the
development
workflow
on
multiple
nodes.
You
know
feel
free
to
share
those
and
add
those
and
there's
a
bit
of
a
discussion
going
pretty
recent
discussion
gone
going
on
to
get
1544
that
so
I'll.
Add
that
to
the
chat
as
well,
because.
C
It
depends
on
what
you
change,
trying,
often
what
you
want
to
test
at
the
end,
because,
if
I
want
to
like
do
stuff
at
the
Monaco
or
most
of
the
time,
you
start
the
simple
darker
and
darker
occupation
cluster,
which
just
a
loss
between
multiple
notes,
at
least
okay
I
think
I,
just
start
up
more
nervous
with,
but
for
our
smaller
changes
like
simple
stuff,
it's
a
team
that
updated
on
changes
or
something
it
is
enough
to
well
start
a
meteor.
We
have
to
really
begin
to
for
their
special
cases,
kind
of
so.
C
A
You
know
I
was
wondering
if,
if
that
works
on
the
Mac
at
all,
you
know
because
it
you
know
with
the
doctor
on
Mac
uses,
like
the
you
know,
they
had
their
hyper
kit
that
has
some
hypervisor
functionality
and
have
no
idea
is
it
that
doesn't
really
launch
a
full-fledged
virtual
machine.
So
I
don't
know
how
it's
like
some
kernel
operations
like
that
K
RBD
does
would
even
be
supporting
an
environment
like
that.
C
A
Possible,
probably
sometime
in
the
near
future,
but
not
this
morning,
cuz
I've
got
other
things
in
the
schedule
this
morning
yeah
this
is
this
is
something
we
can
continue
making
progress.
This
has
been
brought
up
so
many
times
by
so
many
people
and
outside
agreement.
Just
rook-
and
you
know
the
mini
cube
with
people
and
project
in
general.
Has
you
know
state
that?
That's
not
never
gonna
I,
don't
think
I
think
that
they
said
they're
not
going
to
support
multi-node.
So
there's
kind.