►
From YouTube: OpenTracing Spec Council Monthly Call - Aug 4 2017
A
The
call
last
month,
I
started
that
I
was
unable
to
do
it.
I
was
like
I.
Had
a
one-day-old
kenderson
I,
don't
remember
what
it
was
not
a
good
time.
Sick,
try
and
take
the
same
so
I
couldn't
do
it.
I
watched
the
video
which
makes
let's
see
so
boss
had
also
requested
that
we
don't
make
any
like
giant
decisions
or
anything
today,
which
I
think
is
reasonable
obvious.
It
seems
like
a
month
where
it's
difficult
to
find
a
time
that
works
for
everyone,
but
I,
don't
investor
a
big
problem.
A
I
also
wanted
to
introduce
Jonathan,
cobbler
key
and
Facebook,
showing
the
IAB
I.
Don't
know
two
weeks
or
something
Jonathan
just
say
a
few
words
about
your
background.
Tv
plays
you
can
do
in
there
can
ever
hear
me.
B
Yeah
so
hi
I'm
Jonathan
I
work
at
Facebook,
I!
Guess!
If
we're
going
all
background,
I'm
coming
originally
from
a
computer,
graphics
background
doing
cloth
simulation
came
to
Facebook
I've
been
doing
working
on
the
tracing
team
here
for
about
three
years
now,
with
three
years
before
that
working
on
mobile
and
android,
and
so
we've
basically
developed
an
in-house
tracing
system
that
we've
deployed
across
both
our
distribute
back-end
services
or
web
tier
and
our
mobile
applications
and
we've
been
talking
both
with
uring,
been
about
like
problems
in
tracing.
C
A
Yeah
nice
to
meet
you,
oh
sorry,
I,
just
muted.
The
purpose
of
the
industrial
advisory
board
isn't
strictly
to
find
people
who
are
using
open
tracing,
but
just
to
find
people
who
are
using
tracing
and
can
help
us
inform
teacher
future
development
and
so
on,
as
opposed
to
creating
kind
of
a
cycle
in
the
graph
where
we
don't
get
you
guys
from
that
side
and
the
work
that
this
book
is
done
is
pretty
interesting
and
they've
also
been
nice
enough
to
publish
some
of
it.
So,
oh
yeah,
it's
great
to
have
you
here
anyway.
A
I
do
want
to
talk
about
that
spring
cleaning
thing
I'm!
Just
going
down
the
agenda
for
assault,
yeah
yeah,.
D
So
I
feel
like
the
github
issued
section
on
various
projects
have
started
to
get.
You
know
a
little
dusty
and
long
in
the
tooth,
so
I'm
just
going
to
be
going
around
at
the
next
couple
of
weeks
and
for
all
outstanding
issues
that
don't
look
like
they're
actively
moving,
try
to
either
you
know,
get
them
closer
archives
or
or
get
people
to
start
moving
on
them
again.
So
I
did
this
once
before
with
Java,
and
so
just
another
end
of
that.
If
you
see
me
in
there
asking
questions,
that's
why.
A
Listened
to
my
knowledge,
the
most
debate
has
been
going
on
around
with
Aptos
spam
thing.
So
for
context,
I,
don't
know,
I
can't
get
my
month
shape.
That
I'd
say
like
three
or
four
months
ago.
We
merged
in
Pakistan
supported
the
java
issues.
For
those
who
aren't
like
intimately
involved
in
this
stuff
was
basically
the
feature
that
allows
you
to
statically
determine
what
the
current
stand
was
and
a
distributed
trace.
So
there's
a
static,
look
up
front,
a
little
tracer
and
then
the
global
tracer
have
this
new
capability
to
give
you
the
active
stamp.
A
In
order
to
make
that
happen
there,
you
know
where
a
bunch
of
other
API
is
that
have
to
be
modified,
and
that
went
in
with
a
great
deal
of
review
in
the
sense
of
the
PR,
but
literally
was
breaking
github
and
generating
500,
because
they're
too
many
comments,
but
it
wasn't
reviewed
by
everyone
who
care
about
it.
So
it's
sort
of
like
this
issue
where
it
had
a
lot
of
scrutiny,
but
not
as
shows
freed
me
from
everyone's
different
by
it
that
went
out
and
then
I.
A
C
A
Others
were
also
sort
of
like
well
yeah.
I,
know,
I,
wasn't
actually
that
happy
with
like
this
aspect
of
it
either
and
these
things
came
out
or
after
it
had
been
merged
into
master.
So
the
options,
of
course,
would
be
to
say.
Well
sorry,
I'm,
like
we
have
you
had
your
chance
onboard
of
trying
to
fix
it.
A
I
think
obviously
were
trying
to
fix
it
different
and
it
should
review
the
projects
and
how
much
the
feature,
in
my
mind,
is
probably
like
the
linchpin
of
making
tracing
in
general
and
easier
to
use
like
this
is
the
thing
that's
going
to
make
not
just
I've
been
tracing
the
tracing
in
general,
like
more
kind
of
fun
portable
across
package
boundaries.
So
it's
pretty
important
to
get
it
right.
A
I
have
a
PR
that
I
sent
out,
which
the
eyes
LinkedIn
the
dot
strata,
PR
166,
that's
kind
of
like
the
RSC
for
a
way
has
addressed
the
concerns
that
have
been
expressed
with
print
API
and
on
that
thread
and
I
think
also
on
a
specification
thread.
There
are
some
other
suggestions
for
ways
to
mitigate
some
of
the
issues
that
people
are
having
with
the
current
API
I
would
like
I,
don't
think
I
could
underline
enough.
D
I
just
wanted
to
say
I
in
general,
I
am
drafting
a
sort
of
proposal
for
how
we
do
these
API
changes
in
the
future.
Just
like
Ben
was
saying
to
make
sure
we
have
a
sort
of
checklist
or
some
approach
so
that
we
don't
end
up
in
a
situation
where
someone
is
like.
Oh
yeah,
it
matters
what
I
think,
but
I
didn't
say
anything
or
I
didn't
know
those
kinds
of
situations,
and
then
my
other
meta
note
is
yeah.
I.
Think
code
is
helpful.
D
I
feel
like
with
these
discussions,
they're
so
technical
in
arcane,
and
it's
so
easy
for
people
to
speak
past
each
other
like
colloquial
English
and
like
a
giant
PR
thread
just
sort
of
like
runs
out
of
steam,
like
you
get
a
certain
amount
of
consensus,
but
it
seems
like
it's
very
difficult
to
get
over
that
final,
just
with
Coco
English,
and
so
that's
where
some
more
programmatic
way
of
driving
these
things
out
are
driving
discussion
out
past.
A
certain
point,
I
think,
is
necessary.
E
D
It
find
it
specifically
the
issue
where
people
are
like
well,
why
not?
This
simpler
API
and
it's
hard
to
tell
just
from
looking
at
the
code
snippet
of
that
API
interface,
whether
there's
something
that's
been
forgotten
and
that's
why
it's
simpler
because
it
doesn't
work
or
like?
Is
it
really
simpler?
E
So
I
would
suggest,
like
an
action
item,
is
to
maybe
a
focus
on
core
contributors
to
come
up
with
this
list
of
use
cases
and
some
maybe
code,
examples
that
demonstrate
those
use
cases.
And
then,
if
anyone
proposes
saying
like
a
different
API,
then
they
should
show
how
that
API
covers
all
those
examples.
A
I,
don't
think
it
can
be
automated
for
API
changes
because
still
try
to
harness
against
an
API,
that's
actually
literally
in
flux
as
part
of
the
TR,
but
I
do
think
it's
reasonable
to
say
here
are
some
situations
that
come
up
in
real
software
like
how
is
your
PR
going
to
address
this,
and
so
you
have
to
make
some
trivial
modifications
and,
if
they're
not
revealed
that
your
solution
doesn't
order,
access
like
these
are
the
things
that
need
to
be
easy
and
need
to
be
covered
and
I.
Think
that's
a
wonderful
idea.
So.
D
Yeah,
if
you're,
not
just
for
figuring
these
things
out,
but
also
like
like
what
you're
saying
like
compliance
being
able
to
identify
which
tracers
do
or
don't
work
with,
which
version
like
yeah
just
having
that
superstructure.
There
would
also
make
it
easier
for
people
than
to
add
new
test
cases
and
things
as
they
come
up.
Yeah.
A
And
this
has
come
up
on
the
interwebs,
as
well
as
a
as
a
gap
and
for
tracing
Java
pieces
of
seminar.
I
mean
I.
Think
everyone
agrees
that
it's
beneficial
I'm
like
personally,
like
two
swamped.
It's
like
not
even
the
right
word
at
then
I'm
like
time,
bankrupt
completely,
but
I
know
I
really
agree,
maybe
something
that
we
can
maybe
just
pay
someone
to
do
some
things.
It's
not
it's
not
like
it's
that
difficult.
It
just
needs
to
get
done,
and
it's
just
not
going
at
the
tops
of
cute.
E
G
You,
yes,
there
was
yeah
pretty
much
other
tests.
It
got
beat
a
hundred
percent
that
were
applicable
to
any
tracer
I
was
sure
to
put
in
the
harness,
and
then
there
was
a
couple
of
specific
things
that
were
related
to
configuration,
but
yeah
I'm,
going
to
finish
that
pyaari
I
promise
like
this
weekend,
there's
like
just
a
handful
of
little
things
that
I
think
are
blocking
it
from
being
rich,
but
it
she
was.
A
Alright,
so,
let's
see
I
think
looking
through
the
agenda,
there's
a
few
more
things
related
to
this
active
Stan
stuff
that
are
just
kind
of
they're,
almost
a
reference
on
the
nuts
I'm
not
going
to
walk
through
them.
I,
don't
think
that's
very
useful
way
to
spend
our
time,
I!
Think
I'm,
probably
not
the
right
person.
To
summarize
all
the
updates
and
contributor
is
actually
complete,
but
I,
don't
know,
I!
Think
Ted!
You
put
this
very
instrumentation.
Think
you
want
to
talk
about
that
for
a
minute
just
yeah.
D
Yeah
sound
of
a
paddles
not
on
the
call,
because
he's
he's
heading
that
up,
but
this
is
just
an
attempt
to
sort
of
coordinate
the
various
pieces
of
spring
instrumentation
that
are
coming
in.
There
are
certainly
parts
of
the
spring
ecosystem
that
directly
need
to
be
instrumented,
but
then
there
are
other
parts
where
it's
actually
much
more
like
configuration
and
auto
configuration
just
you
know,
configure
this
higher
order
component
to
use
the
instrumentation.
D
You
know
this
lower-order
component
that
it
uses
for
our
PC
or
something
so
and
then
also
this
issue
around
do
all
of
the
dependencies
for
all
these
instrumentations
work
together,
because
all
the
spring
stuff
has
to
work
together.
So
there's
a
a
project
now
in
open
tracing
contributor
in
cloud.
That's
sort
of
this,
this
meta
repo,
it's
kind
of
modeled
on
a
spring
sleuth,
which
is
an
existing
system
that
works
with
Zipkin.
D
Unfortunately,
it's
not
really
feasible
to
try
to
plug
open
tracing
tracers
into
that
system
and
for
you
know
very
reasonable
reasons.
They
don't
want
to
switch
to
open
tracing,
because
it
would
be
a
big
change
for
them.
So
this
is
just
sort
of
a
way
to
kind
of
corral
our
own
ecosystem
of
instrumentation
for
spring.
If
that
makes
sense-
and
we
would
love,
of
course,
a
help
if
you're
interested
in
spring,
please
try
these
things
out
and
help
us
get
the
out
of
configurations
in
there.
A
E
Could
open
sweep
here
this
developer
community
meetings
and
we've
had
to
talk
myself
and
Gary?
Essentially
it
was
a
Jaeger
presentation
and
then
Gary
had
a
example
of
metrics
from
Java
open
tracing
Java
I
was
at
the
scene.
Well,
I
think
it's
like.
The
actual
call
doesn't
have
that
many
listeners.
It
is
more
we're
kind
of
it,
goes
on
a
blog
post
afterwards,
as
a
video,
so
it's
available
yeah
I'm
going
to
share
it
here.
G
D
Cool
I
feel
like
this
is
often
a
core
thing
that
comes
up
with
these
API
changes
of.
On
the
one
hand,
we
want
there's
all
this
situation
where
yeah
you
have
to
do
things
after
you've
started
to
span
just
because
that's
the
way
it
works
out,
but
then
some
tracers
get
concerned
about
things
just.
However,
they
work
they,
they
want
to
front-load
some
work
or
send
this
spam
message
out
and
they
have
difficulty
with
with
things
being
dynamic
or
mutable
I.
D
G
A
Thinking
behind
it,
I
think
well,
part
of
it
was
easier
to
go
to
the
you
know,
it's
easier
to
add
capabilities
and
to
remove
them
or
whatever
that
kind
of
thing,
and
then
the
other
tip
it
was
that
for
some
implementations,
the
type
that,
like
the
way
the
fan
gets
constructed,
it
depends
on
the
context
initialization
so
like
the
most
obvious
example.
Is
that
if
you
were
in
the
tenets
of
Caen
yeah.
G
A
I
mean
it's
a
little
weak
I
actually
had
coffee
with
Rodrigo
Fonseca
few
days
ago
and
talked
about
our
they're
tracing
playing
stuff,
which
I
think
is
a
lot
more
like
elegant
than
any
of
the
funny,
and
he
has
anything
that
happening
to
industry
for
the
most
part,
and
we
were
talking
about
this
this
type
of
familes
over
to,
and
it
doesn't
feel
like,
like
the
desire
to
add
these
context.
References
during
the
span
lifetime
probably
reflects
like
a
weakness
most
and
model
more
than
anything
else.
A
Like
that's
my
that's
my
life,
you
sense
about
it,
yeah
like
it.
If
it's
a
timestamp
with
which
the
thing
is
added
like
matters
at
all
which
I
think
in
some
cases
it
does
like
it's,
it's
almost
like
a
replace.
It's
like
it
really
probably
done
it.
Office
like
and
open
traits
and
terminology
is
like
a
log
event
that
you're
logging
additional
reference
and
not
just
like
a
attribute
understand
like
it
probably
matters.
What
time
it
happened
if
it
wasn't
available
before
the
stand
started.
So
that's.
A
What's
always
hung
me
up
on
us
in
the
past.
Just
feels
like
I
am
like
open
tracing
in
shrines
and
portal,
7
event
where
a
Spanish
arsonist
and
stops,
but
you
get
like
really
sure
and
theoretical
that
I
think
it
should
just
be
one
big
event
stream
with
order
dependencies,
and
then
all
this
becomes
much
more
elegant,
although
first
it
becomes
like
a
lot
harder
to
use.
Well,
I
would
argue
it
becomes
a
bit
harder
to
to
like
reason
about
and
use
an
industrial
code.
A
G
F
The
other
up,
sorry,
one
other
point
of
information
that
would
be
useful,
that
we
don't
currently
capture
in
a
lot
of
cases
is
whether
a
span
is
blocking
versus
not
blocking
the
overall
context,
just
as
a
way
of
like
distinguishing
yeah.
This
piece
of
information
might
be
useful,
or
this
other
thing
might
not.
F
G
I
mean
if
you're
talking
about
that
back
to
the
tracing
plane
stuff,
the
baggage
propagation
up
is
like
the
the
the
main
thing
that
is
different.
You
know,
apart
from
reference
types
and
logging
events,
it's
like
how
to
handle
baggage
propagation
back
up
the
choice
whether
to
handle
it.
I
noticed
that
the
open
tracing
spec.
You
know
it
only
really
says
it
goes
down.
Each
child
inherits
a
copy
and
the
downwards
and
the
output
sounds
like
a
huge
can
of
worms
across
processes.
It's
easier
in
process.
G
Yeah
I
guess
that's
kind
of
outside
the
scope
of
open
tracing
I
assume
is
his
baggage
propagation
app
like
I
did
I
wasn't
trying
to
push
for
that
at
all
that
you
could
see
a
lot
of
use
cases
right.
It's
for
propagating
a
taint
flag
or
all
kinds
of
different
reasons
why
you
could
yeah
use
such
a
facility.
A
That's
a
personal
question:
it
means
that
something
very
posterior
from
like
you
know,
organizations
that
actually
influence
race
and
sort
of
looking
at
Bruce
and
Johnson.
I.
Guess
like
like
to
do
what
you're
suggesting
Chris
would
be
really
nice
like.
Theoretically,
we've
resolved
a
lot
of
uncertainty
that
clock
skew
and
so
on,
but
would
also
require
that,
like
that,
somehow
data
gets
sent
on
the
way
up
the
stack
and
not
to
some
the
way
down.
A
The
stack
through
request,
headers
I,
guess
that
normally
done
through
response
headers
or
something
like
what
do
people
usually
do
a
strategist
about
destroy
like
I?
Don't
have
a
great
sense
of
like
whether
anyone's
doing
things
like
that
for
other
purposes
or
not
it
Facebook
or
tertulia
or
whatever,
but
I'm
I'm
curious
to
hear
about
it.
It
makes
me
my
I:
don't
want
to
buy
a
bike
Jones,
but.
C
I
can
actually
see
that
potentially
people
with
you,
because
Julio
isn't
what
I've
had
to
have
to
get
wrap.
My
head
around
is
that
Twilio
has
our
API
edge,
but
we
also
have
a
carrier
edge
on
the
other
end
and
I
like
his
word
transit,
so
that
that
kind
of
makes
our
systems
very
interesting
to
trace,
because
it's
not
just
like
here's
the
front
door
and
everything
goes
down.
C
C
B
Yeah
on
on
our
side,
don't
I
can't
think
of
any
cases
where
we
propagate
stuck
back
up
for
purposes
of
tracing
there.
We
were
looking
at
cases
where
we
didn't
propagate
that
stuff
back
up
in
the
response
center
for
taking
a
great
with
tracing,
but
we've
generally
shied
away
from
a
lot
of
the
baggage
propagation
stuff,
although
were
we're
thinking
about
like.
B
Basically,
we
want
to
make
sure
that
it's
like,
as
stripped
down
as
limited
as
possible
because
of
just
people
abusing
and
generally
probably
people
abusing
tracing
anyways
and
giving
allowing
them
the
capability
to
sort
of
train
set
stuff
arbitrarily
around
for
the
stack
would
probably
cause
and
should
use
it
even
more
so
yeah.
It's
not
something.
We've
certainly
propagating
something
back
up.
We
haven't
considered
I.
H
A
A
H
I
Working
okay,
yeah
so
so
mostly
all
of
our
tracing
would
be
through
dr
pc,
the
GOP
c
package
right
now,
so
in
any
any
cons,
any
propagation
of
headers
would
be
through
either
contract
or
through
the
dear
pc
editors.
So
but
we're
not
that
tightly
integrated.
Yet
right
I
mean,
and
right
now
we're
looking
at
going
a
using
gear,
if
you
see
basically
across
the
entire
system,
so
all
of
our
tracing,
a
kind
of
injection
would
be
related
to
dr
pc
is
that
is
anything
the
question
that
your
so
you're
looking
for?
Do.
A
You
can
do
beyond
measuring
latency.
If
that's
that
that
require
that
and
will
be
fake
to
see
that
happen,
but
I
also
I
feel
like
it's
sort
of
like
have
to
crawl
before
you
walk
or
something
like
that,
and
a
lot
of
organizations
don't
even
have
basic
coverage
that
that's
always
been
like
my
concerns.
Getting
back
like
the
tear
another
thing,
I,
don't
know,
I
mean
I
like
I,
can't
remember.
If
the
issue
is
actually
includes
a
bunch
of
use
cases
or
not,
but
I
think
that
would
be
really
helpful.
C
B
Yeah
just
to
I
guess
as
I
was
thinking
more
so
I
I
do
think
we
actually
do
sudden.
So
we
do
use
the
response
headers
in
some
cases
and
we're
probably
using
it
more
the
motto
that
we
you
can
deploy:
Sayla,
multiple
flush
scenarios
and
okay,
it
can
support
multiple
flush
scenarios,
and
so,
in
that
case,
what
we
probably
do
is
communicate
an
idea
long
back
with
each
of
the
responses
that
we
may
send,
but
yeah
we've
I
guess
causes
mentioned
like
baggage
in
general.
B
B
D
Can
see
security
implications
that
are
worse
from
bubbling
things
up?
It
would
be
easier
to
to
leak
things
like
grids
or
whatever
going
that
direction,
given
that
it
just
sort
of
would
automatically
be
propagating
all
the
baggage
right,
like
you're,
less
likely
to
hit
the
public
computer
going
down
more
likely
you're
coming
back
up
to
the
client.
G
G
Indeed,
cool!
Well
thanks
for
this,
this
is
all
good
food
for
thought.
I
guess
I'll
update
the
issue,
but
yeah
now
I'm
kind
of
thinking
more.
This
should
just
use
the
log
in
our
face.
The
main
concern
for
trace
is
that
I
don't
want
to
convert
a
whole
lot
of
instrumentation
that
uses
and
propagates
response
headers
to
an
OT
model
without
losing
information
that
we've
been
starting
for
a
long
time
and
M.
A
Yeah
no
I,
don't
remember,
there's
an
issue
yeah,
but
there's
definitely
an
issue
in
my
head
around
the
blogging
of
references
specifically
I'm.
Absolutely
positive
it'll
make
some
people
angry
if
we
start
doing
that,
there's
already
a
bunch
of
angst
around
like
whether
open
tracing
is
like
blowing
up.
A
You
know
the
Western
world
by
having
like
or
the
world
in
general,
by
allowing
logging
into
the
API
and
and
I
think
there's
an
there's,
some
recognition
that
is
valuable
to
log
timestamp
strings
at
times,
but
even
like
the
presence
of
a
key
value
facility
is
like
kind
of
ruffles
and
feathers
and
and
that
maybe
even
some
people
feel
like
if
putting
too
much
of
a
burden
on
its
notations
or
something
like
that.
I
can
I
could
argue
with
that.
But
I
won't.
A
However,
like
you
know,
I,
like
the
idea
of
saying,
like
you
know,
there's
this
reference,
which
is
like
a
special
thing
up
in
figs
and
refers
to
another
trace
or
understand
and
same
crease
whatever
and
I
want
to
just
record
it
for
the
purposes
of
I,
don't
know
whatever,
either
something
having
to
assembly
work
with
you
I
and
like
enshrining
that
like.
If
you
say
you
want
to
log
the
reference
right
now,
it
would
just
be
like
string
to
find
some
implementation
specific
way
and
wouldn't
do
that.
C
Like
being
LC
logs
in
context
and
traces
like
that,
super
powerful
from
my
perspective
and
I,
really
like
it,
the
other
third
function
that
might
I
think
answering
this
might
help
direct
where
the
implementation
goes
is
I,
remember,
I,
think
it
was
Ted
who
did
the
talk
on
measured?
It
was
like
measure
one
or
instrument
once
measure
three
times
or
something
like
that
and
I'm
clocking
a
few
folks,
both
around
folio
and
some
other
companies
and
I.
D
So
you
know
it's
not
like
zipped
in
and
everything
else
suddenly
now
needs
to
deal
with
metrics,
it's
just
like
okay.
Well,
if
you
have
something
plugged
in
for
the
metrics
and
it
handles
the
metrics
and
you
have
something
plugged
in
for
the
logs,
it
does
and
you
don't
well,
we
can
forget
that
so
that's
sort
of
where
I
see
it
going,
please
innate
so
you're
trying
to
talk
so
oh
yeah.
I
I
was
gonna,
wait
till
you're
finished,
so
so,
if
I'm,
if
I
said
anything
wrong,
I
haven't
been
keyed
into
the
open
physical
injury.
So
if
I'm
a
little
off
here,
you
know,
let
me
know,
but
one
if
you
I
can
see
with
with
finding
the
logs
tracing
is
that
it's
going
to
massively
increase
the
data
volume.
It's
going
to
come
out
of
a
trace
application.
I
This
is
something
we
had
we
had
in
the
registry,
where
we
would
kind
of
do
our
tracing
with
logs
and
stuff
like
that,
and
it
ended
up
being
kind
of
operationally
problematic
just
because
of
the
volume
of
logs
that
when
you,
when
you
buy
in
the
logging
to
the
great
thing
you
end
up
with
this,
like
massive
amount
of
data
and
and
having
to
handle
that
inside
of
whatever
target
trading
platform,
will
make
those
platforms
more
complicated.
Such
as
points
on
it
bring
it
up.
D
Yeah
I
agree:
there's
always
this
aspect
of
like
what
is
the
correct
level
of
granularity,
especially
if
you're
going
to
say
these
are
I,
think
it
matters
less
for
your
application
code,
because,
like
kind
of
programming
against
specific
implementations
there
and
it's
okay
to
take
their
send
into
account,
but
certainly
when
making
like
library,
instrumentation
yeah,
there
has
to
be
some
kind
of
rule
about
what
what's
an
acceptable
amount
of
logging.
That
system
can
do.
D
It
can
now
actually
log
which
you
couldn't
do
in
a
library
before
and
so
I
think
that's
actually
huge
and
important
and
I.
Think
again,
if
you
look
at
what
everyone's
doing
with
their
logs
right
now,
they're
trying
to
correlate
them
and
get
correlation
IDs
into
their
logs,
so
it's
sort
of
like
I,
don't
know
that
it's
necessarily
a
different
problem
from
the
original.
How
much
logging
should
I
be
doing
problems
that
everyone
deals
with
I?
Guess
it's
what
I'm
trying
to
say
just
because
you're
now
correlating
it
or
a
tracer
got
involved.
A
Logging
stuff
is
that
I
mean
tracing
I
would
argue
like
there's
only
like
event,
data
and
time
series
data
and
monitoring
and
tracing
is
clearly
event.
Data
at
the
core
and
logging
is
to
so
logging
and
tracing.
Are
you
know,
kind
of
sort
of
the
same
thing
and
the
main
difference
in
practice
seems
to
be
that
trading
systems
understand
a
distributed
transaction,
and
so
they
can
sample
based
on
transactions,
and
so
you
have
like
a
lever
with
which
to
control
log
volume
other
than
verbosity.
A
But
of
course,
if
you
actually
want
to
look
in
an
individual
transaction
unless
it
was
a
warning
or
an
error
or
something
you
probably
want
to
see
like
all
the
details
in
the
distributed
system,
so
I
think
that,
like
tracing
this
gives
you
this
like
transactional
sampling
or
something
like
that
as
a
lever
to
control
the
impact
of
instrumentation
on,
or
you
know
operationally
like
this,
do
saying
no
order.
He
meant,
of
course,
that,
like
it,
you
you
don't
even
need
logs,
you
can
just
you
could
create
a
stand
for
every
functional
call.
A
You
like
overload
your
tracing
system
in
just
the
same
way,
just
if
you
put
more
data
and
you
need
to
be
able
to
handle
more
data.
The
thing
that
I
think
is
nice
about
using
trading
to
some
sport
logging
is
that
you
can
still
get
like
really
fine
grained
detail
about
some
transactions,
and
then
you
can
use
sampling
a
transaction
level
to
control
the
impact
on
the
trading
system
rather
than
verbosity
levels,
which
kind
of
like
limit
the
use
of
your
logging
instrumentation.
A
So
I,
definitely
like
some
speciated
counter
argument
and
then,
of
course,
you
can
have
implementations
that
sort
of
dropping
data
when
it
just
gets
to
be
too
much
or
whatever
we
did.
That
and
apperance
I'm
sure
a
lot
of
commercials,
a
lot
of
industrial
systems.
We've
done
similar
things
to
protect
himself.
It.
C
Was
actually
going
to
say
that,
based
on
my
experience
at
Twilio,
the
lack
of
having
tracing
actually
created
the
problem
with
having
too
much
logging,
but
once
we
added
tracing
in
then
we
could
actually
reduce
our
logging
because
it
was
more
meaningful
or
surgical.
I
guess
you
could
say
because
of
the
context
that
you
could
see
it
in,
but
without
tracing
it
was.
C
It
was
like
log
of
the
world
all
over
the
you
know
and
I
think
our
complexity
is
about
900
micro
services
at
this
point,
so
that
is
a
lot
of
tendons,
a
lot
of
log
data
that
it's
very
hard
to
sample
or
you
can't
really
sample
if
it's
a
log
system,
so
that's
kind
of
how
I
think
about
it.
What
my
experience
been
so
far.
D
And
I
do
have
this.
This
does
bring
up
a
concern
for
me
with
the
current
open
tracing
API
around
observers
and
correlation
IDs,
where
it
does
seem
like
there's
this
great
thing
of
being
able
to
okay
I
got
you
know
my
trace
are
doing
some
things,
but
I
would
also
like
to
shunt
the
logs
off
into
this
log
storage
system.
I
would
like
to
pump
out
some
metrics
to
go
into
my
metric
system,
all
the
usual
kind
of
stuff,
and
to
do
that
through
some
sort
of
observer
API.
You
know
so
someone
can
just
write.
D
You
know
a
Prometheus
open
tracing
bridge
and
you
can
just
plug
that
in
and
configure
it
and
make
it
work.
There's
a
lot
of
interest
in
doing
that.
I
know
URI
and
some
people
have
been
working
on
observers,
but
it
seems
like
the
thing
that
you
really
want,
like
the
whole
point
of
that,
has
to
do
with
basically
span
context,
information
around
the
span
ID
and
the
trace
ID,
and
we
don't
want
to
give
those
out
right
now.
I
can't
really
decide
what
they
would
look
like
and
that's
sort
of
bringing
up.
Well.
D
Should
we
add
this
correlation
ID
to
make
these
things
work,
but
then
what
is
this
correlation
ID
and
how
is
it
feasible
that
we
could
add
that,
but
not
give
you
the
trace
ID?
So
there's
something
that
to
me.
It
feels
confused
in
that
area
where
we're
trying
to
like
hide
information
that
we
can't
really
hide
on
some
levels.
We
want
to
be
successful.
I
don't
have
a
great
solution.
There.
I
won't
try
to
solution
eyes
in
the
middle.
This
call,
but
that
is,
it
I
think,
just
an
area
where
we
need
to.
C
That's
kind
of
the
point
that
I'm
kind
of
make
is:
is
that
then
I
think
the
clarity
is
the
thing
that
will
anchor
the
API
and
Ankur
open
tracing
is
clarity
on
both
of
what
it
becomes
and
and
I
would
encourage.
I
would
encourage
the
board
to
think
about
I'm
an
opinionated
stance
on
the
future
of
what
it
should
be
and
don't
necessarily
let
it
get
there
organically,
because
that
that's
going
to
be
much
harder
to
rein
in
and
so
I
think
it
is
that
at
least
right
now.
The
way
I
think
about
it
is.
C
Base
it
in
on
that
I
think
will
help
clarify
kind
of
the
direction
and
what
should
be
in
so
Kinetoscope-
and
you
know
from
my
two
cents
on
this
I
think
that
was
like
one
of
the
very
first
things
that
people
started
noticing
it
was
like.
Oh,
we
have
a
whole
lot
of
lines
of
code
are
lines
of
instrumentation
in
our
code.
Now,
because
we
have,
we
have
tracing,
we
have
logging,
we
have
all
of
our.
You
know
counter
counters
and
gauges
and
innocent
in
small
micro
services.
C
It
was
like
there's
more
instrumentation
code
than
there
is
business
logic
at
this
point,
and
so
I
think
there
is
a
bit
like
for
developers
going
through
that
instrumentation.
There
is
a
very
obvious,
like
gravitation,
towards
towards
doing
that
now.
I
think
that's
something
to
either
get
ahead
of
to
either
stop
or
you
know
proliferate,
but
it
should
I
think
it
should
come
out
of
this
project
and
that
those
opinions.
D
D
If
we
can
just
solve
that
tiny
piece
of
surface
area,
then
that
opens
up.
You
know
a
sort
of
green
field
for
people
to
go,
try
to
come
up
with
various
solutions
for
this
that
don't
necessarily
involve
changing
the
open
tracing
API.
So
I
see
that
as
like
the
most
fruitful
potential
way
to
move
forward,
I
think
we,
the
idea
that
we're
going
to
add
a
fully
baked
like
stats,
T
metrics.
You
know
interface,
token,
tracing
or
something
I
feel
like
there'd
be
like
a
riot.
We
tried
to
do
that
so.
F
I
have
a
question
going
back
to
your
statement
about,
like
the
ratio
of
business
logic
to
instrumentation.
It
I've
noticed
that
there's
not
a
whole
lot
of
stuff
in
those
open
tracing
communities
that
focuses
on
auto
instrumentation.
Most
of
the
instrumentation
is
very
manually
applied
and
requires
a
lot
of
manual
work
on
the
the
application
developer.
It
is
there.
Is
there
a
reason
for
that,
or
is
there
an
appetite
for
allowing
some
increased
level
of
auto
instrumentation.
A
F
A
Like
as
a
matter
of
personal
opinion
like
I'm,
a
little
bit
skeptical
about
the
automatic
instrumentation
of
distributed
systems
working
at
least
for
now,
if
some
of
them
figure
out
a
way
to
do
it,
like
that's
great,
like
truly
and
more
of
a
technical
concern,
I
have
just
almost
like
an
existence.
Proof
people
get
vendors
and
like
ATM
and
stuff,
like
that,
you've
attempted
to
do
this
in
distributed
systems
in
practice.
They've
usually
ran
into
like
some
pretty
profound
problems,
especially
the
process
boundary,
and
it's
not
like
it
can't
be
done.
A
I
don't
want
to
like
I,
don't
want
to
discourage
anyone
from
trying,
but
I.
Think
part
of
the
concern
has
been
that
it's
so
difficult
to
actually
do
it
without
having
so
many
false
positives.
That,
like
your
system,
is
trustworthy
and
that's
that's
been
the
concern
that
I
pad
up
the
other
thing
I
would
say
is
like.
A
If
we
wanted
to
take
that
approach,
it
might
make
more
sense
to
focus
on
standards
around
wire
formats
and
inter-process
communication
standards
and
stuff
like
that,
and
not
so
much
on
instrumentation,
which
then
becomes
much
less
important
since
it's
not
like
a
like
all.
The
instrumentation
is
like
this
kind
of
black
box
thing
and
I
think
there
is
some
interest
in
moving
in
that
direction
for
other
reasons,
and
it
might
have
to
add
a
benefit
of
allowing
for
the
agent-based
or
to
automate
an
instrumentation
to
fit
in
with
open,
trace
and
compatible
with
tracers.
A
D
C
Having
used
systems
that
automatically
do
this
and
seeing
what
kind
of
method
becomes
at
scale
and
then
having
to
make
developers
do
this
like
I,
could
see
it.
I
could
see
both
ends
because
on
one
end
of
the
spectrum
we
had
so
much
instrumentation
that
it
was
useless
on
the
other
end
of
the
spectrum.
Now
that
we
have
a
cat
hurry
from
trying
to
get
everything
to
adopt,
and
that
is
that
sense,
so
I
certainly
can
see.
We
see
the
two
sides
of
that,
but
I
would
say,
given
the
choice
between
those
two
worlds.
C
G
G
If
one
occurred,
you
know
close
the
span
from
some
other
count
with
the
timing
in
it,
and
then
you
know
log
look
the
result
of
the
query
and-
and
it
was
like
eight
lines
and
then
he
was
like
by
like
adjusted
in
one
line
because
there's
some
fancy
rails
monkey-patching,
but
probably
two
lines
would
be
the
minimum
you
would
imagine
for
the
medical
benchmark
and
that
was
kind
of
cool
and
he's
like
now.
My
app
seein
can
pick
which
of
the
three
different
reporting
channels
to
use
for
that
specific.
G
G
I
think
it
was
more
like
a
pseudo
code
like
there
is
in
this
vinner
project
that
you're
open
tracing
bridge
of
some
kind,
but
if
it's
very
alpha
stage
right
now,
but
it
was
just
a
kind
of
glimpse
into
the
future,
potentially
a
way
to
do
two
or
three
of
those
concerns
with
them.
With
a
with
the
consistent
single
API
that
I
thought
was
pretty
cool.
A
So
I'm
taking
a
look
at
the
closet.
If
you
have
two
minutes,
I
do
want
to
end
on
time.
People
probably
other
things
to
do.
I
wanted
to
mention
one
other
idea
that
I
think
will
do
I.
Think
it's
really
helpful
to
have
is,
and
I
especially
want
to.
Thank
you
know:
Jonathan
Stephen
Bruce
joining
it's
great
and
have
perspective
from
people
who
are
consuming
this
type
of
technology.
I
I
thought
it
might
be
fun
for
the
best,
the
world's
not
for
us
to
schedule.
A
Like
I,
don't
know
it
would
be
quarterly
by
month
year
system
like
that,
but
schedule
like
some
sort
of
talk,
maybe
like
a
30
minute
presentation
or
maybe
even
an
hour-long
presentation
where
people
would
sort
of
prevent
like
tracing
in
practice.
A
company
X
type
of
thing,
I
know
when
Adrian
is
or
demonstrating
workshops.
A
Those
have
been
some
my
favorite
just
like,
with
hopefully
with
the
veneer
taken
off
and
just
explaining
what
did
work
and
what
was
really
difficult
as
how
these
felt
or
address
and
that's
rip
and
I
found
that
it
varies
a
bit
from
company
to
company
I
thought
it
would
be
like
a
nice
thing
to
host
that,
and
you
know
record
the
video
and
make
it
available
for
future
reference.
Some
other
folks,
because
other
people
think
that
sounds
valuable
or
anyone
who
wants
to
say
like.
Oh
absolutely
not
like
shouldn't,
do
that.
A
I
think
that's
interesting
yeah,
it
is
it's
like
tracing
the
kind
of
effective
technology
and
I
think
it's
getting
a
lot
of
buzz
right
now,
but
I
don't
think.
There's
a
lot
of
documentation
out
there
of
the
pain
points
around
it
and
I
think
it
would
be
helpful
like
just
for
the
industry,
it's
like
record
where
it
goes
awry
and
like
where
the
problems
tend
to
be
I.
A
Think
most
I
honestly
has
to
do
with
instrumentation,
but
but
I
would
love
to
get
that
the
stories
have
been
recorded,
and
maybe
we
can
even
revisit
you
know
the
same
company
once
every
18
months
and
see
how
things
would
change
or
not
all
right.
Well,
we'll,
try
and
set
something
like
that
up.
I,
don't
know
who
the
initial
victim
will
be,
but
it
should
be
informal
and
everything,
but
I
think
it
would
be
interesting
cool.