►
From YouTube: Use Your Favorite Developer Tools in Kubernetes With Telepresence - Abhay Saxena, Datawire
Description
Join us for Kubernetes Forums Seoul, Sydney, Bengaluru and Delhi - learn more at kubecon.io
Don't miss KubeCon + CloudNativeCon 2020 events in Amsterdam March 30 - April 2, Shanghai July 28-30 and Boston November 17-20! Learn more at kubecon.io. The conference features presentations from developers and end users of Kubernetes, Prometheus, Envoy, and all of the other CNCF-hosted projects
Use Your Favorite Developer Tools in Kubernetes With Telepresence - Abhay Saxena, Datawire
Are you new to building cloud-native applications? As an engineer, it can be a tough transition if you're used to a quick inner development loop. With the overhead of image build/push/pull, the ability to see your code changes in action immediately probably seems impossible. But Telepresence can help! And it works well with many of the tools you're using already. In this talk Abhay Saxena will demonstrate how to use Telepresence with some popular IDEs and debuggers, including VSCode for Node and Rust services, IntelliJ for a Java service, as well as some command-line examples with Delve for a Go service and Pdb for a Python service. Abhay will also look at how to get Telepresence to play well with Kubernetes client libraries and sidecar containers.
https://sched.co/UaeA
A
Hello,
everyone
I
am
obvious,
like
Senna,
I
am
a
developer
at
data
wire
and
I
am
the
lead
developer
of
telepresence
there's
another
telepresence
talk
earlier
today
done
by
my
good
buddy
Daniel
Bryant.
So
he
talked
about
the
introduction.
The
overview
I'm
gonna
talk
about
specifically
about
debugging
a
service,
that's
part
of
an
application
that
runs
in
a
kubernetes
cluster.
So
let's
talk
about
developing
software
right,
we,
my
guess
a
lot
of
us
are
probably
software
developers
out
there.
We
have
a
process
that
we
follow
right.
We
have
a
piece
of
code.
A
We
changed
that
piece
of
code
because
we're
adding
a
feature
or
we're
fixing
a
bug.
Something
like
that.
We
build
it.
We
run
it.
Sometimes
we
inspect
it
with
some
of
our
tools
right,
a
running
piece
of
code
can
be
somewhat
opaque.
It's
you,
you
really!
You
can
see
the
inputs,
you
can
see
the
outputs,
but
what's
going
on
inside
is
not
necessarily
so
clear.
Well,
we
have
fancy
tools
that
help
us
look
inside.
We
have
what's
your
stochastic,
sampler,
profilers
and
debuggers,
and
so
forth.
That
lets
you
see.
A
What's
going
on
make
your
software
a
little
less
opaque
right,
but
you
change
your
code.
You
run
your
code,
you
change
your
code,
you
run
your
code,
that's
the
normal
software
development
loop.
Now
when
you
traduce
kubernetes
things
change
a
little
bit
right.
You
change
your
code.
You
build
an
image.
You
push
the
image
to
a
container
registry.
You
poke
kubernetes
in
some
way.
Maybe
you
update
your
deployment
or
something
like
that.
Kubernetes
pulls
the
the
image
and
restarts
your
pods
and
well
there
you
go
now.
A
You
can
see
your
code
running
right,
it's
it's
a
pretty
long,
somewhat
slow
loop
and
let's
talk
about
debugging
in
particular,
because
that's
what
I
want
to
focus
on
here,
debugging
software's
hard
right,
you
have!
You
have
to
be
pretty
clever,
to
write
a
good
piece
of
software
and
then
to
debug
that
software.
You
have
to
be
even
smarter
than
that
and
then,
when
you're
running
in
kubernetes,
the
software
is
running
far
away
from
you
you're
kind
of
isolated
from
it.
You're,
typically
not
able
to
use
most
of
your
normal
tools.
I
mean!
A
A
It's
not
an
easy
thing:
it's
not
an
easy
to
replicate
and
it
doesn't
always
work
the
way
you
want
and
then
so,
if
you're
not
able
to
use
your
tools
right,
you're,
basically
left
with
prints
style.
Debugging
right,
you
can
add
some
log
statements
and
you
can
run
your
code
and
then
you
can
see
coop
control
logs
and
see
what's
going
on
there
and
it's
not
the
most
interactive
approach,
so
telepresence
can
help
with
all
this.
The
short
version
is
that
telepresence
lets.
A
You
run
your
code
on
your
laptop,
but
your
code
thinks
it's
running
in
the
cluster.
It
means
your
code
can
access
the
cluster
and
the
cluster
can
access
your
code,
and
this
makes
it
possible
to
use
all
of
your
tools.
I'm
gonna
talk
about
IDs
and
things
like
that,
and
that's
basically,
what
I'm
gonna
do
here:
I'm
going
to
do
a
series
of
demos
showing
you
how
to
use
telepresence
how
telepresence
lets
you
use
your
typical
development
tools.
A
Okay,
so
we're
gonna
go
through
an
example,
application
that
represents
what
might
be
in
a
typical
kubernetes
cluster.
So
we
have
some
sort
of
request
that
comes
in.
Maybe
it
comes
in
to
an
API
gateway,
and
that
goes
to
some
service
and
there
might
be
some
chain
of
services
and
eventually
a
request
comes
to
our
service
right.
This
is
what
causes
our
code
to
run.
A
request
has
come
in
to
our
service
and
our
service
is
probably
going
to
need
information
from
other
services
in
the
cluster.
A
So
it's
going
to
call
out
to
other
things
there,
let's
zoom
into
our
service
and
see,
what's
going
on,
yeah
I
know
the
slide
says
my
service
but
I'm
sharing
this
debugging
process
with
all
of
you.
So
it's
going
to
be
our
service.
Ok,
so
we
have
to
get
information
from
other
services
in
the
cluster
and
then
we
have
to
perform
the
interesting
computation
whatever
it
is,
we're
doing
whatever.
The
reason
is
that
someone
would
call
our
service.
A
We
have
to
do
that
and
then
finally,
we
return
the
result,
so
that's
the
shape
of
a
typical
service
in
your
typical
kubernetes
application.
Ok,
let's
think
about
what
can
go
wrong.
You
have
a
service,
it's
been
running,
it
seems
to
be
running
fine
and
then
it
starts
breaking.
It
starts
giving
wrong
answers
or
it
starts
crashing
or
something
that.
Why
would
this
happen?
What
can
go
wrong
in
these
sorts
of
situations?
A
Well,
we're
relying
on
other
services
in
the
cluster
right
and
while
we
own
our
service,
we
don't
necessarily
own
all
of
our
dependent
service,
the
services
we
depend
on
other
teams,
work
on
those
they
might
be
fixing
bugs
they
might
be
adding
features
and
so
forth.
They
can
make
mistakes
too
right.
So
it
might
be
the
case
that
something
that's
worked.
Fine
suddenly
starts
failing
and
now
here's
where
it
can
be
worse.
Sometimes
you
have
a
situation
where,
whatever
it
is,
failing
only
fails
some
of
the
time
hey
I
ran
it
in
CI.
A
It
worked
fine
there.
It
works
fine
on
my
system,
but
it
fails
some
of
the
time
in
production
what's
going
on
there,
so
it
can
be
really
difficult
to
debug
a
situation
like
that.
Okay,
so
I'm
gonna
start
going
through
these
examples
enough
slides.
So
this
first
example
is
going
to
be
a
Java
example
and
I
have
a
couple
of
others.
Okay,
so
it's
hop
over
to
the
editor
is
those
are
little
notes
for
myself.
A
So,
while
the
demos
I'm
doing
work
perfectly
well
on
a
gke
cluster
or
an
on-premises
cluster,
wherever
your
kubernetes
cluster
is
for
the
purposes
of
this
talk,
I
am
using
a
local
cluster.
It
doesn't
change
anything
except
I.
Don't
have
to
rely
on
the
conference,
Wi-Fi,
so
I
hope
you'll.
Forgive
me
that
a
little
bit
of
cheating,
perhaps
it's
not
really,
cheating,
because
it
all
works
exactly
the
same
way
and
the
other
thing
is
I
have
a
tendency
to
use
K
instead
of
coop
control,
it's
just
the
coop
control
command,
it's
just
a
bash
alias.
A
So
let's
take
a
look
at
what's
going
on
in
this
cluster
I'm
gonna
do
a
coop
control
get
deployments,
and
this
is
what
we
have.
We
have
Ambassador.
This
is
my
API
gateway.
Sorry
for
the
small
text,
I'm
going
to
try
to
read
out
loud
everything,
that's
important.
So
ambassador
is
the
API
gateway.
When
I
use
curl
commands
to
access
the
cluster,
that's
what
I'll
be
hitting,
and
then
we
have
a
bunch
of
other
services
running
in
this
case.
A
We're
going
to
be
looking
at
the
Java
service
and
the
Java
service
and
all
of
these
other
services
all
call
into
a
base
service.
That's
also
running
in
the
cluster
okay.
So
when
I
access
the
cluster
I'm
going
to
use
the
curl
command,
I'm
gonna
curl
edge,
slash
Java,
and
this
is
the
output
that
shows
up.
So
let
me
read
this
to
you.
It
says
hello,
coupe,
con
na
greetings
from
Java
and
then
there's
this
code.
It's
a
base64,
encoded
string
and
then
there's
some
additional
information.
A
Has
this
nice
box
around
it
because
that's
we
want
a
pretty
answer
for
it
for
our
service
output,
so
I'm
gonna
run
this
curl
that
I
just
did
in
a
loop,
I'm
gonna
curl
edge,
slash
Java
in
a
loop
and
then
echo
to
get
a
blank
line,
and
then
sleep
for
one
second,
okay,
and
you
can
see
the
output
scrolling
along
and
now,
if
we're
a
little
patient,
I'm
gonna
pause
it
at
just
the
right
moment.
We'll
see
that
there's
a
bug
in
this
code.
A
Well,
there
you
go
so
I've
paused
the
the
output-
and
you
can
see
this
time
specifically
the
outputs,
not
right
it
doesn't
line
up
correctly.
The
box
isn't
right.
The
code
is
too
short.
Something
is
wrong
now
yeah.
This
is.
This
is
a
pretty
trivial
example,
because
we
want
to
be
debugged
able
to
debug
this
during
the
course
of
the
talk,
but
I
mean
an
unexpected
output
like
that
can
cause
all
sorts
of
things
to
break
in
a
real
system
right.
So
the
goal
is
to
debug
this
error
in
our
IDE.
A
So,
let's
hop
over
to
our
IDE
okay,
this
is
Intel
IntelliJ
IDEA,
and
this
is
the
Java
code.
I'm
gonna
walk
you
through
it
real
quick.
This
is
the
main
function
and
it's
running
on
port
8000
and
every
access
to
slash
is
going
to
run
this
code.
This
code
does
a
call
to
get
value
and
I'm
gonna.
Take
a
quick
look
at
get
value.
This
is
an
IDE.
I
can
just
hit
a
key
and
jump
to
that
get
value
function,
and
this
is
basically
the
moral
equivalent
of
a
curl
to
the
base
service
right.
A
So
our
service
calls
the
base
service
as
one
of
the
first
things
it
does
and
then
once
we
get
that
value,
we
do
a
bit
of
computation.
In
this
case
the
computation
is,
we
do
a
base64
encode
of
the
answer
we
get
from
get
value
and
then
we
construct
our
little
box
and
you
can
see
basically
we're
kind
of
zooming
when
we
draw
this
box
that
the
the
hashcode
is
gonna
be
the
correct
length,
and
so
if
the
hashcode
is
not
the
correct
length,
you're
gonna
see
that
error
that
you
saw
okay.
A
A
Let's
not
do
it
that
way:
I'll
just
type
it
out,
telepresence,
I'm,
gonna,
swap
deployment
and
the
deployment
I'm
swapping
is
the
Java
deployment
and
I'm
going
to
run
some
sleep
command
just
to
keep
telepresence
running
right
because
we're
gonna
run
the
code
in
the
IDE
in
the
debugger.
Okay,
so
telepresence
is
running.
If
we
hop
back
to
our
curl
loop,
a
look,
the
curl
loop
is
failing
now.
Why
is
it
failing?
Well,
it's
because
we've
replaced
the
code
that
was
running
in
the
cluster
with
a
proxy.
A
That's
redirecting
those
requests
down
to
the
laptop,
but
the
laptops
not
running
anything.
So
that's
why
we're
seeing
errors
I'm
hop,
I've,
hopped
back
to
the
IDE
and
I'm
gonna
hit
the
debug
button
now
and
that's
gonna
run
the
code
in
the
debugger.
We
wait
for
idea
to
do
its
thing
and
hop
back
to
the
terminal.
Okay,
so
now
you
can
see
the
code
is
running
and
in
particular
you
can
see
it's
running
on
my
laptop.
The
way
you
can
see
that
is
that
the
host
name
is
now
not
some
Java
blah
blah
blah.
A
It's
chill
chill
is
the
name
of
my
laptop.
My
prior
laptop
had
some
cooling
issues.
Some
thermal
management
problems,
this
laptops
much
better,
so
I
named
a
chill
anyway.
We
can
see
that
it's
running
on
my
laptop
I've
unpause
it
again,
so
you
can
see
it's
updating
and
again.
If
we
are
patient,
we
will
see
that
the
version
of
the
code
running
on
the
laptop
has
the
same
bug
there
we
go.
I
will
highlight
this.
You
see
once
again
the
box
isn't
lined
up,
because
the
output
is
wrong.
A
Ok,
so
now,
if
we
go
to
the
IDE,
how
are
we
gonna
figure
out
why
the
box
is
wrong?
Why
why
the
output
is
wrong
right?
Let's
set
a
breakpoint
somewhere
around
here
where
we
can
see
what
that
hash
code
is
so
we've
hit
the
breakpoint
and
I
guess.
I
have
to
scroll
to
the
side
a
bit:
well,
okay,
lo
and
behold-
buy
it
by
a
matter
of
chance.
I've
hit
the
bug.
I'm
gonna
hit
continue,
and
let's
see
this
without
the
bug.
A
Okay,
so
we
see
that
the
hash
code
is
this
16
character,
string
right
what
we've
been
seeing
in
our
box
and
how's
that
hash
code
it
arrived
well
there's
a
base64
encoding
of
the
answer
to
get
value
and
that
get
value
is
returning,
salty
nine,
nine
zero
point,
seven,
okay,
and
if
we
keep
hitting
continue,
we
can
see
that
number
change.
Every
time
we
hit
the
breakpoint,
that's
nine,
nine
three
point:
four:
nine
nine
three
point:
seven
etc,
and
you
can
see
on
the
curl
loop.
The
curl
loop
is
paused
because
we've
hit
that
breakpoint.
A
The
last
answer
we
got
was
at
sixteen
seconds
passed
and
then,
if
I
hit
the
continue
button,
it
runs
through
and
pauses
again-
and
you
see
now.
The
last
thing
we
saw
was
a
twenty
nine
seconds
passed
right,
so
we
are
controlling
the
code
that
is
running
through
the
curl
loop
now
by
chance.
I've
hit
this
okay.
My
luck
is
really
good
here,
but
I'm
gonna
pretend
that
I
didn't
hit
that
by
chance
right,
we
could
say
here
hitting
the
button
and
not
hit
that
by
chance
is
theirs
with
very
high
likelihood.
A
A
So
now,
when
I
hit
continue,
it's
going
to
keep
running
until
we
hit
that
breakpoint
and
you'll
be
able
to
see
the
curl
loop
runs
and
runs
and
runs,
and
we're
going
to
have
to
be
patient,
because
that
bug
is
intermittent,
even
though
I
happen
to
hit
it
by
chance
a
few
times.
Alright,
we
seem
to
have
stopped
let's
head
over
here
and
we
see
a
look.
The
value
that
we're
getting
is
salty,
nine,
nine
nine,
it's
not
salty
nine,
nine,
nine
point
three
and
I
guess
in
this
case
it
would
have
been
salty
nine.
A
Nine
nine
point
zero,
but
something
changed
at
the
other
end
and
that's
no
longer
the
case.
So
using
our
conditional
breakpoint
we've
been
able
to
get
the
system
to
run
and
show
us
the
error.
That
is
the
value
of
the
debugger
being
available
to
you
to
run
on
the
service
that
would
otherwise
need
to
run
in
kubernetes.
Okay,
so
I've
hit
the
stop
button
and
I'm
gonna
close
out
here,
because
I
think
we
should
take
a
moment
to
figure
out
what
we
just
did:
okay
and
just
hopping
back
to
a
curl
loop.
A
A
Okay,
let's
take
a
breath
and
figure
out
what
just
happened.
Okay,
we
just
went
through
the
swap
deployment
workflow.
Essentially
we
replace
the
code
that
was
running
in
the
cluster
with
a
proxy
that
sent
every
request
down
to
the
laptop.
Then
we
ran
the
code
on
the
laptop
in
the
debugger
step
through
the
code
to
see
what
was
going
on
and
every
time
we
ran
the
code
and
let
it
run
to
completion.
A
The
result
would
in
fact
go
back
up
to
the
cluster
back
through
the
API
gateway
and
down
to
my
curl
command,
where
it
shows
up
on
my
on
my
console.
So
this
swap
deployment
workflow.
Let
me
debug
my
application
as
if
it
were
running
in
kubernetes,
but
stepping
through
it
in
my
IDE
now
swap
deployment
means
that
we
have
the
power
of
the
ID
at
our
disposal.
Sorry,
that
wasn't
what
I
meant
to
say.
A
Let's
move
on
to
more
demos,
I'm
gonna
do
another
demo,
a
similar
demo
using
node.js
and
JavaScript
okay,
so
I'm
gonna
set
up
the
same
curl
loop,
but
I'm
gonna
curl,
the
node
service.
Okay
and
I
am
going
to
hop
over
to
Visual
Studio
code,
which
does
a
great
job
with
JavaScript,
and
the
code
is
basically
the
same
thing.
We
call
get
value
to
access
the
base
service,
we
run
a
base64
encoding
on
the
value
and
then
we
emit
this
output.
Hello
cube
kannan
and
greetings
from
node
GS
and
then
the
code,
the
hash
value.
A
A
A
All
right
so
now
we
can
see
that
the
hosts
chill,
which
is
my
laptop,
is
serving
these
requests.
Thanks
to
telepresence
and
now
I
had
unintentionally
set
a
breakpoint.
Now
let
me
intentionally
set
a
breakpoint
right
here
and
we
hit
that
break
point,
and
we
can
once
again
see
well.
Okay,
I
guess
you
can't
see
it,
but
I
will
tell
you
that
this
value
that
we're
using
to
compute
the
hash
is
salty
977
wow.
It's
even
tiny
for
me
point
eight!
Okay!
Now
we
already
know
what
the
bug
is.
A
Let's
take
advantage
of
another
feature
of
debuggers
I.
Don't
have
to
wait
around
for
the
situation
of
of
the
error,
condition
happening.
I
can
just
edit
this
myself.
The
debugger
lets
me
do
that
it's
one
of
the
features
of
being
able
to
use
a
debugger,
so
I've
edited
the
value.
I
have
in
fact
edited
the
value
too
late.
The
hash
value
has
already
been
computed,
so
editing.
The
value
now
doesn't
help.
That
was
silly.
A
A
Once
again,
I
have
created
the
bad
situation
manually,
so
this
is
one
of
the
other
powers
of
the
debugger
and
it
really
lets
you
cut
through
a
lot
of
waiting
around.
If
you
once
you
figure
out
how
to
create
the
bad
situation
and
why
things
are
failing.
Okay,
so
I've
quit
out
of
debugger
same
as
before,
I'm
gonna
quit
telepresence
and
things
will
get
switched
back
to
normal
and
then,
once
again
the
curl
loop
is
going
to
be
curling
into
the
cluster
okay.
A
So
let's
talk
about
one
more
thing:
here's
another
situation
that
arises
you
have
multiple
containers.
A
common
case
of
this
is
I,
have
an
application
that
needs
to
use
a
cloud
database.
Typically,
the
way
you'd
use
a
cloud
database.
Is
you
have
a
sidecar
or
I
guess
a
proxy
container
that
connects
to
the
cloud
database
for
you
and
you
just
treat
the
container
as
a
local,
Postgres
or
local,
my
sequel
or
Maria
DB,
sorry,
whatever
it's
called
these
days.
A
So
when
you
have
an
application
like
that,
when
you
have
a
service
like
that
with
its
own
database,
sometimes
if
you're
running
into
problems
you
might
want
to
see,
what's
going
on,
can
I
just
inspect
the
database
if
it
were
a
local
application,
you
would
just
spin
up
your
database,
your
favorite
database
tool
and
just
look
at
the
connect
to
the
database
and
take
a
look
now
if
the
database
is
running
via
cloud
sequel.
Sidecar,
etc,
it's
a
little
more
complicated
to
do
that,
but
telepresence
makes
it
easy.
A
Let's
do
this
right
now,
I'm
going
I
have
a
let's
look
at
this
pod.
This
is
coop
control
get
pod
and
the
pod
is.
It
has
two
containers
and
I'm
going
to
run
a
swamp
deployment
on
that
pod,
telepresence
swap
deployment,
whoops
I'm
gonna,
swap
this
deployment
and
I'm
gonna
swap
the
app
container,
not
the
database
container,
because
the
data
could
base
container
needs
to
run
in
the
cluster
and
I'm
gonna
run
PG
CLI,
which
is
a
very
nice
console
based
Postgres
client.
Some
of
you
may
already
know
it.
A
It
gives
you
a
nice
little
tab,
completion
and
drop-down
menus
of
like
what
you
might
want
to
type
next
and
so
forth.
Now,
there's
a
problem,
it
says
connection
refused.
Is
the
server
running
on
localhost
and
accepting
TCP
connections
on
five
four
three
two?
Well,
it
certainly
is,
but
it's
running
in
the
pod
right.
So
when
my
application,
when,
when
the
code
running
on
my
laptop,
tries
to
connect
a
localhost,
this
is
not
kubernetes
networking.
A
This
is
localhost
networking
and
that's
not
going
to
reach
the
pod
automatically,
so
we
had
to
implement
a
workaround
the
flag
to
help
work
around
this
problem.
So
I
want
to
use
this
PG
CLI
to
inspect
my
database.
So
I'm
gonna
set
up
this
two
pod
flag,
which
basically
tells
telepresence
that
I
want
to
take
pipe
localhost.
A
Five,
four
three
two
and
forward
that
to
the
pod
and
there's
an
equivalent
from
pod
option
as
well
to
go
in
the
other
direction
and
telepresence
tells
you
forwarding
localhost
five,
four
three,
two
two,
the
pod,
okay
and
now
you
can
see
PG
CLI-
has
started
up.
I
can
use
all
of
my
great
commands.
This
is
a
really
nifty
program.
It
like
lets.
You
complete
all
these
things
and
so
I
ran
the
describe
tables
command
and
of
course,
there
are
no
tables,
because
once
again
conference
Wi-Fi
I,
don't
want
to
depend
on
that.
A
So
I'm
running
an
empty
post
res
container
instead,
but
this
would
work
exactly
the
same
way
with
a
cloud
sequel
container.
So
there
you
go,
I
was
able
to
use
another
one
of
my
favorite
development
tools
or
in
this
case
I
would
be
using
it
for
debugging.
My
database
and
by
using
telepresence
I,
was
able
to
access
the
database
running
in
the
cloud
or
through
the
cluster
okay.
A
So
that's
the
multiple
containers
example.
So,
let's
talk
about
swap
deployment
right.
We
have
seen
that
I
can
use
it
to
you
to
go
through
my
code,
my
service,
via
an
IDE
or
use
another
tool
like
a
database
tool
to
access
stuff
in
the
cluster,
and
this
lets
me
do
all
sorts
of
things
that
would
otherwise
be
very
difficult
with
a
kubernetes
service.
A
Now
there
is
a
trade
off
swap
deployment
is
relatively
brain-dead,
meaning
it
just
forwards
every
single
TCP
connection
down
to
you
down
to
the
laptop-
and
this
is
great
if
you
use
some
weird
custom
protocol
or
if
your
Redis
or
if
your
post
stress
yourself
and
of
course
it
works
for
HTTP,
just
fine,
but
there's
a
downside
to
this
because
telepresence
doesn't
understand
your
protocol.
It
has
to
literally
forward
every
single
connection
to
your
laptop.
A
So
if
there's
anybody
else
using
this
cluster,
using
this
application,
they're
kind
of
stuck
while
you're
using
swap
deployment,
they
can't
really
use
the
application,
because,
if
you're
stepping
through
the
debugger,
that
service
is
not
running
for
anybody
else,
so
that's
the
trade-off
with
slop
deployment.
What
if
we
could
do
better
right?
What
if
we
had
if
we
could
introduce
one
constraint,
a
significant
constraint,
but
do
better
as
a
result,
so
the
constraint
we're
going
to
introduce
is
that
your
application
has
to
accept
HTTP
requests
as
its
input
now
hcp
includes
things
like
G
RPC.
A
So
it's
not
a
super
horrible
constraint.
It's
it's
still
a
pretty
significant
constraint.
The
side
effect
is
telepresence
can
now
know
what
your
protocol
is
and
it
can
look
at
it
can
make
decisions
based
on
the
headers
in
the
request.
So
what
we're
going
to
show
here
is
intercepting
a
particular
subset
of
requests
and
sending
just
that
subset
down
to
the
laptop
and
letting
the
rest
of
the
requests
flow
to
the
pod
business
as
usual.
A
Okay,
let's
go
do
this
so
I'm
using
a
different
control
command
right
now,
called
edge,
control
and
edge
control
is
part
of
the
Ambassador
edge
stack
which
is
going
to
be
released
by
data
wire
on
December
10th
and
we're
going
to
integrate
this
with
telepresence,
so
edge
controls
shows
that
there
is
one
intercept
will
deploy.
What
that
means
is
this
deployment
has
opted
in
to
being
intercepted.
Essentially,
it
means
that
I
have
up
the
deployment.
A
A
Okay,
I'm
going
to
intercept
the
Python
deployment
I'm
going
to
name
this
intercept
dev
I'm
going
to
match-
let's
say
the
X,
dev
header
and
I'm,
going
to
look
for
X
dev
equal
to
oh
I,
don't
know
one
two
three
and
when
I
get
that
X
dev
one
two
three
header
I
want
to
redirect
that
subset
of
requests
to
port
8000
on
my
machine
okay.
So
what
did
I
just
do
I'm
going
to
list
these
intercepts
and
it
says
the
intercept
called
dev
is
intercepting
requests
to
the
Python
deployment.
A
1,
the
X
dev
header
says
1
2
3.
It
will
redirect
this
to
my
machine
port,
8000,
ok
and
let's
run
something
on
my
machine
port
8000,
so
that
it
has
somewhere
to
go
I'm
just
going
to
run
the
Python
version
of
the
same
old
service,
because
I
wrote
5
different
versions
of
this
service,
and
this
is
handy.
Ok.
A
So
if
we
go
back
to
our
curl
loop,
let's
set
up
the
curl
loop
to
hit
the
Python
deployment
and
we
see
I've
already
run
the
intercept,
but
these
curls
are
all
hitting
the
cluster
they're
hitting
Python
blah.
That's
the
pod
in
the
cluster
and
if
I
go
over
to
this
terminal,
I
can
run
the
same
curl
manually
and
we
see
it's
hitting
the
cluster
right.
Python
blah
is
the
host.
But
if
I
add
this
header,
X
dev
1
2
3,
now
the
curl
is
being
served
by
chill.
A
My
laptop
and
I
can
even
hop
over
to
this
Python
script,
and
you
can
see
there
is
a
single
request
being
logged
now,
once
again,
basic
curls
go
to
the
cluster
curls
with
a
header
go
to
my
laptop
right,
and
the
curl
loop
in
this
terminal
continues
to
go
to
the
cluster.
So
this
is
really
a
big
deal
if
you
think
it
through
right,
you
have
a
clustered,
that's
serving
a
hundred
developer,
so
that's
kind
of
silly,
a
lot
of
developers
and
they're
each
working
on
different
pieces
of
a
big
application.
A
I
can
intercept
one
or
more
deployments
and
just
use
my
dev
header
to
get
those
requests
come
down
to
my
laptop.
Meanwhile,
everybody
else
can
keep
going
on
business
as
usual.
Another
developer
might
be
adding
a
feature
and
she
can
intercept
the
same
deployment
and
get
a
different
subset
of
requests
going
to
her
laptop.
So
this
allows
you
to
do
a
much
more
powerful
set
of
things
I'm
going
to
kill
this
curl
loop
without
affecting
everybody
else
in
the
cluster.
A
It's
the
trade-off,
of
course,
is
that
you
can
only
do
this
with
HTTP
and
just
to
clarify
if
my
service
needs
to
reach
something
in
the
cluster
like
a
post
grass
or
a
Redis
or
some
other
custom
protocol.
That's
fine!
That's
no
problem!
The
point
is
for
for
for
the
intercept
to
work,
to
intercept
requests
to
come
down
to
my
laptop.
That
has
to
be
HTTP.
So
it's
a
trade-off.
It
is
a
constraint,
but
it's
not
a
horrible
one.
We
are
really
excited
to
roll
out
intercept
in
the
very
near
future.
A
December
10th
is
just
a
couple
of
weeks
away,
so
we're
reaching
the
end
here.
Let
me
summarize
so
telepresence
that's
where
we
saw
the
swap
deployment
workflow
at
the
beginning,
that
is
a
CNC
F
sandbox
project.
We
could
use
your
help,
we
could
use
pull
requests,
features,
bug,
fixes,
documentation,
everything
telepresence
do
and
obviously
we
are
on
github
and
then
the
last
thing
I
showed
is
the
intercept
workflow.
A
This
is
a
part
of
Ambassador
edge
stack
by
data
wire,
ambassador
edge
stack
is
being
released,
December
10th
and
you
can
find
out
more
on
get
ambassador
dot
IO.
Let
me
just
show
you
because
I
didn't
update
the
slide
and
that
if
you
go
to
get
Ambassador
I,
oh
there
is
a
big
banner
at
the
top.
That
tells
you
how
to
get
to
edge
stack,
but
in
any
case
so
that's
being
released,
December
10th
and
we're
going
to
have
telepresence
integration
for
the
intercept
workflow
at
that
point,
that's
about
it!
A
C
A
A
Excellent
question
so
telepresence
the
swap
deployment,
workflow,
Maps
environment
variables
and
volumes
down
to
your
laptop
now.
Volumes
are
a
little
tricky
because
it
can't
put
the
the
pods
root
filesystem
on
top
of
your
root
file
system,
so
it
mounts
under
a
subdirectory.
So
there's
a
little
bit
of
work
to
do
there.
You
might
use
sim
links
to
to
connect
things
to
the
right
place
or
you
might.
A
If
you're,
using
a
docker
run,
you
can
actually
use
volume
mounts
to
mount
things
in
the
right
place
when
you
need
them.
Environment
Muriel's
get
passed
through
if
you're,
using
an
IDE
IntelliJ
has
a
a
dot
n
file.
Format,
plug-in
and
TelePresence
will
write
an
end
file
for
you
that
you
can
bring
right
into
into
IntelliJ,
so
the
intercept
workflow
doesn't
cover
that
as
well
yet,
but
we're
still
working
on
it
next
question:
you're.
Just
adding
to
that
previous
question.
A
A
So
I
actually
left
out
an
important
thing.
Thanks
for
the
question,
so
can
we
use
something
other
than
HTTP
for
the
intercept
workflow?
The
intercept
workflow
is
powered
by
envoy
proxy.
That's
really
important.
I
should
have
said
that
I
meant
to
say
that
so
anything
that
the
aught
that
Envoy
proxy
understands
in
the
future.
We
should
be
able
to
intercept
on
that.
At
the
moment.
We
have
HTTP
working
really
well,
and
the
nice
thing
about
having
Envoy
proxy
there
is
you.
A
Can
you
can
use,
intercept
basically
without
worrying
envoy
proxy
can
handle
the
load
of
your
cluster
and
forward
the
majority
of
your
requests
through
to
the
pod
business
as
usual.
So
there's
really
no
concern
about
opting
in
to
using
the
intercept
workflow
but
yeah
it's
built
on
Envoy,
so
whatever
envoy
I
can
do,
we
can
get
there
would
intercept.
E
A
Telepresence
is
two
pieces.
There's
the
part
that
runs
on
your
laptop
and
then
there's
the
part
that
runs
in
the
cluster
and
the
way
swap
deployment
works
it.
It
doesn't
really
swap
out
your
deployment.
It
actually
scales
your
deployment
to
zero
and
makes
a
copy
deployment
that
runs
the
telepresence
proxy
code,
the
proxy
image,
so
that
part
has
to
run
in
the
cluster
where
all
of
the
dependent
services
are
so.
A
In
my
example,
here
I
had
the
Ambassador
API
gateway
and
then
the
various
services
that
we
were
working
on
and
then
the
the
upstream
service,
the
base
service
and
all
of
those
ran
in
the
same
cluster.
When
I
ran
the
swap
deployment,
then
telepresence
was
also
running
in
that
same
cluster.
That's
the
way
it
connects
to
everything
else.
Okay,.
A
So
we
we
sort
of
trick
around
that
when
we
do
a
swap
deployment,
we
we
replicate
the
deployment
and
scale
the
original
to
zero.
The
replica
deployment,
though
that
is
using
the
the
proxy
image
we
turn
off
a
bunch
of
things.
We
turn
off
things
like
liveness
and
readiness
checks.
So
if
your,
if
your
cluster
insists
on
those
I'm,
not
sure
if
you
can
set
that,
but
if
your
cluster
insists
on
those,
then
things
become
more
complicated.
A
A
Obviously,
I
just
don't
type
fast
enough
to
debug
things
in
five
seconds,
so
in
my
mappings
for
for
a
map
for
the
Ambassador
mappings
that
point
to
these
services,
I
set
the
timeout
to
something
long
or
five
minutes,
or
something
so
that
I
could
debug
so
yeah,
it's
it's
not
there,
there's
always
going
to
be
other
things
that
that
you
run
into
when
you're,
taking
a
live
running
system
and
pausing
it.
So
you
can
step
through
the
code
and
you
just
you
hit
you
run
into
them
and
you
work
through
them.
A
Well,
telepresence
does
not
require
our
ambassador
I.
It's
just
convenient
to
use
an
API
gateway
to
get
to
the
services.
Intercept
is
specifically
part
of
the
Ambassador
edge
stack.
So
when
we
really
solve
that
on
December
10th
I'll
be
honest,
I'm
not
sure
exactly
how
we're
going
to
slice
all
those
up.
So
please
keep
an
eye
out
for
that,
but
functionally
you
don't
have
to
be
running
ambassador
to
use,
intercept
so
there,
but
the
the
product
suite
is
tied
together.
It's
all
built
on
envoy,
it's
using
related
technology,
awesome
Thanks!
Thank
you.
H
A
Header
set
the
right,
cookie,
whatever
it
is,
and,
and
so
you
could,
you
could
argue
or
I
could
argue
that
that's
not
downtime,
because
it's
a
developer,
specific
request.
Another
approach
that
maybe
we
could
use
is
shadowing
right.
We
could
say
ambassador,
send
the
the
requests
to
the
real
service
but
send
a
shadow
copy
to
my
telepresence
proxy.
That
might
be
another
way.
A
We
could
do
it
in
the
future
where,
just
because
your
intercept
just
because
you're
receiving
the
request
doesn't
mean
that
the
original
service
doesn't
get
the
request,
that's
another
way
we
could
pursue
it.
That's
not
a
feature
right
now.
I
feel
like
intercept,
addresses
most
of
that
concern
because
you're
not
taking
down
the
whole
pod.
Okay,.
H
One
last
question:
yeah:
we
have
this
use
case
where
you
know
we
have
talked
to
compliance
where
developers
just
can't
see
anything
in
prod.
You
know
they
can
look
at
the
logs,
maybe
Splunk
logs,
but
other
than
that,
like
they're,
not
allowed
physical
access
to
prod.
Could
this
help
with
that
or
it's
just
a
circumvention.
A
If
I
mean,
if
you
want
to
run
one
of
these,
you
don't
want
to
run
swap
deployment
against
prod
because
it
brings
down
project
exactly
what
you
just
said
right
but
intercept
I'm,
not
gonna,
say
it's
okay
to
run,
intercept
in
fraud,
mostly
because
we
haven't
had
it
for
that
long
and
we
we
just
don't,
have
experience
right.
That's
the
honest
truth,
but
it's
just
not
been
around
long
enough
for
me
to
say
that
being
said,
it's
built
on
envoy
envoy
is
super
solid.
A
H
A
This
we
should.
We
should
talk
about
this
offline,
because
I
want
to
I,
want
to
write
out
an
example
like
a
specific
example
and
and
think
it
through
I'm,
not
it's
the
lights
in
my
face,
I'm,
not
thinking
as
well
as
I.
Usually
do
so
my
apologies,
but
let's
let's
talk
about
that
after
I'm
off
the
stage.
Please.
Thank
you
another
question.
Another.
I
A
Well,
so
telepresence
doesn't
support
UDP
at
all,
multicast
or
otherwise.
So
we're
not
there
yet
I
understand
that
envoy
is
gonna
support
UDP
in
the
very
near
future.
It
may
have
landed
already
envoy
moves
really
really
fast.
So
it's
it's
practically
a
full-time
job
to
keep
up
with
its
feature
set
but
yeah
when,
when
envoy
has
UDP
support,
telepresence
will
gain
it
yeah.
It's
it's
a
it's
something
that
people
have
asked
about
before.
At
that
point
we
can
start
thinking
about
what
do
we
do
with
multi
tasks,
but
yeah
we're
not
there
yet,
but.
J
J
So
obviously,
wouldn't
be
pure
local,
but
without
needing
to
stand
up
all
of
my
dependent
services
in
a
local
cluster
or
some
kind
of
local
instance,
and
basically
just
do
like
a
peer
greenfield
development
on
some
application
without
managing
like
deployment
resources
and
all
about.
In
my
cluster
I
I.
A
Mean
it
you,
you
could
do
that.
Telepresence
relies
on
the
existence
of
a
kubernetes
cluster
to
do
the
basic
stuff
that
it
does.
If
I
mean,
if
your
goal
is
to
write
something
that
runs
in
kubernetes,
then
telepresence
will
make
your
environment
look
kubernetes
ish,
you
don't
have
to
swap
a
deployment.
You
can
run
telepresence
with
a
new
deployment
option.
It'll
just
make
a
new
deployment
that
doesn't
do
anything.
You
can
say:
I
want
a
new
deployment
that
exposes
this
port
and
what
just
happened
well
anyway,
I'm
probably
over
time.
A
So
telepresence
will
let
you
create
the
the
kubernetes
effect
without
having
a
deployment
to
swap
I.
Think
that
answers
the
question.
I
was
running
these
examples
and
a
local
cluster
to
avoid
conference
Wi-Fi
but
yeah.
This
is
typically
something
you
would
run
against
your
like
your
developer
cluster,
that's
running
in
GK,
or
something
like
that
on
Prem.
Whatever
last
questions
that
I
think
I
am
everything
changed,
so
I
suspect,
I'm
being
kicked
out.
Yeah.