►
From YouTube: Meshery Development Meeting (June 24th, 2020)
Description
Meshery Development Meeting (June 24th, 2020)
A
B
A
C
So
I
guess
I'll
start
so
just
couple
of
pointers
that
I
want
you
to
know.
Oh
one
was
I
guess
this
sort
of
more
in
line
would
be
layer,
5,
ng
site,
but
I
just
want
to
know
that,
like
we
had
sent
across
a
couple
of
more,
you
know,
Kings
to
you
so
I
mean
like
in
the
call
there
we
had
like
sort
of
decided
which
but
everything
you
know
we
should
go
ahead
with
so
the
one
that
was
mentioned
by
EIU's.
C
Should
that
one
like,
because
we
sort
of
you
know
collectively
decided
okay,
that
that
is
a
good
option
to
go
through.
So
should
we
go
ahead
with
that
one?
Or
would
you
recommend
some
others,
because
I
mean
we
do
search
for
other
ones,
but
I
mean
that
seemed
to
be
quite
good.
The
one
that
like
I,
you
shot
said.
A
C
A
I
did
ask
Karen
whose
last
name
is
Heba,
took
a
hit
but
tackle
I.
Think
it's
actually
a
made-up
name.
She
made
up
that
name
her
her
real
name
is
something
else's
it's
funny
anyway.
She
is
in
UX
and
focus.
Is
there
and
has
in
the
past,
helped
with
some
colors
and
a
bit
of
logo
just
logo
design
I
just
rolled
rolled
on
to
my
anyway
I'll,
be
asking
her
and
pushing
her.
A
If
we
don't
get
meaningful
feedback,
I
might
go,
pull
someone
else
in
because
I'm
not
sure
what
is
what
what
like
third
color
or
fourth
fourth
color
is
the
the
most
appropriate
compliment
from
a
scientific
model
perspective
from
the
for
the
perspective
of
what's
complimentary.
Here.
What's
of
the
same
hue
and.
C
A
A
C
A
D
Don't
really
have
wants
to
talk
about
just
waiting
on
the
operator
side
of
things
so
that
we
can
kick
that
off.
I've
been
using
the
only
side
from
the
operators
have
been
using
the
cube
builder
at
my
current
job
and
it's
okay.
It's
not
that
bad
I
mean
operator.
Sdk
has
a
lot
of
fluff
over
cube
builder,
but
it
it
usually
runs
a
couple
of
versions
behind
the
the
cube
builder
master
overhead.
D
A
D
I
think
depends
on
whether
we
want
to
act
on
custom
resources
or
we
just
be
probably
serve
as
a
web
server
G
RPC
server
and
take
decisions
based
on
requests
coming
in.
So
you
want
to
go
with
an
imperative
model.
You
want
to
go
with
in
declarative
model.
So
if
you
stick,
if
you
want
to
do
degenerative,
then
we
have
to
go
with
an
operator.
If
you
want
to
do
imperative,
then
it's
going
to
be
a
controller.
A
And
thing
with
all
the
changes
of
the
scheduling
decisions
in
some
respects:
the
notion
that
what
you
want
yeah
when
we
move
into
a
declarative
model
it'll
one
of
the
use
cases
there
is
to
really
tech
to
an
adapter
such
that
today.
When
we're,
we
were
using
that
doing
that
declarative
model,
we're
doing
it
through
an
adapter
that.
A
D
Yeah,
if,
if
you're
talking
only
about
mesh
thing,
then
I
agree,
that's
a
controller.
If
we
were
like
you're
saying
talking
about
the
adapters
state,
reconciliation
is
pretty
important
to
make
sure
everything
is
up
and
running.
That's
where
you
you'd
want
an
operator
so
that
you
get
to
the
final
state
where
the
mesh
is
deployed,
and
it's
watching.
D
It's
basically
watching
for
any
events
like
someone
deleting
the
mesh
and
deleting
the
mesh
without
without
actually
deleting
the
custom
resource,
then
we
want
to
make
sure
we
can
stand
that
back
up
back
up
again.
So
that's
where
we'll
have
to
do
some
state
machines
in
like
state
reconciliation
but
yeah,
I,
agree.
Mesh
Inc
is
just
essentially
just
a
watcher
and
it's
just
querying
stuff
from
the
API
server.
So
that
can
be
a
standalone
controller.
A
Yeah,
that's
that's
a
good
question
or
like
way
of
thinking
about.
It
is
understanding
that
yeah
understanding
that
we
have
a
massive
vision
and
roadmap
for
things
that
we
do
want
to
declare
and
state
that
we
do
want
to
manage
within
our
configuration
that
we
do
want
for
to
manage
as
kubernetes
objects
that
that
there
is
an
inevitability
of
of
the
of
the
fact
that
will
bring
out
custom
resources.
A
A
Does
it
feel
ugly
and
discontinuous
when
you
have
to
when
you're
working
with
a
project
like
ma
cherie,
and
you
have
to
go
over
and
part
of
the
installation?
Is
that
it
drops
in
a
custom
controller
and
then
another
part
of
the
installation
is
that
you
install
an
operator
and
there's
a
few
CRTs
or
does
or
do
people
not
really
blink
and
I?
There's
there's
a
question
about
how
it
feels
to
the
user.
Then
there's
also
a
question
about
well.
D
Yeah
I,
don't
think,
there's
any
problem.
Taking
that
route
I've
seen
a
lot
of
frameworks
that
have
their
controller
is
doing
background
jobs
and
they
have
the
operator
that
just
acts
on
the
custom
resource
so
I
think
decomposition
is
a
better
design
than
shoving
everything
into
the
controller
and
some
other
issues
with
controllers
are
scale.
If,
if
we
were
to
expose
a
web
server
from
our
operator.
D
From
the
operator,
then,
it's
not
really
gonna
scale
based
on
requests
coming
in
or
HPA
metrics
cuz.
It's
it's
essentially
a
leader
follower
model,
but
the
followers
are
passive
or
not
active.
So
failover
is
it's
going
to
be
slower,
yeah,
yeah!
That's
what
sort
of
you
make
our
decisions
controllers
can
use
horizontal
body
auto-scaling.
We
can
look
at
different
metrics
based
on
what
can
I
request
they're
coming
in
and
how
well
is
performing
in
terms
of
CPU
and
memory,
but
the
operator
won't
be
able
to
do
that.
D
A
It
becomes
pretty
clear
that
for
that
first,
priority
of
mash
think
that
that
really
is
just
a
custom
controller
and
that
yeah
even
beyond
that
I
guess
the
second
part
of
that
other
question.
Where
I
mean
I,
guess,
there's
more
than
two
questions:
you're
sort
of
articulating
a
third
consideration
around
the
actual,
the
actual
architectural
constraints
of
using
you
know
using
it
an
operator,
SDK
versus
and
so
there's.
A
A
D
If
you
are
worried
about
latency,
if
this
is
this,
is
our
slope
at
home?
You
don't
consider
we're
not
concerned
about
latency
and
taking
your
network
hop
then
I,
don't
think
that's
a
problem.
If,
if
we
are
looking
for
performance,
we
want
to
co-locate
everything
the
way
sto
did
with
this
Tod
then
sure
we
can
put
it
all
in
one
operator.
A
B
A
You
wouldn't
do
it
wouldn't
have
mattered
to
the
human
user
human
operator
in
front
of
the
UI.
The
latency
sensitive
operations
are
those
that
yet
the
design
is,
the
design
of
that
feature
is
predicated
upon
a
performance
first
architecture,
meaning
that
hey.
These
are
things
that
we're
trying
to
run
in
the
data
plane,
because
we
can
execute
them
real
time
and
it's
all
about
the
performance,
and
so
when
we
actually
get
to
those
anyway,
you
look
at
all
game.
Let.
D
Me,
let
me
say
this
too:
we
start
for
the
controller,
that's
a
standalone
deployment,
but
it's
essentially
a
binary
that
you
start
up
right.
We
can
stick.
We
can
stick
that
container
into
our
operator
SDK
deployment
as
well.
When
we
need
performance
and
then
you
have
the
local
host
hop,
which
is
definitely
more
performant
than
taking
the
kick
neural
network
off.
D
D
I,
just
I
just
meant
will
be
yeah
exactly
just
a
network,
hop
you're
going
over
a
local
host
now.
So
obviously
you
have
better
performance.
You're
not
taking
latency
hits.
There's
no
packet
drops
so
say:
I
mean
it's
not
going
to
be
too
hard
to
move
the
controller
container
into
the
operator
deployment
spec.
So
right,
you
start
with
a
separate
deployment
with
a
single
container
right.
Now,
that's
just
running
mesh
sync
and
then,
once
you
start
building
the
operator,
you
move
that
container
into
your
deployment
spec
for
the
operator.
All.
A
C
Hey
so
yesterday,
I
was
attending
the
linux
foundation's
open
JS
of
a
conference
that
that
is
taking
this
yesterday
and
today,
so
like
I,
was
in
a
meeting
with
the
Google
cloud
engineers
and
I
like,
according
to
our
conversation
with
remember
that
you
are
told
that
the
reason
that
we
haven't
really
made
a
lot
of
progress
in
the
distributed
load.
Testing
of
the
Envoy
planes
has
been
that
you
waiting
for
Google's
input
right.
So
basically
I
emailed
and
I
I
mean
starting
with
this
person
by
the
name
of
Brett
Brett.
C
Who
is
a
Google
cloud
engineer?
So
he
told
me
that
there
are
I
mean
he
knows
people
within
his
team
who
are
working
on
the
Envoy
planes.
So
he
told
me
to
like
message
like
email
him,
the
details
of
the
program
so
that
they
can
probably
approach
you,
and
you
know
the
work
can
get
started
on
that
on
that
front.
So
I
mean
this
email
him
yesterday,
so
hopefully
I
mean
we
should
be
hearing
from
him
on
this
front.
It
was
I
guess,
as
he
said,
that
we
haven't
really
moved
forward
with.
A
C
Help
you
yeah,
but
they
also
say
that
I
mean
they
were
a
couple
of
more
projects
that
they,
where
they
are
actually
using.
You
know
distributed
testing,
so
I
mean
not
specifically
for
the
Envoy
Road
things,
but
something
like
cloudBees
is
also
something
that
uses
the
stupidity.
So
they
say
that
if
you
wanted
some
in
like
inspiration,
you
can
probably
look
at
a
couple
of
more
projects
that
are
using
distributed
testing,
not
specifically
with
relation
to
the
service
meshes
or
with
the
own
way.
But
yeah
I
mean
they
sort
of
follow
similar
architecture.
A
Okay,
yeah
I
mean
some
respects.
It
makes
a
lot
of
sense
that
they
would
be
doing
distributed
testing
because
core
to
part
of
their
service
is
distributed.
Artifacts
and
the
exchange
of
the
caching
of
artifacts
in
different
geographies
I'm,
guessing
yeah.
They
don't,
they
don't
have
artifactory.
What
do
they
have?
Cloudbees
has
ax
I
mean.
C
A
Good
good
there's
a
couple
of
learns
in
our
slack,
actually
the
ones
that
are
responsible
for
Nighthawk,
and
so
this
is
good.
It
I
think
we
I
guess
the
short
of
is
I
think
we
have
who
we
need,
but
it
never
hurts
to
have
additional
contacts
and
more
input
and
other
perspectives
wish
to.
Thank
you
for
that.
If
I
resisted
to
to
get
the
rubber
touching,
the
road
on
a
two-week
conversation
is
or
three-week
is
I,
don't
know
that
we
even
like
one
of
the
things
to
potentially
I.
A
Don't
core
to
what
mesh
sink
can
become
is
a
is
an
architectural
diagram.
There's
a
logical
model
that
says
that
that
takes
constructs
inside
of
a
system
line.
That
basically,
is
something
like
an
SMI
its
let's
sort
of
slightly
not
entirely,
and
that's
to
say
it
there
are.
These
things
within
is:
do
there's
destination
rules,
virtual
server
services
that
there's
all
these
constructs.
A
B
A
D
So
Shalini,
my
thought
is
in
time,
as
I
mentioned,
as
some
IV
is
the
same:
abstractions
like
resources
are
defined
traffic
management
resources
that
define
security
resources
that
define
policies,
so
we
use
similar
abstractions
as
SMI.
You
start
with
that,
and
some
things
are
going
to
be
more
specific,
then
in
go.
You
would
have
interfaces
that
would
be
implemented
by
sto,
implemented
by
a
linker,
D
and
they'll
have
their
own
constructs
so
say:
I
think
we
can.
We
can
do
that.
Smi
is
a
good
place
to
start
because
it
identifies
the
common
areas
between
meshes.
A
A
We
should
start
like
we
should
that
if
we,
if
today,
if
we
had
a
custom
controller
today
that
had
a
one
for
want,
that
was
just
a
direct,
an
untranslated
capturing
of
the
fact
of
everything
that's
deployed
in
in
in
a
kubernetes
environment,
so
basically
all
the
kubernetes
info
and
all
of
the
sto
info
available
in
in
this
model
in
which
we've
got
watches,
you
know
like.
Oh
you
know,
not
everything
is
going
to
be
of
interest
to
us,
but
but
more
or
less.
D
D
A
Yeah
then,
as
we
work
through
that
as
we
work
through,
that
we're
worried
yeah
somewhere
in
there
you're
able
to
draw
kind
of
this
jagged
line
or
across
each
of
them,
where
each
of
them
they've
got
all
these
features
and
all
these
constructs
and
there
there
will
be
a
universal
there
will
be.
You
know,
a
common.
A
A
Some
of
those
are
note
agents,
and
so
there's
a
fair
bit
of
commonality
of
the
core
constructs
that
that
are
going
to
get
us
much
further
down
the
road
then,
and
by
the
time
that
we're
by
the
time
that
we're
going
deeper
and
providing,
for
example,
configuration
analysis
of
best
practices
on
a
given
mesh.
That's
there
is
going
to
be
some
commonality
of
best
practices,
but
quite
a
few
of
them,
the
vast
majority
of
them
and
want
the
analyses
or
the
best
practices
that
are
really
worth
there.
D
A
Really
and
then,
and
in
flying
is
with
an
operator
a
stick
is
with
it
so
by
just
lat
finished
that
up
real
quick
I
know
for
Dhruv
and
Dheeraj,
or
that
some
of
this
was
abstract
and
that
we
didn't
provide
a
lot
of
context.
My
apologies
we're
doing
that
in
some
respects
on
purpose,
because
we're
trying
to
do
a
bit
more
of
a
stand-up
style
meeting,
it
is
one
thing
to
note:
is
that,
as
we've
been
going
through,
this
service
mesh.
A
Performance
specification
on
the
dish
we've
been
running
into
we've
begun
to
run
into
a
little
bit
of
the
same
types
of
questions
on
well
on
whether
or
not
we
should
be
leveraging
some
things
from
SMI
and
it's
actually
helpful.
If
we
take
a
moment
to
describe
this
because
I
don't
have
a
because
my
answer
today
is
I,
don't
think
SMI
helps
us
it's.
Let
me
describe
the
issue
and
that
is
that
in
the
service
mission
in
SPS
today.
Currently
there
are
two
protos
that
describe
this.
A
A
Well,
it's
one
thing
to
say
that
we're
gonna
generate
HTTP
load
and
just
assume
that
that's
a
bunch
of
get
operations,
but
maybe
it's
a
post
with
or
maybe
it's
got
a
cookie,
and
maybe
it's
got
some
other
and
as
we
go
to
capture
that
inside
of
this
specification,
music
well
SMI
has
a
traffic
spec
which
is
intended
to
identify.
You
know
that
it's
the
same
thing
when
you
go
look
at
that
SMI
first
of
all,
SMI
is
not
and
has
never
moved
very
quickly
of
what
I
see
of
the
traffic
spec.
A
A
So
it's
not
a
it's,
not
something
directly
leverageable
I
could
be
mistaken
and
maybe
buried
in
one
of
the
repose
is
something
that
we
could
maybe
be
be
more
directly
inspired
by
or
or
leverage
so
yeah.
If
you're
looking
for
an
excuse
to
kind
of
dig
into
SMI,
naveen
is
in
part
like
one
of
the
questions
I
wanted
to
bring
up
with
you.
A
Of
the
time
that
you've
spent
with
of
the
time
you
spent
with
SMI
I'm,
not
sure,
have
you
come
across
the
traffic
spec
much.
A
B
E
So
sorry,
so
so
like
wrong,
yep
expect
we
work
week,
what
it
is
it
like,
they
support,
gzip
it
out
and
they
should
have
it
out,
but
not
the
GRDC
or
any
other
protocol.
Yet
and
along
with
that,
as
far
as
I
remember
like
I
worked
on
F
expect
I,
guess
three
or
four
weeks
earlier,
so
it
was
just
destination
and
it
was
just
which
service
was
the
destination,
so
it
was
not
like.
You
are.
What
endpoint
did
it
hit
or
something
like
that?
So
I'll
have
to
look
into
that
like
PI.
A
Which
to
me
isn't
nearly
what
we,
when
you
think
about
the
measure
you
I
the
performance
section
and
the
fact
that
people
will
define
some
specific
headers
cookies
that
they
they
might
want
to
include.
This
doesn't
capture
that.
Moreover,
even
if
it
did,
it's
not
like
it's
saving
us
that
much
time
to
just
go
write
that
down
into
a
different
proto.
E
Yeah
like
cookies
and
everything
is
there
but
like
it
is
just
HTTP
and
TCP,
and
we
are
bound
by
that.
Like
no
gr
piss,
your
son
like
we
would
have
to
either
be
a
design
for
waiting
for
there
isn't
everything
and
the
URL
part
needs
to
be
there.
Currently,
it
just
say
is
okay.
This
is
the
this.
Is
the
router
end
of
lit
and
there
is
no
URL
part
to
here.
A
A
A
B
A
Dheeraj
I
was
curious,
is
whether
or
not
it
was
your
first
meeting
I
wasn't
sure
just
because
just
like
in
the
US,
where
you,
every
third
person
you
meet,
has
the
name
Mike
you
meet,
has
the
name
T,
Russia
or
I.
Wasn't
sure
if
that
you
were
someone
else,
or
do
you
want
to
take
a
moment
to
introduce
yourself
if
you
would
be
nice
to
get
tonight.
B
F
A
F
Yeah
I've
been
working
on
adding
the
JSON
payloads
for
terroristic
years
and
right
now
what
I
am
working
on
it.
I
have
using
RF
c80
pools,
you
know
orange,
which
is
the
basic
format
of
replacing
the
current
URLs
next
EP
8
and
now.
First
of
all,
in
this
week,
I
have
worked
on
open
flow,
which
is
which.
F
Because
I
am
mini
net
for
that
purpose,
using
a
VM
and
then
they
was
basic
set
of
switches
have
been
introduced.
There
is
a
command
in
prehnite
itself,
which
is
by
school
eminent
controller.
We
use
the
IP
of
the
opendaylight
controller,
now
opendaylight
runs
on
that
normal
local
host
127.0.0.1
and
then
using
that
IPP
basically
communicate
with
the
open
tonight
and
this.
F
What
we
call
is
a
minute,
because
it
supports
open,
put
switches
which
is
virtual
switches
for
that,
and
that
and
audial
works
on
the
different
projects,
including
open
flow
and
and
next
aspect
is
that
I
have
been
working
on.
I
am
going
to
work
on
a
net
con
and
MD
cell,
which
is
which
will
be
I.
Will.
A
F
A
Okay,
here's
something
to
if
you
would
so
d-rush
thanks
for
coming
and
that
this
is,
and
thanks
for
sharing
about
some
of
your
recent
focus,
you
were
the
networking
centric
focus
that
you're
doing
right
now
is
quite
relevant
and
interesting
too
this
the
focus
and
mesh
land.
It
is
interesting
and
funny
how
big
the
networking
space
is
and
how
how
large
the
of
an
area
focus
that
you're
talking
about
with
a
no
DL
is
button.
A
Yeah
that
particular
one
is,
did
I
share,
which
one
I
share:
the
SMI
one,
which
is
a
bit
of
a
red
herring,
probably
probably
that
that
traffic
specs
link
is
I,
think
something
that
we're
saying
we're
going
to
intentionally
ignore.
But
but
you're
you're
right
like
that
it
within
context
of
the
mastery
project.
We
do
want
to
ultimately
identify
a
technology
that
we
would
use
to
do,
traffic
mirroring
or
that
traffic
capture
traffic
replay
and-
and
we
haven't
so
so
as
you're
going
through
odl
and
you're.
A
B
E
And
like
graphic
excesses
about
like
complete
traffic,
spec
I
guess
is
complete,
like
the
test
cases
have
been
written,
we
need
to
like
shift
to
maybe
a
so
many
conformance
weapon,
think
about
the
architecture
on
how
these
just
will
actually
run
through
G
RPC,
using
a
raptor
patter
or
not.
Maybe
something
else.
E
I,
like
I,
don't
know.
Actually
these
are
only
test
cases
and
the
only
test
case
that
is
complete,
as
of
now
is
the
epic
XS
n
def
expect,
which
that
left
expect.
Oh
just
what
it
does
is
it
just
checks,
the
CR
DS,
whether
these
are
like
applicable
or
not,
but
what
I
can
play?
What
we
can,
what
I
can
show
is
how
the
test
will
run
or
what
are
the
solution
and
how
cutter
land
everything
blends
in
I
guess
I
can
do
that.
A
B
A
For
my
part,
I've
been
focused
a
little
bit
on
SPS
to
help
make
that
specification
a
bit
more
robust
within
SPS
there's
a
couple
of
protobuf
or
proto
files
that
we've
been
defining
I've,
and
so
today
there
are
two
in
this
spec
I
have
intentionally
not
scheduled
a
recurring
meeting
with
in
the
CNC
F
to
advance
this
spec,
because
I
don't
feel
like
I
have
because
I
need
some
I
need.
Another
I
need
someone
to
champion
it.
A
I
can't
my
plate,
overflows
and
I
will
I
don't
want
to
engage
very
publicly
and
broadly
on
something
that
we're
not
going
to
advance
quickly.
So
too
many
priorities
at
the
moment,
and
so
that's
probably
something
I
should
talk
to
Kanishka
about
when
he
comes
back
from
exam,
so
Kanishka
cars
out
for
a
week
on
exams
he's
been
helping
advance
this
quite
a
bit.
A
Kanishka
car
has
been
taking
the
newer
specifications
that
we've
been
writing
here
and
making
and
updating
the
implementation
inside
of
measure
e,
so
that,
when
measure
e,
when
you
define
a
performance
test
and
measure
II
that
that,
on
the
back
end,
it's
described
like
this,
that
you
could
import
a
file.
You
could
import
and
export
it.
The
definitions
of
those
tests
configs
in
this
format,
so
Dhruv
of
part
of
our
conversation
about
service
mesh
I'm.
Sorry,
our
conversation
was
about
performance
test
profiles.
A
The
back
end
of
that
has
been
worked
on
recently
and
looks
a
bit
like
this,
but
we
don't
yet
and
so
I
guess
just
for
context.
Some
things
that
I'm
going
to
do
before
we
meet
next
week
is
will
have
a
Gatsby
theme
in
hand.
We
will
hopefully
have
communicated
some
of
the
decisions
that
Nitesh
has
made
about,
starting
with
it
the
cash
package,
starting
with
a
custom
controller
and
and
informing
a
deep
and
vinayak
and
asking
that
they
help
in
getting
going.
A
D
A
D
Is
a
it's
a
request
response
driven
controller,
so
so
we
need
to
introspect
that
out
what
the
request
bodies
are,
what
the
response
response
objects
are.
What
kind
of
api's
are
we
exposing
and
then
based
on
how
the
how
that
spec
is
laid
out
it'll
be
much
clearer
on
how
we
want
to
implement
their
handlers.
A
Got
you
okay
and
something
I
hate
to
toss
this
in,
hopefully,
doesn't
really
does
it
complicate
this
a
lot,
but
is
graph
QL
starts
to
come
to
mind
as
a
potential
as
a
highly
flexible
interface.
That
might
be
make
some
sense
for
us
here
that
there
are
any
number
of
questions,
any
number
of
queries,
any
number
of
requests
that
in
the
future
we're
gonna
want
to
make
based
on
stuff
that
I
can't
articulate
at
the
moment,
which
is
like
you
know,
for
a
given
node
for
a
given
service.
A
A
D
A
A
Okay,
we'll
get
there
we're
getting
that
thanks,
guys,
I'm,
sorry
that
I
have
been
a
little
bit
off.
It's
need
to
go,
get
more
sleep,
so
Dheeraj,
very
nice
to
have
you
there's
a
couple
of
people.
I
want
to
introduce
you
to
in
the
community.
So
if
I
don't
do
that
very
soon,
please
remind
me:
cuz
I,
think
there's
some
other
folks.
Who've
got
some
similar
focus
as
you
might
be
nice
to
meet
them.