►
From YouTube: Kubernetes SIG Testing 2018-04-10
Description
A
A
D
C
C
Okay,
so
what
problems
does
API
snoop
hope
to
solve?
Well,
a
couple
of
things.
So
first
thing
is
that
I've
heard
that
to
kubernetes
test
coverage
is
currently
calculated
using
88,
logs
and
I.
Guess
like
what
would
try
and
do
is
calculate
that
API
coverage
in
a
more
generic
and
reusable
way,
we're
also
the
components
working
group.
What
is
working
on
raising
EDI
tests
currently
that
test
coverage,
raters,
11%
and
I?
Guess
you
know
this?
C
We
can
redirect
a
pair,
so
we
redirect
API
requests
through
the
proxy
at
the
moment.
It's
using
IP
tables
and
that
only
affects
certain
pods
with
annotations
and
then
we're
planning
to
reuse,
MIT
and
proxy,
which
is
an
existing
project.
So
we
don't
have
to
start
from
scratch
and
then
we
hope
to
create
our
reusable
open.
Api
inspection
logic
is
a
plugin
and
so
yeah
I'm,
that's
just
a
brief
overview.
Is
there
any
sort
of
questions
before
the
before
I
start
demoing
I.
B
Wouldn't
use
frequency
as
the
target
of
what
to
determine
as
the
highest
priority
things
to
have
tests
on.
Obviously,
that
stuff
is
typically
been
tested
really.
Well,
usually,
the
coverage
you'd
want
to
determine
from
either
this
signor
cigarette
picture.
What
is
the
highest
praise
aims
to
evaluate
okay,.
A
C
So
so,
and
I
mean
a
hippie
hacker
he
couldn't
be
here
this
morning
because
of
he
has
a
doctor's
appointment
about
20
minutes
who's
actually
in
transit
already
to
the
doctor,
and
it's
completely
related.
It's
the
same
project.
Basically.
C
So
this
is
like
a
perfect
concept
to
just
show
that
it's
possible
and
then
now
we're
like
like
what
what's
the
X
for
API
snoop.
So
we're
reflex,
refactoring
that
complexity
and
then
building
a
reliable
hound
chat
that
we
can
deploy
and
then
that's
when
we
start
to
write
the
MIT
module
and
collaborate
with
testing
for,
but
sega
architecture
is
interesting
like
maybe
we
could.
A
C
E
Too
hi
I'm
Mitra
I
wanted
to
introduce
some
people
on
the
call
we
have
some
vendors
from
globe
and
for
joining
us
and
we'll
be
actually
writing
conformance
tests
to
improve
our
coverage.
So
I
see
them
asking
questions
on
our
suggesting
slack
from
time
to
time.
If
they
do,
please
help
them
out
with
respect
to
confirmands
testing
and
got
any
questions
they
might
have
I
think
I'll,
say
I,
remember
seeing
Jose
on
the
call.
If
any
of
you
are
here,
can
you
say
hello?
Yes,.
F
A
C
B
C
C
C
G
C
The
moment
it's
also
setting
an
annotation
on
some
of
the
templates
as
well,
which
is
what's
then
being
picked
up
by
another
part
of
the
architecture
hold
on
like
there's
a
container
and
they're
called
pod
watch,
which
is
basically
looking
for
those
annotations.
So
once
again,
it's
like
I
mean
this
is
kind
of
the
complexity
of
the
perfect
concept
that
we're
looking
to
simplify
but.
B
So
a
quick
design,
choice.
Question.
Is
there
any
reason
why
you
didn't
just
put
a
proxy
or
and
load
balance
for
the
front
end
just
route
all
traffic
through
that
little
balancer,
you
do
like
certificate
termination
and
I'll
be.
C
You
know
HTTP
requests
the
outside
world
so
that
it
would
sort
of
lock
the
points
down
so
that
they
would
only
talk
to
a
certain
HTTP
come
things.
So
what
we
did
is
we
modified
that
instead
to
to
intercept
traffic
between
the
between
the
part
and
the
API
server,
so
it
was
more
of
you
using
of
an
existing
solution
rather
than
like
a
like
a
design
decision
as
such
yeah.
C
C
C
C
C
C
You
sorry
and
come
and
call
there
so
then
we
can
deploy
our
example
app
and
then
all
we
have
to
do
after
that
is
basically
annotate
the
pod
and
that
we
deploy
like
the
example
pod
and
then
the
traffic
will
be
intercepted
by
this
API
snow.
So
I'll
just
do
that
quickly.
So
here
we
are
we're
creating
the
kubernetes
api
certificate.
C
C
C
C
B
C
Then
the
pot
is
now
annotated
and
we
should
see
the
results
here
we
go.
So
this
is
the
traffic
being
intercepted
on
that
particular
pod
and
we
can
see
the
authorization
barrier
etc.
It
doesn't
keep
the
response
from
requests
currently
because
in
my
team
web
keeps
this
all
in
memory,
but
the
request
and
response
can
be
intercepted
as
well,
so
this
proofs
that
it
can
be
done
using
this
method
and
basically
so
we'll
use
this
as
a
base
for.
A
C
C
C
Like
your
client
side
or
your
server
side
would
be
unable
to
like
it's,
it's
like
they
could
get
filtered
out
earlier
in
the
process.
You
know,
like
you've
got
logic
between
you
know
when
things
are
intercepted
on
the
server
side
and
whatever
mechanism
you
have
behind
that
that
logs
things
in
the
same
I
guess
is
on
the
client
side,
I
guess
also.
The
other
point
is
that
if
you
do
it
this
way,
then
it
requires
it's
like.
C
C
You
just
need
to
basically
suit
yourself
in
between
what
what
the
server
and
the
client
and
then
you
can
and
sip
the
traffic
and
then
use
that
to
check
for
conformance
against
your
open,
API,
spec
and
then
I
guess
one
of
the
ideas,
as
it
might
have
said
before
us,
the
idea
of
reuse
so
like
yes,
this
can
be
used
for
kubernetes,
but
it
could
also
be
used
for
other
open
source
projects.
C
G
I
mean
I,
don't
know
it.
Sort
of
just
stubby
is
sort
of
I'm.
Think
stubby
uses
HTTP,
but
there's
some
other.
You
know
mechanics
around
turning
HTTP
endpoint
into
a
service,
so
it
might
be,
you
know,
is
used
fairly
extensively
in
you
know
around
Google,
so
I
could
sort
of
see
if
there's
a
proto
that
says
what
my
you
know,
similar.
B
G
C
A
Okay,
cool
Thanks,
thanks
for
your
time,
around
yeah
I
mean
like
when
I
talk
to
hippy
I
I
got
the
sense
that
this
would
be
super
useful
for
pretty
much
any
open.
Api
based
service,
not
just
kubernetes
and
so
I-
think
that's
a
really
lot
of
a
goal,
just
yeah,
so
it
I
don't
know
that
we
really
got
to
see
a
proof-of-concept
last
time.
You've
seen
at
this
time,
that's
great
I
think
figuring
out
how
to
like
start
to
automate
this
moving
forward.
A
The
more
time
the
lower
that
percentage,
the
more
often
humans,
were
subverting
the
submit
queue
and
pressing
the
merge
button
on
their
own,
and
it's
like
way
up
on
the
90
percents
now,
which
is
super
cool,
that
kind
of
happened
shortly
after
we
turned
off
post
Simmons,
blocking
the
queue
shocker
I
think
the
other
question
I
had
just
with
coop
con
in
mind,
is
I,
always
loved
demoing.
Our
big
query
based
metrics,
but
I
feel
kind
of
bad.
A
When
we
get
to
this
dashboard,
then
I,
can
you
probably
can't
see
if
the
numbers
are
really
tiny
but
like
you've
had
jobs
that
are
failing
for
over
200
days
over
100
days,
I
feel
like
it
would
be
really
nice
to
get
rid
of
these
to
make
them
go
away
prior
to
Cape,
Cod
I,
don't
know
who
we
need
to
ask
about
that.
The
majority
of
these
are
gke
or
GCE,
belated
jobs,
not
entirely
so
maybe
those
Thera
folks,
we
could
poke
inside
a
Google
and
I.
G
So
for
the
approval,
breakouts
and
then
the
config
breakout,
which
is
happening
immediately
after
that,
do
we
want
to
I
feel
like
since
the
meeting
is
scheduled
on
the
Google
hangout.
We
should
probably
people
interested
in
that
should
migrate
over
to
the
hangout,
but
I
wanted
to
discuss
whether
we
wanted
to
do
that.
I'm.