►
From YouTube: 2021-08-17 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
B
Yeah
because
we
went
like
like
first
week
of
june
or
something
I
can't
remember
now
it's
awful
the
last
two
years.
I've
just
been
like
a
blur.
It's
like
all
time
bleeds
together,
but
if
you
look
at
sequential
pictures
of
me,
it's
kind
of
like
I
didn't
have
all
this.
B
We
could
probably
sing
some
songs
or
something
that
are
like
copyright
infringement
or
something
like
that
and
then
it'll.
Maybe
it'll
delete
like
this
whole
section.
Who
knows.
B
I
wonder
if
anybody
ever
watches
I'm
like
I
don't
look
at
the.
I
don't
go
back
to
look
at
the
analytics
of
our
videos
at
all,
because
they're
not
segmented
right,
so
it's
kind
of
it's
not
like
it's
just
it's
just
called
like
open,
telemetry
and
then
the
date
and
I'm
like
where's
the
hours
I
don't
know.
C
Cool
all
right,
so
the
first
thing
is
that
as
people
start
using
the
sampler
interface,
they
find
that
they
want
access
to
more
and
more
stuff.
So
apparently-
and
I
think
I
missed
this-
the
instrumentation
library
was
added
to
this
should
sample
a
method,
and
this
is
so
that
you
can
kind
of
adjust
the
sampling
rate
based
on
different
instrumentation.
So
maybe
you
want
to
give
like
a
higher
priority
to
you
know
things
on
the
boundaries
like
an
inbound,
http
request
or
an
outbound
one,
for
example.
C
C
So
maybe
maybe
I
didn't
miss
anything.
It's
happened
in
the
last
week,
but
yeah
this
is
kind
of
an
additional
one
is,
is
adding
the
resource
so
that
the
sampler
can
make
decisions
now,
based
on
the
resource
data.
D
Go
ahead,
as
I
said,
sounds
good.
We'll
just
need
to
update
our
implementation
if
they
want
us
to
pass
that
stuff
in
right
sounds
fine.
C
Yeah
yeah
and
because
we
haven't
officially
1.0
yet
I
think,
like
these
are
like
one
one
dot,
something
I
don't
know
what
number
actually
is
the
the
spec
is
that
right
now,
I'm
not
exactly
sure
how
to
how
everybody
is
is
tracking
this
stuff.
But
at
some
point
in
time
I
don't
know
it
would
would
seem
like
you
would
want
to
incrementally
go
through
the
numbers
and
add
the
functionality
that
came
in
each
dot.
C
But
I
don't
know
I'm
just
kind
of
guessing
that
that
is
what
we
would
end
up
doing.
C
Well,
yes,
so
I
guess
there
was
just
a
clarification
on
the
spock
around
span
names.
We,
I
think
we
tell
you
what
we
prefer
for
a
fan
name,
what
are
goodspan
names,
but
we
don't
really
tell
you
what
are
bad
names
and
things
that
you
should
avoid,
or
at
least
not
comprehensively.
C
So
I
think
there
is
a
yeah.
They
wanted
to
clarify
that.
C
That
empty
string
is
not
expected.
I
think
it's
what
this
is
supposed
to
say.
Definitely,
a
null
is
probably
not
expected,
and
maybe
there
was
some
talk
about
having
like
a
default
name.
C
E
C
C
Well,
there's
there's
two
parts
about
it,
so
one
is
kind
of
like
the
current
span,
like
what
decision
went
into
it
being
collected
just
so
that
like
when
your
back
end
receives
it,
it
can
kind
of
like
you
know,
if
you
wanted
to
know
like
yeah,
with
the
probability
you
can
see
like
how
many
stances
is
it
kind
of
representative
of
and
then
kind
of
you
know,
compute,
compute
and
extrapolate.
C
You
know
the
the
full
set
of
spans
from
there
and
then
the
other
part
is
propagating
to
the
next
service,
like
the
kind
of
trace
level
adjust
to
count
and
or
the
trace
level
kind
of
sampling,
probability
and
kind
of
sampling
strategy,
so
that
it
can
kind
of
continue
to
sample
with
the
same
same
logic,.
D
Does
the
the
adjusted
represent
the
adjusted
count
represents,
like
number
of
like
spans
versus
unsampled
spans
within
a
you
know?
Is
that
what
adjusted
count
is
this?
I
think
that's
supposed
to
be
better
than
just
maybe
propagating
the
probability
rate
itself.
It's
like
you
propagate
it
and
just
encounter
something,
but
it'll
be
good
to
see
something
implemented
like
I've
been
saying.
C
Yeah,
I
think
there
are
some
yeah
there
are
some.
I
think,
that's
what
a
lot
of
the
discussions
around
is
like
what
is
like
the
more
useful
value
and
like
how
to
actually
represent
it.
I
think,
like
I'm,
probably
over
simplifying
things
by
just
saying
a
probability,
but
I
think
ultimately
like
that
is
the
essence
of
the
thing
that
we're
capturing
there.
I
think
the
the
discussions
are
possibly
around
representing
that
in
some
slightly
different,
possibly
more
usable
ways.
D
Make
sense,
does
anyone
yeah?
Does
your
colleague
think
that
that
will
get
merged
like
that?
Is
he
have
you
spoken
with
josh
about
it.
C
D
C
But
yeah
there
was,
I
guess,
there's
open
pr
for
the
the
metrics,
the
metric
exporter
requirements
for
the
sdk.
There
is.
C
I
guess
pr
for
some
details
for
the
aggregator
and
the
metrics
back.
These
are
all
kind
of
just
like
different
components
of
the
metrics
sdk,
so
it
looks
like
things
are
moving
along
and.
C
C
Issues
with
the
the
getter
implementation,
basically
you're
supposed
to
be
able
to
pass
in
a
header
key
and
get
a
value
and
http
is
a
horrible
thing.
I
mean
it's
not
a
horrible
thing,
but
it
has
like
these
interesting
things
that
happen
that
sometimes
we're
actually
from.
I
think
we're
shielded
from
this
in
ruby.
Most
of
the
time
where,
like
headers,
can
appear
multiple
times,
but
there
is
there's
this
rfc
7230.
C
That
says
they
can
be
combined
into
a
single
header
order
matters,
but
you
just
like
concatenate
them
with
a
comma
and
that's
a
valid
representation
of
a
header,
and
I
feel
like
this
is
what
we
get
in
ruby.
I
feel
like
most
of
the
libraries
give
us
hash
access
and
data,
but
I
think
some
some
languages
and
libraries
will
just
kind
of
give
you
like
an
array
of
like
headers
with
like
key
value,
and
this
this
ends
up
mattering
when
that's
how
you
access
headers,
but.
C
Yeah,
I
think
he
was
initially
proposing
that
you
could
return
a
single
item
if
there
there's
only
a
single
header
and
you
can
return
an
array
of
items
if
there
are
multiple
headers.
I
feel
like
this
would
cause
many
problems
in
a
lot
of
typed
languages,
and
generally
I
don't
know
I
feel,
like
generally,
a
return
type
should
be
consistent.
Hopefully
it
just
makes
things
easier.
C
So
I
at
least
chimed
in
that
just
knowing
what
I
know
about
the
spec,
it
did
have
like
a
a
clause
where
it
said
when
you
receive
multiple
trade
state
headers.
They
must
be
combined
into
a
single
header
according
to
rfc
7230,
and
then
I
kind
of
just
cut
and
paste
it
in
this.
D
Great
explanation
thanks
I'm
I
feel,
like
I
learned
something
about
http
today,.
C
Yeah,
it
seems
like
maybe
tristan
still
thinks,
there's
some
complications
with
us.
So
I
guess
the
discussion
will
be
continued.
C
And
then
there
was
just
a
brief
mention
of
these
new
cigs
spinning
up
and
I
think
really
it's
just
that
they
exist
and
people
should
come
and
kind
of
the
times
for
them.
C
So
yeah,
I
think
we.
C
C
11
30
to
noon
pacific,
it's
a
short
one
and
then
the
instrumentation
semantic
conventions
working
group
is.
C
It
seems,
like
there's
a
lot
of
overlap
between
like
what
what
these
groups
are
doing,
although
I
do
feel
like
some
of
them,
are
meant
to
be
like
a
little
bit
more
targeted,
but
it
seems,
like
I
don't
know,
we'll
see
how
things
play
out,
but
I
feel
like
there's
a
huge
like
potential
for
there
to
be
too
many
cooks
in
the
kitchen
on
certain
things
and
like
different
groups,
dive
kind
of
going
in
different
directions
on
stuff,
and
if
there
isn't
that's
like
three
meanings,
I
feel
like
I
don't
know
I
I
don't
know
what
your
schedules
look
like,
but
I
don't
if
I
pack
three
more
meetings
into
mine,
like
I'm,
probably
going
to
be
in
tears.
C
If
I
could
even
do
it,
I
wish
it's
probably
not
it's
probably
already
a
non-starter
but
anyways
enough
about
my
life.
I
think
I
think,
there's
a
good
chance
that,
like
that
not
everybody's,
going
to
be
going
to
all
those
meetings
and
unless
there
is
at
least
like
one
person
going
to
all
three
that
can
kind
of
like
you
know,
keep
everybody
on
track.
Things
could
get
interesting.
So
so
there's.
A
That
yeah,
I
was
actually
intending
to
start
checking
out
the
instrumentation
sig
says
that
it's
not
me
the
open
calendar.
I
guess
they
repurposed
that
one
calendar
event.
It
was
like
the
pacific
time
spec,
so
I'm
gonna
jump
in
there
and
I
think
I
might
start
lurking
there
just
to
see
what's
going
on
what
kind
of
things
are
being
discussed-
and
maybe
I
can
bring
stuff
back
here
if
they're
applicable.
C
Yeah,
that
would
be
awesome.
Actually
it
would
be
great
if
anybody
wants
to
show
up
to
to
any
of
these,
but
I
feel
like
the
instrumentation's
like
seems
like
a
pretty
high
value
one
and
it
I
think
it
kind
of
has
been
going
on
like
secretly
under
under
the
name
of
the
pacific
time
zone
spec
meeting,
so
they
are
probably
just
a
little
bit
more
active.
C
The
topics,
which
is
the
reason
why
I
think,
there's
like
a
messaging.
Why,
specifically,
they
call
out
messaging,
I
think
their
goal
is
to
have
some,
like
you
know,
very
distinguished
people
in
the
messaging
realm
kind
of
come
in
and
give
their
guidance
on
on
things.
C
Which
might
be
why
that
one
was
a
little
bit
shorter,
as
well
kind
of
yeah
just
trying
to
get
some
cooperation
between
people
who
are
kind
of
like
domain
experts.
I
guess
in
some
of
these
areas
and
then
kind
of
like
people
that
are
kind
of
hotel
experts
and
try
to
meet
in
the
middle
to
come
up
with
a
sensible,
sensible
strategy.
C
They
want
to
discuss
about
that
or
any
next
topic
that
we
should
move
on
to.
A
Maybe
for
like
a
next
topic,
I
did
get
the
rc3
out,
despite
the
fact
that
it
would
tell
you
that
it
failed.
I
was
having
some
issues
with
ci
we're,
depending
on
the
toys
gem
that
I
guess
daniela
zuma
has,
and
he
did
a
release
recently
that
broke
the
release
process.
So
I
just
I
pinned
the
retry
release
to
the
previous
version
to
see
if
that
would
get
it
working,
it
does
work,
but
there's
still
some
issues
around
publishing
docs.
A
So
if
you
actually
go
to
the
issues,
I
think
I
opened
one
or
two
one
of
them
is
a
little
bit
targeted
at
daniel,
and
I
know
he
hasn't
been
around
a
little
while
see
if
he'll
come
back,
but
the
failing
to
push
docks
to
github
the
third
from
the
top
there.
That's
that's
just
describing
what's
actually
failed,
because
the
failure
release
issue
that
it
auto
opened
is
just
a
bunch
of
these,
so
it
supposed
to
do
with
like
branch
protection.
A
I
don't
know
what
changed
recently
or
why
this
is
suddenly
an
issue
myself.
I
was
working
on
the
release.
Pr
and
I
couldn't
push
to
the
branch.
So
I
had
to
like
un
check
a
sneaky
button
somewhere
in
our
settings
to
get
it
to
go,
but
I
don't
know
if
we
have
anyone
from
github
github
who's
like
super
familiar
with,
like
the
configuration
of
refo,
because
I
haven't
played
with
any
of
this
stuff
actually
in
a
very
long
time.
So
I
don't
know
why
it
would
have
just
changed
on
us.
B
I
could
certainly
ask
about
that,
but
I
don't
have
admin
access.
I
don't
think
to
the
repository
right,
so
I
can't
check
any
of
these
settings.
We'd
have
to
probably
pair
up
on
it.
A
I
just
changed
the
settings,
so
we
could
like
override
that
and
just
push
anyways,
which
might
not
be
the
most
desirable
thing
for
now,
but
ci
doesn't
have
that
same
superpower,
I
suppose,
and
so
when
it
tried
to
push
up
to
branch,
it
failed,
but
the
gems
did
all
get
released,
which
is
nice
so
like
the
the
rc3
and
all
the
the
changes
that
have
been
in
since
the
the
last
release
have
all
been
pushed
out.
So
they're
all
available
at
rubygems.
A
Yeah,
so
it's
like
it's
not
in
like
a
super
great
state
but
like
we
are
still
pushing
out
gems,
which
I
think
means-
and
that's
kind
of
wanted
to
lead
me
to
the
next
thing-
is
what's
the
best
way
to
flag
carlos
to
take
another.
Look
at
our.
G
We
were
going
to
reach
out
to
him
once
we
were
ready
for
another
look,
so
we're
ready
so
yeah,
just
ping.
It.
A
A
A
A
That
was
my
latest
like
I
went
through
that
yesterday
and
it
was
a
bit
frustrating
there's
another
issue
that
I
pinged
daniel
on
just
regarding
that
toys
gem
so
see
what
he
wants
to
do
there
and
then
I
think
francis
dropped
something
in
our
chat
here
to
talk
about.
But
I
don't
know
if
anyone
has
any
comments
on
this
before
we
move
on.
C
I
saw
two
issues
coming
in
my
only
final
comment
on
this
was
just
yeah.
It
seems
like
definitely
something
changed
like
I
would.
I
think
daniel's
input
would
be
good,
like
I
feel
like
from
time
to
time,
like
the
the
hotel
community
makes
some
changes
to
kind
of
like
all
the
repos,
and
I
think,
like
asking
around
to
see
if
anything
like
that
has
happened,
might
might
help
us
out.
C
C
G
Okay,
so
yeah,
so
it
looks
like
this
went
in
last
week
and
then
there's
another
open
issue
for
adding
resource
to
the
samplers
interface.
G
So
yeah
they
split
the
two,
but
the
the
issue
is
the
same.
For
me,
the
proposal
seems
to
be
that
these
should
be
added
to
the
sampler
interface,
so
basically,
instrumentation
library
and
resource
should
be
passed
into
the
should
sample
method.
I
can't
remember
offhand
what
we
call
it.
We
call
it.
Oh
we
we
just
have
yeah.
Actually,
sorry,
we
have
a
required
interface,
which
is
should
sample
with
question
mark
okay,
so
these
are
proposed
as
optional
parameters.
G
C
So
if
they
are
optional,
you
should
be
able
to
just
omit
them
and
it
would
be
okay,
so.
G
C
G
G
G
Keyword
argument
right,
that's,
I
thought
it
would
be
something
like
that.
So
how
will
we
support
this
kind
of
thing?
I
mean
it's
going
to
it.
I
guess
it's
going
to
be
a
breaking
change
right.
They
they
suggested
like
that,
the
spec
or
these
issue,
so
these
pr's
were
suggesting
you
know
you
could
add
them
as
optional
parameters,
and
you
could
do
something
where
there
was.
G
You
know
this
object
that
wrapped
up
all
the
possible
parameters
to
sampler
and
that,
like
that,
would
be
passed
into
should
sample,
but
that's
not
the
way
we
expect
our
our
interface
for
this
so
like
pushing
this
in
is
going
to
be
a
breaking
change.
G
B
E
B
D
D
G
D
G
I
mean,
I
don't
think
we
want
this
breakage
twice,
so
you
know
they've
split
the
prs
up
into
instrumentation
library
and
resource
as
the
two
new
parameters.
So
I
mean
we
should
wait
for
the
second
one
to
land
and
there
is
some
objection
to
it
at
the
moment.
So
we
should
wait
for
this
all
to
be
sorted
out
and
we
should
do
this
post
1.0,
but
it
will
be
a
a
breaking
change.
C
So
I
will
say
that
I
do
believe
that
every
other
language
will
be
a
breaking
change.
We
we
can
use
ruby
and
just
place
it
in
the
method
class.
C
So
if
you
can
get
a
handle
to
a
method
to
the
sample
method
of
your
sampler,
like
you
can
kind
of
like
do
some
clever
runtime
introspection
of
the
thing
and
figure
out
if
it's
going
to
work
or
not,
that
would
be
an
escape
hash
for
us,
but
I
feel
like
nobody
else.
Has
this,
and
maybe,
if
we
got
into
this
we'd
find
it
would
not
work.
C
G
C
C
But
yeah,
like
that
aside,
I
think
ruby
gives
gives
us
a
lot
of
this
stuff,
like
I
think
other
languages
are
totally
they're
going
to
have
to
do
something
different.
Okay,.
G
No,
I
think,
we're.
I
think
the
suggestion
is
that
we'd
do
the
introspection
and
based
on
whether
it
supports
the
extra
args
or
not
we'd,
call
the
method
differently.
G
D
We
have
it's
like
with
our
exporters.
I
think
we
have
like
a
validate
exporter
or
validate
something.
We
have
some
sort
of
like
helpers
like
that,
so
yeah
we
could
always
do.
I
actually
been
poking
around
around
there,
because
I
was
looking
at
the
metrics
reporter
but
yeah
it's
I
feel
like
it's
bad
exp.
We
should
people
will.
If
I'm
on,
I
mean
I
guess
the
question
is
like
in
practice.
D
G
C
Yeah,
I
feel
like
I've,
I've
seen
some
of
this
stuff,
like
in
rail
source
code,
like
I
can't
give
you
an
example,
but
just
in
the
past,
just
like
reading
through,
where
between
like
versions
where,
like
arguments,
have
changed,
that
they
are
doing
some
clever,
clever
introspection
to
kind
of
like
make
things
work
and
then
logging
like
a
deprecation
warning
that
hey
things
are
not
going
to
work
like
this
soon.
So
I
think
they're,
there's
precedent
for
this
sort
of
thing.
B
I'm
just
wondering
who's
writing
custom
samplers
at
this
stage
and
how
many
of
those
implementations
out
there
and
who's
going
to
be
impacted
by
it.
Maybe
that's
a
reason.
B
A
We've
got
one
coming
up
soon,
like
we're
going
to
be
doing
a
custom
sampler,
probably
in
the
next
few
weeks.
We
haven't
done
it
yet,
but
it's
it's
coming.
D
G
D
B
That
one
took
the
took
the
words
out
of
my
mouth
matt
and
for
what
it's
worth.
F
F
Do
you
find
that
you
like
sampling
in
process,
rather
than
sending
to
collectors
in
the
collectors
sample.
G
For
us
right
now,
we're
mostly
from
for
most
things,
we're
not
sampling,
so
we're
we're
collecting
everything.
We
have
a
couple
of
applications
where
we
are
doing
head
sampling.
We
are
doing
some
a
little
bit
of
kind
of
uniform
tail
sampling,
but
we
actually
want
to
move
that
back
to
head
sampling.
It's
probably
going
to
be
easier
for
us
to
easier
for
us
to
manage
it's
going
to
provide
a
better
experience
for
people,
because
they'll
have
anyway
there'll
just
be
fewer
complaints.
G
So
that's
that's
a
irrelevant
thing
for
us
yeah.
We
we
want
to
do
a
little
bit
of
headsampling
for
sure.
G
In
process
yeah
yeah,
okay,
yeah,
one
of
the
things
that
was
lost
in
the
move
from
open
census
and
jaeger
has
this
as
well.
But
the
move
from
open
senses
to
open
telemetry
was
the
ability
to
push
sampling
configuration
from
the
collector
into
the
application
process,
and
that
would
be
really
nice.
So
we've
had
to
work
around
that
issue.
We've
had
to
come
up
with
our
own
solution
for
that
for
our
monolith
like
our
largest
application.
G
D
G
We
we
have
something
called.
G
So
jaeger
had
an
ability
or
has
an
ability
to
pass
configuration
at
runtime
to
applications
and
opencensus
had
the
same
capability.
C
I've
definitely
heard
people
asking
about
things
like
being
able
to
like
push
push
configuration
to
collectors,
and
so
I
think
I
think
it
is
a
need.
I
will
try
to
like
ask
around
and
see
if
it's
like,
like.
I
get
the
impression
that
it's
like
it
is
a
thing
that
people
are
talking
about,
but
it's
kind
of
like
future
a
little
bit.
C
G
Yeah,
it's
evolved
a
lot
since
then,
but
it's
more
that
the
protocol
has
changed
so
the
open
census
protocol
supported
this
otlp
does
not.
C
G
And
funnily
enough,
like
the
open
census,
collector,
never
actually
supported
this.
It's
just
a
protocol
had
support
for
it,
whereas
jaeger,
a
collector
does
actually
have
support
for
this,
and
some
folks
are,
I,
I
think
I've
seen
reference
to
some
folks
using
jaeger
in
production
specifically,
so
they
have
this
capability.
F
Yeah
I
was
asking
because
we
have
twinkles
in
our
eyes
about
trying
to
get
making
the
the
collector
sampler,
smarter
and
don't
know
how
far
we
ought
to
go
with
that.
G
I
think
there's
a
bunch
of
people
who
want
to
do
that
kind
of
thing.
Right,
josh
mcdonald
has
some
work
happening
at
lightstep
for
this.
I
think
you
know
there's
that
varrot
thing
that
he's
working
on
they're
also
wanting
to
use
are
they
proposing
using
some
of
that
for
the
exemplar
work
as
well?
And
you
know,
grafana
did
some
tail
sampling
stuff
as
well,
so
yeah,
there's
there's
enough
interest
in
the
community
for
some
kind
of
tell
by
sampling
or
something
smarter
in
the
collector.
F
G
Yeah
yeah
yeah,
totally
yeah,
but
yeah
we're
moving
our
stuff
to
regional
collector
clusters
instead
of
having
clusters
living
in
so
instead
of
having
collectors
living
in
the
application
clusters,
partly
to
allow
us
to
explore
some
of
this
stuff.
G
So
kind
of
yeah,
okay,
yeah,
it's
messy
running
collectors
across
you
know
40
50,
whatever
eventually
hundreds
of
application
clusters
yeah
it's
much
nicer.
If
you
just
say
you
know
what
we've
got
like.
Our
team
supports
this
cluster
of
collectors
in
each
region
and
we
control
everything
in
that
cluster.
F
G
Stand
alones,
so
I
mean
the
way
open.
Telemetry
proposes
as
the
kind
of
default
deployment
is
that
you
have
the
collector
running
in
agent
mode
as
they
refer
to
it
as
a
daemon
set,
and
then
you
also
would
have
a
pool
of
collectors
running
in
a
cluster
and
that
would
be
auto
scaled
or
something
we're
instead
running.
We've
gotten
away
from
the
demon
set
a
long
time
ago,
because
managing
those
team
and
sets
is
just
so
painful
where
we're
actually
moving
the
collectors
out
of
the
cluster
application
clusters
as
well.
G
So
instead
they're
just
in
a
regional
cluster,
and
we
have
envoy,
you
know
doing
the
routing
to
the
the
collectors
and
failover
between
regions.
If
you
know
a
collector
cluster
goes
down.
G
Cool,
it
just
means
that
you
have
to
do
some
of
it,
like
you
can't
do
like
resource
augmentation
like
adding
infrastructure
tags
to
resources.
D
G
Something
yeah
you
can't.
You
can't
do
that
because
you
don't
have
an
agent
or
a
collector
in
the
same
cluster,
which
means
the
applications
themselves
need
to
get
all
that
infrastructure
information
and
add
it
to
their
resource.
B
This
sounds
like
a
great
topic
for
open,
telemetry
tuesdays,
and
I
think
that
we
should
probably
connect
ted
with
y'all
to
talk
about
that
experience,
because
that
feedback
is
very
helpful
like
we
haven't,
deployed
the
collectors
yet
and
it's
becoming
more
and
more
of
a
need
right.
B
It's
showing
itself
as
a
need
one
for
attribute
filtering,
but
you
know
managing
different
configurations
for
different
applications
already
making
me
sick
and
then
getting
resource
allocation
specifically
for
these
right,
because
one
of
the
biggest
concerns
that
I
had,
I
was
trying
to
like
offload
some
of
the
processing
and
throwing
either
daemon
sets
into
the
mix
or
host
agents
and
taking
cpu
and
memory
allocation
away
from
my
applications
is
not
like.
I
I'm
not
gaining
anything
by
moving
the
client,
the
sdk
processing
into
the
collector
when
they're
both
sharing
this
compute.
G
G
Really
yeah,
our
original
architecture
had
a
very
simple
custom
agent,
if
you
like,
running
as
a
demon
set,
and
that
was
very
lightweight
added
some
data
but
forwarded
things
on,
so
it
really
just
added
information
to
the
headers
and
then
the
collector
downstream
would
handle
actually
adding
it
to
the
the
resource
bundles.
G
When
we
moved
to
the
open,
telemetry
collector
and
running
as
an
agent,
I
think
our
memory
requirements
for
that
went
up.
4X
and
cpu
went
up
4x
and
we
were
still
hitting
cpu
limits
with
it,
because
the
collector
needed
to
deserialize
everything
and
then
re-serialize
it
again
to
forward
it.
So
that
was
a
huge
pain
point
for
us
and
on
our
largest
application.
G
You
know
the
agents
were
just
getting
swamped,
they
couldn't
keep
up
and
we
were
getting
back
pressure
into
the
application
and
it
was
just
really
painful.
So
we
had
to
move
away
from
that.
Pretty.
B
Early,
I
think
I
had
a
colleague
once
I
refer
to
that
as
the
iterator
sandwich,
whereas,
like
I've
got
an
iterator
and
I
iterate
over
a
collection
and
then
I
send
it
over
to
transform
it.
I
iterate
over
it
again
and
then
the
other
thing
takes
and
iterates
it,
and
it's
like
damn
we're
just
trying
to
pass
bytes
over
the
wire.
What.
D
B
C
C
I
think,
like
the
more
awareness
that
people
have
of
open,
telemetry
ruby,
even
in
the
open
telemetry
community,
like
I
think
it's
good
and
I
think
like
having
having
those
discussions
you
know,
would
would
be
valuable
for
others
and
kind
of
puts
on
the
map
to
something
mean
yeah.
G
Yeah
I'd
be
interested
in
doing
that.
Tuesdays
do
tend
to
be
a
little
horrible
for
me,
but
yeah
it
would
be.
It
would
be
interesting.
I
think.
C
Cool
I'll
I'll
make
a
connection
and
scheduling
can
be
figured
out.
C
Cool,
we
are
down
to
the
final
five
anything
that
we
should
cover.
B
Not
really
anything
just
just
some
personal
things
as
the
school
year
kicks
in,
I
hope
to
have
more
capacity.
But
unfortunately,
from
my
organization
we
have
diminished
capacity,
and
so
I'm
the
only
person
left
working
on
hotel
at
all
and
I'm
doing
mostly
application
side
stuff.
So
that's
why
my
contributions
are
fewer
and
fewer
to
the
project
itself,
but
also
going
to
be
transitioning
a
little
bit
of
my
focus
to
golang
pretty
soon.
B
So
I'm
still
going
to
attend
these
meetings,
but
I'm
even
less
than
what
I've
been
giving.
You
might
see
a
fraction
of
that
as
well.
So
that's
kind
of
what
wanted
to
give
you
all
a
heads
up.
If
you
don't
see
me
or
if
I
disappear
or
whatever.
A
That's
fine.
I
hope
you
don't
disappear,
but
even
if
you're
sharing
issues
you're
running
into
with
the
work
you're
doing
at
the
application
level
like
stuff
you're
running
into
problems,
even
if
you're
just
surfacing
problems
just
like
that's
super
valuable,
so
don't
don't
don't
diminish
the
value
of
that.
Just
like
letting
us
know,
there's
a
problem
is
actually
really
really
helpful
because
realistically
we're
kind
of
in
the
same
boat
of
our
tech
stack.
So
if
you
find
something
before
us,
that's
great.
B
Yeah
there
there's
some
experimentation
going
on
with
ruby3,
so
you
know
I
haven't
been
paying
close
attention
to,
but
I'm
going
to
try
to
get
the
rc3
released
next
week
and
try
to
get
that
up
on
at
least
some
pre-production
environments
so
that
I
can
you
know,
report
back
any
issues
that
we
might
be
running
into,
but
outside
of
that,
that's
about
all
the
capacity
I'll
have
to
be
able
to
give.
F
Also
hoping
to
spend
more
time,
I
know
I
missed
the
past
couple
meetings,
but
we're
we're
concentrating
on
sun
setting
some
of
our
multitude
of
open
source
libraries
so
that
they
become
less
of
a
maintenance
burden
and
take
up
and
don't
eat
up
so
much
of
our
time
and
as
that
frees
up,
I'm
hoping
to
be
able
to
pay
more
attention
and
contribute
to
hotel
ruby.
F
G
A
Regarding
the
rc3,
we're
actually
releasing
like
a
new
version
of
our
rapper
gem
today,
so
we'll
be
pushing
people
over
to
the
latest
release
this
week
as
early
as
this
week.
So
hopefully
we
can
run
the
bugs
before
ariel
does.
A
Hopefully
I'm
I'm
a
little
excited
and
a
little
bit
nervous
about
all
the
new
rails.
Instrumentation
like
action
view
and
active
record
is
now
the
default
there.
So
our
whole
fleet
that
start
getting
their
dependable
bumps
going
through
we'll
start
seeing
it.
So
I'm
excited.
F
I'm
excited
for
that
too,
because
one
of
the
major
some
of
our
customers,
who
are
using
our
ruby
libraries,
the
main
difference
between
it
and
hotel,
ruby,
are
the
instrumentations
and
our
instrumentation
was
my
naive
question.
I
opened
months
ago
that
that
you
corrected
me
on
robert
with
the
whole,
like
active
support,
notifications
and
sort
of
a.
F
Dark
art
and
can
blow
up
on
you
like
we
did
the
naive
thing
of
just.
We
probably
we
subscribe
to
that
and
hope
that
there
aren't
any
subscribers
that
are
going
to
throw
exceptions,
but
lots
of
people
like
just
like
providing
an
array
of
strings
of
names
of
events,
and
suddenly
things
are
lit
up.
So
I
foresee
trying
to
help
out
with
adding
more
of
those
instrumentations
that
are
wraps
of
the
as
notifications.
A
A
But
I
I
do
think
like
ultimately
supporting
that
that
traditional
use
case
that
people
expect
is
something
that's
going
to
be
happening
soon,
like
it's
just
a
matter
of
taking
the
time
to
extract
out
the
work
andrew
did
with
the
active
support
notifications
for
action
view
and
just
generalizing
a
little
bit
and
making
it
so
it's
accessible
to
everybody.
Well,.
D
So
well,
we
have
a
minute.
I
was
wondering
if
I
was
meaning
to
file
an
issue.
Something
I
wanted
to
kind
of
push
up
is
working
on
like
a
troubleshooting
section
of
the
docs
for
just
like
common
issues,
but
also
one
thing:
I've
noticed
is
like
our
sort
of
like
diagnostic
or
debug
level.
Logging
is
maybe
a
little
light
on
like
a
lot
of
it
comes
through
those
stats.
You
know
a
lot
of.
D
But
realistically
speaking
in
the
case
where,
like
you
know,
there's
like
an
export
error,
it's
also
maybe
possible
that,
like
these
stats,
the
metric
or
whatever
will
not
get
committed
or
get
drop.
You
know
whatever
we'll
get
lost
on
dvd
the
magic
up.
I'll
see
that
so
I
wasn't
sure
I
was
thinking
about
maybe
trying
to
add
like
a
back
door
to
say.
Like
just
log,
you
know,
here's
all
your
drop
span
ids
or
something
like
that.
D
I
wasn't
sure
if
that's
something
there's
trade-offs
to
it
and
that,
like
you
can
blow
up,
you
know
you
can
start
logging
extremely
aggressively
or
whatever,
and
maybe
there's
workarounds
around
that
like
you
can
throttle
your
locks.
But
do
you
think
that's
something
that
would
be
useful
if
I
was
working,
I
think.
A
Right,
so
we
that
actually
that's
come
up
a
couple
times
and
just
last
week
someone
kind
of
was
asking
about
it.
So
details
aside,
one
of
our
environments
goes
to
standard
out.
So
if
someone
like
goes
into
a
console,
they
just
see
like
export
failure
drop.
These
fans
drop
these
fans
and
they
just
see
this
like
blast
of
noise,
because
we
we
do
log
that
right,
but
it
doesn't
say
why
it
says:
export
failure.
A
So
one
of
the
things
I
think
tim
wanted
to
work
on
either
this
week
or
probably
this
week
is
adding
some
visibility
around.
Why?
Because
we
have
some
granularity
around
spans
being
dropped,
but
then
there
is
that
kind
of.
A
I
don't
want
to
call
it
a
catch-all,
but
it's
like
there's
that
else
condition
that
just
says
export
failure.
It
doesn't
really
tell
you
why
it
failed,
like
you
can
dig
into
the
code
and
you
can
see
that
it
failed
in
this
section
of
code,
but
it's
not
clear
which
one
of
those
like
five
situations
actually
occurred.
D
So
yes,
I
agree,
I
think
it's
good
work
to
do.
Maybe
there
was
already
an
issue
and
I
just
didn't
read
it.
No,
I'm
sorry.
If
that's
no.
A
No
there's
no
there's
no
issue
for
it.
This.
This
is
more
recent
that
we've
been
getting
a
little
bit
of
feedback
internally
so
and
I
think
it's
good
feedback,
because
an
application
owner
is
like
how
do
I
dig
into
this,
and
I
was
like
they
looked
at
the
code
and
they're
like.
I
have
no
idea
what's
going
on
here.
So
that's
something
we
need
to
address.
A
I
don't
know
the
exact
approach
that's
going
to
be
taken
yet,
but
I
think
tim
is
going
to
take
it
on
and
then
we
could
probably
talk
about
it.
Then
yeah.
D
Yeah,
that's
good.
I
can
split
out.
I
have
so
like
kind
of
a
messy
power
with
a
bunch
of
stuff
in
there,
so
I'll
split
out
some
of
the
troubleshooting
stuff
with
the
diagnostic
or
debug
vlogging
stuff,
and
maybe
try
to
your
stuff
work.
Awesome.
That's
glad
we're
on
the
same
page.
Anyway.
Sorry
I
forced
everyone
to
run
two
minutes
over
with
that.
F
Some
of
the
ways
that
we
do
that
is
emitting
both
a
log
entry
with
like
span
lost
with
a
with
a
given
reason,
and
then
in
some
of
our
code
pads.
We
also
have
a
no
op
stats
d
implementation,
where
you
could
provide
a
stats
to
implementation.
If
you
wanted
it
and
and
the
same.
F
D
I
can
you're
right
then
robert
added
something
sort
of
a
little
lesson,
not
even
stats
to
you
necessarily,
but
some
a
client
like
that.
Yeah.
F
I
don't
think
whether
we'd
want
to
do
it
in
hotel.
Ruby
is
a
separate
question.
A
D
G
Right,
I
think
what
we
want
to
do
is
kind
of
define
the
interfaces
for
that
kind
of
thing
and
maybe
provide
a
gem
of
you
know
a
few
different
ways
of
doing
this,
so
that
you
know
people
can
just
grab
one
of
those
and
plug
it
in
it's
like.