►
From YouTube: Seamless Integration - Take Kubernetes into your Existing Monitoring Stack - Shaojun Ding, IBM
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
Seamless Integration - Take Kubernetes into your Existing Monitoring Stack - Shaojun Ding, IBM
Kubernetes has become a key scheduling and orchestration systems for containers and microservices nowadays. Along with this trend companies are willing to move their new workloads onto Kubernetes. However, for most of the companies they still have a lot of legacy systems which cannot be migrated to kubernetes. Typically, there are already well-established monitoring solutions for these legacy systems. So there comes a challenge that how you can make sure kubernetes can be integrated seamlessly with the existed monitoring stack. In this presentation I will walk you through how we integrate kubernetes with our existed monitoring system and shows some best practices for it: for example, how to have a central dashboard for viewing the whole system status, how to maintain a uniform alerting mechanism, etc. I will also show you how we introduce prometheus into our existed monitoring stacks.
To learn more click here: https://sched.co/FuK4
A
Good
morning,
a
good
afternoon
welcome
back
for
this
topic,
so
we'll
focus
on
seamless
integration.
So
you
take
your
bonus
into
your
existing
monitor
sent
here
today.
I
will
cover
the
following
aspects.
First
of
all
the
background
why
we
have
this
scenario
and
the
generated
problems
in
this
scenario,
so
we
have
all
these
problems:
how
to
solve
all
these
problems,
and
then
I
will
have
a
brief
summarization
for
the
application
scenarios.
A
We
worked
as
the
CTO
team
for
the
account
for
the
IBM
cloud
for
offering
these
solutions
and
now
I
work
for
the
open
source
lab
of
IBM
and
a
focused
on
service
map
service.
Mesh
I
also
have
lots
of
experience
in
terms
of
the
development
of
needleless.
So,
first
of
all,
let's
have
a
look
at
the
background
of
the
related
problems
as
a
mission
before
in
my
previous
Department,
which
was
a
cloud
CP
or
CPU
waiting
to
provide
the
monitoring
capability
when
it
comes
to
monitoring.
Well,
this
is
actually
a
legacy
problem.
A
We
have
the
IT
system
and
the
monitoring
is
necessary.
You
can
imagine
this
for
any
production
system
or
any
IT
system.
If
you
don't
have
monitoring,
then
this
would
be
like
a
man
without
eyes.
You
don't
know
the
current
status
of
your
system.
This
is
very
risky
and,
from
my
perspective,
I
think
management
covers
three
key
parts.
One
is
the
data
collection,
upfront
data
collection.
A
In
different
virtual
machine
and
your
workload,
you
need
different
agents
to
collect
data,
so
this
is
the
first
corporate
and
second
covered
is
for
the
exact
data.
You
need
to
store
the
data.
Your
metrics
need
to
have
a
place
to
store
the
data,
so
you
have
the
extraction
of
the
data
and
storage
of
data.
You
want
to
apply
this
data
this
on
the
data
expected.
You
will
have
analysis
the
specifications,
this
will
trigger
alerts
or
you
have
the
part
to
have
the
comprehensive
demonstration
to
help.
You
understand
the
current
status
of
the
system.
B
A
A
Each
consent
matrix
to
collect
it,
de-clutter
can
publish
magics,
install
all
the
matrix
in
the
data
store
and
the
agents
can
trigger
the
data
from
the
data
agents
and
data
store
I'm
going
to
put
it
just
sent
out
to
the
dashboard.
Now
this
is
a
perfect
operation
and
very
stable,
but
in
recent
several
years
we
have
their
popular
micro,
so
this
our
Cuban
'its
container
and
the
orchestration.
So
our
team
wants
to
migrate
its
workload
to
some
physical
and
virtual
machines
into
the
clusters.
A
As
you
can
see
here,
we
have
lost
the
workload
of
some
of
the
teams
have
been
migrated
to
this
cluster
and
after
this
migration,
the
manager
will
also
have
some
changes.
Some
teams
will
natural
thing
for
this
project.
We
native
monitors,
Prometheus,
and
this
team
do
not
want
to
continue
to
use
the
conventional
or
the
legacy
monitoring
and
they
want
to
use
the
premises
so
so
the
integrate
for
media's.
Another
issue
is
with
the
our
team.
We
have
lots
of
the
requests.
We
hope
that
for
the
metrics
we
could
have
a
central
place
for
storage.
B
A
Scenario
has
lots
of
players;
this
will
be
very
concept.
Complex
and
a
monitoring
team
has
lots
of
pressure
because
of
this.
They
don't
know
how
to
fit
for
these
complex
situations,
because
lots
of
legacy
systems
haven't,
got
the
migration
so
how
you
can
have
the
uniform
orchestration
chant
for
the
legacy
and
the
new
systems.
C
If
it
isn't
well
weld
deployed
well,
you
can
just
run
it,
but
now
you
put
it
into
the
container
and
pod,
but
sometimes
some
containers
can
run
several
seconds
and
then
just
die
away
so
compared
to
the
law
Ronnie
matrix
how
to
differentiate.
This
is
short
lived
in
long
Ronnie.
How
to
monitoring
them
is
a
big
challenge.
Second
channel
was
economic.
C
The
workload
is
running
on
the
server
we
tested
need
to
monitor,
CPU
and
memory
and
disk,
but
now
we
find
that
due
to
that
Cuba
Cuban
aids
has
its
own
functionality.
Maybe
sometimes
your
pod
is
run
on
this
part,
but
out
later
it
just
move
to
another
node.
So
in
this
case,
how
do
you
monitor
them
how
to
detect
the
problem?
So
this
will
be
a
challenge.
The
third
challenge
will
be
the.
B
C
Services
from
a
monolithic
application
to
the
macro
services,
so
now,
when
there
is
a
request,
it
just
circle
across
the
several
macro
services
like
a
service,
a
B
and
C.
So
how
to
the
cap?
The
problems
will
be
a
challenge,
but
now
you
don't
know
which
money
service
has
some
problems,
so
you
have
to
figure
out
the
choosing
between
the
different
micro-services
to
find
the
what
is
going
on.
This
will
be.
The
third
challenge.
C
B
C
B
C
B
C
C
B
C
Terms
of
a
key
keeper
nuts
here
are
ours:
OSHA
is
a
solution.
One
and
each
survey
team
has
its
own
problems.
Some
of
the
teams
work
well
with
us
how
they
can
maintain
their
traditional
system.
Maybe
some
are
transferred
to
the
key
cabinet,
so
in
this
case,
hope
that
can
they
rep
maintain
what
it
is.
So
in
this
case,
if
we
can
help
them
reduce,
they
were
clouded
so
I
was
motion,
one
will
be.
The
self-maintained
will
call
the
city
monitor
internally,
any.
C
B
C
C
B
B
B
B
C
Can
unload
the
customer
demand?
So
you
see
this
has
become
complicated.
Maybe
let's
say
this
could
be
true.
Maybe
there
are
some
monitor
scripts
running
and
I
support
them
to
Rama
script
on
their
pod.
I,
don't
get
irrelevant
n
matrix.
By
doing
this,
we
can
dab
such
a
magics
to
customize
your
own
healthy
cluster.
B
B
C
Whole
cluster,
we
know
we
can
get
the
results.
You
said
you
for
the
whole
cluster.
We
will
do
the
magic
integration,
okay,
get
a
demo
container
data,
we
do
the
categorization
and
we
know
the
treatment
of
each
part
and
the
for
the
cluster
level
and
not
level.
We
know
all
of
this
data
is
called
and
we
can
so
that
we
can
get
a
sausage
for
part.
We
can
know
there
was
a
usage
for
knows
who
can
get
them
now
instead,
as
posted
as
no
number
part
number
classic
lots
of
events.
B
B
B
C
Give
this
into
the
servers
team
will
also
find
some
problems.
As
you
see
on
the
structures
graph,
we
only
rely
on
the
divisor
on
the
bottom
level,
maybe
some
things.
Sadly,
there
is
no.
No,
they
say
the
visor
maintains
they
can't
do
other
trousers.
It
can
do
a
demo
Center,
they
can
do
the
city
level.
B
B
C
B
B
B
B
C
The
promises
storage
is
a
burden,
for
them
means
a
little
bit
hard
for
them
to
maintain
the
promises,
so
they
old
love
it
is
run
and
monitored
on
the
out
in
the
old
way.
So
after
anything
to
that
in
the
promises
how
we
can
get
a
new
solution.
So
in
this
case,
if
we
get
the
solution
to,
we
have
the
promises.
Adapter,
one
of
their
promises
has
its
own
searching.
Language
is
called
a
certain
language,
so
this
prom
adapter
ask
use
the
search
language
of
promises.
B
C
B
B
B
D
C
C
Parameters
will
connect
it
with
the
targets
and
get
the
data
and
the
pull
the
data,
but
in
terms
of
parameters
when
a
the
pull
and
also
push
so
it
will
not
occupy
too
much
comatose
availability
and
resources.
So
by
doing
this
way,
just
a
comma
weight
is
a
good
solution,
but
you
have
to
change
the
configuration
of
the
raw
material
in
some
to
some
degree.
B
B
C
B
B
B
B
C
E
B
B
C
Maybe
if
you
go
back
to
the
old
Caesar,
we
welcome
this
approach,
but
how
to
reduce
your
cost,
how
to
maintain
your
professor,
our
systems,
how
to
cut
the
promises
while
maintaining
your
resources
and
your
data.
So
there
is
our
solution
for
is
called
exporter
only
if
their
permit
is
not
available
by
the
honor
of
the
matrix
and
the
exporter
can
imitate
so
we're
at
accumulator
controller.
So
this
controller,
this.
A
B
A
A
And
the
address
you
will
got
the
address,
so
you
can
feel
the
targets
dynamically,
not
statically,
and
if
you
have
the
service
go
down
or
go
up,
then
I
will
have
all
this
information.
I
also
have
these
workers
to
retrieve
the
data
from
the
targets,
but
also
you
need
to
transfer
lots
of
formats,
for
example,
for
the
labor
of
Prometheus.
You
can
transfer
this
into
the
desired
format.
You
need.
B
B
B
B
A
B
A
Summary
about
pros
and
cons
of
different
monitoring
strategies
and
use
cases.
First
of
all,
self-maintained.
The
pros
are
no
extra
madness
efforts.
You
can
only
rely
on
the
resources
that
we
offer
you
so
no
extra
cost
for
maintenance
and
premises
because,
as
you
need
self-development.
The
years
ago,
I
was
mentioned
before
smart
work
float
migrated
to
Cuba
nice.
A
If
you
want
to
use
the
legacy
since
summer,
no
promises
exist,
so
you
can
select
dissolution
and
second
solution
is
police's
with
remote
right.
If
your
team
has
parameters
and
if
you
select
the
solution,
no
impact
on
performance,
what
comes
is
you
need
to
change
the
configuration
it
is?
You
need
to
enable
the
Prometheus.
B
A
In
the
remote
right
community
that
was
just
able-
and
now
we
think
that
it
has
been
very
stable,
so
the
subversions
forum
users
are
very
low
and
when
it
out
great-
and
he
also
need
to
maintain
your
parameters
so
generate
constant.
A
use
case
relating
to
the
prometheus
and
third
solution
is
trauma.
The
security
query.
A
The
coins
include
in
some
scenarios
the
team
wants
to
take
the
metrics,
the
key
metrics
out
for
a
long-term
storage,
as
I
mentioned
before
for
some
metrics
they
don't
need
to
change
and
they
just
need
small
or
limited
amounts
of
the
change.
The
metrics
no
significant
inputs
for
performance,
but
if
you
need.
B
A
B
A
A
A
B
A
Our
team,
the
internal
alerts,
as
you
know,
if
you
do
not
have
analysis
or
processing
for
alert,
New,
York,
O&M
team
or
have
lots
of
pressure
in
the
morning
session.
We
have
learned
lots
about
how
to
operate,
alert
and
do
not
false
alerts.
We
have
optimized,
for
example,
we
have
a
correlation
engine.
What
does
it
mean?
One
alert.
B
A
Time,
Sunday,
if
you
have
a
hundred
and
you
can
locate
some
analysis
and
you
can
find
that
all
this
are
generated
by
one
trouble.
You
need
to
have
the
troubleshooting,
and
sometimes
you
also
have
the
altar
healing
alert,
no
need
for
manual
work.
You
just
need
script
text
to
help
the
articulate
through
these
angels,
so
they
can
enjoy
a
benefit
to
healthy
alerts,
generated
infrastructure.
The
Prometheus
alert
also
support
web
hood.
B
A
A
A
A
A
A
B
B
B
B
B
B
B
B
A
This
you
can
see
here
we
have
all
these
metrics
also
for
the
cluster
level
metrics.
If
the
Prometheus
want
to
search
the
cluster
level
data,
then
you
need
to
have
the
aggregation
and
this
significantly
impact
the
performance.
But
here
you
don't
need
to
do
this
and
this
part
can
help
you
to
finish
the
aggregation
and
no
significant
performance
impact.
B
A
B
B
B
B
A
B
B
A
B
A
B
B
A
B
B
A
B
D
B
A
Resolve
wasn't
to
promise
a
new
matrix
to
fix
it
to
resolve
it.
It's
like
the
matrix
came
back
to
the
normal
state.
If
I
don't
tell
it,
it
can
also
detect
that
it
has
gone
back
to
the
enormous
day
right
because
it
collected
yes
after
the
self-healing.
Yes,
its
matrix
will
come
that
normal.
So
what
about
the
significance
also
resolved?
A
Why
doesn't
matter
because
matrix
also
has
the
latency
or
delay
we
want
to
resolve
it
on
in
time,
because
of
timing,
so
all
to
resolve
how
its
judge
you
have
finished
the
alter
hearing,
not
only
in
prometheus,
we
have
the
alert
engine.
The
Prometheus
alert
will
not
automatically
be
sent
to
our
O&M
stuff.