►
Description
En los ultimos años la aparición de service mesh como Linkerd o Istio han traido al ecosistema de aplicaciones cloud native nuevas prestaciones enfocadas en tolerancia a fallas, especialmente con el uso de patrones como Retry o Timeout los cuales al día de hoy se interesectan con patrones de tolerancia a fallas que tradicionalmente son cubiertos con Microservice Chassis como Spring Boot o MicroProfile.
En esta charla mediante el uso de Linkerd y Java exploraremos cuales son las diferencias entre ambos abordajes, sus ventajas, desventajas y como pueden utilizarse correctamente para implementar patrones de tolerancia a fallas en sistemas distribuidos.
Contacta a Victor en:
- https://www.linkedin.com/in/victororozco/
- https://twitter.com/tuxtor
A
A
A
A
A
A
Very
good
afternoon,
everyone
thank
you
for
continuing
to
try
the
talks
we
have
on
tv
community
day.
It
is
a
pleasure
for
us
to
restrict
you
and
for
me
it
is
also
a
home
to
introduce
you
to
our
guest
presenter.
At
this
time,
What
is
Víctor
Orozco.
Who
brings
us
an
interesting
theme
of
tolerance,
Fire
Víctor,
nice
to
meet
you,
time
is
yours,.
B
B
Are
the
patterns
found
in
microservices,
point
I,
and
if
we
consume
all
the
patterns
that
are
available,,
we
will
find
which
patterns
of
trust
or
tolerance
to
Faults
such
as
circuit
breakers
play
an
important
role
in
this
type
of
architecture
and
at
the
same
time,
patterns
such
as
trace
metrics
in
distributed
and
exception.
Racing
are
part
of
the
patterns
of
observe
ability.
Let's.
Remember
that
these
patterns
do
not
require
the
use
of
a
specific
product,.
B
But
yes,
it
is
recommended
that
while
we
create
our
cloud
night
architecture
and
in
some
way
we
go
covering
this
either
with
rulers
or
5
vernet
is
there
are
n
ways
to
create
fault,
tolerant,
architectures,
with
cloud
native
applications,
and
in
this
context
we
have
to
analyze
a
little
bit
of
history.
To
that
we
know
where
they
stand.
B
Let's
remember
that
cloud
night,
architectures,
and
especially
with
rulers,
are
of
a
recent
appearance
here.
What
I
have
is
a
timeline
of
concepts
that
I
consider
important
for
the
scenario
that
the
knicks
and,
for
example,
we
have
that
in
2002
aparicio
spring
in
2009,
appeared
noyes
that
are
today
two
of
the
most
popular
talk
its
to
create
applications
and
the
boxing
movement.
As
we
know
it
appeared
later,
then
the
cloud
applications
to
what
I
call
cloud
night
and
first
generation
had
a
boom
in
2012.
B
Many
of
you
will
remember
that
netflix
went
to
a
conference
to
explain
how
they
make
their
cloud
native
architecture,
how
they
could
go
from
their
infrastructure
to
w
and
parallel
to
their
conference,
in
which
many
of
the
microservices
fundamentals
were
established.
They
released
the
code
of
river
district
about
k
that
are
projects
that
belong
to
the
netflix
meta
project,
or
you
want
and
that
in
fact,
sprint
boot
as
a
base
combining
it
with
spring
to
have
a
punctual
tunnel
that
completed
in
flame
of
creation
of
applications
that
the
nets
now
because
I
make
this
distinction.
B
Netflix
and
His
approach
to
create
fault,
tolerance,
microservices
and
observe
ability
is
prior
to
the
appearance
of
dock
and
long
before
the
appearance
of
cover
next,.
So
we
could
say
that
from
2002
to
2012,
fault,
tolerance
patterns
and
observe
ability
were
founded
without
the
need
for
existence
of
cover
nets,
then
the
first
generation
chassis
micro
services,
like
nose
and
sprint,
covered
these
patterns
directly
in
the
code.
But
from
the
appearance
of
docker
and
the
eventual
appearance
and
popularization
of
rulers.
B
They
gave
a
second
option
to
have
fault
tolerance
patterns
and
observe
ability
that,
in
this
case,
would
be
the
monthly
services.
So
today
there
are
two
ways
to
do
the
same
thing
and
software
architects
have
to
decide,
which
is
the
best
technical
decision
to
choose
one
way
or
another,
or
combine
them
and
reach
a
result.
B
That
we
deem
appropriate
now
well
I'm,
going
to
demonstrate
this
more
than
anything
with
a
live
demonstration,
but
I
just
want
to
highlight
the
following
netflix,
or
that
is
an
ecosystem
where
we
have
different
tools,
such
as
distress
for
resilience,
which
is
a
library.
In
java.
Your
google,
which
also
it
is
a
library
for
dynamic,
routing
or
load
launch
with
irrigation,
serve
discovery
with
a
horeca
server
and
training
council
kings,
and
all
this
was
originally
created
for
microservices
and,
in
fact,
the
popularity
today
of
sprint
boot
for
the
creation
of
microservices.
B
In
the
following
way,
when
we
call
from
a
service
to
a
service,
we
go
through
various
stages.
For
example,
if
I
do
not
depend
on
the
service
data
being
live
from
service
to
I
can
store
cache,
and
as
long
as
the
service
continues
to
be
powerful,,
the
call
will
give
me
the
same
results,
assuming
that
I
cannot
use
cache,,
which
would
be
the
first
defense
mechanism,
I
go
towards
the
circuit
breaker
pattern,,
which
is
basically,
I,
have
communication
rules
when
I
have
different
communications
between
microservices,.
B
If
any
of
this
communication
is
wrong,,
the
chain
of
responsibility
would
bring
down.
All
the
microservices.
What
'circuit
breaker'
does
is
that
through
programmable
rules,
I
open
the
circuit,
and
even
if
the
communication
between
a
service
falls,
a
service
c
d
y
f
knows
that
it
has
fallen
and
I
can
guarantee
that
the
entire
system
does
not
collapse,
and
parallel
to
the
opening
of
the
circuit,
I
can
execute
an
alternative
method
called
fullback
to
give
an
alternative
response.
With,
for
example,,
giving
the
previous
data
or
simply
saying
the
system.
It
is
currently
overloaded.
B
I
tried
later
later,
so
it
will
break.
Another
important
pattern.
Is
to
return,
which
is
an
exclusion
or
a
simpler
pattern
in
which
we
set
a
hard
limit
to
say:
I
am
going
to
have
a
maximum
of
1,2000
concurrent
clients
and
out
of
this,
if
the
2001
client
arrives,
I,
send
it
back
to
full
back
if
I
comply
with
the
fault
tolerance
rules,
I
already
passed
the
cache
and
executed
the
method.
I
still
have
one
last
point
of
failure,
which
is
that
the
method
goes
wrong.
B
For
some
reason,
the
database
crashed
all
three
in
java
ran
out
the
event
queue
in
naut
ran
out,
but
still
when
my
code
is
fault,
tolerant,
I
can
offer
a
form
method.
That
should
still
give
me
an
answer.
The
idea
is
not
that
systems
fault
tolerant,
they
do
not
fall,,
it
is
that
they
fall
in
a
controlled
way
or
in
other
words
they
fall
with
style
without
cover
net.
It
is
generally
we
used
to
do
this
at
the
code
level.
B
For
example,
in
java
we
had
to
do
it
with
strix
resilience
and
macro
profes
folk
toleras,
always
complementing
with
metrics
like
sprint
metrics,
but
my
duty
and
micro
profile
me
tics.
If
not.
On
the
other
hand,
it
has
its
own
fault,
tolerance,
library
like
well
with
metrics,
like
perón
client
and
in
neutral
formats
like
open
metrix,
and
all
this
works
like
that,
at
least
until
the
year
2012
2013,
when
it
went
to
the
emergence
of
docker
and
eventually
the
popularization
of
cuber.
B
Next,
as
it
established
itself
towards
popular,
it
was
going
to
be
bland,
it
was
going
to
be
the
first
generation
of
ranchero.
Vernet
is
many
cloud
maker
architectures
that
were
born
in
that
era.
Al
nowadays
implement
fault
tolerance
and
see
ability
that
are
complementary
across
libraries
in
this,
so
software
developers
and
software
architects
need
to
be
aware
of
at
least
the
following:
five
'circuit
breaker'
patterns,
dumper
read
thrive
time
out
and
firework
that
are
usually
combined.
B
Prof.
I
can
establish
that
if
something
goes
wrong
in
this
method,,
please
execute
the
callback
method
that
could
go
wrong,
for
example,,
a
time
out
rule
that
I
don't
comply.
With,
I
send
it
to
full
back
and
that's
where
the
story
ends.
execution.
If,
at
a
certain
moment,
the
method
is
executed
correctly
because
it
never
reached
the
fullback
code,
this
could
be
programmed
manually
even
using
iffhs
interceptors
and
all
the
mechanisms
of
its
programming
language.
B
But
the
common
thing
when
we
use
a
micro
service
chassis
like
sprint
boot,
micro
profile
or
nou
with
libraries,
is
that
we
do
it
through
declarative
rules
or
in
the
case
of
not
through
intermediate
functions
and
extensions.
Now
this
worked
perfectly,
but
it
has
some
disadvantages.
Number
one
is
platform
dependent,
which
implies
that
programmers
have
to
be
aware
how
this
works
in
java,
how
it
works
off,
eventually
how
it
works
in
python.
So
we
have
to
have
very
good
programmers.
B
Second,
it
is
based
on
interceptors,
most
of
the
time
or
based
between
middleware
or
in
the
case
of
nose,
which
would
imply
that
there
is
more
time
and
consumption
of
cpu
to
execute
the
fault,
tolerance
rules,.
So,
in
the
long
run,
this
makes
our
applications
slower,,
which
matters
a
lot
depending
on
the
scale
of
the
applications
we
manage
and
thirdly,.
This
is
lingo
bergé,
because
I
have
to
learn
a
lot
of
new
things
here.
B
It
comes,
then,
the
promise
of
cover
next
I
am
not
going
to
talk
too
much
about
the
promises
of
rulers,
because
that
is
why
they
have
this
whole
conference
with
different
talks.
But
one
of
the
important
promises
of
rulers
is
that
rulers
are
resilient.
In
other
words,
I
deliver
a
deployment
plan
to
rulers
and
rulers.
According
to
all
the
interactions
between
apis
controllers
and
others,
seeks
to
make
a
reconciliation
that
all
the
time
the
plan
that
I
gave
to
rulers
in
a
manifest
is
being
synchronized,
and
the
stability
of
my
infrastructure
is
guaranteed
by
itself,.
B
Then
I
delivery,
let's
say
so
it
heals
automatically
like
wolverine,
but
if
there
is
a
problem
between
two
services,
if
the
call
between
the
service
goes
to
the
ve
service,
it
is
wrong.
Gubern,
&
es
by
itself
does
not
guarantee
that
the
services
are
fault
tolerant
between
calls
it
may
services
are
ok
with
the
correct
metrics,
so
they
don't
trigger
a
message
reset
on
hoover
nets,
but
the
call
doesn't
work.
So
this
is
where
the
services
come
in
lo
service
month.
As
many
of
you
have
probably
already
used,
they
are
network
level
interceptors.
B
So
one
of
the
reasons
why
q
vernet
is
used
as
its
body
unit
by
bots
and
not
containers
is
because
more
than
one
container
can
live
inside
bots
and
this
container
complementary
forks
are
generally
used
for
log
management.
Observe
ability
tolerances
to
failures,
then,
are
complementary
containers
and
to
execute
tasks
such
as
their
engineering
when
we
simulate
falls
or
traffic
through
interceptors
and
proxies
being
the
most
important
at
least
today.
This
is
quite
popular,
especially
for
new
architectures
link.
B
Enrique
is
popular
for
give
fault
tolerance
to
already
existing
clusters,
and
it
is
easy
to
use,
honestly.
That
is
my
opinion
and
I
conclude
that,
for
historical
reasons,
it
was
the
first
as
of
today
because
it
already
has
competition,
but
you
can
see
many
deployments
still
with
contact
today.
We
are
going
to
a
Small
example,
so
that
you
understand
how
fault-,
tolerant
environments
can
be
complemented
with
linkedin.
B
The
Dellinger
Díaz
architecture
is
quite
similar
to
other
cover
next
operators,
in
which
we
have
a
control
plane
with
the
correct,
bots
and
deployments
to
collect
metrics
from
each
one
of
them.,
the
productive
posts
or
posts
of
my
application
that
use
link
erdi.
Next,
then,
in
a
traditional
post,
suppose
I
have
my
application
with
ya.
Goes
parallel
to
my
application
with
key
I
have
a
site
car
or
link
kerry
proxy
then,
instead
of
my
microservices
being
communicate
with
each
other
directly
through
the
application.
B
Jim
kerr
di
proxy
works
as
an
interceptor
but
interceptor
at
the
network
level
with
which
it
can
be
constantly
monitoring.
The
health
status
of
my
application
can
report
it
to
the
linkedin
control
plane
and
parallel
to
reporting
it
to
the
control
plane.
Dellinger
and
I
can
even
report
it
to
a
metrics
collector,
as
promised
visualization
like
graph
anna.
So
the
value
proposition
of
a
service
is,
in
a
nutshell,
is
to
provide
me
observe
ability,
but
among
its
important
features
that
many
times
we
do
not
exploit
properly.
B
Is
that
if
I
identify
an
error
at
the
network
level
or
in
other
words,
fault
tolerance
at
the
network
level,
I
can
intercept
that
error
and
using,
let's
say
so,
the
appropriate
approach
to
complement
the
fault,
tolerance
that
comes
within
a
micro
service
chassis.
Now,
as
this
is
a
short
presentation,
I
want
to
focus
on
the
demo.
The
source
code
of
this
demo
is
already
available
on
youtube
and
I
am
going
to
focus
on
the
following
scenario.
Suppose
I
have
an
application.
B
A
sp.a
hentai
script
in
angular
jay
is
generally
this
application
will
talk
to
through
a
login
a
piquet
with
if
it
were
something
complicated
to
a
back-end
service.
This
backend
service
could
be
written
in
java,
but
in
turn,
this
back-end
service
could
depend
at
the
http
level
of
another
service.
We
call
local
service,
which
is
find
it
written
even
in
another
programming
language.
It
could
be
coaching
at
the
moment
that
ca
service
here
is
a
dependency
chain
in
which
jota
service
b
would
be
brought
down.
B
If
I
manage
to
make
the
call
tolerant
between
j
service
b
and
ca
service,
I
could
guarantee
a
path
in
such
a
way
that,
even
though
each
service
has
failed
intermittently,
I
can
still
give
a
response
to
the
front
end
in
intense
in
service
to
a
mobile
application
to
something
in
react
and,
above
all,
this
demo
is
programmed
in
various
programming
languages
to
demonstrate
that
it
does
not
matter
which
language
you
give
them
If
you
like,
or
you
use
it,.
You
will
be
able
to
use
it
correctly.
B
Now,.
How
is
the
demo
structured?
In
the
demo,?
We
have
three
projects,
and
here
are
some
instructions
to
use
link
erding.
Now,.
The
first
thing
about
this
demo
is
that
I
have
a
meme
space
on
which
I
can
work.
For
example,
we
could
apply
the
name
space
called
polyglot
point.
We
call
and
I
am
connecting
at
this
moment
to
an
instance
of
oracle
close
so
that
you
can
also
try
it
at
home
within
this
environment.
So
I
could
deploy
my
first
service
called
car
service
and
how
This
is
not
a
presentation
of
a
programming
conference.
B
I
am
not
going
to
go
into
much
detail
in
the
code,
but
I
want
us
to
see
at
least
one
of
the
classes
in
cotel
the
most
important
class
in
code
link
for
this
demonstration
is
the
following,
and
it
has
a
Quite,
simple
logic
when
the
user
calls
bar
and
the
ready
method
is
going
to
be
executed.
Today,,
what
it
does
is
read
an
array
of
famous
computing
phrases,
for
example,.
We
could
have
to
read
Víctor
Orozco
in
Thompson
James
Gosling,
with
some
of
his
most
famous
phrases.
important
now.
B
Well,
if
I
do
not
call
the
list,
because
the
list
is
not
made
to
fail,
if
not
I
do
not
to
diagonal
freixes
diagonal
zero
to
obtain
the
phrase
zero,
for
example.
This
second
method
is
executed
in
this
second
method.
I
randomly
generate
a
failure.
This
failure
that
I
generate
will
fail
50%
of
the
time.
This
method
50%
of
the
time,
gives
a
correct
answer,
but
the
other
50%
of
the
time
it
will
throw
an
exception.
I
will
show
it
to
you
quickly
in
the
terminal.
So
this
is
a
java
project.
B
Now,
note
that
here,
I
am
still
not
using
fault.
Tolerance.
I
have
a
service
that
intentionally
fails,.
So
if
I
go
to
that,
point,
less
iarc
targets
a
service,.
What
is
a
service?
in
outage?
This
service
is
programmed
at
the
conferred
time
and
starts
quickly
because
it
is
a
framework
for
cloud
mit,
I
call
localhost.
B
80,
which
stops
hello
to
see
if
everything
is
ok,
the
micro
service
responds
so
I
will
quickly
show
you
the
clock.
We
could
have
changed
it
then,
or
the
song
from
guatemala
that
I'm
working
on
being
this
law
down
in
my
system
running
on
the
real
ip
address
of
my
computer,
it's
just
a
ping
service
to
find
out
what
our
communication
is
towards
now.
Well,
if,
instead
of
calling
hello,
I
call
freixes
pass
the
following:
I
always
receive
the
list
of
phrases,
no
matter
how
many
times
I
call
it.
B
If
the
service
is
fine,
I
receive
the
service,
the
famous
phrases
in
jason
format,
but
if,
for
some
reason,
I
call
phrase,
one
sometimes
I
get
results,
sometimes
not
so,
if
I
do
We
see
in
verb
mode,
it
happens
that
sometimes,
when
I
call
a
bar,
my
result
is
200
and
the
service
is
responding
successfully.
Let's
assume
everything
is
fine
with
my
database
with
my
communication,
but
if
I
call
again
in
verb
mode,
sometimes
it
works
again
get
200.
B
200
does
not
want
to
fail,
and
sometimes
I
do
not
get
error
500.
What
this
service
tells
me
intermittently
gives
a
good
response.
Its
memory
consumption
is
ok,
its
disk
space
consumption
is
ok,
its
cpu
consumption
is
ok,
but
it
may
be
that
the
database
failed
I
needed
to
open
a
file
from
a
jan
efe
is
that
it
fell.
In
other
words,
a
lot
of
things
can
go
wrong.
This
service,
in
particular,
is
designed
so
that,
if
it
does
go
wrong,
it
gives
a
500
error,
and
that's
enough
now,
let's
explore
your
client.
B
I
am
going
to
close
this
terminal
and
it
left
my
service
running
in
kotla
and
notice
that,
while
it
was
failing,
it
was
generating
a
red
log.
What
happens
then,
with
jota
service?
It
sees
that
its
client
j
serve
sees
in
this
case.
It
is
a
java
application.
This
is
it
steam
or
not.
It
is
code
link
and
it
has
the
particularity
that
it
is
programmed
solely
and
exclusively
to
consume
the
service
as
it
consumes
it
through
a
client
in
the
rest.
B
So
at
this
moment
I
am
using
micro
profile
with
the
implementation
of
redhat
warchus
to
consume,
and
what
What
this
interface
represents
is
to
the
other
service,
again.
I
had
two
methods:,
a
method
to
show
all
the
phrases
and
a
method
to
get
a
particular
phrase,.
If
it
fails
or
not,
the
calling
service
doesn't
know,,
it
will
know
until
the
moment
it
invokes
it.
In
this
case,
when
exposing
a
face,
demo,
I
consume
the
data
repository
and
two
things
can
happen,
here,
one
that
can
be
success
and
one
that
can
be
failure.
B
I
received
the
phrase
such
a
wisdom
by
the
author,
but
remember
that
this
can
go
wrong
and
it
is
a
chain
failure,
because
if
this
failure
is
going
to
fail
because
the
one
that
generates
the
information
is
incorrect-
and
this
is
where
fault
tolerance
comes
in
via
micro
service
chassis,
the
micro
framework
in
this
case
is
its
own
micro
and
with
rooms.
It
still
allows
me
to
offer
an
alternative
method
that
says,
or
something
went
wrong.
The
system
is
in
mode
maintenance.
B
Please
try
again
later.
This
is
implementation
dependent
to
fully
understand
how
I
could
implement
fault
tolerance.
I
would
have
to
have
studied
java.
This
is
a
super,
simple
method
to
be
understood
in
the
chat,
but
in
other
words,
if
my
team
is
made
up
of
microservices
in
java
göttling
python
for
different
purposes,
I
would
have
to
learn
in
each
one
how
to
do
afaex
tolerance.
This
second
is
done
in
rooms,
so
we
are
going
to
run
another
bñs
service
and
see
how
it
runs
correctly
or
incorrectly.
A
B
And
limit,
but
before
reviewing
it
I'm
going
to
launch
then
to
the
full
demo:
ok.
Well,
we
have
two
services,
a
service
that
fails
intermittently
and
a
service
that
at
the
time,
is
a
service
created
with
four
so
here.
For
some
reason
it
has
a
small
recital,
I
have
yet
to
see,
but
each
of
these
has
its
own
deployment
plan
that
I
am
going
to
show
in
this
moment
on
screen.
B
First,
we
are
going
to
deploy
car
service
a
car
service,
a
has
three
elements:
it
has
an
internal
service
that
is
created
as
an
ip
cluster
and
an
external
service
for
redundancy
worth
exposing
it
to
the
outside
of
my
platform
in
q.
See
nets
as
this
is
a
demonstration
in
Alive.
I
already
had
compiled
the
image
that
is
ready
directly
in
do
kerkhof,
and
this
image
is
going
to
be
published
directly
in
oracle
cloud
within
the
cnime
space
that
I
created
previously
so
I
entered
that.
He
said
yes,
that
the
apply
efe
caser
guys.
B
In
java,
my
service
in
java,
that
is
moderately
fault,
tolerant
and
has
its
own
deployment
plan
in
the
same
dynamic,
has
a
service
that
is
internal
that
is
made
with
ip
cluster
has
a
balancer
load,
which
is
external,
so
it
gets
its
own
ip
address,
and
the
room
service
image
is
ready
and
deployed
again
on
touch
hop,
so
this
demo
can
run
so
I
say
a
site.
The
apply,
efe
j
serve
be
up.
We
check,
then
j
service
will
be
provisioned
that
they
consume
here
service.
B
B
It
tells
me
hello,
it's
called,
which
is
another
identifier
that
is
used
in
the
others,
I'm
in
that
local,
open
heidi,
that
jones
notice
that
they
already
have
the
virtual
key
machine
inside
the
container
in
dock
a
running
at
the
address
ca
service
aps,
which
is
the
internal
ip
address
that
is
given
within
cv
birmex.
If
I
execute
then
freixes
whenever
it
executes
phrases,
I
receive
for
a
response.
This
works
correctly
and
in
fact
you
can
test
this
ip
address
directly
and
at
the
moment
it
is
online
However.
If
I
call
price
it's
one.
B
Sometimes
it
doesn't
work
and
I
get
the
guinness
error,
sometimes
yes,
sometimes
no
and
depending.
If
you
are
trying,
it
helps
the
ratio
of
50%.
It's
a
service
down
again.
This
public
service
only
used
it
with
Demonstrative
purposes.
Internally,
j
service
b
would
have
to
consume
and
in
fact
it
does
that
directly
here,
vernet
sorry
to
each
service
through
the
internal
network
of
cover.
Next,
ok,
first
service
is
online.
B
B
If
I
call
diagonal
hello,
it
tells
me
the
rest
and
if
it
does
not
find
diagonal,
why
bits?
But
it
did
call
freites
two
things
happen
and
j
service
be,
did
not
find
a
ca
service
to
the
same
micro
service.
Chassis
causes
this
error.
Sometimes
it
will
work.
What
does
it
say:
I
received
the
phrase
Gral,
well,
rock,,
pure
wisdom
by
Víctor
Of,.
B
Sometimes
he
received
another
phrase,
I'm,
already
more
without
the
guns
and
knives
for
twice
as
much,,
but
eventually
Hot
and
Service
falls
off
because
here
Service
and
is
not
able
to
give
it
a
response
at
the
time
it
falls.
So
this
would
directly
impact
a
front
end
because
a
500
error
arrives,
but
the
front
end
instead
of
receiving
the
full
500
at
least
says
the
system
is
under
maintenance.
Now,
well,
if
we
know
that,
eventually,
if
I
try
enough
times,
the
call
from
now
goes
to
cars,
this
is
going
to
work.
B
I
could
automate
this
execution,
so
in
fact,
I
could
do
it
using
a
rule.
It
retracts
directly
in
the
micro
service
chassis,
but
but
it
has
to
be
done
and
yes,
but
if
I
use
another
tool
in
this
case,
it
would
be
the
service
month.
I
can
automate
this
step
at
the
network
level
when
it
is
detected
that
there
is
a
communication
problem
in
the
network,,
it
would
gain
ability
to
observe,
knowing
that
this
error
exists,,
but
most
importantly,
instead
of
the
user
having
to
reload
to
obtain
the
correct
result,
I,
could
making
the.
B
Micro
service
chassis,
complemented
by
the
service
month
and
service
month,
given
the
error
condition,
allows
me
to
see
why
my
implementation
is
failing
now.
My
cluster
and
the
cluster
I'm
using
at
the
moment
already
have
the
link
ervey
installed.
Because.
The
installation
takes
a
little
time
if
I
had
not
done
it
in
view,
but
link
erdi
has
the
particularity
as
the
first
option.
It
gives
me
the
display
option,
so
we
say
linkedin
via
facebook
and
it
sent
it
to
the
background.
B
Now,
how
am
I
going
to
do
to
implement
linkedin
I
do
have
this
in
a
goat,
because
it
is
a
lar
command,,
well,
I
would
inject
all
the
deployments
in
my
current
cluster
with
a
tag
so
that
the
deployment
is
enriched
and
the
hervilla
link
can
monitor
them,.
So
I
will
leave
these
instructions
here
on
whatsapp.
B
What
I
do
then,
is
I
get
the
entire
deployment
plan
of
the
polyglot,
myspace
I
convert
it
to
call
I,
make
it
go
through
the
link,
earl
executable
and
in
my
system
and
I
replicate
it
again
with
this
I
win
that
the
deployments
that
originally
do
not
have
a
site
card
eventually
have
it
and
I
can
now
monitor
the
status
of
my
application,.
I
can
now
see
what
the
network
traffic
is,
dependencies
between
microservices,
and
know
what
is
happening
in
my
application,,
so
I
am
going
to
return
to
the
linkedin
dashboard
and
notice.
B
What
happened:
no
service
It
was
like
a
month
to
a
service,.
It's
been
like
a
month,.
This
eventually
I
can
place
it
as
annotations
and
metadata
in
my
deployment
plans
directly
in
the
posts
and
notice
that
someone-
and
they
are
asking
at
this
time.
It
was
not
me
here.
You
can
do
everything
after
you
want
about
the
ip
address
and
how
they
are
reloading.
The
ip
address,
thanks
for
your
help,
right
now
and
in
real
time,
I
am
seeing
that
each
service
is
failing
per
axis.
Recharge
tells
me
the
system
is
in
maintenance
mode.
B
I
received
the
java
phrase
is
under
maintenance.
You
guys
trust
the
ip
address
this
is
made
to
fail.
Sometimes
it
gives
a
good
response.
Sometimes
it
doesn't
and
the
more
requests
it
makes
to
you
as
it
is
intentionally
programmed.
It
will
approach
50%
of
failures,
one
out
of
two
times
the
more
it
is
used.
This
will
fail
to
win
then
with
the
introduction
of
the
month.
First
of
all,
look
at
ability,
as
well
as
I,
can
see
it
in
real
time.
B
B
For
example,
here
I
have
to
see
a
service
profile
in
which
I
say
when
it
is
invoked
through
the
internal
domain
name
of
q.
Vernet
is
here
service
to
please
obtain
a
retry
budget,
so
it
says
that
it
has
at
least
10
retries
per
second
and
the
considerable
minimum
of
failure
to
make
the
retry
is
20%.
I
already
know
that
this
fails
50
percent
intentionally.
It
is
done
like
this,
so
we
can
try
to
make
my
micro
service
fault
tolerant
now,
through
service
month,
so
I
say
each
service
to
service
prof.
B
The
point
we
call
and
what
this
does
is
give
tolerance
rules.
It
fails
linkedin
and
while
linker
deep
takes,
the
new
rule
will
happen
as
follows:
ok
keep
giving
the
ip
address
note
that
now
it
does
not
fail.
I
know
that
the
code
is
wrong.
I
already
demonstrated
that
the
call
is
wrong,
but
with
the
correct
tools
helping
me
both
the
micro
service,
chassis
and
the
service
month,
I
can
give
my
users
a
continuous
experience
of
what
is
also
known
as
reactive
applications
in
real
time.
B
If
I
can
see
within
linkedin
that
there
is
a
dependency
where
j
service
depends
on
caser
vista
and
although
the
calls
you
make
work
correctly
abroad,
Internally,
link
early
continues
to
report
this
service
for
the
moment,,
which
is
in
a
month,,
although
it
may
be
that
your
service
profile
shows
users
that
something
internal
is
failing,,
fixes
it
eventually,,
but
you
are
failing
with
style,.
So
these
are
the
two
ways
I
wanted
to
show
you.
In.
This
small
talk
on
fault,
tolerance
with
micro,
service
chassis
and
service,
month,
well,
I
had
a
fourth
presentation,
so
I.
B
A
B
More
than
schemes
I
would
say
that
they
are
patterns
personally.
I
think
that
the
minimum
that
all
michael
services
include
are
these
four,
these
five
sorry,
the
meis
service
in
general,
only
includes
retracts
and
time
out
so
make
an
advanced
fault.
Tolerance,
rule
It
implies
at
least
as
of
today,
because
this
is
a
recording.
You
can
see
the
one
that
a
year
may
have
already
changed
if
you
want
'circuit
breaker',
just
like
you
probably
have
to
see.
B
A
B
It
would
be
a
perfect
storm,
because
if
I
have
the
fault,
tolerance
rules
and
guber
net
is
doing
the
constant
synchronization
of
the
plan
versus
the
current
and
the
moment
that
it
is
going
to
fall
is
that
when
the
micro
recovers'
net
is
no
longer
endured,
because
my
application
was
wrong
or
I
know
that,
ideally,
this
should
not
fall.
I
know
that
the
microservices
are
going
to
fall,
but
the
observability
tools
would
also
give
me
this
guide
to
know.
B
Look
here
deployed
while
service
that
only
responds
well
20%
of
the
time
everything
has
been
fixed
well
with
which
rulers
it
is
restarting
every
time,
but
ideally
when
it
has
a
lot
of
governing
traffic
is
going
to
restart
so
much
that
it's
not
going
to
be
able
to
recover
past.
And
yes,
I've
seen
a
couple
of
situations
in
the
scenario.
A
B
The
case
of
income,
the
income
is
particularly
used
to
have
a
centralization
suppose
that
my
domain
goes
up
orozco
dotcom
months.
I
could
have
the
application
1.1
euros,
co
point
as
an
application,
2.1
euros
with
application,
3
point
v
orozco.
If
I
do
it
through
the
oad
balance,
to
publish
up1
up
2
and
up
3,
these
balances
would
have
to
have
their
own
ip
address,
as
I
did
in
the
demo,
but
having
their
own
ip
address
implies
cost.
B
So
in
one
scenario,
most
common
I
would
deploy
an
income
operator
like
contour
or,
for
example,
in
English
yen,
which
is
the
most
common
and
I
would
have
a
single
ip
address,
application,
1
point
euros,
2
point
and
application.
3.1
orozco
would
arrive
at
the
same
ip
address
and
before
falling
directly
like
step
from
this
moment,
they
would
fall
to
english
english.
B
B
The
most
popular
one
existed,
I
have
seen
more
niche
deployments
than
linkedin,
but-
and
here
is
the
but
when
you
are
going
to
deploy
and
this
site,
if
it
is
a
design
consideration,
because
this
has
its
own
enter
your
own
site
cards,
then
it
requires
more
planning.
If
you
are
a
team
that
has
been
using
coherent
for
a
long
time,
surely
and
the
site
is
going
to
adapt,
but
not
that
one
It
is
the
reality,,
so
I
often
gave
advice
to
teams
that
already
had
rulers
and
they
learned
it.
The
hard
way.
B
A
A
Also
remember
those
who
are
watching,,
those
who
can
participate
in
the
raffle
for
a
book
by
making
the
hashtag
on
Twitter.
acb
gti
falls
that
then
there
you
can
participate
to
win
a
book
and
also
thank
you
very
much
reminding
you
that
in
the
other
tracks,
we
have
more
talks
so
that
you
continue
tuning
in
and
well
nothing.
Nice
was
a
pleasure
and
thank
you
very
much
for
your
time.
Thank
you
very.