►
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
Deep Dive: NATS - Colin Sullivan & Wally Quevedo, Synadia
The NATS project and its ecosystem have been continuously evolving since joining the CNCF and in this session we will share a retrospective of what is the current state of the art and overall direction of the project in terms of security, performance, reliability, and integration with other cloud native tools such as Kubernetes. If are you actively evaluating NATS and want to learn the latest best practices, are writing clients for it, production user or interested in becoming a contributor, then you should definitely attend. For those new to NATS, we suggest also attending the NATS introduction session.
To learn more: https://sched.co/Grdm
A
Hello:
everyone
thank
you
for
attending
the
deep
dive
for
Nats.
My
name
is
Colin
Sullivan
I'm,
product
manager,
Nats,
here's
wall,
equal
Vado,
engineer
and
just
to
show
of
hands
who's
already
using
Nats.
Today,
all
right
good
good.
What
about
in
production
all
right?
Ok,
so
I'll!
Let
Wally
go
ahead
and
get
started
with
the
deep
dive
I'll
chime
in
here
and
there
and
then
at
the
end.
We
actually
have
Derrick
here
creator
of
Nats
to
answer
questions.
B
Ok,
thank
you
for
attending
the
deep
dives
nets
for
this
about
the
evolution
of
the
Nets
project.
To
give
you
a
brief
history
of
the
Nets
project,
it's
an
8
year
old,
probably
already
first
comment
was
seen
to
in
2010.
First
release
was
part
of
cloud.
Foundry
in
2011
got
open
source
with
the
open
sourcing
of
Cloud
Foundry.
Then
he
got
real
reasoning
to
go
in
2012
and
so
in
October
you
reach
the
v1
version
of
the
nat
server
in
2017,
and
finally,
this
year
we
joined
the
CN
CF
in
March
2018
as
incubation
project.
B
So
2018
is
a
really
big
year
from
that.
If
you
look
out
or
get
history,
you
can
see
that
there
have
been
significant
contributions
and
in
today
to
the
server
and
the
because
a
spike
of
commits
and
from
several
paths
PRS
are
landing
into
master.
So
the
project
evolved
quite
a
bit
this
year
and
there
were
three
different
releases
for
the
v1
series
of
the
earth.
Nats
we
had
releases
of
features
such
as
a
graceful
disconnection
from
the
clients
that
we
call
it
the
brain
mode.
B
The
no-neck
annoy
commode
to
stop
receiving
your
own
messages
and
your
subscriptions.
There
were
enhancement
to
the
permissions
such
as
allow
and
deny
hot
configuration
reload
by
sending
the
hop
signal.
Major,
an
internal
overhaul
with
the
outbound
architecture
for
out
performance
improvements
and
also
improvements,
the
sublist
cash
and
for
the
v2,
which
is
the
week.
B
B
And
it's
a
it's
so
that
we
can
achieve
the
vision
that
we
have
of
scenario
phone
for
nuts,
so
that
as
a
global
utility
to
be
able
to
connect
devices
and
services.
And
this
is
what
the
maintain
is
have
been
up
to
all
of
this
year.
So
in
this
talk,
I'm
going
to
share
about
the
latest
features
from
that
they're
coming
soon.
The
next
release
give
us
some
demos
of
these
features
and
we
leave
some
time
at
the
end
for
the
a
Q&A
questions.
B
So
first
want
to
start
with
the
graceful
and
disconnect
which
is
the
drain,
both
these
clients
site
enhancement.
There
is
no
changes
for
to
the
server
for
this,
so
this
allowed.
You
is
a
new
API
call
from
the
clients
it's
called
drain
that
allows
to
the
clients
to
be
able
to
detach
from
the
cluster
gracefully
and
usually
before
drain.
B
You
would
use
the
close
connection
to
try
to
flush
all
the
internal
buffers
from
the
server
for
the
client,
and
that
would
then,
after
that,
you
would
terminate
the
TCP
connection,
but
that
would
also
cause
dropping
some
of
the
messages.
So
the
way
that
drain
works
is
that
it
announces
to
a
server
that
is
no
longer
interested
into
the
subjects
and
but
it
still
gives
me
some
time
to
process
the
messages
that
have
already
been
delivered
by
the
server
to
the
client,
and
it
is,
they
were
not
changed
to.
B
The
client
protocol
is
using
more
of
the
protocol
using
the
unsub
protocol
to
share
announcer
the
server
that
to
stop
at
unnecessary
to
stop
delivering
messages
on
a
certain
subject,
and
this
you're
familiar
with
the
reconnection
logic
from
Nats.
There
are
closed.
It's
a
terminal
state
before
I.
If,
before
reaching
the
closest
state,
which
is
like
the
client
will
give
up
trying
to
connect
to
a
server,
it
will
try
to
reconnect
a
number
of
times
flipping
from
between
the
reconnecting
and
disconnected
states
and
also
with
the
drained
mode.
B
This
means
that
there
is
a
new
state
that
is
called
the
draining
state
during
this
draining
state.
It's
also,
the
client
will
reach
the
terminal
state
of
the
close
connection,
but
it
will
process
any
in-flight
messages.
So
the
way
this
work
is
by,
let's
say
from
another
client
delivering
a
message
this
to
another
client.
It
could
be
a
queue
subscription,
for
example,
the
client
that
is
going
to
become
a
drain
announced
to
the
server
via
unsub
that
to
stop
receiving
messages
and
after
it
it
finishes,
processing
all
those
messages.
B
A
This
is
the
the
primary
use
case
is
for
a
really
smooth
transition
when
you're
using
load
balance
cue
subscribers,
so
a
load
balance.
Cue
subscriber
has
your
subscriber
as
part
of
a
group
and
the
server
randomly
sends
messages
between
each
of
these
subscribers
providing
a
layer,
seven
load
balancer
in
the
past.
When
you
want
it
to
downscale,
you
had
to
turn
one
of
these
subscribers
off.
You
disconnected
any
messages
that
were
in
flight
could
be
lost,
causing
your
application.
A
B
Something
important
highlight
here
is
that
use
the
drain
mouth
works
as
synchronously,
so
in
case
you
want
to
exit
the
process
as
soon
as
the
connection
is
closed,
you
still
have
to
use
the
closed
Handler
and
that
and
and
once
it
reached
to
the
tag
that
has
reached
the
closest
state
and
from
there
you
can
have
the
logic
to
either
recreate
the
connection
and
maybe
just
exit.
The
process.
B
Here
is
an
example
of
the
top-left
corner.
There
is
a
clients
and
it
may
be
requests
per
second,
and
we
have
send
the
termination
signal
to
one
of
the
few
subscribers
that
was
responding
to
that
subject
and
because
we
sent
a
termination
signal
which
causes
a
sense
closed,
called
closing
that,
for
that
cue
subscriber,
then
some
of
the
requests
were
dropped
and
you
can
see
in
comparison.
B
We
using
drain
we'll
be
sending
a
different
signal
quit,
which
is
what
I
defined
in
the
handler
to
the
same
cue
subscriber,
and
you
can
see
that
it
trapped
the
signal,
but
there
were
no
anyway,
there
were
any
any
requests
and
dropped
or
whatsoever.
So
all
the
requests
were
processed
and
then
the
process
exited
from
the
server
side.
There
is
also
an
enhancement
to
be
able
to
gracefully
shutdown
note.
This
is
the
lame-duck
mode.
B
These
can
be
triggered
by
sending
the
user
to
signal
that
this
will
make
the
server
slowly
disconnect
the
clients
that
were
connecting
to
it.
This
helps
avoided
Authority,
thundering,
hurt
issue
by
slowly
migrating
the
traffic
from
the
server
to
lend
the
clients
to
they
reconnect
to
another
one.
So
yeah.
A
One
issue
we
ran
into
and
we
heard
of
people
running
into,
is
when
you
have
tens
of
thousands
of
connections.
There
are
hundreds
of
thousands
of
connections
in
there
using
TLS.
That
TLS
handshake
is,
is
incredibly
CPU
intensive
because
you're
doing
cryptography
right.
So
what
would
happen?
Is
somebody
would
shut
down
an
ad
server
and
all
the
clients
would
go
over
to
a
new
server
and
that
server
would
get
CPU
bound?
A
B
In
these
demand,
you
can
see
that
again
there
are
a
couple
of
you:
subscribing
receiving
the
requests
from
a
number
I,
think
twenty
requesters
and
by
sending
user
to
signal
to
the
server
which
is
the
seed
server.
In
this
case,
the
other
server
was
discovered
and
the
clients
start
to
reconnect
to
the
other
one,
but
the
one
on
the
corner
they
claim
the
corner
is
not
dropping
any
requests.
Then
everything
with
well
one
request
at
the
end,
but
ok
very
decreases
the
failed
request.
A
lot.
B
Another
significant
feature
that
made
into
the
Nats
version
to
each
version
is
at
the
gateway
is
a
good
gateway
which
allows
you
to
create
super
clusters.
This
is
allows
you
to
create
a
global
and
that's
network
based
on
renoble
spline
based
technology.
Make
you
subscribers
are
aware,
by
using
the
round-trip
time
to
be
able
to
know
the
coordinates
of
whether
they
should
be
the
ones
to
handle
our
requests.
Yeah.
This.
A
Is
actually
a
really
cool
feature,
so
you
network
these
clusters
together
using
our
creating
super
clusters
internally,
we
call
them
gateways
and
you
can
have
skew
subscribers,
so
we
just
demonstrated
the
cue
subscribers
that,
where
the
server
will
randomly
send
messages
to
either
subscriber,
however
many
subscribers
you
have
well
now,
if
they're
on
different
gateways,
you
have
a
preference
for
the
local
cue
subscriber.
So
that
means,
if
I
have
a
service,
that's
performing
something
for
me
and
it's
local.
A
Let's
say
I'm
in
Japan
and
I'm,
connected
to
AWS,
it'll
use
this
Singapore
service
and
if
something's
and
let's
say
I've
got
another
one
in
New
York.
If
I
have
an
application
in
New,
York
it'll
use
the
New
York
service,
but
if
any
one
of
those
goes
down
so
let's
say
the
Singapore
service
goes
down,
then
the
application
in
Japan
will
start
using
the
service
in
New
York
as
a
backup.
B
We
define
in
five
two
two
two
and
just
to
give
you
a
demo
of
the
basic
pub/sub
functionality
using
gateways
at
the
top.
We
have
the
one
in
Amsterdam
and
telnet
client
connecting
another
to
New
York
at
the
bottom
is
Bangalore.
There
are
two
subscriptions
of
the
hello
subject.
Then
the
client
from
Bangalore
at
the
bottom
is
sending
a
hello
world
and
is
received
by
the
other
two
clusters
in
the
remote
and
data
centers.
B
You
can
see
the
public
IPS
that
are
advertised
on
from
the
gossip
are
different
in
the
three
examples
and
to
show
us
well
the
way
that
you
subscribers
work
on
their
decent
apology.
We
have
a
couple
of
subscribers
cue
subscribers
that
are
connected
to
that
Amsterdam
data
center
they're
receiving
the
request
from
another
client
that
is
also
connected
to
the
Amster
and
data
center.
The
one
at
the
bottom
is
connected
to
the
New
York
cluster,
so
in
case
we
kill
one
of
the
cue
subscribers.
B
A
So
accounts
are
they're
a
great
way
to
isolate
your
communications
by
organization.
You
can
do
it
by
geography
by
organization.
However,
you
want
to
set
it
up,
but
if
you
have
two
clients
that
are
connected
using
credential
in
the
same
account
they
can
communicate
but
to
clients
using
even
the
same
name
of
the
credentials
and
different
accounts
won't
be
able
to
communicate.
So
this
introduces
multi-tenancy
in
Nats.
A
It
allows
you
to
to
bifurcate
your
technology
from
the
business
driven
use
case
in
the
old
days
you
had
data
silos
and
that
was
driven
by
software
requirements
and
hardware
requirements
that
no
longer
exists
in
the
cloud,
but
you
still
want
to
sometimes
isolate
data
right.
You
want
to
isolate
your
clients.
A
This
provides
a
secure
way
to
do
that,
and
it
also
makes
it
easier
on
operations
because
they
only
have
to
maintain
one
Nats
deployment,
not
a
gnat's
deployment
for
every
context
you
want
to
isolate
so
accounts
are
also
decentralized
in
that
you
can
have
administrators
for
different
accounts
operating
on
the
accounts
and
the
accounts
can
share
data,
but
only
when
the
two
accounts
mutually
agree
that
data
can
be
shared.
So
we
have
secured
streams
and
services,
you
can
export
streams
and
services.
A
service
is
a
request
reply.
So
that's
your
your
certificate
service.
A
That's
maybe
where's
my
my
geo,
where
some
what's
my
IP,
whatever
you
can
think
of,
but
that's
the
communication
that
goes
back
and
forth.
We
have
streams
that
are
a
stream
of
data
and
what
happens
is
when
an
account
exports,
a
stream
and
another
account
imports
a
stream.
Only
then
will
the
data
flow
and
that
allows
you
to
decentralize
the
administration
of
accounts,
so.
B
A
B
It's
a
couple
of
snippets
of
how
the
crossing
protocol
looks
like
because
he
now
the
what
used
to
be
with
an
Italian
way,
accounta
becomes
a
part
of
a
global
account
and
the
broccolis
are
quite
different.
So
yet
we
have
the
services
and
extreme
definitions,
services
at
RPC
endpoints
as
convention
and
the
extremes
just
and
data
flow
between
accounts
and
but
again
from
the
client
perspective
you
don't
have
to.
There
are
no
API
changes
required
to.
A
B
B
Password
here
is
how
we
are
defining
that
the
sign
area
account
is
exporting
a
stream
name
cloud
network
status
and
also
service.
That
is
private,
though
they
have
stats,
and
it's
churning
to
the
sharing
that,
with
the
scenes
of
CN
CF
account,
then
there
is
a
CN
CF
account
that
has
a
couple
of
users,
Alice
and
Bob
as
a
password
be
cryptid,
and
it
is
important
from
this
an
ad
account
a
subject
called
cloud
network
status
and
is
mounting
that
on
under
design
area,
dot,
streams
prefix.
B
So
you
will
have
to
publish
to
the
senator
and
dot
streams,
total
network
status
to
receive
the
messages.
Then
there
is
an
import
for
service
called
private
dot,
dev
stats-
and
it's
not
in
that
on
this
sign
area
services,
a
dot,
dev
stats,
something
very
cool
about
the
way
that
this
data
sharing
works.
Is
that
the
request
response
is
completely
anonymous,
so
you're
familiar
with
the
way
that
the
request
response
works
in
Nets,
it
uses
a
female
subjects
and
you
need
their
identifiers
within
the
same,
a
single
account.
B
So
you
can
still
can
see
the
inbox
and
the
unique
identifier
for
that
particular
request.
But
once
that
our
request
traverses
to
another
subject
and
other
account
subject:
namespace
it
becomes
transformed
and
anonymized
into
a
different
subject,
so
that
the
receiver
of
the
original
request.
It
is
not
aware
of
what
it
was
the
original
subject
that
was
used
for
making
the
request
so
and
that
really
helps
with
the
isolating
and
the
subjects
a
different
and
unique
identifier.
B
A
So
yeah
a
couple
of
the
new
features.
We
have
our
n
keys
and
we
alluded
to
this
previously
at
coop
con
and
earlier
this
year,
but
we've
got
n
keys,
it's
an
ATS
identity.
It's
an
e
d,
two
five,
five
one,
nine
key!
That's
made
easy.
Eighty,
two,
five,
nine
four
one:
nine
keys
are
really
fast,
they're
very
resistant
to
side-channel
attacks,
and
then
you
can
sign
and
verify
with
them.
A
So
we've
set
up
the
net
server
to
be
able
to
accept
so
we've
set
up
the
net
server
so
that
when
a
client
connects
it
sends
a
nonce
which
is
random
series
of
bytes
that
are
salted
and
what
the
client
does
is
sign.
This
nonce,
with
its
private
key,
sends
it
back
to
the
server
and
the
server
verifies
it
with
the
clients
public
key.
What's
really
cool
about
this
is
the
NAT
system
never
has
to
access
or
store
your
private
keys
ever,
which
is
which,
which
gives
you
many
advantages
in
administrating,
a
zero
truss
system.
B
And
this
is
an
example
of
the
flow
of
authenticating
using
the
N
keys
and
JWT
functionality.
Now
you
can
see
that
in
the
info
protocol
you
get
unknowns
and
the
client
library
then
like
this
knowns,
and
is
you
see
it's
going
to
use
one
of
its
private
and
n
keys
to
sign
and
send
us
signatures
passed
as
part
of
the
connect,
payload
and
there's
also
a
JWT
that
is
passed
to
the
server
in
unit
2
for
its
identity?
So
the
gel
abilities?
B
B
B
B
A
Year
are
we've
gotten
a
lot
of
requests,
as
people
are
moving
to
the
cloud
to
integrate
with
other
messaging
solutions,
so
we've
got
someone,
that's
really
interested
in
us
developing
something
with
them
q-series
and
more
people
that
are
interested
in
us
developing
some
sort
of
bridge
with
Kafka.
So
that's
something
to
look
forward
to
we'll
have
data
at
rest,
encryption
and
NAT
streaming.
We've
got
project
jet
stream,
which
is
net
streaming
version.
2
we've
taken
a
lot
of
the
things
that
we've
we've
seen.
A
People
run
into
using
NAT,
streaming
and
and
other
streaming
products
as
well,
and
we're
learning
from
that
and
we're
developing
jet
stream.
We're
going
to
have
native
MQTT
support
coming
up.
Mqtt
is
ubiquitous,
it's
it's
everywhere,
so,
Nats
being
so
lightweight
and
in
the
NAT
server
can
even
run
on
a
Raspberry.
Pi
makes
sense
that
we
would
integrate
with
that
protocol,
so
we're
gonna
have
native
support
in
the
NAT
server
for
MQTT.
B
A
C
C
Our
first
attempt
was
good,
I
think,
but
I
think
we
can
do
a
lot
better
and
keep
that
philosophical
component
being
very
easy.
You
know
just
kind
of
work
so,
even
though
the
encryption
stuff
and
the
authentication
stuff
that
Wally
and
Colin
were
talking
about,
you
know
the
line
in
your
code
that
you're
writing.
It
says
Nats,
connect
and
usually
say
URL
now
it
says:
Nats
connect,
URL,
slash
user
credentials
and
points
it
to
a
file,
and
that's
it,
but
what's
happening
underneath
the
covers
is,
is
that
we
dynamically
create
callbacks
that
load.
C
These
things
wipe
the
memory
every
time
you
connect
and
reconnect.
So
we're
trying
to
make
a
lot
of
these
things
very
sophisticated,
but
easy
from
a
developer
and
an
operation
standpoint.
We
care
deeply
about
our
users
and
our
customers,
so
you're
not
going
to
see
a
that
streams
dead.
You
know
you
know
move
over
type
stuff,
but
we
do
believe
that
there's
some
compelling
things
in
jet
stream
and
I've
told
the
team
that
I
don't
want
to
be
handcuffed
by
Stan
Stan's.
C
C
So
the
the
question
is
the
decoupling
between
producers
and
consumers
and
the
notion
that
Nats
protects
itself
at
all
costs
right.
It's
a
very
resilient
system
in
that
it's
very
selfish
about
itself,
which
actually
is
a
really
good
tenant.
From
my
perspective
of
predictable
distribute
systems,
does
the
supercluster
technology
kind
of
help
that
so
natchez
is
an
overlaid
technology?
C
So
back
in
the
90s,
there
was
a
transition
from
products
that
used
to
work
on
that
tip,
co
and
Colin,
and
others
from
the
team
were
there
as
well
from
UDP
broadcast,
unicast
and
multicast
to
tcp/ip,
and
the
reason
was
is
that
inside
of
the
network
elements
when
you
actually
looked
inside
of
like
a
Cisco
router,
everything
was
outside
the
fast
path
except
tcp/ip.
Alright,
so
you've
seen
a
lot
of
these
overlay
networks,
cuz
they
actually
perform
better
or
less.
C
You
have
a
massive
fan
out
and
multicast
is
a
is
an
administration
nightmare,
so
that's
as
an
overlay
network.
So
if
you
have
a
publishing
rate,
that's
coming
inbound
and
the
consumer
can't
keep
up.
It
doesn't
matter
how
many
links
hops
and
everything
in
between
it's
just
gonna,
we're
gonna
cut
it
off
now,
when
you
look
at
how
the
server's
actually
fan
out
right,
superclusters
and
gateways
obviously
allow
you
to
have
more
components
working
very
efficiently.
What
superclusters
were
specifically
designed
for
was
a
global
scale
technology,
where
round-trip
time
and
interest
graph
propagation
were
optimized.
C
So
right
now,
clusters
propagate
all
interest
graph,
meaning
I,
say
I'm.
It's
leaning
listening
to
foo
everyone
in
that
lo
small
cluster
knows
about
that
in
gateway
technologies.
We
actually
don't
share
that
right,
because
you
could
have
millions
of
accounts
and
your
accounts
over
on
the
west
coast
and
someone
else
on
the
east
coast
and
if
we're
just
going
back
and
forth
nonstop
telling
each
other
about
what
you
guys
are
interested
in,
but
you'll
never
talk
to
each
other.
C
It
doesn't
make
a
lot
of
sense,
and
so
it
has
a
spline
based
technology
where
it
has
inbound
outbound
connections
per
server.
So
it's
a
full
mesh
graph,
but
it
actually
uses
a
lot
less
socket
connections
and
it
doesn't
have
a
single
point
of
failure
where
there's
a
one
selected
leader
in
the
cluster
to
get
to
the
other
side.
C
And
what
we'll
do
is
we
do
optimistic
sends
and
then
the
other
side
can
prune
very
quickly
based
on
account
interest
or
subject
interest,
and
it
also
has
the
ability
to
when
we
get
into
a
degenerative
case
where
the
pruning
is
actually
generating
more
traffic
than
if
we
just
told
you
what
we
are
interested
in.
We
can
switch
to
perfect
knowledge
mode
on
the
fly.
C
At
the
time
that
said,
oh,
the
World,
Wide
Web
is
going
to
be
greatest
thing
since
sliced
bread
and
I
was
one
of
those
people
had
no
way
of
understanding
that
we
would
be
here
today
and
the
opportunities
that
existed
just
by
being
connected.
That's
essentially,
all
that
happened.
They
were
just
being
connected
and
we've
got
about
4.5
billion
people
online,
maybe
about
5.8
with.
If
you
look
at
2g
type
cell
connections
and
stuff.
C
We
have
to
have
isolation,
but
then
we
want
to
promote
secure
sharing
because
we
think
the
opportunities
for
that
are
just
going
to
be
mind-boggling,
things
that
we
can't
even
understand,
but
we've
seen
some
amazing
messaging
and
communication
technologies,
but
they
always
exist
in
silos,
even
in
companies
that
say
we're
all
in
on
blank
whatever
that
is
you
go
inside
and
you
go.
How
many
blanks
do
you?
Have
they
go?
Well,
forty.
C
You
know
we
go
sidle
over
here,
it's
kind
of
like
the
database
thing
and
so
we're
trying
to
promote
this
notion
of
always
on
digital
dial
tone
geo,
where
single
URL,
every
cloud
provider
edge
technology,
all
the
way
down
to
the
smallest
type
endpoint
devices.
That's
just
always
there
and
it
always
works
and
it's
a
trusted
utility.
It's
a
global
communication
utility
for
anything.
C
That's
digital
and
that's
really
ambitious,
I
understand
that,
but
that's
okay,
you
know
I
think
we,
the
team's,
been
working
really
hard
for
last
year
we
turned
the
lights
on
we're
all
tired
and
crossing
our
fingers,
but
we
know
it's
just
the
beginning,
but
we
do
believe
these
opportunities.
We're
also
there's
so
many
resources
that
you
can
connect
to
and
either
have
streams
that
you'd
want
to
share
out
or
services
right
where
you
will
take
requests
from
the
outside.
C
C
So
the
the
question
was
mqtt
I'll
support
in
the
server
and
Colin
actually
went
over
it
a
little
bit.
It
was
on
the
slide.
Yes,
we
are
committed
to
it.
We
believe
that
the
opportunity
for
connecting
everything
exists
further
out
towards
the
edge,
then
more
going
back
into
the
clouds,
not
saying
we're
not
doing
that.
We
just
did
that.
We
just
launched
the
product
actually
g8
it
today,
so
it's
open,
but
we
believe
that
the
massive
opportunity
exists
there
for
telemetry
sensor
type
data,
command-and-control
eventing
things
like
that.
C
C
We
will
allow
native
MQTT
clients
my
game
plan
right
now
is
that
go
all
the
way
up
to
5.0
SN
version,
which
has
the
compression
tricks
and
things
like
that
for
the
sensor
networks,
but
also
the
3/3
clients
which
are
the
most
prevalent,
can
connect
directly
into
g
net
ste
server
and
have
it
work
seamlessly
between
itself
and
then
all
the
other
nats
clients,
and
so
we're
we're
excited
about
that.
We're
going
to
be
pushing
on
that.
It's
no
small
feat,
and
it
also.
If
you
are
aware
of
MQTT,
they
have
multiple
QoS
levels
right.
C
So
we're
going
to
write
an
extremely
lightweight
and
performant
micro,
Python
client
for
Nats
that
we're
going
to
try
to
embed
into
that
and
hopefully
that'll
be
coming
soon
and
will
be
embedded
on
microcontrollers
that
the
people
want
to.
You
know
enable
so
not
sure
when,
but
we
are
going
to
do
it,
it
will
be
next
year.
Sometime
I
hope
my
hope
is
its
first
half
I
think
we
can
do
that,
but
I'm
always
optimistic
and
team
always
tells
me
I.
Usually
you
know
too
optimistic
any
others.
Yes,.
C
I
think
we
actually
do.
We
have
a
code
of
conduct
and
we
have
a
governance
piece.
That's
an
a
separate
repository,
that's
like
an
umbrella
repository
for
all
of
it,
because
we
have
a
lot
of
repos
with
clients
and
servers
stuff.
So
absolutely
I'm,
I'm,
not
a
huge
fan
of
the
clas,
so
I
think
we
CN
CF
allows
you
do
a
CLA
or
sign
it,
and
you
know
types
of
things.
So,
if
you
just
say
s
on
your
git
commit
we're
good
from
from
our
perspective,
we
were
before
considering
joining
the
CN
CF.
C
C
C
So
the
question
is
with
the
global
communications
utility:
what
is
the
relationship
with
the
telephony
operator
around
the
world
such
right?
So
right
now
Nats
is
all
it
requires
is
tcp/ip
in
the
future.
We
are
looking
very
closely
at
the
development
of
quick
into
HTTP
3,
which
essentially
moves
away
from
tcp/ip
back
into
UDP
UDP
unicast,
which
is
back
a
long
time
ago,
as
things
I
did
reliable,
unicast
on
UDP
for
TIBCO,
so
I.
Remember
that
fondly
so
what
I
mean
by
that
is.
Is
that
doesn't
matter
the
substrate
underneath
the
covers?
C
So
that
being
said,
NGS,
which
is
the
global
utility
that
we
launched.
We
are
currently
the
only
operator,
but
by
design
it's
supposed
to
be
a
federated
model,
meaning
that
we
can
easily
see
other
people
coming
in,
whether
it
be
cloud
providers
or
telecom
operators
and
say
we
want
to
run
some
servers
that
serve
our
customers
very
well.
But
the
goal
from
the
original
design
was
is
that
you,
as
a
user,
can
connect
to
any
servers
an
ad,
a
server
verizon
Google,
whoever
might
be
running
those
in
the
future
and
your
system
just
works
right.
C
We
don't
want
to
have
to
go
through
the
ATT
Verizon.
You
know
not
being
able
to
talk
together
way
back
when
when
it
first
came
out
so
so
right
now
we
have
no
bearing
on
the
physical
substrate.
That
being
said,
we
are
since
we're
looking
at
more
of
the
very
low
level
microcontroller
sensor
type
networks
where
they
might
be
battery-powered
and
stuff.
They
don't
want
you
setting
up
massive
tcp/ip
connections
and
draining
the
battery,
so
I've
been
investigating
stuff.
C
I,
don't
know,
I,
don't
have
any
answers,
you
know,
but
we
looking
at
short,
throw
graph
technology
with
the
blue
to
stuff
flora.
You
know
those
types
of
things
and
so
we're
going
to
kind
of
try
to
figure
our
way
through
that,
but
right
now
there's
no
physical
substrate
requirements
for
Nats,
just
a
tcp/ip
stack
sure.
Yes,.
C
What
do
you
have
now
and
we're
starting
to
see
companies
wanting
to
do
mixes,
so
the
pendulum
seems
to
always
swing
between
best
debris
to
all-in-one
and
so
we're
getting
a
lot
of
people
saying
hey?
Could
you
coexist
with
an
MQ
series?
Could
you
coexist
with
a
Kafka
and
the
reason
I
think
people
like
that
is
because
Nats
is
not
only
easy
to
develop
against
it's
extremely
easy
to
operate.
One
of
our
core
tendencies
is
that
itself
heals
itself.
C
It
puts
itself
back
together
it
gossips
to
the
clients
where
the
new
servers
are,
so
they
don't
have
to
be
relying
on
DNS
service,
and
you
know
another
type
of
IP
discovery
stuff.
So
a
little
towards
the
edges,
you
have,
these
very
large
companies
do
want
to
move
faster
and
they
have
requirements
that
kind
of
look
like
those
tech,
but
if
they
feel
almost
like
before
cloud
technology
where
they
put
in
a
ticket-
and
they
have
to
wait
for
weeks
before
they
get
anything
to
move
forward.
C
So
the
last
time
we
were
in
New
York,
we
said:
well,
we
just
set
up
right
now
and
I
think
we
had
30
minutes
of
problems
with
the
firewall
to
get
in
and
out
of
just
the
company.
Once
I
was
done,
it
was
about
four
minutes
and
their
whole
system
was
set
up
for
them.
So
people
like
that
and
like
that
they
can
experiment
play
and
then
they
realized
that
they
want
to.
They
can
bet
on
it
in
production,
I
mean
that's,
it's
not
perfect
software,
but
it's
pretty
darn
good
I
mean
it.
C
It
runs,
it
does
what
it
says
it
does
and
so
we're
seeing
a
lot
of
people
approach
us
on
the
and
story,
not
an
or
story.
They
just
say
hey.
We
have
this,
can
you
interoperate
with
that?
And
so
we
put
together
a
Kafka
bridge,
we'll
open
source
it.
It
was
just
kind
of
a
throw
together
POC
for
the
the
meeting
in
New
York,
but
anything
we
do
unless
a
client
says
it
can't
be
open
sourced.
C
C
C
So
what
we're
trying
to
make
then
that
server
be,
is
so
extremely
light
and
performant
and
you
can
have
as
many
as
you
want
all
the
way
out
to
the
edges
that
it's
almost
like
a
network
element,
meaning
they
exist,
but
you
don't
even
know
they
exist
there
right
to
answer
your
question
directly.
We
have
not
heard
of
anything
when
people
have
come
to
us
saying
please
interoperate
with
this,
whether
it
be
0,
mq
or
whatever
those
things
are
I've
had
a
couple.
People
mention
0
mq.
C
My
big
thing
is
when
Colin
was
talking
about
the
accounts
and
the
metaphor
of
containers
for
your
communication
space.
We
literally
can
pick
those
up
and
move
them
anywhere.
So
the
whole
notion
of
subject
based
addressing,
which
is
what
we
do
with
pub/sub
is-
is
that
there's
no
ties
diapies,
none
whatsoever.
C
If
you're
doing
point-to-point
right,
someone
has
to
tell
you
where
that
other
end
point
is,
and
you
have
to
connect
to
that
thing
right,
whether
it's
tcp/ip
or
unicast
or
whatever,
and
we
believe
the
world
is
moving
such
that
IPS
are
just
simply
a
mechanism
to
do
communication
that
should
be
treated
as
ephemeral.
I
grew
up
when
it
wasn't.
You
know
I
walk
into
any
bank
in
Wall
Street
and
go
my
stuff's
running
on
that
rack
down
on
that
floor.
You
know
that
just
doesn't
seem
to
exist
anymore
and
the
more
people
fight
against
it.
C
It
makes
it
harder
and
harder
to
design
some
of
these
newer
systems.
I
I
think
so
like
with
NGS.
We
struggled
with
you
know.
How
do
we
mix
the
the
older
world
from
the
new
world
to
bootstrap
this?
So
we
have
a
single
DNS
URL
right
that
filters
itself
geographically,
and
then
it
pins
you
to
the
cloud
provider
that
you're
coming
from,
because
it's
a
religious
thing
you
don't
want
to
be
coming
from
Google
and
be
blacked
into
a
you
know,
Amazon
or
whatever.
C
C
So
even
as
we're
rolling
up
our
own
system
and
doing
rolling
updates,
the
clients
are
automatically
informed
of
all
the
new
set
as
the
old
sets
going
away
and
if
DNS,
hiccups
or
you're
in
a
hotel
where
the
DNS
cache
says
that
we'll
do
our
own
rules
we'll
hold
onto
it
for
eight
hours,
it
won't
matter
right
we're
trying
to
be
resilient
against
that
as
well.
So
is
there
any
specific
technology,
I
mean
there's
some
good
ones
for
sure.
I
mean
I,
know
a
bunch
of
those
people.
C
C
A
C
Sure
sure,
so
the
question
is:
how
does
Nats
fit
into
kind
of
the
the
microservices
talks
around
this
conference
and
things
like
that?
I
have
a
lot
of
opinions
on
that
for
people
who
have
embraced
Nats
and
really
looked
at
it
as
something
it
has
three
simple
patterns
all
right:
it
has
pub/sub
request,
reply
and
then
load
balancing
where
it's
totally
no
configuration
at
all
right.
Just
it
works
it
just
figures
itself
out.
C
I
was
stuck
for
so
long
doing,
HTTP,
where
I
had
to
know
the
other
end
point
and
then
I
had
a
sidecar
and
then
I
had
another
sidecar
for
that
side
car
then
we
had
to
do
security,
so
then
we
did
this
and
for
TLS
certs,
and
then
you
know
next
thing
you
know
you've
got
a
PhD
project
just
to
do
a
communication
thing
between
two
things
that
might
move,
and
we
just
don't
think
it
should
be
that
hard,
I,
really
don't,
and
so
every
time
people
are
like.
Oh,
we
just
need
G
RPC.
C
You
just
need
this
I,
always
kind
of
give.
The
same
example,
I
said:
okay
developer
mindset
usually
is
in
requests.
Reply,
I'm
gonna,
send
a
message
to
him.
He's
gonna.
Send
me
a
message
back,
even
if
that
layer,
Nats
write
abstract
sout,
where
he
is
where
I
am.
If
there's
one
of
him
10
ahem,
don't
care.
What's
interesting
to
me
is:
is
that
people
kind
of
forget
a
little
bit
about
okay
well
day
two,
when
we
go
into
production,
what
happens?
Well,
yeah
an
Allison
I
need
to
actually
load
balance
him.
C
Oh
well,
we'll
stick
a
log
here
and
then
we're
gonna
flip
the
law
of
Diwali
and
now
all
these
five
minutes
behind,
but
we
got
to
figure
it
out
in
30
seconds,
or
else
we
lose
ten
million
dollars
and
I'm
going
just
stop
sticking
that's
in
I
mean
I
really
do
believe
that
and
I
believe
that
it
was
just
pub
sub-systems
pub
subsystem
with
Nets
Colin.
Never
changes,
I,
never
change.
Opps
goes
wow,
we're
slowing
down,
let's
crank
up
ten
more
of
him
hey,
we
need
compliance.
Great
Wally
come
in
we've,
never
even
stopped.
C
We've
not
been
rican.
Bob
we've
not
been
turned
off.
We
just
keep
doing
our
own
thing.
I
just
think.
It's
such
a
powerful
pattern
for
our
world.
That's
getting
more
and
more
complex
right.
Software
systems
will
never
be
simpler
than
they
are
today
and
yet
we're
still
using
these
tools
that
were
like
hey
we've
got
this
solution.
C
If
you
look
at
it
and
maybe
I'm
biased,
but
if
you
look
at
it's
like,
we
make
this
assumption
at
the
very
lowest
level
that
we
use
HTTP
to
communicate,
microservices
and
then
look
how
many
layers
up,
especially
in
this
conference,
are
being
built
to
try
to
get
around
that
limitation
and
just
makes
no
sense
to
me
whatsoever
the
security
things
hard.
So
we
leaned
in
really
hard
to
try
to
make
this
thing.
I
mean
forward.
C
Look
is
secure
where
you
get
a
high
level
cryptographic
person
to
come
in
and
just
try
to
beat
the
snot
out
of
it
and
say
this
is
actually
pretty
good
and
our
developers
still
go.
Wow
I
just
had
to
add
another
thing,
saying
user
credentials
to
my
connect
line.
That's
it.
We
believe
in
that
type
of
philosophy
of
czarina.
Do
things
like
that
and
not
do
sidecar
after
sidecar
after
operator
operator
see
or
you
know
whatever
that
stuff
is
so,
as
you
can
tell
I,
have
an
opinion
on
I.
C
C
So
the
question
is:
what
about
the
additional
services
that
the
silversmiths
technologies
offer
I
think
there's
there's
two
that
are
a
really
important,
Auton
dynamic
and
secure
authorization
and,
like
you
said
monitoring,
and
so
with
the
code.
That's
in
master.
Now
it's
not
easy
to
find,
but
we're
working
on
the
docks.
We
were
wet,
painting
it
towards
the
end
to
get
the
GA
with
our
product.
C
The
system
is
actually
tracking
everything
for
you
and
so
inside
of
NGS.
Now
you
can
just
say:
hey
I'm
gonna
send
requests
to
NGS
that
usage
and
it'll
come
back
and
tell
you
how
many
messages
sent
by
sent
vice
or
C
bytes.
You
know
done,
and
every
user
is
almost
kind
of
ephemeral.
So
in
the
old
version
of
Nats
typologies,
people
were
just
it
was
a
free-for-all.
Everyone
could
send
anything
and
receive
on
anything,
and
what
even
our
own
team
is
starting
to
see
now
is.
C
Is
that
as
we're
moving
and
getting
these
functionalities
such
that
you
know,
a
cluster
now
is
driven
by
a
trusted
set
of
operators.
The
operator
sign,
accounts,
accounts
then
signed
for
users,
and
then
users
have
those
permissioning
and
they
can
be
turned
off
instantaneously
globally
worldwide,
with
you
know,
clique,
that
we're
starting
to
get
a
feel
for
saying
this
is
something
that
you
can
take
from
dev
into
production
and
feel
safe,
secure,
it's
locked
down.
C
If
somebody
routes
that
machine
over
there,
they
have
credentials
that
can
send
to
one
subject
on
one
account
right
and
so
that
notion
of
accounts
plus
the
user
permission.
Inc
allows
people
to
take
the
exact
same
code
from
dev
where
it
might
be
totally
open
and
you
move
into
production
and
all
sudden
it's
the
code
doesn't
change,
but
who
I
am
my
identity,
changes
and
all
sudden,
it's
totally
locked
down?
So
we
can
get
better
at
giving
tooling
to
surface
some
of
those
technologies
to
make
it
easier
for
you
to
implement
them.
I.
C
Just
don't
think
you
need
the
15
layer
stack
to
just
figure
out
how
to
communicate,
but
I
do
believe
in
monitoring
and
security
and
authorization.
So
when
we
catch
up
on
some
sleep,
I
think
we're
gonna
start
looking
at
what
are
the
opportunities
for
servers
mesh
if
people
want
them
like
I
said
I,
you
know
I
I'm
too
old
to
worry
about
people
who
are
saying
well,
I
can
do
everything
with
HTTP
and
I,
say:
okay,
no
problem.
C
Yeah,
so
the
the
question
is
any
conversations
on
distribute,
tracing
open,
tracing
Zipkin.
That
kind
of
thing.
We
definitely
have
some
customers
that
have
been
talking
to
us
and
want
that
put
in
philosophically
I
believe
that
the
core
system,
underneath
the
cover,
should
treat
payloads
as
opaque,
meaning.
You
know
it's
the
thing
about
it,
our
phones,
don't
say
you
have
to
speak,
English
or
French
or
Italian
when
you
pick
them
up
right,
they're,
just
transport,
our
payload
agnostic
and
the
the
so
we're
looking
at
it
and
we'll
do
it
with
envelopes.
C
So
you
just
do
a
layer
of
abstraction
in
and
your
your
client
language
of
choice
instead
of
a
Nats
publish
you'll,
do
a
my
cool
publish,
which
will
just
say,
put
a
sequence
number
on,
and
we
do
it
ourselves
so
system
events
that
all
the
server's
are
sending
out.
Now
they
put
an
envelope
around
that
identifies
themselves
and
the
sequence
number
from
that
server.
C
So
that
end
points
without
having
to
do
any
type
of
crazy,
at
least
once
exactly
once
can
detect
gaps
record
innate
with
each
other
philosophically
I'm
happy
to
figure
out
how
to
make
you
know,
customers
happy
I've
been
doing
these
things
for
91.
So,
what's
a
long
long
time,
I've
never
used
the
server
chasing
I
was
at
Google
for
six
years.
They
said
you
can
do
it.
You
know
and
I
go
yeah
I,
don't
want
to
I.
Don't
need
that
I!
C
D
A
C
C
Okay,
well,
we'll
hang
out,
thank
you
guys
so
much
I
really
I,
especially
at
the
end
of
the
conference
and
everyone's
tired,
but
we
really
appreciate
the
users
and
stuff
and
the
ecosystem
is
really
important
to
us.
So
thank
you
and
again,
if
you
have
any
other
questions,
we'll
hang
come
on
up
and
beat
beat
me
up
so.