►
From YouTube: 2022-12-01 meeting
Description
Instrumentation: Messaging
A
B
B
A
B
B
D
So
we
have
quite
some
items
on
the
trend
or
maybe
let's
first
get
those
out
of
the
way
that
might
not
take
that
long.
Maybe
let's
start
with
you
what's
wrong
your
vacation
announcement.
B
Oh
yeah
I
just
wanted
to
to
let
you
know
that
I
will
be
mostly
away
in
December,
I'll
take
a
longer
vacation,
I
might
might
join
in
middle
of
December
or
so,
but
I
I
want
I
won't
pass.
So
if
you,
if
you
don't
see
me
then
but
I
didn't
disappear,
I'm
just
I'm
just
visiting
for
a
while,
but
that's
about
it.
B
D
And
then
next,
maybe,
let's
quickly
chat
about
the
attributes
PR,
because
I
think
they
are
actually
not
not
many
more
open
comments
on
the
pr.
C
Oud
so
there's
one
open
discussion
and
I
was
I
wanted
to
ask
participants
to
check
if
there
is
anything
to
address
there.
I
don't
believe
so.
Foreign.
C
Yeah
I
think
there
are
some
points
from
Bruno
I.
Don't
think
he's
here
and
I
can
think
people
are
Paulo,
yeah,
I'm,
sorry.
D
About
what
goes
into?
What
goes
into
the
message
and
what
goes
into
the
destination
or
source.
D
Yes,
so
I
can
I
will
have
a
I'm.
Sorry,
I
didn't
wasn't
able
to
do
it
last
week,
but
I
will
have
a
look
at
the
two
and
I
think.
If
that
is,
if
they're
not
not
any
more
questions
here
from
me,
then
I
actually
think
I
can
approve
the
pr.
D
Maybe
other
people
can
have
we'll
have
a
look
again
because
I
think
that's
the
only.
B
Yes,
I
just
looked
at
it
before
the
I
wanted
to
do
it
sooner,
but
I
I
managed
to
do
it
today.
Yes,
but
it's
just
just
Cosmetics
I
think
nothing,
nothing
major,
not
nothing
blocking
it.
C
I'll
try
to
address
it
today.
So
are
you
still
around
tomorrow,
ciao.
C
Okay,
okay,
so
I'll
ping,
you
on
slack
and
then,
if
you
can
just
approve
it
after
you're
happy
with
it,
it
would
be
great,
yes
and
I
think
that
the
important
part
is
Dwayne.
Can
you
please
also
review
it
because
you
you're
requesting
changes,
and
this
is
the
signal
for
the
technical
committee
members
not
to
review
it,
because
we
don't
have
a
consensus.
C
So
if
you
can
take
another
look
and
see
if
there
is
anything
that
has
to
be
fixed,
please
leave
your
comments.
Otherwise,
can
you
please
approve
it.
E
D
Yes,
so
let's
try
to
get
some
some
reviews
on
here.
I
will
also
take
time,
either
today
or
tomorrow,
and
give
this
another
pass,
and
hopefully
we
can.
We
can
approve
this
and
then
we
are
other
people
outside
our
group
can
start
reviewing
this
PR
awesome.
D
So
we
have
done
a
next
Point
regarding
span
structure
links,
but
maybe
we
should
take
this
point
before
the
questions
around
batch
processing
with
Apache
knife.
Five
I
think
that's
how
it's
pronounced
I,
guess
that
is
from
either
Brian
or
Greg.
D
D
Because
I'm,
maybe
let's
just
get
a
quick
idea,
what
what
exactly
this
is
about,
and
then
maybe
it
anyway
relates
to
questions
about
span,
structured
links
that
we
are
that
we
are
going
to
discuss
here.
D
D
Takes
some
time
so
if
you,
when
you're
dialing
Brian
just
interrupt
us
and
we
will,
we
can
go
over
to
to
the
to
The
Pantry
point
and
in
the
meantime
we
can
start,
we
can
start
talking
about
those
band
structuring
links.
I
mean
we
have
two
documents
here.
The
one
is
the
Otep.
It
is
open
for
some
time
and
the
other
one
is
amir's
document
about
messaging
links.
I
think
Luke
Miller
had
added
some
comments
to
the
messaging
links
document
and
I.
D
Think
one
of
the
first
questions
here
I
would
just
love
to
answer
is
better
in
terms
of
the
a
spec
process,
whether
we
can,
whether
we
can
extend
this
existing
span
structure
Otep
so
that
it
captures
the
requirements
that
Amir
that
you
put
forth
in
your
in
your
Google
talk
here.
A
So
what
I
want
to
avoid
is
the
situation
we
had
with
the
previous
thought
app
that
we
added
a
lot
of
things
to
it
and
it
got
so
big,
and
there
were
so
many
comments
that
it
was
a
very
hard
to
follow
and
since
their
links
discussion
is
not
like
it
is
related,
but
it's
not
like
we
can
walk
on
it
separate
from
the
spin
structure,
so
I
want
to
suggest
to
to
do
them
in
a
parallel
that
way.
C
C
To
visualize
it
in
the
back
end
right,
so
it
gives
like
the
backend
perspective.
It
does
not
need
to
end
up
in
the
spec
necessarily
right.
We
can
just
extract
I.
Think
your
high
level
statements
are,
some
of
them
can
go
directly
to
this
pack
and
I.
Think
it's
just
a
few
sentences
that
we
need
to
add.
C
Maybe
we
can
just
work
on
the
pull
request
exactly
directly
to
this
pack,
because
this
few
sentences,
or
to
the
current
Sun
structure
PR
but
I,
guess
it's
related
to
attributes
to
the
big
extent,
plus
the
link
usage
and
then
maybe
it
does
not
belong
constant
structure.
I,
don't
know
what
do
you
think.
A
So
I
really
want
to
do
it
as
fast
as
possible
like
if,
if
we
can
just
open
a
BL
and
get
it
merged,
but
I
think
the.
As
you
mentioned,
the
the
changes
to
the
speckle
like
the
the
not
a
lot
of
changes
to
the
spec,
but
the
philosophy
behind
why
we
need
to
to
do
each
thing.
It
needs
to
be
documented
somewhere,
I'm,
not
sure
if
the
spec
is
the
right
place
to
to
do
it,
but
like
in
the
spec
we're
going
to
say
you
must
do
this.
A
You
should
do
this,
but
the
reason
that
we
want
these
changes
like
if
somebody
wants
to
to
understand
the
big
picture,
then
not
sure
what
we.
What
is
the
right
place
to
document
this.
C
I,
don't
think
we
have
the
president
for
other
specs
like
this
like
an
up
until
I'm
a
tree.
It
may
be
blogs
or
some
like
your
personal
something,
but
there
is
nothing
like
this
for
HTTP
database
and
I
agree.
It's
useful.
C
D
It's
definitely
useful
to
have
like
the
back-end
perspective,
but
I
think
then
the
main
question
for
actually
opens
back
is
to
ask
you
what
kind
of
what
requirements
do
we
have
to
fulfill
on
the
instrumentation
side
that
we
can
kind
of
fulfill
this?
What
is
described
in
armies
document
here
and
I
think
that's
what
needs
to
go
into
the
spec,
but
but
definitely
armies
document
is
important
too,
because
that
kind
of
describes
why
those
changes
or
by
those
kind
of
parts
of
the
spec
are
there.
A
So
we
can
also
have
like
a
blog
post
on
the
open,
Telemetry
website
or
or
maybe
add
it
to
the
to
the
VR
description.
So
if
somebody
wants
to
wait
again
see
where
it's
added
and
read
in
the
description
but
yeah
I
already
started
to
work
on
it
and
I
hope,
I
will
have
the
updates
soon,
because
this
document
was
just
to
for
you
for
the
sick.
Just
to
give
you
like
the
points
of
what
I
want
to
do,
but
I
would
want
to
formalize.
B
I
just
wanted
to
to
ask,
for
example,
the
document
is
great,
would
would
it
benefit
be
beneficial
for.
B
If,
at
the
top
of
the
document,
you
had
the
specific
action
points
that
you
would
have
to
change,
for
example
in
the
spec,
so
so
you
have
the
whole
document
explaining
the
goals,
everything
but
then
maybe
at
the
top
or
at
the
bottom,
whatever.
B
If
we
could
have
like
the
Practical
changes
that
the
that
needs
to
happen
in
the
spec,
then
we
could
look
at
it
and
say:
okay
like
this,
this
is
small
enough
and
it
can
go
directly
to
this
spec
as
a
PR
or
no,
it's
probably
better,
to
have
to
have
a
note
app
and
then
the
whole.
Let's
say
context
and
in
story
of
of
why
we
came
up
with
this,
then
it's
recorded
as
an
Otep,
along
with
the
let's
say,
change
that
going
to
be
going
to
the
spec.
B
I'm
not
sure
if
this
helped,
but
but
I
I
read
it
once
last
week.
The
the
document,
but
I
was
a
bit
like
okay.
But
what
what
like
actual
practical
changes
are
going
to
happen
in
the
in
the
spec
like.
D
B
All
so
all
of
this
all
of
these
items
that
will
will
change
then
to
on
the
spec.
A
Yeah,
it's
it's
I,
didn't
phrase
it
for
the
spec.
It
was
just
to
leave
yeah
we're
going
to
talk
about,
but
I
I
will
walk
on
a
formalizing
it
and
give
it
some
structure
and-
and-
and
we
can
discuss
it
next
week
or.
B
Yeah,
because
if
it's,
if
it's
just
a
little,
not
not
a
lot
of
change,
then
I
agree
with
me
that
we
can
just
go
directly
to
the
pr
and
you
can
get
it
done.
I
guess
quickly,
especially
now
that
it
was
recently
clarified
that
we
can
introduce,
let's
say
breaking
changes
so
and
we
are
the
owners
of
that
space.
So
we
should
be
able
to
move
things.
I
guess.
A
Yeah
I
don't
think
the
breaking
changes
in
this
spec
is
only
new.
Yes,.
B
Yeah,
so
it
should
be
even
even
better
than
not
not
breaking
I
want
to
say,
but
in
recently
I
noticed
that
there
was
some,
not
pushback,
but
there
was
some
some
PRS
and
symmetic
conventions
being
like
delayed
because
there's
still
this
discussion
around
if
things
are
gonna
affect
metrics
and
and
how
it
is
relates
to
Resource
and
so
on
that
there's
some
stuff
going
on
behind
I
guess
the
related
to
semantic
conventions
but
yeah.
But
it
should
be
fine
and
I
think
for
required.
B
I
think
it
does
make
sense
and
I
think
being
the
best
or
taps
were
used
for
something
like
that
right
for
recording,
like
the
rationale
behind
things
and
and
and
why
but
I
not
sure
if
that's
gonna
take
a
lot
of
time
to
put
the
or
tap
and
then
review
and
then
get
external
people
to
review
as
well.
So
if
it's
just
for
context
on
the
spec
Pi,
then,
for
example,
your
document
should
be
I,
guess
a
good
context.
A
D
Yeah,
that's
what
I
said
I
mean
I
would
suggest
that
maybe
you
could
work
out
actually
like
the
text
that
should
go
into
the
spec,
so
you
see
okay.
Where
would
we
need
to
add
something
to
the
spec?
D
What
wording
would
we
need
to
add
and
then
based
on
that
we
can
or
or
you
can
or
we
can
see
how
to
best
proceed,
whether
that
is
just
like
a
change
that
we
can
simply
put
into
PR
or
whether
the
changes
would
be
kind
of
big
or
fundamentally
enough,
that
it
needs
an
Otep
or
a
different
approach.
G
B
D
Thanks
Amir,
so
let's
then
continue
discussing
on
this
next
week.
E
F
So
we're
developing
a
integration
with
apache95
and
basically
just
have
a
few
questions
regarding
the
spec
when
it
comes
to
batch
processing
figured
we
kind
of
show
some
use
cases
to
you
guys
and
see
what
your
thoughts
are
on,
how
we're
handling
it
just
to
make
sure
we're
in
line
with
the
spec
and
for
those
that
don't
know,
Apache,
nifi,
more
or
less
is
kind
of
a
combined,
a
queuing
mechanism
and
then
consumer
and
producer
all
in
one
package,
so
I'll
kind
of
I'm
going
to
share
my
screen.
D
Go
ahead,
I
mean
first
I
I
can
give
you
some
short
context
about
like
the
group
here
and
what
they're
currently
working
on,
because
when
we're
talking
about
the
messaging
semantic
conventions,
there
is
those
semantic
conventions
that
are
currently
part
of
the
hotel
specification
and
be
in
this
group.
Here
we
are
actually
working
on
refactoring
those
semantic
conventions
and
and
bringing
them
to
a
state
so
that
we
can
call
them
stable
instead
of
experimental.
So
we
are
currently
in
the
process
of
making
significant
and
breaking
changes
to
the
existing
semantic
conventions.
D
There's
one
PR
App
that
we
looked
at
in
the
beginning
for
attributes
and
there's
another
Otep
that
we
plan
to
to
continue
discussing
either
today
about
the
span,
structure
and
I.
Think
all
those
changes
also
that
we
are
currently
working
on
influence
like
the
bettering
scenarios,
so
it
would
be,
would
be
great
to
get
your
input
and
also
to
see
your
kind
of
pain
points
with
the
current
spec,
because
that
might
help
us
too,
to
validate
any
things
that
we
came
up
with
so
far.
F
F
Okay,
great
so
effectively
what
you're
looking
at
here
is
just
a
few
knife
eye
processors
and
to
kind
of
make
it
simple
you
can
think
of
this
as
a
producer
in
terms
of
if
we're
talking
about
Kafka
as
a
killing
mechanism.
So
as
a
producer,
this
line
represents
a
topic,
and
this
represents
the
topic
name.
F
So
you
have
producers,
consumers
that
then
produce
as
well.
So
just
kind
of
keep
that
in
mind
when,
when
I'm
kind
of
going
over
this
scenario,
but
so
if
you
can
think
about
this,
we
we
would
receive
a
file,
a
single
record
and
then
that
record
would
contain,
let's
say
three
lines
in
it
and
so
we're
splitting
based
on
new
line.
So
now
you
would
have
one
record
that
then
gets
split
up
into
three.
F
So
that's
what's
denoted
by
this
quote-unquote
topic,
name
right
here
and
then
of
course,
we
pass
down
the
original
file,
so
the
file
that
has
the
three
records
into
its
own
topic.
F
So
we're
just
trying
to
understand,
because
when
we
looked
at
links,
bands
I
don't
think
it's
necessary
for
this
scenario,
one
to
many
and
so
we're
just
kind
of
showing
this
as
kind
of
this
is
what
we
would
expect
as
we're
testing
this.
This
out.
Implementation
I
just
want
to
make
sure
that
this
is
kind
of
in
line
with
where
you
guys
are
headed.
D
Well,
digital
Force
seems
like
like
a
more
or
less
standard
kind
of
multi-level
messaging
workflow,
because
you
have
like
the
first
message
that
is
then
consumed
and
then
in
the
process
of
consuming
this
message,
you
create
two
more
messages
that
are
then
consumed
at
the
later
stage.
D
About
the
patreon
I5,
because
I'm
not
familiar
with
that
is,
is
about
to
nifi.
Is
that
is
it
a
messaging
broker
or
is
it
kind
of
an
in-memory
message,
processing
system.
F
It's
a
kind
of
it's
pretty
you're
rich
and
that
it's
stateful,
so
basically
you
have
queuing
built
in
so
we
can
move
data
around
data
could
be
anything
could
be
text.
Video
audio,
you
know
whatever
you're
interested
in
moving,
and
so
it's
drag
and
drop.
F
So
it
comes
with
a
GUI
and
then
it
has
these
things
called
processors,
and
these
processors
are
typically,
you
have
three
types,
mostly
Ingress
egress
and
then
you
know,
do
work
kind
of
a
unit,
work
level,
and
so
each
one
kind
of
offers
different
capabilities,
and
so
you
can
kind
of
route
your
data.
F
D
F
Okay
yeah,
so
rather
than
me,
writing
a
consumer
like
if
I'm
doing
Kafka
streams
rather
than
me
writing
the
code.
The
code
already
exists,
and
it's
just
a
matter
of
configuring
it
and
then
it
just
kind
of
runs
and
then
can
pick
up.
Data
handles
back
off
retry
things
like
that
automatically
for
you.
A
F
A
A
A
A
F
Well,
so
you
know
a
simple
example,
but
it
you
know
if
I
had
more
detail,
it
would
allow
me
to
kind
of
look
at
the
lineage
of
how
long
it
took
something
coming
in
to
where
it
ends
up
Final
Destination.
So
it's
not
necessarily
a
transaction
more
more
or
less
we're.
Looking
at
the
total
time,
it
takes
something
to
go
through
a
system,
and
so
this
Jaeger
example
kind
of
references,
the
flow,
if
you
will
so
you
have
this
first
span
the
root
which
is
I,
receive
something,
and
then
this
one.
F
So
there's
four
here
that
say:
split
text,
which
is
the
name
of
this
one
right
here.
So
one
of
these
is
the
original
file
that
gets
passed
through
based
on
the
output
of
this
processor.
So
the
the
quote-unquote
topic
that
pushes
out
and
then
also
the
same
level,
will
work
for
the
three
individual
messages
that
get
generated.
Based
on
that
three
line
file:
getting
split
into
three
individual
messages.
F
A
F
Yeah
so
effectively,
this
right
here
kind
of
shows
how
long
it
took
to
receive
the
file.
This
one
shows
how
long
it
took
to
split
the
file,
and
then
each
one
of
these
is
kind
of
showing
how
long
it
takes
to
sit
on
the
Queue.
If
you
will,
before
it
kind
of
it's
sitting
in
here,
and
then
it
goes
to
the
next
level,
and
so
that's
just
representing
that
amount
of
time.
E
D
Yes,
that
is
actually
an
open
discussion
point
that
we
have
when
we
factor
into
spec,
whether
we,
whether
we
want
to
go
consistently
with
links
in
our
cases
or
whether
we
want
to
have
parent-trial
relationships
in
the
in
the
world
to
many
cases.
D
So
we
did
not
100
settle
on
the
debt,
because
the
current
spec,
yes
says,
has
examples
in
there
that
say:
okay
for
those
money
to
one
to
many
cases,
use
parent
trial
relationships
and
only
if
you
must
use
the
links
like
if
you
have
a
situation
where
you
would
have
multiple
parents
in
batting
scenarios,
whereas
this
first
draft
of
our
span
structural
tip
that
we
came
up
with
suggests
to
use
links
in
all
cases
to
be
consistent.
D
D
A
D
D
This
message
and
I
mean
I'm
a
bit
confused.
Why
there
is
a
orders,
one
two,
three
four
split
text
for
for
each
split
yep,
but
you
don't
have
to
do
that,
but
you
don't
have
like
actual
duration
for
the
process
of
the
splitting,
but
you
have
kind
of
these
four
durations,
and
that
means
that
those
are
the
durations
of
putting
the
split
messages
into
the
queues.
F
But
I
guess
would
you
expect
if
I
have
one
to
many,
that
I
would
generate
the
four
individual
records
that
get
put
into
the
queues
and,
if
so,
is
how
we
have
it
in
terms
of
the
the
body
of
work
being
split
text
for
each
message?
Is
that
accurate
or
does
it
need
to
have
split
text
one
that
then
has
subsequent
ones
like
go
underneath
or
is
this
unreasonable.
D
I
mean
to
generally
talk
here
about
the
the
messaging
instrumentation
here.
The
spans
you
have
here
are
basically
related
to
your
operation.
You
have
listened
to
be
split
text
and
then
your
Vlog
attribute
of
Avatar
messaging
semantic
conventions.
They
basically
ask
you
to
create
spans
that
are
specifically
related
to
messaging
operations.
So,
for
example,
you
have
like
you
have
your
maybe
listen.
Http
span
that
kind
of
measures-
the
duration
of
gathering
this
this
file,
but
then
you
would
have
a
separate
span
for
kind
of
that.
D
D
So
basically,
you
would
have
just
this
one:
a
distinct
scent
or
published
span,
then
that
covers
storing
the
Span
in
the
queue
and
then
on
the
other
side.
Before
you
do
the
split
text,
you
would
have
like
an
operation
called
receive
which
basically
receives
this
message
again
from
the
queue.
Then
the
send
previous
sent
that
the
receive
will
be
linked,
and
then,
after
this
receive,
you
will
have
like
this
for
this
other
split
text
operation.
D
F
It
does
but
also
I
think
that
might
get
pretty
busy,
especially
when
there
could
be.
You
know
many
of
these
I
guess
because
I
don't
know
what
kind
of
messaging
systems
are
expected
for
the
for
the
specs
of
you
know
how
many
quote-unquote,
consumers
and
and
producers,
but
I,
presume
that
it
would
get
pretty
busy
pretty
quick.
D
F
Yeah,
this
one's
probably
a
little
more
advanced
that
deals
with
link
spans,
but
so,
when
I
talk
about
listening
HTTP,
you
can
just
think
of
that.
As
someone
sends
you
a
file
using
curl
for
example,
and
so
you
have
a
like
a
web
server
that
basically
accepts
a
file
coming
in,
pushes
it
into
a
queue
now
merge
content
is
the
opposite
so
effectively.
You
have
many
files
queued
up.
Denmark
content
puts
those
files
into
one,
so
you
can
think
of
it
as
I'm
zipping
files
into
a
single
file.
F
So
same
same
thing:
three
files
sit
in
the
queue
and
this
thing
pulls
down.
Three
files
merges
it
into
a
single
record
coming
out
and
same
concept
of
original.
So
these
are
your
individuals
messages.
So
there
will
be
three
here
then
merge
is
the
single
file
coming
out
so
I'm,
just
showing
this
point
of
view
we'll
get
into
the
more
advanced
for
Trace,
but
you'll
see
here
it
generates
four
Force
bands
or
four
distinct
traces.
Rather,
and
so
you
know
looking
at
this,
you
have
again
three
coming
in.
So
that's
three
separate
traces.
F
D
And
basically
your
merge.
Your
merge.
Your
merge
span
is
then
linked
to
all
those
three
listen.
Http
spends.
So
here
you
have
links.
F
So
yes,
my
question
to
that
is
shouldn't.
There
also
be
links
on
the
individual
that
are
getting
merged
into
one,
because,
if
you're,
using
Something,
Like
Jaeger
or
one
of
these
display
tools,
you
lose
trackability,
so
I
see
these
and
then
that
to
me
that's
the
end
of
the
trace,
but
realistically
I
imagine
there
should
be
links
here.
That
then
allows
me
to
look
further
down
the
flow.
D
So,
let's
see
what
you
have
here
in
the
in
the
demo
three
here
you
have:
can
you
go
back
to
the
third
slide?
That
was
not
cute,
so
here
you
have
merge
content
as
a
child
of
recent
http.
D
F
Yeah,
so
what
happens
is
there's
going
to
be
four
merge
contents,
so
this
one
looks
at
the
original
coming
through,
so
it's
basically
just
a
pass
through
if
you
will
and
then
the
new
one,
a
new
Trace,
the
parent
is
the
merge
content.
So
this
one
right
here
is
the
root
of
the
new
trace,
and
so
that
is
merge
content
that
then
links
to
those
three
records
that
came
in.
A
D
B
B
So
if
I
got
it
right,
you
have
three
files
that
you
mentioned
in
the
beginning
right,
so
you
have
the
first
file
coming
in
listening,
HTTP,
best
image
contact
and
then
pass
as
this
to
the
blog
attribute
on
the
left,
and
then
you
have
this.
You
can
have
this
parenting.
Child
relationship
like
this,
because
it's
just
like
an
RPC
call
normal
it.
F
The
question
or
suggestion
I
have
is
Grant
I'm,
not
showing
the
full
view
here,
but
effectively,
there's
or
both
they're
supposed
to
be
four
of
these.
On
this
view,
one
of
them
just
links
to
or
I
guess.
Let
me
let
me
rephrase
that
is.
Does
it
make
sense
to
on
this
View
have
a
link
to
the
new
traces
that
get
generated
as
if
I'm,
drilling
down
and
I
and
I
see
the
file
coming
in
I
lose
the
ability
to
see
where
it
goes
from
here.
A
B
Is
the
only
way
that
you
can
have
a
a
link
for
to
all
the
files
that
come
in
is
is
on
the
on
the
merge
married
one
that
the
other
two
show,
because
you
you
receive
a
message
and
then
like
the
three
whatever,
then
you
can
add
them
as
as
links
but
and
in
the
other
case,
every
fire
that
you
pass
through.
It's
it's
a
single
trace,
so
not
sure
if
how
you
can
have
the
other
ones
as
well
there.
In
that
case,.
G
F
Potentially
next
week
or
the
week
after
it
sounds
like
we
got
the
first
one
down
pretty
good,
we'll
of
course
look
at
the
timings
and
then
I
guess
the
last
question
due
to
time.
F
This
is
just
a
simple,
straightforward
flow.
There's,
no
merging
there's!
No
splitting
it's
just
a
single
pass
through
the
question
I
have
here
in
terms
of
a
messaging
system,
how
you
handle
failures
back
offer!
You
try.
Does
this
look
accurate
in
this
example?
It's
not
able
to
save
a
file
to
disk,
and
so
it
backs
off
for
30
seconds
and
then
tries
again.
So
we
just.
We
think
this
is
accurate,
but
we
just
wanted
to
at
least
show
it
to
you
all
to
make
sure
that's
expected.
B
G
G
B
D
I
think
that
looks
okay
for
me,
because
that's
the
that's
the
best
you
can
do
here
actually
because
you're
doing
like
some
asynchronous
processing
and
if
you
do
synchronous
processing,
you
can
Bubble
Up
the
error
to
the
very
top
layer.
But
here,
as
you
do
asynchronous
processing
yeah,
the
best
you
can
do
is
just
having
the
error
on
the
on
the
on
the
on
the
lowest
I.
Think
here
it's
producer
level
where
it
happens.
F
I
guess
I'll
just
make
one
last
comment
when
it
comes
to
Links
bands.
The
other
concern
we
have
is
there
could
be
hundredths.
Maybe
more
of
links
and
I
know.
There's
a
limit.
I
know
you
can
also
increase
the
limit,
but
I
know
with
batch
processing
that
might
be
a
concern
of
losing
losing
out
and
then
I
guess.
The
other
comment
is
I'm
sure
it's
in
there,
but
if
he
Upstream
decides
not
to
sample
I
assume
the
new
Trace
needs
to
honor
that
sampling
and
make
sure
that
it
generates
the
trace.
F
D
He
has
sampling
in
message
scenarios
and,
in
particular
context
with
links
that
is,
that
is
a
topic
of
also
ongoing
discussions.
I
can
put
the
link
here
in
the
chat
to
an
to
an
to
a
related
issue.
D
But
but
yes
you're
right
when
there
is
like
sampling
here
in
place,
then
the
situation
gets
more
complicated
and
you
are
likely
to
have
actually
broken
are
broken
traces.
D
D
The
heavy
batching
scenarios
with
links
that
you
have
lots
of
links,
that
is,
that
is
also
valid,
a
valid
concern,
and
we
don't.
We
don't
have
currently
yet
any
best
practices
regarding
it,
because
you
can
definitely
have
matching
scenarios
you
can
find,
which
you
can
maybe
have
even
thousands
of
messages
in
a
batch,
and
we
don't
have
worked
out
best
practices
yet
of
how
to
best
kind
of
capture
those
okay,
so
somehow
I
think
it
needs
to
involve
sampling,
but
but
there
is,
there
is
no.
A
A
Yeah
and
then
do
you
think
it's
it's
it's
valid
if
like
to
increase
the
limit
of
the
number
of
spends
in
the.
F
I
personally
think
there
shouldn't
be
a
limit,
but
I
know
that's
crazy
due
to
Heap
space
and
and
whatnot,
because
you
could
set
a
static
limit,
but
that
really
depends
on
to
is
using
nifi
in
this
case,
because
you
could
change
this
processor,
stop
it
enable
or
update
the
properties
to
say
you
know
batch
by
10
000
records,
and
if
you
didn't
change
that
static
variable,
then
you
kind
of
lose
out
on
that
visibility.
So
it's
one
of
those
of
a
need
to
be
something
dynamic
or
something
a
little
smart
based
on
memory
footprint.
A
D
Yes,
and
and
one
last
question
of
your
Prime,
this
instrumentation
that
you
show
here
is
to
intent
to
ship
this
as
part
of
a
petri
knife
file.
Or
is
this
just
like
the
instrumentation
of
one
given
nifi
application.
A
And
just
to
make
sure
I
understood
correctly.
So
when
you
create
the
spins,
so
there
was
one
option
to
create
a
span
for
pulling
the
message
from
the
queue
and
then
processing
the
message
and
then
sending
it
to
the
next
queue.
So
you're
saying
that
you
think
it
will
be
too
busy
and
you
could
sell
just
a
single
span.
That
represents
the
processing
time
without
the
sending
and
the
receiving
like.
F
I,
don't
know
how
beneficial
it
is,
because
this
is
just
a
single
transaction
if
you
will
in
terms
of
unit
of
work,
so
there's
no
timing
on
how
to
receive
from
q
and
whatnot,
because
it's
effectively
sitting
in
here
and
it's
basically
a
pointer
to
a
file.
So
there's
no
worry
about.
You
know
throughput
in
terms
of
setting
data
over
the
wire,
because
everything
is
local
to
a
single
instance.
F
So
for
this
I,
don't
think
it's
timing
per
se,
but
but
also
the
I
think
people
mostly
care
about
how
long
it
takes
for
that
single
processor
to
do
work
and
hard
for
them
to
visualize.
F
They
do
care
about
the
queue
and
we
do
have
a
math
attributes.
I
mean
I
know
this.
One
just
shows
error
attributes
that
we
were
capturing,
but
we
do
capture
the
quote-unquote
cues
that
the
files
are
in
so.
A
F
F
Yeah,
not
necessarily
because,
for
example,
merge
content
could
accept
from
another
queue,
so
there
could
be
any
number
of
cues
coming
in
and
then
this
one
could
pull
from
n
number
of
cues.
So
it
could
be
effectively
an
array.
So
we
kind
of
just
simplify
it
and
say:
oh,
it's
going
out
to
one
queue
and
then
that's
one
body
of
work.
A
Can
it
also
like
do
a
single
processing
and
send
it
to
multiple
queues?
The
the
one
message
that
was
processed
the
one
result
of
the
processing.
F
A
So
you're
processing
it
one
time
you
send
it
to
multiple
queues,
but
then
you
have
multiple
spans
for
the
processing
did
I
understand
correctly.
F
A
So,
on
the
bottom,
they'll,
like
followers,
spend
or
time
they
all
represent
a
single
process,
operation
right.
F
F
So
if
it's
doing
three
files,
what
it
will
do
is
that
body
of
work
for
those
three
files
as
a
single
unit,
but
it
won't
consider
it
complete
until
all
three
files
are
successfully
put
onto
the
queue
effectively
we're
measuring
the
time
it
takes
for
each
record
individually,
but
granted
the
whole
operation
would
be
the
longest
poll
of
of
these.
D
G
D
Would
be
very
grateful
if
you
could
share
your
further
progress
on
this.
That
really
is
helps
us
a
lot
to
see
a
to
see
a
those
use
cases
I
mean
to
summarize
I
think
you're
not
conforming
to
the
semantic
conventions,
because
it's
too
noisy
for
you
to
have
all
the
individual
spans
that
that
are
required
there.
We
will
take
this.