►
From YouTube: 2023-01-17 CNCF TAG Observability Meeting
Description
OpenTelemetery Profiling Update
Meeting notes and more: https://github.com/cncf/tag-observability
B
B
C
C
A
C
C
C
C
C
B
B
B
Ryan
perry,
(pyroscope):
hey?
yeah,,
it's
not
worth
going
through
the
whole
thing.,
but
basically
so
you
know,.
We
are
tomorrow
actually
meeting
with
the
collector.
sig..
This
presentation
was
prepared
for
that,
and
we're
having
some
discussions
today.,
and
so
the
idea
is
that
you
know
we
want
profiling
to
be
treated
in
a
similar
way
to
the
other
signals
in
hotel.,
so
logs,,
metrics,
and
traces,
and
being
able
to
be
correlated
with
those
signals.
Pretty
easily.
B
B
B
Ryan
perry,
(pyroscope):.
You
know
why
we
want
to
add
it..
What's
the
current
state
making
it
compatible
with
other
signals,
standardizing
profiles?
there's,,
you
know,
a
1
million
different
formats
for
profiles,,
and
so
we
want
to
kind
of
start
to
condense,
those
more
and
more
and
ideally
come
up
with
sort
of
like
1
one
format,
tool
them
all,,
you
know,.
I
would
never
easy
to
create
a
new
format.
but
yeah,
hopefully,
that
works
out.
B
B
Ryan
perry,
(pyroscope):.
You
know
now
that
we
have
this
idea
of
what
profiling
should
be..
We
know
we
want
to
support
p.
prof.
and
jfr.,
which
are
the
2
main
profiling
formats
that
exist
currently.,
and
then
we
know
we
want
to
create
a
slightly
more
hotel
specific
version
of
those
called
we're,
probably
gonna
call
it
oprop.
B
B
Ryan
perry,
(pyroscope):
and
then
and
then
yeah..
So
now
the
part
we're
trying
to
figure
out
is
like,.
What
do
we
do
with
the
collector
like??
How
do
we,
you
know,
support
all
of
these
formats,
and
then
get
them
from
the
collector
to
their
respective
backends.?
What
happens
if
there's
not
a
collector
all
that
kind
of
stuff.,
and
so
we're
planning
on
meeting
with
the
collector.
sig,
basically
to
figure
out,,
you
know.
B
Ryan
perry,
(pyroscope):
yeah,,
we're
trying
to,
you
know.,
ryan
perry,
(pyroscope):,
explain
this
idea
and
then
discuss
these
various
questions,
so
that
when
we
do,
you
know,
sort
of
like
add
profiling
to
the
collector,,
we
do
it
in
a
way
that
is
consistent,
and
that,.
You
know,
keeps
the
hotel
community
happy.
F
B
B
B
B
B
B
Ryan
perry
(pyroscope):,
you
know,
just
have
that
for
the
conversion
thing,
or
maybe
maybe
we
switch
it..
Maybe
we
convert
it..
You
know
what
I
mean,
like.
There's
there's
a
lot
of
options
there.
but
yeah,.
We
we'll
make
sure
that
that
ability
is
available
somewhere.,
it's
just
a
matter
of
where
that's
available.
F
F
F
B
F
B
B
B
Ryan
perry,
(pyroscope):
understand,
like,.
What
are
the
par,?
What
are
the
constraints?
Within.
you
know
of
what
we
can
and
can't
do.
and
basically
we'll,
you
know,
stretch
those
as
far
as
we
can,,
assuming
that
it's
you
know.,
assume
that
that's
allowed,
for
example,
like,
yeah,.
We
should
totally
do
that.,
I
mean,
that
makes
a
lot
of
sense.
honestly,
I'll.
Add
it
to
the.
B
C
C
F
C
B
B
Ryan
perry
(pyroscope):,
but,
however,
they
do
it..
That's
you
know,
kind
of
the
the
ideas
that
profiling
would
do
it
the
same
way..
I
don't
see
any
reason:
why.
you
know,
just
from
a
fundamental
standpoint,
that
we
wouldn't
be
able
to
do
it
the
same
way
in
profiles
as
we
do
in
in
those
other
signals.
B
Ryan,
perry,
(pyroscope):
and
that's
part
of
the
reason.
yeah,
for
this,,
like
new
format,,
it
might
make
it,,
I
guess.
yeah,.
The
the
only
time
it
would
be
hard
to
do
is
if
we
try
to
like,
cram
it
into
like.
jfr.
for
example,,
like
that
gets
a
little
little
tricky.,
but
that's
kind
of
the
reason
for
the
new
format
is,
you
know,,
making
it
easy
to.
C
C
Matt
young:
of
of
processing
that
I
think
matt
young:,
that's
a
that's
a
relevant.
ryan,
perry,
(pyroscope):,
so
so
yeah,.
So
the
the
interesting
thing
with
profiling
is
that
you
know
it,
like
p.
prof.
is,.
You
know
definitely.
very
good,,
you
know..
Obviously
it's
been
refined..
It
was
created
by
google,
like,,
you
know,,
it's
like,,
not
a,.
You
know.
B
Ryan
perry,
(pyroscope):
yeah,,
it's
not
just
like
a
random
format.
and
so,
you
know,
for
us
to
like,,
do
anything
to
like,,
extend
or
change.
p.
prof..
It's
like,
you
know,
definitely
has
to
be
justified.,
but
at
the
same
time,
you
know,
p.
prop,
like
I
said,
wasn't
created,
for
you
know
this
like
continuous
profiling
use
case,,
which
is
now
where
most
you
know,
observability
vendors
are
heading.
What
you
know
the
way
people
are
using
profiling
is
sort
of
shifting
more
from,.
You
know.
B
Ryan
perry
(pyroscope):,
like
profiling,
a
random
process
on
a
machine
or
profiling..
You
know
a
script
that
you're
running
locally
to
a
more
sort
of
like
whole
system,.
You
know,
always
on
profiling
model
and
and
p.
prop,
isn't
the
most
optimized
for
that,
and
it
just
doesn't
make
sense
to
like,
you
know.
change,
p.
prof.
directly.
We
have
one
of
the
p.
prop
maintainer.
that,,
like
say,,
is
part
of
the
working
group
at,
you
know,
and
he's
like,
yeah,.
It's
just
like.
B
Ryan
perry
(pyroscope):,
you
know,
like
not
realistic
at
the
same
time,.
You
know
you
have,,
like
the
go
runtime
which
uses
p.
prof.
and
they
also
said,
you
know,
for
now
p.
prof.
does
what
they
want
it
to
do..
They
don't
want
to
like
switch
to
a
new,,
you
know,
format,
necessarily
yet,,
but
they're
like
they
would
say
that
they
would
be..
We
also
had
someone
from
one
of
the
p.
pro
or
one
of
the
go.
Runtime
maintainers
come
to
our
meeting,
and
he's
like,.
You
know.
B
B
Ryan
perry,
(pyroscope):
data
points
before
them.,
so
in
the
meantime,
you
know,,
we
want
to
still
be
able
to
support.
go,.
You
know
p.
prof.
and
then
still
be
able
to
support
jfr:,
while
also
sort
of
introducing
this
new
format..
So
what
we're
gonna
right,
now,?
What
we're
presenting
is
this
architecture
where
you
know
the
back
end
can
sort
of
decide
what
which
one
of
those.
B
B
B
Ryan
perry
(pyroscope):
to
the
back
end,,
in
which
case
the
back
ends,
would
have
to
decide
like,.
Do
we
accept
p.
prop:
and
do
we
accept?
jfr:
and
you
know,
yeah,?
They
can
kind
of
decide
how
they
want
to
do
that,
like
maybe
they
convert
it
to
oh,
prof.
on
their
side,.
You
know
the
back
end.,
but
it
kind
of
allows
a
transition
period.
B
Ryan
perry
(pyroscope):
to
go
from,
and
also
the
ability
to
kind
of
compare
the
3
like
what
happens
if
you
just
send
it
as
p.
prop.,
what
happens
if
you
send
it
as
so.
prop.?
What
happens
if
you
send
it
to
jfr.
and
being
able
to
see
you
know
which
one's
more
efficient,,
which
ones
you
know
better,
for
whatever
reason
you
know.
B
B
Ryan
perry
(pyroscope):
current
proposed
architecture,
again.
like,,
you
know,.
It
might
turn
out
that
converting
on
the
collector
is
like
too
much,
you
know,
too
heavy
for
the
collector,
in
which
case,
you
know,.
We
have
to
figure
out
some
other
way
to
handle
that.,
but
that's
kind
of
the
the
set
of.
C
C
matt
young:
you
know,
by,
but
by
doing
it
this
way
we
could
kind
of
have
an
apples
to
apples
comparison
on
a
couple
of
different
fronts,,
one
from
existing
collection
and
distribution
mechanisms
for
things
like
p.,
prof.
or
tofr
in
their
in
their
normal
form,
and
then
use
the
collector
to
shovel
bits..
That's
a
nice
comparison,!
You
know.
C
C
Matt
young:,
so
every
time
that
I'm
glad
is
still
part
of
the
conversation
is
in
in
talking
to
various
consumers.
profiling,
stuff.
I'm.,
one
of
them,.
You
know
where
the
where
the
cost
of
conversion
happens
is
gonna,
be
different
or
different.
Scenarios..
Some
people
don't
want
to
do
it
at
the
point
of
collection
at
the
edge.
C
Matt
young:
because
it
makes
sense
to
do
so.
there.
Others
are
going
to
want
to
do
it
very
much,.
Very
strongly
feel
that
it'll
be
done
all
on
the
back
end..
So
I
think,
by
leveraging
hotels,
architecture,
and
letting
people
mix
and
match
where
they
do
these
things
you
don't
have
to
have
those
trade
off
discussions.
Of.
is
this
too
expensive
for
the
collector
or
not
expensive
enough
for
the
collector,
you
know,
and
just
have
a.
C
C
C
Ryan
perry
(pyroscope):
at
this
point.
yeah,,
I
mean,,
I
think
I
think
we're
getting
close
to
that
sort
of
like,.
You
know,
request
for
feedback
from
the
community
type
of
thing
like
how
for
those
who
are
using,,
you
know
the
collector,
for
example,,
you
know,
like,.
Do
you
feel
like
it's
so,,
you
know,
like
it's?
It's
too
busy
already,
you
know,,
like
I,
I
guess.
yeah,,
there's
we're
close
to
that
point..
I
think.
B
ryan,
perry
(pyroscope):.
I
keep
getting
it
backwards.
sig:,
yeah,,
yeah,,
the
same.
and
and
yeah,
I
mean,
and
I
think
after
that,,
then
I
think
we
will
probably
go
back,,
make
a
couple
of
changes,
and
then
we're
getting
pretty
close
to
doing
like
an
official
o
tap,
for
you
know,
a
a
proposal
for
this
yeah,.
For
for
this
new
format,,
you
know,
before
we
just
kind
of
did
a
proposal.
B
Ryan
perry
(pyroscope):
about
the
proposal,,
you
know.
just
kind
of
like,
hey?
we're
going
to
propose
something..
This
is
like
the
direction
we're
going,,
but
this
one
will
be
more
of
a
sort
of
like
technical
otep,.
You
know.
with
these
are
the
fields,.
You
know
kind
of
what
heinrich
mentioned.
Earlier,
like
those
types
of
questions..
Those
that
will
probably
be
when
those
become
more
are
definitely
be
where
those
become
more.
B
C
F
F
F
B
Ryan
perry
(pyroscope):,
you
know.
hack,
it
into..
If
you
do
have
p.
pro.
for
example,
p.
prop,
has
labels.
you
can
sort
of
put
like
the
span,
id,
or
or
whatever
it
is
in
those
labels.,
but
again,.
That's
like
just
not
the
most
efficient
way
to
do
it,,
especially
if
you're
needing
to
kind
of
like
access,
that
that
key
easily
and
so
or
that
label
easily..
And
so
one
of
the
you
know,
proposed
things..
It's
like
a
fairly
simple
change,,
but
it's
just
making
that
a
field
all.
B
B
F
F
F
B
Ryan
perry,
(pyroscope):
dashboard
side
like
where
you
then,
you
know,
once
you
send
it
somewhere,
where
you
view
it.,
but
I
guess
theoretically,
that
should
probably
be
where
you
view
the
profile
is
probably
the
same
place.
you
send
the
profile,,
and
so
that
is
one
thing
that
we
did
talk
about..
Another
piece
of
this
was
that
you
know
right
now,
so
many
different.,
you
know
backends
use
different
formats,.
You
know,
pyroscope,
use
a
different
one
from
data
dog,,
which
is
a
different
one
from
elastic,,
which
you
know.
so
on,
and
so
forth.
B
Ryan
perry
(pyroscope):,
I
just
direct
it
to
whichever
back
end.
You
want
similar
to,,
you
know,,
like
the
that's
one
of
the
main,
you
know,,
sort
of
like
tenants
or
missions,,
or
whatever
you
want
to
call
it,
of
hotel,,
is
being
able
to
kind
of
substitute
in
or
out..
I
think
it's
like,
you
know.
yeah,
for
tracing.
You
can
have
just
like
your
tracing
provider.
B
F
B
B
B
C
C
B
B
C
C
C
B
Ryan
perry,
(pyroscope):
yeah,,
I
think
it
kind
of
depends
a
little
bit
on
honestly,.
You
know
how
the
meeting
goes.
Tomorrow,
you
know,.
If
the
collector
sig
is
like.
well,.
This
is
like
so
far
from
you
know,,
something
that
we
would
do..
I
don't
think
that
would
be
the
case,,
but
our
goal
is
to
have,
you
know,
sort
of
like
hit
that
next
milestone,,
which
is
what
you
just
mentioned.
Like,
you
know,
having
it
out
of
phase
where
you
know.
at
least
there's
yeah,.
B
F
Ryan
perry
(pyroscope):
mid
of
april.
yeah.,
so
yeah,,
so
our
goal
is
to
you
know,
at
least
have
something,
or
be
very
close
to
something,,
so
that,,
you
know,
coupon
is
always
like
a
nice
place
to
sort
of
like
regroup
with
everyone
and
and
explain
the
progress
and
all
that
kind
of
stuff.,
and
so
that's
sort
of
what
we're
aiming
for,
timeline
wise,,
but
we'll.
B
C
Ryan
perry,
(pyroscope):
draft
and
ready
for
feedback,
or
some
degree
of
implementation
or
prototyping
done,.
I
guess.
yeah,
the
oats.
Definitely
the
oatsep
side
of
things,.
I
would
say,
yeah,
the
actually
having
it,
you,
know,
sort
of
like,
ready
for
people
to
use
is
a
little
bit
more
ambitious
for
that
timeline,.
But.
B
B
B
B
C
D
D
D
D
D
D
D
D
D
D
F
F
F
F
F
D
C
C
C
C
C
C
C
C
C
C
C
C
Matt
young:,
gentlemen
from
signals
was
one
of
the
people.
that
kind
of
brought
this
up
a
couple
of
times,
and
then
that's
what
motivated
the
question.
so
so
open
telemetry
is
pretty
firm
on
on
a
working
group
like
this
should
not
exist.
Now
in
the
technical
advisory
group.-
you
know
we
have
vendors
and
users,
and
you
know,
and
and
and
project
maintainers
all
come
together..
We
have
a
lot
of
latitude.
C
C
C
Matt
young:
and
we
have
an
example,
working
group
that
was
launched
by
ken.,
so
we
have
sort
of
a
model..
But
what
you've
done
here
in
the
doc,,
from
what
I
can
see,
where,
with
a
cursory
read,
is,
is
up
the
center
line..
So
I
would
suggest
that
we
can
put
out..
I
send
an
email
to
the
tag
observability
list.
C
C
Matt
young:
at
a
high
level,
without
reading
all
of
the
things
working
groups
really
just
have
to
do
a
couple
of
things..
They
have
to
be
time
bounded
right,
so
they
can
go
on
for
a
long
time.,
but
they
do
have
to
have
a
line
of
the
sand.
That
says
this
is
the
rough
scope
of
what
we're
trying
to
do,,
and
this
is
the
timeline,,
and
this
is
what
the
outcomes
we
hope.
C
C
C
Matt
young:
in
in
an
open,
open
source
way..
So
I
think
what
you're
done
is
great
again..
I
would
you
just
send
an
email
to
the
mailing
list,
take
a
couple
of
weeks
of
feedback,
and
perhaps
our
next
tag
meeting
in
2
weeks
from
today,,
which
would
be
the
2
weeks
and
change,
will
be
the
first
to
the
the
first
tuesday
of
february.
We
meet
on
the
person
third
tuesdays,
and.
C
C
D
C
Matt
young:,
maybe
depending
on
how
long
you're
gonna
run
it,
like
maybe
monthly,
updates
to
the
tag.
or
if
you
want
to
come
and
and
give
it
a
quick.,
I
mean
it's
depending
on
how
quickly
you're
going,
and
how
much
feedback
and
exposure
you
want
to
get,
and-
and
we
can,,
you
know
we'll-
have
in
the
notes
this
year.
I'm.
I'll
leave
it,
and
I
are
going
to
endeavor
to
do
a
much
better
job.
Of.
C
B
B
B
Ryan
perry,
(pyroscope):
this.,
you
know,
sort
of
structure
here.
but
yeah,.
They.
They
basically
gave
us
this
for
the
profiling
group,
as
well
of
like
the
things
that
you
need
in
order,,
like
the
check
boxes,
you
need
to
tech,
also.
and
and
if
you
want
to
create
a
working
group
like
getting
a
tc.
member
to
sponsor
it
and
all
that
kind
of
stuff,.
So
it's
a
lot
of
stuff
that
matt
just
said.,
but
I
think
this
is,
and.
B
C
C
C
D
C
C
Matt
young:
this
will
now
be,.
I
think,
the
second
or
third
work
screen
group
that
the
the
tag
is.
this
tag,
is.
look
to
launch..
We
probably
have
critical
mass
now
to
to
do
the
issue
around
formalizing
how
to
do
a
work
group,,
but
some
templates
to
jump.
start
what
you're
about
to
do..
So
perhaps
we
just.
D
C
C
C
C
C
Matt
young:,
but
I
I
I'm
certain
that
either
through
the
end
user
community
there's
some
mechanism
to
do
that.,
you
could
also
do
a
survey
and
put
it
out
to
things
like
the
you
know,,
the
the
back
slack
channel,
and
or
the
mailing
list..
We
have
almost
about
800
people
or
something
like
that,
and
the
tag
is
a
really
channel..
So
if
you
wanted
to
make
a
survey
as
a
tag,
member.