►
From YouTube: 2022-02-09 meeting
Description
No description was provided for this meeting.
If this is YOUR meeting, an easy way to fix this is to add a description to your video, wherever mtngs.io found it (probably YouTube).
B
I'm
about
to
go
on
vacation,
so
I'm
very
excited.
Oh,
where
are
you
going?
I'm
going
up
to
canada
to
do
some
backcountry
skiing?
So
oh
that
sounds
fun.
Yeah!
It's
gonna
be
great,
really
excited,
but
yeah
I'll
be
I'll,
be
out
next
week.
So
I'll
miss
this
meeting.
A
Wondering
the
I
wanted
to
watch
one
of
the
spec
meetings,
but
it
seems
like
they
don't
get
uploaded
to
youtube
until,
like
a
month
later,.
B
But
they
they
haven't
been
getting
uploaded.
We
we
switched
how
we
were
managing
our
zoom
meetings
because.
B
We
just
like
ran
out
of
zoom
rooms,
basically
and
something
about
how
we're
doing
it
now
broke
our
auto
uploading
thing.
So
I've
been
poking
the
the
people
who
set
it
up,
unfortunately,
aren't
like
around
very
much
they've,
been
like
poking
them
again
to
get.
C
B
Get
it
set
back
up,
but
in
theory,
if
we
can
get
it
working
again,
it
should
work
better
than
it
did
before,
because
we
should
now
be
able
to
name
the
individual
zoom
meetings.
B
Exactly
yeah
yeah,
that's
just
just
one
of
those
things,
that's
just
like
a
tedious
task
and
we
don't
have
an
intern
to
do
it
for
us
basically
but
yeah.
Hopefully,
hopefully,
that'll
get
sorted
out
soon
and
like
long
term,
it
would
be
nice
if
we
could
like
get
them
organized
playlists
or
something
but
yeah.
Sorry,
no.
B
B
B
B
A
Man,
my
office,
my
office,
isn't
opening
until
mid-march
it's
gonna
be
a
while.
I'm
gonna
be
stuck
in
my
in
my
bedroom
for
still
some
time.
B
Sorry,
I'm
just
trying
to
get
caught
up
on
these.
These
two
issues,
so
I
haven't
been
looking
at
them,
but
yeah
I
haven't.
I
haven't
been
to
the
light
step
office
in
forever,
like
lightstep
literally
like
leased
a
whole
new
building
and
like
moved
it
to
like
get
like
a
big,
proper
office
like
right
as
the
pandemic
hit,
and
then
it
just
sat
empty
for
like
a
year
and
now
people
are
finally
in
it,
but
I
have
never
once
set
foot
in
it
because
I
don't
go
to
san
francisco
anymore.
C
B
You
know
new
relic
has
a
big
office
up
here.
I
think
that's
that's
what
holds
it
down.
I
don't
know
if
anyone
else
does,
but
just
because
new
relic
is
up
here.
I
think
like
like
just
like
a
bunch
of
observability
people
live
in
portland,
even
though
they
may
have
like
moved
on
to
to
like
other
other
jobs,
but
yeah.
It's
like
a
big
remote
work
community
up
here.
A
Yeah
I
see
yeah
well,
atlassian
has
just
gotten
governor
government
approval
to
build
a
huge
building
in
the
smack
bang
in
the
middle
of
sydney,
but
like
they're,
also
saying
they're
also
really
geared
towards
working
from
home.
So
it's
like.
B
C
B
B
The
problem
is
like
marijuana,
is
legal
up
here
and
basically,
like
all
the
warehouse
space
got
like
leased
out
for,
like
marijuana
grow
operations
when
it
got
legal
and
it's
like
totally
pushed
out
like
anything
else.
Anyone
wanted
to
do,
but
I
that
like
crashed,
I
heard
that
crashed
like
last
year
or
something
so
so.
I'm
hoping
there's
no
warehouse
space
available
in
portland
for
like
cheap
but
we'll
see.
B
That's
fine,
anyways
dennis
I'd
love
to
get
caught
up
on
your
work.
Sorry
I
was,
I
was
out
last
week
and
just
fyi,
I'm
gonna
be
out
next
week
as
well,
but
I
did
see
you
were
getting
some
feedback
from
bogdan
about
links.
I
don't
know
if
that
got
resolved.
C
Correct
yeah,
so
I'm
actually
looking
for
waiting
for
some
response
from
some
of
you
guys
yeah.
So,
basically
the
discussion
there
was
you
know:
bogdan
was
a
bit
hesitant
to
adding
or
actually
using
links
right
now
to
you
know,
to
model
this
this.
These
scenarios
he
thinks
it
might
be
too
early
to
you
know,
include
this
usage
of
this
tooling,
but
looks
like
like
from
my
perspective.
I
don't
see
any
other.
C
You
know
convenient
way
for
us
to
model
this
these
scenarios,
and
actually,
I
remember
back
in
like
a
october
or
something
we
had
this
discussion
in
this
group
just
to
you
know,
figure
out
how
which
which
options
we
have.
Should
it
be
like
a
logical,
physical,
logical
and
physical
and
different
types
of
you
know
spans
that
we
can
create.
So
looks
like
that.
That
time
we
decided
that
links
are
really
useful
and
we
can
leverage
them
right
now,
but
yeah
bogdan
just
joined
last
time
and
provided
this
feedback.
C
So
currently
I'm
trying
to
figure
out
like
what
is
what
will
be
the
next
step.
So
I
do
have
like
a
four
or
even
five
approvals
for
the
for
this
pr.
Four
of
them
are
green,
but
we
do
have
this
feedback.
I
try
to
figure
out
this
with
with
stigrun
who
is
assigning
there
and
basically
he
sit
back
and
just
just
resolve
this.
This
issue
there.
So
I'm
trying
to
figure
out.
I
tried
to
think
bogdan
yesterday,
but
there
is
no
feedback
yet.
C
So
what
do
you
think
like
what
will
be
the
best
way
for
us
to
proceed
and
make
it
make
it
happen
so
like?
I
should
be
well
actually
yeah
so
background.
Didn't
reply
so
yeah?
That's!
That's!
That's
the
current
state.
B
Yeah
I
mean
I,
I
guess
I'm
still
for
using
links
for
this,
like
as
you
as
you
said,
like
they're,
not
they're,
not
parent-child,
they're,
just
they're
they're
siblings.
But
the
thing
we're
trying
to
indicate
is
that
they're
they're
more
than
just
siblings
right,
like
there's,
there's
actually
abounding
context
here
that
isn't
modeled
with
a
parent,
because
that's
obnoxious
and
overkill
the
you
mentioned
the
concept
of
a
session.
So
I've
been
working
with
the
the
client,
instrumentation
rum
group
and
they're
interested
in
adding
a
concept
of
a
session.
B
But
I
think
it's
a
different
scope,
different
scale,
they're,
more
thinking
like
user
session
right
but
yeah,
but
the.
B
B
At
the
end
of
the
day,
a
session
is
like
an
id
number
right
that
you're
going
to
staple
on
to
all
of
these
things,
and
so
it's
like
they
either
have
like
you,
generate
another
id
and
like
attach
it
as
an
attribute
to
these
things.
So
you
can
group
them
or
you
directly
link
them,
and
I
don't
like
I
don't
know
it
just
it
doesn't
like
they
they.
I
guess
they
both
seem
a
little
overkill.
B
B
Under
I,
I
just
have
this
worry
that,
like
links,
are
just
this
vague
concept
and
so
everywhere
we're
using
trying
to
use
links
we're
getting
some
like
pushback
and
to
to
some
degree
like
like.
We
need
to
figure
it
out,
but
but
I
don't
see
why
why
links
would
be
a
big
deal
here.
C
Yeah
so,
basically
like
from
my
perspective,
I
see
that
actually,
the
this
link
concept
or
the
idea
of
using
links
here
came
from
the
you
know,
I
think,
or
like
a
messaging
scenarios
and
looks
like
there
is
no
other
way
to
model
these
kind
of
things
so
even
session
id
is
something
that
probably
would
not
be
really
useful
there
yeah
so
and
like
from
this
perspective,
it
might
be
a
good
idea.
You
know
to
have
this
to
have
this
consistent,
and
you
know
I
just
added
this.
C
This
example
here
trying
to
you
know
what
I
was
thinking
about,
like
a
will
be
just
enough
to
have
this.
You
know
parents
span
or
something
like
that
and
even
like,
if
who,
if
they
have
two
parallel
threats,
it's
not
really
feasible
just
because
they
can
communicate
with
with
the
same
host
and
just
looking
at
these
spans
and
the
siblings.
We
cannot
say
for
sure
which
are
which
each
of
them
kind
of
correspond
to
that
session.
C
So,
like
I
put
this
session
kind
of
like
a
term
from
from
this
perspective,
that
we
can
have
some
some
scenario,
like
some
ongoing
scenario
and
some
spans
related
to
that
scenario,
so
yeah,
basically
that
that's
that's
my
thought,
I
I
think
that's
since
we
have.
We
have
this
like
a
shoot
everywhere.
C
We
are
fine
and
kind
of
going
with
this
with
links
and
going
forward.
It
was
also
a
good
question
from
anurag.
That's
we
can
also,
you
know,
you
know
explicitly
kind
of
add
information
about
this
relationship
like
what
it
is
like.
We
use
these
links
to
actually
build
this
chain
of
spans,
but
we
can
explicitly
indicate
what
what
what
is
what?
What
is
that
like?
C
Is
it
some
kind
of
retry
or
some
async
scenario,
messaging
scenario
or
something
so
going
forward
again,
we
would
want
to
think,
like
you
know,
starting
this
discussion
as
well.
For
now
it
might
be.
You
know
early
just
like
to
start
discussion
within
this
pull
request,
but
definitely
something
that
we
can
move
forward
with
yeah.
B
You
know,
storage
and
database
systems
to
do
with
this
stuff
and,
for
example,
like
the
way
we're
using
links
to
model
messaging
systems
for
like
async
workloads
where
you
it's
like
a
trace
of
traces.
B
There's,
it's
really
easy
to
index
all
the
spans
in
a
trace,
but
links
on
their
own
are
not
like
inherently
easy
to
index
right,
like
at
any
rate
like
we
need
to
something
we
haven't
proposed,
as
part
of
that
group
is
like,
like
a
reference,
implement,
not
even
an
implementation,
but
like
a
reference
data
storage
model
for
this
stuff.
B
And
then
you
go
grab
those
traces
and
find
all
the
spent
like,
like
you
end
up
with
this,
like
iterative
fetch
situation,
and
so
it's
like
clear,
you
need
to
do
a
little
more
data,
mullet
modeling
to
to
make
that
more
efficient
and-
and
we
don't
have-
we
don't
have
something
like
that.
B
Right
now-
and
I
see
like
these,
like
link
attributes
to
some
degree,
is
like
a
place
where
we
would
put
that
info,
but
at
any
rate
like
whatever
we
did,
there
is
probably
not
what
you'd
want
to
be
doing
with
these
links.
Right,
like
these
links,
are
all
going
to
be
stuff
in
the
same
trace
and
you're
almost
you're,
almost
just
using
this
for
like
a
ui,
affordance
or
something,
but
how,
however,
you
whatever
you're
doing
with
them
is
pretty
different
from
what
you're
doing
with
those
those
messaging
links.
You
know
what.
D
D
D
So
you
have
some
kind
of
semantic
bit
of
some
semantic
help
there,
and
I
think
it
seems
to
me
that's
also
what
bogdan's
always
a
problem
here,
that
it's
harder
here
to
kind
of
figure
out.
What
does
this
link
mean?
Because
when
you
have
a
link
between
a
producer
and
a
consumer
span,
you
have
some
kind
of
meaning
there
of
the
link.
Basically,
you
you,
I
think
the
mean
of
the
link
here,
that
there
is
the
same
thing
that
is
produced
here
and
consumed
here.
D
B
B
This
represents
an
attempt
to
do
the
last
thing
again
and
so
you're,
on
the
one
hand,
wanting
to
look
at
like
how
long
these
individual
spans
take,
but
on
the
other
hand,
you
might
be
trying
to
do
some
analysis
around
like
these
fans
and
their
totality,
or
something
like
that.
I
don't
know
dennis
if
you
have
like
an
example
of
like
concrete
use
cases
here
well,.
C
Yeah
exactly
so,
I
do
have
this
this
different
examples,
and
basically
they
just
show
these
two
different
types.
Let's
say
why
it
can
be
used
or
like
when
it
can
be
useful,
so
bogdan
was
mostly
like
a
you
know.
Referring
to
this
scenario,
which
is
the
you
know,
major
one,
I
would
say
when
you
do
have
some
top
level
spam
or
like
a
trace,
already
started.
So
in
this
case
you
you
can
have
them.
You
can
kind
of
correlate
them
under
under
this
top
level
thing,
but
it
should
be.
C
Basically,
there
are
some
some
in
other
cases
when
you
don't
have
any
started
span
right.
So,
for
example,
you
are
building
some
mobile
application
and
you
just
you
know,
starting
this,
the
the
overall
trace
from
from
that
device.
So
in
this
case,
if
you
have
some
mobile
application
and
you
do
have
retries
you-
you
have
totally
separated
traces
right
because
you
just
start
in
a
new
trace.
Every
time
you
send
a
request,
it's
an
outbound
request
and
in
this
case
links
are
only
kind
of
you
know
thing.
C
That's
allow
you
to
to
correlate
those
together.
The
main
the
main
kind
of
the
main
concern
from
bogdan
was
that,
like
even
in
this
case,
we
can,
you
know,
ask
people
to
start
a
new
new
trace,
kind
of
or
starting
to
start
trades
manually
or
you
know,
do
something
about
this.
But
it
looks
like
it's
not
really.
You
know
it
can.
It
can
be
done
automatically
and
that's
the
the
main
main
problem
for
mine
from
my
point
of
view,
just
because
you
know
we
cannot
provide
them
default
behavior
and
they
can
do.
C
They
can
have
all
this.
You
know
goodness
automatically,
but
with
links,
it's
possible
still
right,
so
yeah
that
that's
the
that's
the
kind
of
like
another
example
that
where
links
can
be
useful,
even
if
they
you
know,
try
like
a
all
of
us
to
to
to
stitch
all
these
spans
together
within
different
spans.
C
So,
like
sorry,
within
different
races,
as
we
do
have
in
in
this
example
right
so
traces,
that's
still
like
they
rely
or
they
correspond
to
the
same
to
the
same
scenario,
so
yeah,
basically
that
that's
the
that's
the
overall
thing
and
I
like
from
my
perspective.
We
just
need
to
have
some
way
of
you
know,
building
this
consensus,
so
we
can.
We
can
just
pretty.
B
Yeah
yeah
I
mean
it,
it
seems
I
I
think,
you've
laid
it
out.
Well,
you
know
there
is
a
problem
of,
as
we've
seen
like
you
don't
always
have
the
like.
Holding
on
to
that
link
is,
is
tricky
potentially
like
not
every
so
some
of
these
http
clients
are
too
low
level
right
and
they
are
not.
They
do
not
have
an
internal
concept
of
a
retry,
and
so
we
have
to
keep
that
in
mind
when
we're
modeling
this.
B
B
Like
I
don't
know
how
I
don't
know
what
else
like
if
we're
not
gonna
model
it
this
way
and
we're
not
gonna
model
it
as
and
we're
not
explicitly
not
going
to
include
like
a
bounding
parent
span,
because
we
decided
that's,
that's
a
bunch
of
noise
that
that
we
don't
want
at
this
level.
C
B
C
Exactly
exactly
that's
my
feeling
as
well
like
by
introducing
some,
you
know
additional
attribute
kind
of
session
id,
or
you
know
some
use
case
id.
What
not.
We
should
also
define
this
type
of
you
know
of
id
what
it
should
be.
Should
it
be
good,
should
it
be
just
string
like
a
what
what
what
people
should
put
there
and
it
should
be
really
clear-
and
you
know
feasible,
for
every
platform
to
to
generate.
D
C
And
it
should
be
standard
in
some
some
way
so
yeah.
This
that's
bring
like
a
brings
another
set
of
questions
that
might
not
be.
You
know,
really
easy
to
answer
so.
B
B
You
know
session
grid,
or
is
it
better
to
have
them
as
a
linked
list
like
which
which
data
structure
is
actually
for
the
use
cases?
We
think
people
are
going
to
the
common
use
cases
for
for
dealing
with
retries
in
some
analysis
tool
and
like
which
which
data
structure
is
actually
better.
I
guess
that
would
be
the
way
you'd
answer
it.
C
Right-
it's
not
really
simple
to
to
you
know
to
foresee
it
as
of
now
but
looks
like
since,
since
we
do
have
this,
this
concept
of
links
and
basically
every
platform,
every
every
backend
should
somehow
deal
with
this
going
forward
right
because
it's
already
introduced,
we
started.
We
are
starting
to
use
it
for
different
cases,
including
messenger,
including
this
one.
It
should
be.
You
know
some
discussion
on
which
the
data
structures
are
useful.
C
Here,
like
I
should
you
know,
will
it
be
possible
to
have
some
you
know,
representations
of
this
list,
kind
of
or
trace
of
traces
going
forward
right,
yeah
it
should
it
should
be
done
anyway,
but
that
that's
my
feeling.
B
Yeah-
and
I
I
do
think
actually
and
again,
this
stuff
might
all
be
like
overkill
for
retries,
but
certainly
in
the
case
of
messaging,
like
these
big
asynchronous
workloads,
I
do
kind
of
feel
like
we
have
all
these
links,
but
there
does
need
to
be
again
something
like
a
session
id
like
like
an
attribute
on
the
links
themselves.
B
That's
like,
like
a
link
group
id
again
just
in
terms
of
indexing
right,
like
you,
have
all
of
these
links.
All
these
spans
are
coming
in
and
they're
all
coming
in
with
links,
but
the
links
are
just
pointers
to
other
spans
that
you
may
or
may
not
have
access
to
right
there,
and
so,
as
far
as
like,
how
do
you
build
a
join
table
so
that
when
you
pull
one
thing
like
you
get,
you
have
one
span
id
or
one
trace
id
and
you're
able
to
just
very
quickly
pull
everything
out
of
the
database.
B
You
know
you,
you
need
to
create
some
kind
of
join
table,
and
this
kind
of,
like
linked
list
structure,
doesn't
seem
like
it's
it's
great.
For
that
it
seems
like
you
would
want.
You
would
want
a
group
id
that
was
attached
to
every
link
and
was
the
same
so
that
you'd
be
able
to
pull
all
of
the
stuff
or
something
like
that
I
don't
know,
does
that?
Does
that
make
sense.
C
Yeah
for
a
relationship,
the
kind
of
representation
of
this
stuff
is
definitely
something
we
should.
We
should
have
some.
You
know
some
grouping
identifier.
So
definitely
we
can
just
query
based
on
this,
but
yeah
other
than
the
structure
structures.
Like
you
know,
graph
databases,
it
might
be
much
easier
to
you
know
to
to
model
and
represent
this
structure
from
the
other
other
side.
C
I
also
see
another
kind
of
you
know
issues
when
it
comes
to
links
or
actually
that's
not
only
about
the
links
but
any
type
of
modeling
modeling
approach
we
can
take
here.
For
example,
we
do
have
sampling
right
and
some
expenses
can
some
spends
from
from
traces
or
from
different
traces
can
be
just
simple
out,
and
you
know
those
can
be
the
exact
trace,
the
exact
spence
that
contains
this
links
or
contains
any
kind
of
you
know
additional
attributes
or
whatnot.
C
So
from
this
perspective,
it's
you
know
it's
the
overall
structure
like
the
whole,
the
whole
sampling
concept
complicates,
you
know,
tracing
a
lot
so
and
another
thing
here
like
when
it
comes
to
when
it
comes
to
links.
Another
thing
here
or
another
complication
that
we
can
have
here
is
just
you
know,
different
timing
of
of
this
all
these
traces,
so,
for
example,
trace
which
you
know
happened,
happens
like
a
later
can
arrive
earlier,
and
then,
if
you
have
some,
you
know
link
that's
you
are
expecting.
C
You
know
to
to
be
resolved
at
this
point
of
time
you
might
want
like
it
might
be.
It
might
happen
that
you
don't
have
this
data
yet
right,
it's
kind
of
kind
of
eventual
consistency
or
type
of
so
yeah.
From
this
perspective,
it's
also
like
a
can
be
difficult,
but
I
mean
again
like
all
these.
All
these
problems
should
be
somehow
resolved
on
the
back
ends
on
the
back
ends
right
or
like
all,
the
all.
C
The
vendors
probably
should
start
another
discussion:
how
to
properly
design
this
this
type
of
types
of
data
and
how
to
you
know
what
is
the
best
storage
model?
What
is
the
best
like
a
model
for
querying,
for
you
know
yeah,
visualization
and
whatnot.
B
B
That
could
do
this
right
and
then
it's
it's
on
everyone
to
find
some
tricky
more
advanced
way
to
do
it.
But
I
I
do
feel
like
we're
a
little
bit
in
danger
of
like
foisting
foisting
something
gnarly
on
gnarly
and
expensive
on
people,
maybe
compared
to
what
they
currently
have
and
it's.
C
Right
right,
that's
a
really
good
point
and
yeah.
Basically,
it
probably
should
be
should
have
been
done
when
with
like.
Well,
when
the
all
the
overall
kind
of
links
concept
was
introduced
right,
because,
basically,
that
that's
the
the
root,
but
you
know
just
just
for
just
basically
we
are,
we
are
working
in
in
microsoft.
We
are
working
on
the
on
these
back
ends
explicitly
in
my
team
right
now,
and
we
do
have
some.
C
You
know,
scenarios
when
we
want
also,
you
know,
be
able
to
resolve
this
links
between
different
races,
yeah.
It
looks
like
in
in
upcoming
months,
we'll
be
in
in
microsoft.
We'll
be
trying
to
solve
this.
These
particular
problems.
C
How
how
to
do
this
and
it's
it
doesn't
doesn't
come
like
a
from
the
from
the
retry's
perspective,
but
mostly
about
like
from
from
the
asynchronous
scenarios
when
people
tend
to
use
these
links
to
just
you
know,
link
producer
consumer
and
for
different
phases
and
yeah.
It
looks
like
we
will
be
trying
to
do
something
about
that
yeah,
but
we'll
keep
you
posted
if
possible.
B
Good
and
personally
I
don't
think,
there's
anything
any
I
I
don't
think
I
think
it
is
not
harmful
to
model
these
with
links.
It's
like.
I
don't
think
we
will
regret
that,
because,
like
a
pointer
is
like
a
very
logical
data
structure
to
use
here,
I'm
just
like
feeling
like
it's,
probably
not
sufficient
for
a
lot
of
the
reasons
you
just
gave
a
big
one
being
like
it's
very
easy
for
a
span
to
go
missing
right,
there's
sampling
spans
can
just
get
dropped.
B
You
know
right
like
this
is
a
big
distributed
data
collection
system
and
it's
you
have
to
presume
it's
lossy,
and
so
a
linked
list
is
like
a
little
brittle
there,
but
if
they
all
had
like
you
know,
so,
if
there
was
like
something
additional,
you
could
staple
to
these
things.
B
That
would
increase
your
ability
to
to
ensure
things
like.
Even
if
you
lost
the
causal
relationship,
you
would
at
least
know
all
of
this
stuff
is
grouped
together
somehow,
so
I
guess
what
I'm
saying
is
like.
I
think
we
should
go
forwards
with
links,
but
I
don't.
B
I
think
we
have
to
solve
that
problem,
but
I
don't,
I
think
we
could
deliver
this
without
solving
that,
like.
I
think
it
would
be.
Okay
to
say,
look
we're
gonna
use
a
link
here.
We
agree.
There
should
probably
be
something
like
a
link
type.
You
know
to
differentiate
between
retries
and
like
other
stuff.
There
possibly
needs
to
be
something
else
other
than
just
a
link.
B
As
far
as
being
able
to
index
this
data
for
some
situations,
like
the
big
asynchronous
use
cases,
I
don't
think
that's
a
big
deal
here,
like
the
chances
that
you're
gonna
lose
some
of
these
links
and
not
the
other
ones
are
like
really
low.
That's
like
kind
of
a
big
deal
for
this
situation,
but
so
I
don't
know
that
would
be.
B
My
suggestion
in
this
thread
is
to
say,
like
we
think,
links
are
like
a
core
as
a
core
concept
are
right
here,
but
we
think
we're
gonna
need
to
add
stuff
to
this
later,
but
we
should
actually,
we
should
actually
model
that
cohesively
and
not
just
start,
not
just
start
shooting
from
the
hips
and
be
like,
we'll
just
add
a
link
type
or
something.
You
know.
C
Yeah,
that's
what
I
think
it
might
my
thinking
as
well,
so
it
looks
like
we
just
need
to
make
it
more
nature,
more
major
like
from
the
perspective
of
different,
you
know,
use
cases.
So
now
we
have
these
two
use
cases
like,
I
think,
scenarios
and
retries
maybe
going
forward.
We
might
you
know,
see
another
use
cases
that
we
can
also.
We
can
solve
this
like
using
using
this
structure
or
concept
of
links,
and
then
at
this
point
we
might
want
to
you
know,
have
some
explicit
identification
of
this
type
of
relationship.
C
It
might
be
useful
there
but
yeah
that
going
like
a
I.
I
basically
see
this
as
just
a
first
step
here,
so
it
looks
like
it
should
be
improved
somehow
going
forward,
but
yeah
just
just
for
now.
I
don't
don't
understand
like
what
exactly
we
can
add
here.
So
it
will
be
kind
of
you
know
meaningful
enough
and
stable
enough
going
forward.
So
my
perspective
here
that
we
can
start
using
link
here
links
here
as
a
concept
but
going
forward.
C
We
can,
you
know,
improve
by
clarifying
you
know
some
some
relationship
type
or
adding
some
group
grouping
id
or
whatnot
back
back
to
sampling.
Actually,
I
I
also
like
you
know
the
the
overall
idea
of
any
kind.
D
Maybe
you
can
find
a
case
there,
how
you
were
how
how
the
how
your
proposal
fits
into
those
those
cases,
because
I
I
think
that's
what
that's.
Maybe
what
bogdan
was
relating
to
that
we
say
yeah
kind
of
I
think
in
his
view
it
did
not
fit
in
there,
but
maybe
that
can
be
challenged.
B
Yeah
so,
unfortunately,
I
know
where
links
come
from
and
I
don't
think
you're
gonna
be
satisfied
because
they
were,
they
were
tacked
in.
They
were
think
tacked
in
by
google,
specifically
to
model
kind
of
like
fan
in
fannin
out
workloads
like
scatter
gather
kind
of
stuff,
and
that
was
it
like.
They
just
had
this
this
one
situation
they're
running
into
where
you
may
be
doing
a
kind
of
map
reduce
operation,
in
which
case
you
couldn't
use
parent-child
relationship,
because
just
the
way
things
were
kicked
off
in
parallel.
B
Like
it
just
didn't,
didn't
work,
you
didn't
know
the
ids
you
wanted
ahead
of
time
and
because
you
have
like
a
bunch
of
a
bunch
of
small
scatter
operations
right
and
then
you
have
like
one
gather
operation
that
pulls
them
all
together,
and
so
that's
not
a
parent
child
and
the
gather
starts
after
the
scatter.
B
So
it
can't
be
a
parent
child,
the
other
way
and
so
they're
like
we'll
just
invent
this
thing,
called
links
and
use
that
to
collect
this
stuff
up,
and
that
was,
as
far
as
I
can
tell
like
they're
one
use
case
within
google
that
they
use
them
for
and
they
don't
really
use
them
for
much
else.
C
B
B
Is
that
one
use
case,
which
is
why
he's
probably
looking
at
this
and
he's
like
this-
isn't
like
a
batching
thing
that
you're
doing
here,
but
something
this
community
has
done,
is
looked
at
links
and
looked
at
like
all
these
other
unsolved
problems
that,
like
they
didn't
solve
within
google,
for
whatever
reason
and
we're
like
oh
links
are
like
a
nice
like
cool,
so
we
we
have,
we,
we
have
pointers
and
you
can
attach
attributes
to
the
pointers
cool.
So
we
can
like
do
all
kinds
of
graphing.
B
With
this
thing
great
we're
gonna
use
that,
but,
but
I
think
bogdan's
gonna
find
most
of
these
use
cases.
Weird.
C
Right,
but
at
least
we
have
three
use
cases
right
now
for
links
right.
So
like
a
this
retries
redirects,
I
think
scenarios
and
batching
operations,
so
at
least
three
and
looks
like
just
putting
or
indicating
which,
which
what
what
type
of
what
what
type
of
you
know,
relationship
we
want
to
build.
We
want
to
build
using
the
links,
might
be
useful
for
all
three
of
them.
E
It's
like,
for
example,
that
word
sibling
like
we
can
talk
about
siblings
here,
but
since
we
have
no
way
to
actually
say
that
it's
a
sibling
like
if
the
spec
says
they
should
be
added
as
siblings.
That
doesn't
really
tell
me
anything
so,
like
those
sort
of
details,
you
can
probably
leave
out
for
now
just
add
a
link
with
the
trace
siding
span
id
like
that's,
probably
as
much
detail
as
we
need
for
the
first
version
before
trying
to
flesh
out
how
we
model
those
relationships
further.
C
C
This
one
yeah-
actually
it
was
it-
was
really
useful
for
python
folks,
because
I
also
was
working
on
this
with
python
folks
who
were
helping
to
you,
know
verifying
this
proposal,
and
there
was
an
explicit
ask
from
from
them
like
how
they
should
model
this
spans.
Should
it
be
like
a
nested
parent
spines
or
it
should
be
siblings,
or
some
mixture
of
this
that
that's
what
I.
E
C
C
E
C
Yeah
so
like,
as
of
now,
I'm
I
I
wanted
to
provide
any
any
more
clarifications.
I
actually
recently
rephrased
the
whole
section
here,
providing
these
examples
so
just
making
it
more
straightforward
for
people
who,
just
you
know
who
have
they
are
sdk
and
they
want
to
add
this
additional
instrumentation
for
a
device
in
their
rex.
So
looks
like
I
mean
this.
This
helped
a
lot
for
python
folks.
It
should
be
some
reference
to
their
work
somewhere
here.
C
Maybe
it
will
be
possible
yeah
here
it
is,
and
they
had
some
really
long
discussion
there
about
that.
So
basically
yeah.
So
that's
that's
was
the
main
question,
how
they
should
model
them
and
which
kind
of
which
type
of
spam
they
need
to
create
for
retries
and
redirects.
C
So
basically,
this
discussion
helped
them
a
lot.
I
believe
so
yeah
I
just
says
like
I:
it
makes
sense
to
them
so
yeah,
but
if
you
see
any
any
other
things
that
can
be
clarified
here,
even
more,
I'm
really
open
to
to
make
this
clarification.
B
Okay,
well,
I'm
gonna
respond
to
bogdan
on
this
thread
with
some
of
what
we
just
talked
about
and
make
the
case
that
links
seem
to
be
right
here.
B
We
need
more
than
what's
currently
here,
but
but
what
currently
currently
here
is
like
enough
to
to
move
forwards
and
it
sounds
like
yeah,
microsoft
and
others
are
going
to
you
know
start
you
know
making
use
of
links,
and
so
we'll
get
some
feedback
there
about.
You
know
what
what
we
might
need
in
addition
to
what's
already
here.
B
A
A
All
right
so,
as
we
all
know,
the
otep
dennis's
oceap
was
talking
about
forex
errors
and
we
need
to
kind
of
decide
how
to
handle
client
vortex
errors
in
spans.
A
A
So
I
really
just
raised
this
pr
in
order
to
see
what
people
thought
and
see
if
we
can
get
a
kind
of
consensus
and
decision,
whether
these
should
be
left
unsaid
or
they
should
be
left
as
an
error
and
seems
like,
like
nikita
says
it's
against
against,
it
seems
like
I
I
feel
like
most
people
are
against,
and
most
people
are
saying
we
should
keep
it
as
it
is,
which,
to
be
honest,
I'm
totally
fine
with
that.
I
really
this
pr
is
just
about.
A
Should
we
make
a
decision?
I
know
that
johannes
and
dennis.
A
B
Yeah
I
mean
it
seems,
like
I
haven't
read
through
all
the
comments,
but
it
it
does
seem
like
people
in
general
feel
like
400
should
represent
errors
on
the
client
and-
and
that
is
like,
like
a
logical
reading
of
the
situation
right
like
like
that
is
like
the
400
does
mean
the
client
made
an
error.
That's
what
the
server
says
you
you're
done.
Did
it
wrong?
That's.
B
Says
right,
yeah,
so,
fair
enough,
so
so
that's
reasonable!
You
know
the
only
pushback
I
would
have
against
that
would
be
if
like
in
practice.
This
is
just
like
super
super
obnoxious,
and
I
don't
know
if
that's
the
case,
so
I
guess
that
would
be
kind
of
my
question.
Is
that
like
over
at
atlassian?
Is
that
like?
A
A
B
C
Yeah,
I
just
wanted
to
mention
that
you
know
in
microsoft,
there
is
a
product
called
azure,
monitor
application
insights
which
which
built
on
top
of
like
a
distributed
tracing
thing
and
this
product
has
this
feature
called
app
map
which
basically
just
renders
you
know.
You
know
some
some
kind
of
dependencies
between
services
and
oh,
look
at
these
dependencies.
C
It
marks
it
with
you
know
some
green
or
red
lines,
depending
on
the
errors
that
happens.
You
know
during
the
communication,
and
this
view
explicitly
has
this
kind
of
flag
like
exclude
4x6
as
errors,
just
because,
if
you
add
it
there,
it
generates
too
much
noise
right
because,
like
you
know
it
basically,
the
whole
picture
becomes
red
in
this
case,
because
you
know
all
like
a
in
in
clouds
environment,
you
you
have
a
lot
of
different
retrace.
C
You
have
a
lot
of
different
kind
of
trends
in
failures,
and
you
know
if
you
have
them
like
a
4xx
as
errors.
Basically,
you
see
all
the
all
this
stuff
as
just
red
picture.
So
from
this
perspective,
I
believe
from
the
like
from
the
and
actually
that's
that's
my
personal
experience
when
I
was
using
different
or
instrumented
different
types,
kind
of
applications
or
services
with
distributed
tracing.
C
That's
people,
just
you
know,
see
this
spans
as
or
and
their
status
as
just
noise,
just
because
yeah
it
happens,
but
like
overall,
the
overall
communication
is
just
fine,
because
we
were
using
great
rice
and
you
know
some
some
kind
of
additional
resiliency
patterns
and
we
are
fine
in
general,
so
yeah.
So
from
this
perspective,
I
believe
that
the
overall
idea
was
to
you
know
to
remove
this.
C
I
would
say
kind
of
controversial
statement
that,
since
all
these
are
errors,
like
a
four
four
x
x
as
client-side
errors,
we
need
to
treat
them
as
errors,
but
the
same
the
same
specification
actually
explicitly
says
that
5xx
should
be
treated
as
errors
regardless
right,
but
yes,
server,
side,
and
you
know
it's
it's
like
it's
already
controversial
at
this
point.
So
yeah.
B
C
B
B
A
B
Are
some
specific
places
where
you
kind
of
want
to
look
at
those
right,
because
it
might
indicate
something
something
strange
is
going
on
right
like
like
if
the
the
percentage
of
user
logins
400s
like
goes
above,
like
you're
gonna,
have
like
noise
right,
but
if
it
suddenly
goes
to
like,
like
a
hundred
percent
user,
logins
are
like
400s,
you
know,
some
up
thing
has
happened
and
you
need
to
go.
Take
a
look
at
that
right,
but
you
know
like
in
general:
it's
not
like
the
way
you
the
way
you
set.
C
Don't
know
yeah
personally,
I
feel
that
it
will
be
useful.
You
know
if
at
this
point
like
a
prior
to
to
the
stable
version,
we
just
relax
these
conditions
a
bit
saying
that
you
know
we
can
just
leave
them
as
set.
But
if
you
really
want
nothing
like
you
know,
you
can
do
this.
You
still
can
do
this.
You
can
put
all
the
errors
everywhere.
C
B
D
A
So
you're
saying
about
microsoft
system
kind
of
looks
at
all
four
x
x's
as
errors
and
then
creates
red
red
stuff.
I
can't
remember
the
exact
details
like
the
flip
side
is,
maybe
you
miss?
A
D
C
Right
so
that's
mostly
about
like
a
4xx
right,
and
you
know
when
it
comes
to
the
the
the
case
when
that
map
with
rich
net
mention.
Basically,
when
you
have
100
of
errors,
then
you
have
a
lot
of
you
know
many
different
indications
of
this
obvious
problem.
C
So,
for
example,
you
will
be
having
like
a
5xx
on
your
server
side
as
well,
and
you
know,
even
if
you
do
sampling,
then
you
still
be
having
some
kind
of
spike
in
you
know
on
different
levels,
because
basically
we
are
just
talking
about
http
level
of
things,
but
in
reality,
http
is
just
just
a
you
know
some
some
level
like
and
we
have
many
different
levels
of
communication.
C
So
from
this
perspective,
like
a
that's
http
level,
just
from
my
perspective
generates
a
lot
of
noise.
But
if
you,
if
you
have
this
error
and
it's
something
really
bad
happens,
then
all
different
levels
will
be
adding
your
disinformation
and
I
believe
it
will
not
be
possible
to
miss
the
fact
that
you
have
some
something
bad
happening
in
your
system.
A
Yeah,
I
suppose
that
is
true
for
like
this
scenario
that
chad
mentioned,
like
you,
know,
100
401s,
or
something
like
that,
but
like
often
there's
just
one
span
or
two
spans
that
it's
edge
case
and
it
is,
it
is
an
error
as
well
like.
I
guess
it
wouldn't
be
the
end
of
the
world
if
it
was
left
unset
as
an
error
either
but
yeah.
That's
just
my
my
thoughts
as
well.
A
C
Yeah
just
found
this
this
one
open
issue
in
application:
insights.net
libraries,
which
is
just
like
a
sdk
part
of
application
and
sites
with,
and
people
just
asking
to
filter
out
kind
of
normal
for
xx.
C
Okay,
that's
exactly
the
case,
but
yeah,
but
maybe
we
can
have
you
know
some
another
discussion
like
offline
discussion
with
christian
with
with
bogdan,
to
try
to
resolve
this,
or
we
can
just
try
to
reiterate
this
offline.
So
what
do
you
think
will
be
the
most
efficient
way
for
us
to
make
progress
here.
B
Well,
I'm
sorry
I'm
going
to
respond
to
this
just
kind
of
some
I'm
going
to
respond
to
this
in
favor
of
your
pr
james.
By
just
proposing,
I
think
what
we
just
discussed,
which
is
like
look,
people
need
to
configure
this
stuff.
They
shouldn't
configure
it
at
the
instrumentation
level
to
be
clear,
like
all
this
configuration
should
happen
in
collectors
and
farther
down
the
pipe.
It's
like
totally
obnoxious
to
to
to
implement
this
everywhere.
It's
like
instrumentation
configuration
in
general.
B
I
think
we
want
to
say,
like
we're,
not
going
to
take
that
approach
in
open
telemetry
because
it
sucks,
but
but
that
said
what
is
like,
the
default
people
are
looking
for,
because
it
would
be
a
bummer
to
have
these
on
by
default
and
then
have
like
90
of
users
like
wasting
cpu
cycles,
to
suppress
them
in
collectors.
C
And
okay,
it
sounds
some
sounds
reasonable,
yeah,
just
just
to
have
yeah
as
as
much
flexibility
as
possible
when
we
go
to
the
first
stable
version,
because
you
know
it
actually
gives
us
more
flexibility
going
forward.
D
Yeah,
just
one
point
I
mean
I
I
I
because
I
had
somebody
actually
reaching
out
with
with
a
problem
very
similar
to
that.
Also
with
four
fours
like
it
was
a
like,
a
library,
instrumenter
and
the
library
used
http
under
the
hood,
and
basically
they
hit
like
an
endpoint
and
when
a
404
came
back.
Basically
it
was
success.
Actually,
so
they
just
checked
if
something
does
not
exist,
and
they
asked
me
how
to
model
that
with
open
telemetry
and
how
to
say,
you
are
currently
unfortunate.
D
You
can't
with
the
comment
state
of
the
spec
and
I
think,
by
making
those
four
fours
not
errors.
Still
every
instrumental
has
the
flexibility
to
kind
of
set
an
error
on
the
parent
span
of
the
http
span.
I
mean
you
have
that
you
you,
you
can
just
set
an
error
on
the
parent
span
and
then
you
have
a
trace
with
the
error.
It
is
not
true.
Vice
versa,
like
when
4
4
creates
an
error
as
an
instrumental
you
have.
You
have
no
way
to
suppress
this
error.
D
That
is
just
there
and
then
you
have
to
rely
on
the
backhand
to
do
something
there,
but
you,
when
you
ship,
a
library
you
have
no
way
of
kind
of
suppressing
those
errors,
but
when
it's
the
other
way
around,
if
it's
not
an
error,
when
you
are
instrumental,
you
can
still
put
an
error
in
the
parent
span
and
thus
create
the
behavior
that
you
want.
So
I
think
actually
there's
more
flexibility
with
having
four
force
being
not
errors.
C
C
All
right,
so
the
plan
probably
would
be
due
to
finalize
these
two
discussions
next
week,
but
ted
will
be
not
here
so.
B
I
won't
be
here
but
but
move
on
without
me,
I
I
want
I
wanna.
My
hope
is
that
this
this
group
will
will
grow
up
and
interact
enough
with
the
the
tc
members
that
I
can
get
hit
by
a
bus
and
that
wouldn't
the
end
of
things
here,
not
that
I
want
that
to
happen
yeah,
but
I'm
gonna
leave
my
comments
and
I
will
try
to
follow
up
on
them
and
I
will
have
like
I
will
be
on
slack.