►
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
Keynote: 5 Years of etcd: Past, Present, and Future - Brandon Philips, CTO CoreOS, Red Hat & Xiang Li, Senior Staff Software Engineer, Alibaba
In July 2013 the etcd project was announced to solve a critical problem for CoreOS: how to safely coordinate unattended automatic software updates across a cluster of Linux nodes. Today hundreds of companies use etcd as part of their production systems to hold critical data for their Kubernetes clusters, network systems, monitoring systems, and much more. But, how did etcd evolve into a critical system in the Cloud Native ecosystem? This talk will cover the 5-year history of etcd from the stumbles, use cases, and technical evolution. We will also dive into the trade-offs that were made as the system went from theoretical prototype to a production ready workhorse. And touch on where the project is going as it moves into the CNCF. There is no expectation of prior knowledge of etcd or how it works. The ideal audience member has a love for distributed systems, and operations.
To Learn More: https://sched.co/Gsvf
A
All
right
good
morning,
everybody
so
good
to
see
you
all,
but
today
we're
gonna
give
a
little
bit
of
background
on
the
last
five
years
of
SCD,
as
Liz
mentioned,
it's
the
consensus
database
that
powers
kubernetes
and
many
other
projects,
and
we
also
have
some
exciting
news
to
share
about
the
projects:
well,
I'm,
Brandon,
Phillips,
CTO
of
core
OS,
formerly
and
currently
a
member
of
technical
staff
at
Red
Hat.
This
is
Zhang
Li.
A
He
was
formerly
an
engineering
manager
at
core
OS
founding
engineer
at
C
D
and
senior
staff
at
poly,
Bob
McLeod,
all
right
so
first
I
want
to
say
hello,
Det
CD,
how
many
of
you
ever
use
at
CD
directly,
not
in
relation
to
kubernetes,
all
right,
all
right,
all
right,
so
Etsy,
deep,
very
simple!
You
can
download
it
really
quick
on
your
laptop
run,
it
you
can
store
keys
into
it
and
then
later
retrieve.
A
The
keys
very
important
attribute
of
a
database
get
stuff
back
out,
and
this
is
how
you
interact
with
it
from
the
command
line
and
the
important
property
of
that
CD
is
that
it's
replicated
it's
consistent
and
partition
tolerant
and
it
uses
this
algorithm
called
raft.
And
so
a
MTD
cluster
looks
something
like
this.
A
You
have
three
to
five
members
and
they're
homogeneous,
meaning
that
there's
really
no
like
difference
in
the
attributes
that
the
members
in
the
cluster
and
then
the
fancy
property
is
that
there's
automatic
leader
elections
so
that
if
one
or
two
of
these
machines
that
are
running
up
to
the
fails
in
the
cluster,
you
can
continue
to
read
and
write
to
the
database
as
you'd
expect
and
then
obviously
it's
the
database
used
by
kubernetes,
and
so
you
can
use.
If
you
use
a
TV
CTL,
you
can
see
things
you
might
be
familiar
with.
A
It'd
be
used
kubernetes
inside
of
the
database.
Now
we
built
at
CD
five
years
ago
to
solve
this
particular
problem
that
we
had
at
core
OS,
which
is
we
were
seeing
this
view
of
servers
as
something
that
would
automatically
update
themselves,
and
obviously
you
don't
want
to
automatically
update
all
your
servers
at
once.
That
would
cause
downtime,
you'd
be
upset,
and
so
we
needed
something
that
could
hold
on
to
an
atomic
mutex
and
make
sure
not
too
many
machines
we're
rebooting
at
once.
A
And
so
you
do
something
like
this,
where
every
time
the
Machine
wanted
to
reboot
it
asks
for
permission
and
then
something
would
stop
you
from
rebooting,
too
many
machines
and
that's
what
we
built
that
TV
for
and
it
started
like
all
good
things
with
the
readme.
We
wanted
to
build
something
that
was
simple:
secure,
fast,
reliable
and
persistent
and
Alex
Bobby
Chun
and
myself
throw
down
this
readme
like
all
good
startups
and
a
Palo
Alto
garage
somewhere,
and
so
just
to
reiterate.
@Cv
is
a
database.
It's
replicated
it's
open
source.
A
We
built
it
at
patched
it
to
using
the
go
language
and
it
has
a
lot
of
stores
on
github,
most
important
metric.
Any
software
project-
it's
not
testing
everybody,
it's
not
language,
selection,
stars
and
the
project
has
been
brought
to
this
point
so
far
by
huge
amount
of
effort
from
this
top-level
maintainer
ship
team
and
so
I
want
to
give
them
a
quick
round
of
applause.
A
B
Thanks
Brendan
hello,
Iowa
I'm,
going
to
talk
about
the
especially
developing
journey
of
the
five
years,
so
in
the
past
five
years
over
400
contributors
have
checked
in
more
than
14,000
Khamis
into
the
product.
Together,
we
made
more
than
150
releases
with
three
major
ones:
ICD
alpha
as
a
tool
and
as
it
is
three
so
as
the
Alpha
is
our
fourth
release.
With
the
goal
of
being
the
cloud
native
distributed
consensus
system,
it
provides
a
standard,
HD
restful
api,
which
is
very
easy
to
integrate
with
and
to
interact
with
to
be
cloud
native.
B
B
So,
a
year
half
later
we
released
a
c2
with
the
goal
of
being
solid,
so
I
City
is
one
of
the
fourth
system
to
adopt
wrapped
in
production.
A
consensus
system
originally
developed
at
Stanford,
so
wrapped
is
designed
to
be
easy
to
understand
and
simple
to
implement,
however,
implement
interrupts
reliably
efficiently
and
in
full
production
system
is
still
hard
enough.
So
after
many
attempts
and
discussions,
we
will
add
that
we
have
to
review
the
coop
as
it
is
a
rough
material
from
scratch.
B
We
need
a
much
more
careful
design
to
enable
deterministic,
behavior
and
rigorous
testing.
So
with
this
goal
in
mind,
we
created
a
city
was
a
new
qualified
City,
so
now
I
see
developed
has
become
one
of
the
most
feature-rich
and
reliable
disability
consensus
or
ISM
implementations,
so
it
powered
many
popular
systems
like
ty,
Kelly,
DeGraff,
cohosh,
DB
and
hyper
lighter
and
many
more
so
as
city
3
is
our
current
release
with
focuses
on
efficiency,
reliability
and
usability.
B
So
a
new
storage
backhand
is
developed
for
ICD
3
to
support
incremental
snapshots
which
allow
users
to
millions
of
key
into
a
city
efficiently
and
reliably.
Multi-Version
concurrency
control
and
transactions
are
also
introduced
in
today's
release
to
simplify
application
development.
So
remote
snapshot
is
also
added
for
disaster
recovery
and
for
backup.
So
the
team
is
working
on
new,
exciting
features
like
learner
promotes
proxy
to
further
enhance
as
these
reliability
and
usability
so
I
say.
B
So
kubernetes
is
a
major
user
of
a
city
and
we
have
been
working
very
closely
with
the
kubernetes
community
and
the
tools
products
benefited
from
each
other,
the
least
and
would
feature
from
a
CD
includes
the
design
of
kubernetes
and
the
kubernetes
requirement
for
ICD,
especially
around
scalability
of
durability.
Reliability
pushed
as
the
development
forward,
actually
a
few
as
a
team
engineers
are
also
cooperate.
He
spent
a
nurse
so
we'll
continue
to
see
a
prayer
future
of
the
collaboration
between
two
products.
B
Besides
kubernetes
many
other
cloud
native
technologies
have
a
double
SAT,
for
example,
I'm.
Sorry,
a
dispute,
I'm
serious
database
from
uber
uses
a
CD
for
serving
story
and
what
melody
the
management's
business
Mexico
solution
from
YouTube
uses
a
CD,
2
polity
and
many
commies
are
running
our
city
in
production.
Today,
nearly
all
Club
riders
are
running
a
city
for
their
infrastructure
and
for
their
user-facing
products,
for
example
at
Alibaba.
B
It
runs
the
city
for
the
internal
cost,
cost
manager
Sigma
on
our
cloud
in
ruins,
a
city
for
the
cloud,
continuous
services,
so
five
years
is
not
long
for
an
infrastructure.
Software
and
I
City
has
achieved
so
much
for
five
years
and
as
a
community,
we
are
proud
that
the
product
has
benefited
so
many
users
and
developers
so
with
that
I'm
going
to
hand
over
to
Brannigan,
and
he
what
all
about
these?
As
in
the
future,
you
CN
CF,
hey.
A
A
There's
a
lot
of
interesting
war
stories
that
we've
written
blog
posts
about
discovery,
ta
vio
is
used
as
a
bootstrapping
service
and
getting
a
more
robust
on-call
system
for
that,
and
then
some
things
we've
been
wanting
to
do
and
we're
looking
forward
to
do
in
the
next
year
is
a
third-party
security
audit
and
a
third-party
correctness.
Audit
building
a
distributed
system
is
really
difficult.
A
We've
done
the
best
that
we
can
with
our
own
team
and
having
somebody
with
zero
understanding
of
the
internals
of
the
system
will
help
build
some
confidence
all
right,
so
I
want
to
have
a
few
advertisements
at
the
end.
If
you
want
to
learn
more,
we
have
an
intro
to
Ed
CD
course.
Tuesday.
Today,
at
11:40
a.m.
we
have
a
debugging
at
CD
at
2:35
p.m.
we
have
a
roadmap
session
at
1:45
p.m.
and
finally,
a
deep
dive
Thursday
at
10:50
a.m.
please
check
out
these
sessions.