►
From YouTube: 2023-02-07 CNCF TAG Observability Meeting
Description
(Vijay Samuel) "Why and How eBay Pivoted to OpenTelemetry"
https://opentelemetry.io/blog/2022/why-and-how-ebay-pivoted-to-opentelemetry
The Open Telemetry initiative is providing observability practitioners a vendor neutral and an open standard by which applications can ship telemetry data into observability platforms. eBay is a well documented adopter of Elastic Beats. eBay has been an extensive contributor to the Beats project and contributed its Kubernetes autodiscover module which can dynamically monitor pods on Kubernetes clusters. However, with the rest of the observability ecosystem converging onto Open Telemetry, we felt that it is essential for us to adopt the same. This presentation focuses on:
* scale problems seen in large Kubernetes installations
* benefits of using annotation based discovery at scale
* our journey of replacing Metricbeat with Open Telemetry Collector without impacting metric collection
* bridging critical gaps such as:
* dynamic configuration reloading
* semantic differences between Prometheus receiver and stock Prometheus
* challenges faced in ensuring data parity post migration and solutions
---
(Chris Larsen) Observability Query Standardization WG Charter Review (doc)
---
(Matt Young) CNCF Landscape Graph
Meeting notes and more: https://github.com/cncf/tag-observability
E
E
E
E
E
E
E
G
D
D
D
D
D
D
D
D
D
D
D
D
D
D
D
D
D
D
D
D
D
D
D
D
D
D
D
D
D
D
D
D
D
D
D
D
D
D
D
D
D
D
Vijay
samuel:,
there
are
some
objects
that
are
not
scoped,,
which
means
that
for
the
entire
cluster
you
need
to
bring
that
entire
metadata
in
like
namespace,
or
deployments
or
stateful
sets.,
so
this,
where,,
while
you
do
it
from
every
node
on
very
large
notes..
The
amount
of
pressure
that
you
put
on
the
api
server.
D
D
D
D
D
D
D
D
D
D
D
D
D
D
D
D
D
D
D
D
D
D
D
D
D
D
D
D
Vijay
samuel:,
so
through
that
process
we
vijay
samuel:
ensured
that
we
submitted
pr.,
sending
the
prs
and
bob
reported
bugs
to
say
that.
okay,
the
this
is
what
we
are
seeing
is
this
expected
for
open
shell,,
which
a
collector,?
If
it's
not
expected.,
then
we
went
in
and
submitted
the
ts.
that
are
required
to
actually
go
and
fix.
It.
D
D
D
D
D
D
D
D
D
D
D
D
D
B
B
B
D
D
E
E
D
D
D
D
D
D
D
Vijay
samuel::
do
you
use
a
one
tier,
open,
telemetry,
collector
or
2
tier.?
How
does
how
do
you
set
up
bread
metrics??
How
do
you?,
where
does
the
tail
sampling
fit
in
like
the
these
are
some
things
that
we
are
improvising
as
we
see.
okay,,
this
probably
works
better,,
or
this
would
probably
cost
a
lot
less
than.
D
D
E
E
D
D
E
E
C
C
C
C
C
C
C
C
C
C
C
C
E
E
E
E
C
C
C
F
H
F
F
F
F
E
E
E
E
B
E
E
E
E
E
E
E
E
G
G
G
Matt
young:,
this
is
gonna,
go
really
fast,,
but
it
aims
to
answer
questions
like
this.
If
I'm
an
it
decision
maker-
and
I
want
to
use
2
or
3
projects
next
quarter,,
who
are
all
contributed
to
them.
who
do
they
work
for?,
who
funded
the
companies,
they
work
for,
and
what
else
do
they
fund
like?
If
I
care
about
looking
at
flow
of
capital
for
the
ecosystem?
or
if
I
care
about
what
projects
have
had
contributor
bases
that
are
decaying
a
little
bit.
and
what
else
have
they
gone
on
to
go??
G
G
Matt
young:
we're
talking
here
about
a
labeled
property,
graph,
namely,,
the
relationship
between
things
have
types
and
can
also
have
metadata.
right?,
so
you
can
go
between
the
2..
It's
just
that
legal
property.
graphs
also
sometimes
end
up
more
giving
you
a
more
useful
structure,
because
already
has
a
lot
of
noise,,
because
every
little
property
of
every
little
relationship
has
to
be
a
tone.
G
G
Matt
young:
projects
for
and
scope
for,
tags.
projects
can
have
roles..
There's
good,,
you
know.,
you,
you
get
the
idea..
This
is
a
big
giant
knowledge
graph
that
uses
open
cipher
as
the
query
language.
and
if
you're
not
familiar,,
it's
simple
for
graphs..
It's
a
textual
kind
of
description
that
lets
you
write,
match,
queries
and
things
like
that.
G
Matt
young:
there's
a
layer,
I'm
going
to
briefly
cover
elements
of
the
tech
stack
here
in
the
last
1
min.,
there's
a
layer
that
neil
for
j
open
source
that
goes
from
an
arbitrary
graph
to
all
query
to
an
arbitrary
open,
cipher
query..
This
is
useful
for
nested
mutations
in
particular,
and
solving
the
n
plus
one
problem.,
if
you've
ever
messed
around
with
graphical
resolvers,
with
relational
backing
stores..
So
that's
kind
of
neatly
solved.
I'll
put
in
the
deck
kind
of
just
a
brief
overview..
G
G
Matt
young:,
you
know
same
thing..
You
can
also
layer
in
authorization
into
the
graphql
sdl:,
so
this
makes
for
a
nice,,
clean,
design..
There's
a
notion
of
graphical
federation.
apollo
has
it..
There
are
other
open
source
mit
alternatives,
I
put
in
here
to
allow
for
a
super
graph
to
be
composed
of
sub
graphs..
Now
each
of
these
sub
graphs
can
have
whatever
behind
it,,
but
typically
a
graph
database.
in
this
case,.
I
think.
G
G
Matt
young:
2
to
3
digits
order
to
magnitude,
to
value
better
performance,,
and
I
can
actually
scale
to
larger
data
volume
to
charge..
It
does
query,
planning.
it
reassembles,
results
and
things
like
that..
So
we
have
a
nice
architecture
to
make
a
a
composable
set
of
sub
modules
that
this
project
defines.
G
Matt
young:
to
metadata,
from
github
to
all
of
this
other
data
that
we
have
aggregated
in
the
landscape
already..
We
want
to
have
a
sub,,
a
module
or
composable
architecture,,
so
that
organizations
can
mix
and
back
to
public
and
private
data
internally
right.
so
use
some
public
data
sets
and
and
have
a
composable
way
to
kind
of
put
legos
together,,
but
also
so
they
can
be
ci,
and
they
can
be
developed
in
parallel..
G
Matt
young:
that
ensures
that.,
matt
young:
and
things
like
that.,
so
it's
quite
nuanced
and
quite
nice..
So
the
idea
was,,
I
kind
of
defined
a
project.
and
here's
a
a
rough
snapshot..
I've
defined
a
bunch
of
work.
streams
in
many
cases,
approved
out
mvps
of
the
design
to
prove
that
it..
It
works
along
with
some
alternatives.
G
Matt
young:,
I
basically
got
a
well
formed
project.
now,
that
just
needs
contribution,,
because
my
ego
is
not
that
big,
and
there's
only
so
many
hours
in
the
day,,
but
I
really
wanted
to
make
something
that
people
could
engage.
With.
finally,,
we'll
talk
more
next
time
when
we,
we
might
have
another
guest
speaker
from
klein.
G
G
Matt
young:
domain.,
the
domain
of
this
project
is
a
glorified
rolodex
and
or
analysis
tool
for
the
cnc.
f.
and
tags
and
project
maintainer,,
and
whoever
else
is
interested
in
that
in
that
data,
and
it's
meant
to
be
non
confrontational
and
not
an
observability
tool,.
But
a
way
for
people
who
are
interested
in
the
governance
of
open
source.
G
G
Matt
young:,
so
that's
kind
of
why
I've
kind
of
carved
it
out
with
this
domain
and
this
tech
stack.
but
again,.
I
do
believe
this
tech
stack
has
brought
up
the
ability,
and
we'll
be
talking
more
about
that
on
next
time
on
the
fourteenth
of
valentine's
day,.
So
we
can
have
that
we
love
graph
state.,
but
anyway,.
I've
left
links
in
the
back.
G
G
Matt
young:,
that's
apache..
There
is
an
mit
set
of
projects
out
of
the
guild
to
have
a
full
stack,,
I'm
using
some
of
them
in
this
project,
like
the
code
jen.
one
of
the
nice
things
about
that
is
for
interface
definition
languages..
You
can
just
pick
your
language
and
generate
language
findings,
just
like
you
would
with
grpc
and
proto..
So
there's
an
mit
variant
to
all
of
this..
I
started
prototyping
with
the
out
for
j,
because
it
was
expeding
them.
G
Matt
young:
the
graph
data,
science
to
etl
pipelines
and
stuff,
like
that,
that
are
all
in
scope
for
this
project,
and
it's
not
so
narrowly
defined
that
everything's
designed..
There
are
some
basic
ideas
that
have
been
proved
out,
and
a
lot
of
alternatives.
There's
still
a
lot
of
choices
to
make,
and
a
lot
of
really
cool,
innovative
design,
work
that
can
be
done.