►
From YouTube: What's New with Fluentd & Fluent Bit
Description
Fluentd and Fluent Bit are two highly popular open source projects for data collection and log forwarding in the realm of observability. Fluentd's flexibility and scalability have led to seamless integration with diverse applications and data sources, while Fluent Bit's lightweight and efficient log forwarding have made it a preferred choice for modern observability pipelines. But Fluent Bit can process more than just logs. The recent release of Fluent Bit v2 added major new integrations with OpenTelemetry, Prometheus, WebAssembly plugins and more.
On this episode I hosted Eduardo Silva Pereira, one of Fluentd project maintainers and creator of Fluent Bit. He also is the founder of Calyptia, the Fluent company. Eduardo shared with us the latest updates of Fluentd and Fluent Bit, as well as valuable insights into the future roadmap of these projects.
You can read the recap post: https://logz.io/blog/whats-new-with-fluentd-fluentbit/?utm_source=devrel&utm_medium=devrel
Resources
=========
Fluent Bit: https://fluentbit.io/
FluentD: https://www.fluentd.org/
Prometheus supports OTLP: https://horovits.medium.com/83f85878e46a
PromCon '23: https://promcon.io/2023-berlin/
Kubernetes 1.27: Query node logs using the kubelet API: https://logz.io/blog/a-practical-guide-to-kubernetes-logging/#Kubernetes_1_27
Kelemetry project: https://www.cncf.io/blog/2023/07/27/kelemetry-global-tracing-for-kubernetes-control-plane/
Chapters
========
00:00 - episode intro
02:11 - Fluentd and Fluent Bit background
09:51 - Should I choose FluentD or Fluent Bit?
13:26 - developing an active engaged OSS community
17:18 - enterprise needs and building commercial offering with Calyptia
19:54 - Fluent Bit v2 updates
29:22 - plugins, filters and processors in Fluent Bit
38:23 - A sneak peak into the planned announcements for KubeCon Chicago
44:16 - where to follow the community and Eduardo
47:43 - Prometheus now supports OTLP
48:57 - PromCon will take place in Berlin, 28-29 Sept.
50:11 - OpenTelemetry Semantic Conventions is separated from the Specification
53:38 - New in Kubernetes 1.27: Query node logs using the kubelet API
54:43 - Kelemetry: global tracing for Kubernetes control plane
57:45 - episode outro
Dotan Horovits
============
Twitter: https://twitter.com/horovits
LinkedIn: https://www.linkedin.com/in/horovits/
Mastodon: @horovits@fosstodon.org
Eduardo Silva Pereira
==================
Twitter: https://twitter.com/edsiper
LinkedIn: https://www.linkedin.com/in/edsiper
A
A
Just
a
few
months
ago,
I
saw
the
announcement
of
fluent
bid
version,
2
a
major
release
with
lots
of
new
stuff
and
later
I,
caught
up
with
Eduardo
and
heard
some
cool
updates
and
decided
it's
way
overdue.
Let's
put
it
this
way
to
get
an
update
about
it
and
share
it
with
you
with
our
audience.
Eduardo
Silva
is
one
of
fluent
D's
project
maintainers
and
a
creator
of
fluent
bit
as
well
as
a
co-founder
of
caliptia
and
is
perfectly
the
right
person
to
take
us
through
it.
A
A
Yeah
pleasure
having
you
actually,
we
had
your
fellow
founder
anuragupta
on
the
show.
I
think
it
was
over
a
year
ago,
maybe
closer
to
two
years
now,
but
so
much
has
changed
that
it's
definitely
a
way
over
with
you
to
sit
down
again
and
and
catch
up
so
great
to
have
you
here
on
the
show.
Thank
you.
Thank
you
happy
to
be
here
yeah,
and
you
know,
let's
start
with
maybe
with
with
fluent
D.
A
You
know
the
the
veteran
brother
and
a
graduated
project
of
the
cncf
being
a
cncf
Ambassador.
It's
also
surprised
that
these
projects
are
cncf
projects.
It's
a
graduated
project
over
12
000
stars
and
more
than
1300
Forks
on
GitHub
I
checked
before
the
show
say
someone
use
it
I
know.
Even
in
my
company,
at
logs
IO,
we
we
use
that
many
users
that
I
that
I
encounter
that
I
work
with
so
it's
definitely
highly
popular.
A
B
One
of
the
challenges
when
you
start
deploying
applications,
you
know
you
make
sure
that
your
application
is
kind
of
working,
but
then
your
journey
doesn't
stop
right
there.
After
that,
you
might
want
to
analyze
the
data
being
generated
by
your
applications
can
be
your
login
information
errors,
warnings
and
sometimes
you
know,
be
able
to
take
actions
based
on
this
analysis,
but
when
you,
if
you
will
go
back
to
2014-15
with
all
the
height
of
a
distributed
systems,
kubernetes
and
Docker
was
very
popular
at
that
moment,
and
people
start
decoupling
the
applications
right.
B
This
information
started
to
go
multiple
places.
Also
before
this
journey,
we
got
similar
problem
where
we
got
different,
for
example,
Hardware
devices
like
firewalls
or
applications
in
Java
applications
in
PHP
or
any
kind
of
fancy
language.
At
that
time,
everybody
was
generating
logs
in
a
different
format
through
a
different
type
of
transport.
Some
of
them
were
using.
The
network.
B
Udp
like
system
is
pretty
common,
other
the
file
system,
and
if
your
final
goal
is
to
analyze
data
right,
that's
what
you
want
to
do.
The
next
question
is
how
you
might
realize:
oh
I
need
to
take
this
data
and
centralize
it
in
some
place.
So
I
can
write
my
own
queries,
and
this
is
a
challenge.
How
do
I
deal
with
data
from
multiple
sources
from
multiple
destinations
and
then
moving
forward
at
scale
right
and
fluent?
B
D
was
a
prick
that
was
born
in
2011
at
Treasure
data
at
that
moment,
because
they
they
used
to
have
like
a
Hadoop
as
a
service,
and
they
wanted
to
make
sure
that
customers
could
ingest
data
into
the
platform
as
any
vendor
right
as
much
data
you
just
you
know
you
get.
We
get
a
bigger
bill
at
the
end
of
the
month
and
everybody's
happy,
but
at
least.
B
And-
and
one
of
the
challenge
was
hey
how
to
move
all
this.
This
data
that
is
existing
different
place,
data
centers,
on-prem
well
previous
Cloud,
to
a
place
for
analysis
and
fluency
was
one
of
the
first
open
source
agents
that
was
around.
That
was
able
to
talk
to
a
different
formats,
collect
data
in
different
ways,
but
also
provide
capabilities
to
process
the
data
before
the
data
hits.
B
The
Final
Destination
I
think
that
one
of
the
smart
decisions
that
was
by
made
on
by
Sarah
the
creator
of
fluency
was
like
hey
I'm
gonna,
make
it
with
a
plugable
architecture.
So
right
now
it's
from
Treasure
data,
but
then
anybody
else
can
write
their
own
plugins
to
collect
data
from
different
sources
or
different
type
of
filters
or
output
destinations.
B
Today
there
are
more
than
a
thousand
plugins
for
fluency
available
Within
by
the
community.
Right.
If
you
go
to
the
GitHub
repository
fluency,
then
not
many
changes,
it's
quite
small,
but
if
you
go
to
their
ruby
gems,
a
website
and
you
will
find
the
fluent
plugins,
you
know
gems,
available
yeah,
thousands
and
and
fluently
solved
that
problem
and
well.
The
technology
was
was
chosen
at
that
time
was
a
Ruby.
B
Actually
one
of
the
stories
that
I
just
learned
recently
is,
like
everybody
say:
oh,
what
is
reading
Ruby
Ruby
is
not
that
fast
for
a
system
level
application
service,
but
actually
Ruby
was
choosing
to
write
the
POC
of
fluency,
because
the
final
goal
was
to
write
it
in
C
plus
plus
at
that
time,
but
it
worked
out.
Well,
it
started
working
people
started
running
plugins
and
okay,
it
continued
as
a
working
out
and
then
when
I
joined
the
company
we
got
this
Mission
a.
B
B
Right
and
the
same
problem
that
you
you,
if
you
abstract
from
the
the
staff
of
deploying
applications
and
getting
data
out
as
information
of
what's
Happening,
will
happen
with
devices
too,
because
devices
will
freely
run
software
right.
It's
like
your
car.
Your
card
is
full
of
software
with
sensors
and
you
construct
that
information
and
fluently
was
not
suitable
for
that.
B
It
Was,
Written,
rubies
a
bit
heavy
right
and
you
had
to
download
I,
don't
know
a
couple
of
dozens
of
gems
to
make
it
work.
So
I
started
one
bit
so
film
bit
was
born
to
solve
this
problem
of
iot
at
that
time.
But
at
the
same
time
you
might
recall
that
a
kubernetes
was
you
know,
was
born
in
2014.
The
cncf
was
created
and
every
is
it.
B
This
hype
started
around
the
stupidest
systems,
a
kubernetes,
but
if
the
same
story
happens
again,
you
get
a
new
platform,
you
you
deploy
your
applications
and
then
how
do
you
analyze?
Your
data
and
fluently,
was
the
first
choice
we
donated
the
project
and
with
flu
and
bet
on
it
right
because
they
were
kind
of
same
family
projects.
B
A
B
A
B
Oh
sorry,
no
I
mean
to
say
that
fluency
was
created
to
work
in
normal
Linux
boxes
like
in
VMS
bird
metal
machine
flu
embed
was
decided
to
work
on
embedded
Linux,
but
then
we
pivot
the
project
for
cloud.
We
created
kubernetes
filter
processors
same
as
fluency,
so
we
started
to
bring
this
parity
right
that
existed
to
for,
to
create
data
pipelines
mostly.
A
So
today,
if
you
depict
it
today
so
before
it
was
very
clear
to
say,
okay,
you
recommended
fluently
for
for
a
classic,
and
this
is
for
iot
today.
How
would
you
explain
the
these
projects,
the
differentiation,
maybe
the
best
fit
when
new
users
that
might
be
a
bit
confused?
What
would
be
your
guidelines
in
terms
of
of
these
two
projects.
B
Yeah
we
have
a
confusion
around
the
marketing
of
the
project
and
organization,
because,
if
you
go
to
GitHub,
you
will
see
that
the
organization
is
not
called
flu
and
D
actually
is
fluent
right
and
fluency
was
the
prey
because,
oh
it
was
the
demon
said
and
I
think
that
fluent.com
was
taken
so
hey.
Let's
call
the
project
fluency
and
then
fluent
was
born
under
the
fluent
organization
right,
so
yeah
fluency
keeps
being
the
the
parent
project
is
what
people
knows
and
and
fluent
build
has
been
always
like
the
small
brother.
B
In
the
same
in
the
same
family
and
yeah,
we
have
some
ongoing
conversations
to
not
Rebrand
rename
the
projects
but
making
sure
hey.
The
organization
is
fluent
but
not
fluently,
because
this
generate
confusion
and
as
I
come
as
developers.
We
always
wanted
that
users
has
the
choice
right.
They
can
decide
what
type
of
agent
to
to
use
for
their
own
problems.
A
Yeah,
but
but
again,
when
you
look
at
these
two
projects-
and
you
come
a
new
user
that
just
sees
the
the
the
the
the
entire
ecosystem
and
Landscape
of
the
cncf
is
confusing.
I
had
this
discussion
also
here
on
the
show
with
with
the
CTO
of
the
cncf
chrysana
check,
but
even
when
you
look
specifically
about
the
let's
call
it
the
fluent
family
and
and
new
users
that
are
not
in
the
nitty-gritty
details
but
trying
to
understand
okay.
A
B
Yeah
I
will
talk
by
using
our
from
our
users.
Experience
I
think
that
all
of
them
that
were
long-term
fluently
users,
majority
of
them,
including
Club
providers,
today
switch
it
fully
to
fluent
that
and
I.
Think
one
of
the
problems
that
you
have
in
the
in
the
market
and
infrastructure
is
like
every
year
you
are
generating
20
to
30
percent.
A
Nice,
nice
actually
I,
am
curious,
I'm
also
passionate,
not
just
about
open
source,
but
also
about
communities,
and
the
the
community
around
fluently
is
actually
amazing
and
I'm
honoring
as
one
of
the
people
who
built
out
this
community.
How
do
you
get
this
a
way
of
an
active
community
that
is
not
only
consuming
the
project
but
also
contributing
back,
creating
plugins,
maintaining
and
evolving
the
plugins
in,
to
the
extent
that
you
say
hey,
even
if
the
core
team
has
switched
to
fluent
bid,
the
the
Lively
Community
is
carrying
on
with
the
flu
and
data.
B
Yeah
I
would
say
that
first,
a
community
has
been
created
organically
right.
It
was
not
meant
to
me:
oh
we're
going
to
create
a
community
and
we
did
XYZ.
It
was
created,
I
think
that
we're
starting
in
the
early
days
of
the
project
and
the
company
that
was
involved
at
treasured
data.
They
were
very
supportive
in
open
source
from
Day
Zero
and
they
told
us
hey
whatever.
We
need
for
the
project
to
succeed
as
a
side
project.
There
was
no
business
plan
around
it
right,
hey.
B
If
you
want,
we
can
go
to
a
conference
to
talk
about
the
project,
engage
with
the
users
that
will
add
more
value
to
the
project
and
years
after
that,
I
started
my
started.
We
started
this
company
called
caliptia,
which
is
Enterprise
chapter
of
fluent,
and
we
take
the
same
approach:
Day
Zero.
We
are
going
to
hire
maintainers
we're
going
to
invest
engineering
resources
on
the
projects
and
I
think
the
community
value
a
lot
that
we
keep
connected
with
them
right.
It's
like
yeah.
We
are
wearing
our
business
hat,
but
then
we
go
to
the
community.
B
We
got
this
community
community
meetings
every
two
weeks.
We
hear
too
about
the
pains
from
the
users
and
always
trying
to
see
and
think
hey
how
we
can
collaborate,
how
we
can
solve
this
problem
that,
maybe
you
are
not
you
facing
it
right,
they
might
be
I,
don't
know
100
companies
under
the
same
problem,
because
that
I
don't
know.
We
will
think
I
think
that
people
behind
the
fluent
and
maybe
having
other
projects
too.
We
try
to
innovate.
We
try
to
set
the
pace
of
how
this
needs
to
evolve.
B
A
No
but
I'm
just
curious,
because
I
see
so
many
young
projects
that
are
struggling
with
buildings,
so
they
have
their
own
core
team,
especially
projects
that
arise
from
vendors.
So
they
have
the
engineering
of
that
vendor
contributing
obviously,
but
getting
A
Wider
Circle
that
will
be
actively
involved
is
not
trivial
and
it's
amazing
to
see
that
around
the
fluency.
So
I
was
wondering
if
you
have
some
some
tips
for
the
newer
projects
trying
to
build
out
such.
B
B
We
try
to
connect
with
Prague
who
are
competitors,
products
that
are
competitors
because
that's
a
way
to
get
more
value
for
the
end
user,
but
yeah
if
an
advice
would
be
a
learn
about
the
pain
points
of
the
users
is
really
easy
to
get
in
love
with
technology,
creating
a
really
cool
engine.
But
if
that
doesn't
add
a
lot
of
value
for
the
user
right
now,
they
would
say:
oh
it's
cool,
but
maybe
I
don't
need
it
right.
Yeah
build
something.
A
B
B
So
what
we
build
is
like
a
limited
pipeline
solution
that
maybe
you
can
solve
the
same
with
flowing
bit
by
having
five
Engineers
working
24x7
right
to
mind
because
maintain
data
pipelines
and
to
solve
the
problem
for
different
stakeholders
in
the
company
for
a
long
period
of
time
is
complex.
So
our
product
comes
to
simplify
all
these
complexity.
B
Around
configuration
data
pipelines
how
to
bring
the
business
in
the
sorry,
the
business
logic
into
the
pipeline,
because
it's
not
just
someone
moved
data
from
A
to
B-
is
also
extracted
data
process,
the
data
and
send
it
out-
and
you
will
be
surprised
that
a
old
company
has
a
very
crazy
needs
around
this.
For
example,
you
can
do
very
simple
processing,
like
other
key,
remove
a
key,
but
then
you
have
I,
don't
know
a
financial
institution
that
tell
you
you
know
I'm
moving
a
lot
of
transactions
here
and
in
the
transactions
in
the
records.
B
Okay,
you
can
drop
this
record
this
key
and
they
tell
you
oh,
but
they
have
a
more
custom
business
requirement.
It's
like
if
this
is
an
American
Express
I
have
to
do
XYZ
if
it's
a
MasterCard,
another
type
of
logic,
if
it's
an
American
Express
a
different
type
of
logic
or
send
the
data
to
a
different
backend.
So
when
you
get
to
this
very
Enterprise
use
cases
right.
This
is
where
a
product
helps
you
to
reduce
time
right
and
solve
this
problem
because
nobody
wants
to
be
configuring
pipelines
the
whole
day.
You'll
find
a
goal.
B
As
I
said,
the
beginning
is
to
analyze
data,
and
now
we
find
users
that,
with
our
product
we
are
able
to
move
certain
analytics
part
that
used
to
be
happening
in
the
back
end.
Sorry,
the
final
like,
for
example,
Splunk
elastic
they
are
moving
certain
analytics
to
inside
the
data
pipeline.
So
they
can
have
you
know
a
smarter
decision
is
don't
send
this
data
or
drop
it
or
send
it
to
a
different
backend
yeah.
A
Nice,
it's
it's
an
interesting
angle,
so
you
mentioned
the
that
most
of
the
Innovation
that
that
you
and
the
maintainers
put
these
days
go
to
fluent
bit
the
the
younger
brother,
the
younger
sibling
of
fluentd.
So
let's
talk
about
a
bit
about
and
then
I
I
was
there,
as
I
mentioned
hearing
about
fluent
bit
version,
2
a
major
release
and
everything.
So
can
you
tell
us
in
in
Broad
Strokes
about
the
new
new
features
that
were
introduced
to
front
bit
in
the
major
release.
B
Yeah
slightly
on
in
the
major
version
one
we
started
to
extend
the
scope
we
used
to
just
do
login
right,
but
internally.
If
you
look
at
fluency
and
flu
and
bit
for
us,
the
data
is
quite
of
agnostic
right.
The
data
could
be
whatever
we
want
right
and
we
just
care
about
hey
how
to
connect
certain
analytics
or
telemetic
type.
Sorry,
Telemetry,
backends,
in
a
different
way.
B
Right
right
now,
I
just
got
my
Wi-Fi
I
just
might
lose
my
Wi-Fi
right,
it's
very
normal
to
get
a
network
outage
to
get
unresponsive
back-end,
and
the
thing
is
that
how
reliable
you
can
be
in
order
to
make
sure
that
that
data
doesn't
get
lost
and
doesn't
it
doesn't
affect
your
production
environment
and
you
can
still
continue
operating
right
and
there
are
different
approaches
approaches
for
that.
But
when
we
extended
the
this
scope
we
found
that
also
people
who
were
using,
for
example,
I,
don't
know
if
you
have
you
heard
about
the
agent's
fatigue.
B
I,
don't
know
if
that's
the
right
pronunciation,
where
people
has
multiple
agents
for
different
things
and
people
starting
to
approach
us
and
they
tell
us
hey.
Would
you
be
able
to
implement
this?
For
example,
Prometheus
not
exported
feature
as
a
flu
embed
plugin
when
they
asked
was.
Why
is
that?
Because
hey
you
can
do
collection,
you
can
send
the
data
you
can
process
the
data,
but
I
don't
want
to
manage
multiple
agents.
Oh
that
makes
sense.
So,
for
example,
we
took
Prometheus,
not
exporter,
and
we
do.
B
We
did
a
couple
of
copy
paste
from
Golan
into
C
code,
instant,
fluent
bid,
and
that
now
is
a
plugin
and
we
see
the
same
Matrix
same
level
same
descriptions,
so
you
can
remove
One
agent
and
get
the
same
permitted
experience
through
fluent
bid
and
continue
using
your
same
dashboards
and
I.
Think
that
is
a
huge
value,
because
you
also
reduce
the
overhead
of
the
server
right
and
people
get
to
know
value.
This
is
the
same
agent
same
experience.
I,
don't
need
to
learn
something
new.
B
B
Telemetry
is
mostly
used
for
traces,
I
would
say:
yeah,
Matrix
and
metrics
are
there
right
now,
logs
is
still
is
Market
stable,
but
I
just
knew
a
few
people
who
are
using
open,
Telemetry
logs
today,
so
I
always
say
that
might
need
more
feedback,
but
at
least
from
the
flu
and
bit
side
we
extended
the
scope
to
support
all
the
payloads
from
a
open,
Telemetry
and
The
Innovation
piece
comes
here
is
not
just
we're
going
to
extend
to
support
multiple
Telemetry
data
data
types.
It's
about
hey,
our
users
doesn't
have
a
single
type
of
environment.
B
Where
just
does
Lux,
they
also
do
Matrix
the
traces
and
they
have
Prometheus
and
they
are
using
open
Telemetry
in
their
endpoint
and
they're,
using
different
things
yeah,
and
they
struggle
to
connect
the
things
and
it's
really
hard
and
by
experience
since
the
early
days
of
fluency
there's
no,
you
were
never
finding
production,
just
one
single
tool.
That
does
everything.
Actually?
A
So
it's
definitely
interesting
to
hear
the
support
for
time
series
data
that
you
have
built
into
fluent
bit
and
especially
the
support
for
exporting
it
is
for
ingesting
sorry,
Prometheus
format,
because
I
think
this
is
today,
especially
in
the
cloud
native
world,
the
the
de
facto
standard.
So
it's
definitely
something
that
will
enable
many
of
the
users
to
use
fluent
bit
also
in
the
metric
style
side
of
the
house
any
by
the
way,
any
any
other
updates
that
you
wanted
to
to
share
with
the
audience.
A
B
Sure,
yeah
and
well
right
now
this
year
we
will
be
just
fully
influenced
version
two
and
iterating
extending
a
more
connectors.
For
example,
the
Enterprise
is
asking
for
these
connectors,
like
Google
Chronicles,
it's
like
a
special
server
for
security
things.
We
just
a
ship
that
one
we
are
doing
a
lot
of
work
in
the
windows
area,
right
people,
always
love
windows,
but
hey.
B
If
you
go,
we
talked
to
a
couple
of
banks
once
in
a
while
and
they
have
100
000
deployments
with
Windows
and
running
from
a
bit
today
right
and
they
primarily
did
it
for
collecting
logs
and
now
everybody
hey
can
I
collect,
also
the
metrics
from
from
Windows
servers
yeah.
So
with
the
same
thing
that
we
did
for
Linux,
we
did
it
for
Windows,
so
we
now
can
collect
native
metal
from
Windows
and
and
well
we're
solving
that
that
problem
at
a
at
a
very
high
scale.
B
Yeah
Windows,
is
not
a
pleasant,
a
you
know,
operating
system
to
work
with
to
be
really
honest,
but
yeah
we
got
thousands
of
users
a
lot
of
them.
Had
this
problem,
hey
it's
something
that
we
have
to
tackle.
Somebody
has
to
do
it
right
and
I
think
that
we
are
in
a
really
strong
position
to
do
it
and
happy
to
support
all
of
that
and
another
Windows
developer
right,
there's
other
people
from
the
community
who
contributes
all
these
places
and
in
traces.
Well,
we
didn't
talk
about
traces
yeah.
B
B
A
B
Yeah
today
you
can
use
the
open,
telemeter
collector
right.
You
can
do
so.
You
can
centralize
all
your
traces
in
there,
but
at
least
for
everything
else,
whether
it's
log
symmetric
I
think
that
the
flu
bit
has
been
working
really
great.
We
get
really
good
feedback
yeah,
it's
really
exciting
to
get
in
achieving
10
billion
dollars
from
Docker
had
from
containers,
but
also
that
means
more
bugs
more
community.
That
needs
a
more
support
right.
It's.
B
It
becomes
a
tough
job
from
a
maintenance
perspective,
because
every
line
that
you
write
and
not
talking
just
by
me-
I'm
talking
about
mostly
maintenance
rights
times
more
code
than
writing
today
and
yeah.
This
will
go
to
production.
It
will
be
deployed,
100
000
nodes
and
make
it
make
sure
that
you
don't
break
any
anything
right.
Yeah.
A
Way
what
you
said
about
Windows,
that's
a
a
known
problem,
I
think.
Actually,
just
a
week
or
two
ago,
my
team
released
a
few
fixes
for
open
Telemetry
on
on
the
Windows
side,
because,
let's
face
it
mostly
most
of
the
maintainers
on
the
open
Telemetry
are
not
Windows
people
and
even
bear
things
like
you.
A
When
you
ran
it
as
a
service,
because
I
guess,
whoever
did
the
testing
just
read
it
locally,
but
not
as
a
service,
and
things
may
get
messed
up
when
it's
a
service
format,
so
Windows
always
gets
I,
guess
less
attended
in
these
ones,
but
very
important,
and
someone
needs
to
to
pick
up
also
there.
So
that's
that's
a
known
known
issue.
A
Definitely,
and-
and
you
mentioned
the
the
rich
Suite
of
of
plugins
and
I'm,
just
wondering
with
fluentd-
you
talked
about
it
before
and
lots
of
contributions
from
the
community,
the
ruby
gems
that
you
mentioned.
Can
you
tell
us
a
bit
about
the
Community
behind
the
plugins?
How
much
is
the
core
team
and
how
much
is
is?
Is
it
active,
Community
contributions
around
that.
B
Yeah
sure
a
flomat
has
a
different
plugin
mechanism
that
flew
in
the
fluency
or
basis
based
on
Ruby
was
kind
of
a
small
core
and
you
pull
the
gems,
so
they
become
plugins
for
flow
index
and
extensions
inflam
made
when
they
probably
born
a
personally.
It
took
the
opposite
approach
because
it's
like
okay,
this
sounds
too
heavy
I
found
that
people's
struggling
to
build
a
bundle
or
something
that
works
for
them.
Hey,
let's
build
all
the
plugins
inside
the
agent
right
kind
of
can
call
them
binary
where
you
have
everything
the
same
thing
but
hey.
B
We
are
in
scene,
so
there's
some
some
challenges.
So
what
we
did
was
kind
of
write,
almost
everything
by
hand
and
from
scratch,
HTTP
clients,
or
how
to
make
an
Essence
security.
Well,
we
use
open
SSL,
of
course,
we're
not
going
to
do
a
ring
band,
crypto,
libraries
and
and
I
think
that,
from
that
perspective
and
the
way
to
build
blinds
influence
today,
even
in
C
is
really
simple,
because
it
the
API
abstracts
a
lot
of
the
complexity
of
networking
failures.
B
Tls
handshakes
events
base
mechanisms-
all
of
that
is
Works
behind
the
scenes.
We
have
a
really
cool
architecture
with
threads
core
routines
and
it
works
great,
but
when
people
want
to
extend
beyond
that-
and
they
don't
know
C,
for
example,
one
of
the
first
pain
for
them
was
hey.
I
need
to
create
more
filters
for
my
own
use
case,
like
I
explained
you
in
with
his
credit
card.
That's
why
Lua
was
born.
The
luafita
was
born,
so
you
can
write
your
own
modification
script.
B
That
goes
into
a
configuration
and
just
works,
and
now
we
get
to
a
different
point
in
the
project
where
people
has
like
I
have
seen
configuration
with
40
filters,
chained
together,
it's
insane
and
it
was
like
okay.
This
is
wrong.
It
should
yeah.
This
will
become
the
next
bottleneck
and
what
we
did
influence
to
that.
Oh,
we
implemented
this
concept
of
processors
right.
What
is
different?
B
That
in
in
previous
versions,
like
normally
from
a
high
level,
you
have
input
filters
and
outputs
period
now,
inputs
and
output
plugins
can
run
in
separate
threads,
so
they
can
consume
other
CPUs,
right
or
CPU
cores,
but
all
the
main
eval
Loops
that
orchestrate
everything
plus
filters
was
in
the
main
thread.
If
you
have
more
filters,
yeah
you
get
more.
That
will
become
your
next
bottleneck.
B
If
you
extend
too
much
so
what
we
did,
it
was
to
implement
the
concept
of
processors
that
runs
together
with
the
input
plugins
or
the
output
plugins
in
a
very
agnostic
API,
you
can
reuse
the
same
filters,
but
they
can
run
in
separate
threads.
So
now
you
can
scale
up
by
moving
that
workload.
Processing
to
different
CPU
cores
and
different
threads
and
processor
has
been
a
really
good.
You
can
reuse,
for
example,
the
the
Lua
filter
in
in
processor,
mode
and
processor.
B
The
difference
with
filters
is
like
they
don't
go
through
a
routing
mechanism,
so
they
just
run
from
top
to
bottom
in
the
configuration
and
those
are
just
supported
in
the
new
jaml
schema
configuration
format
that
we
have,
what
we
call
a
classic
configuration
mode,
and
now
we
just
pushing
also
a
a
party
with
demo.
You
know
I
think
that
has
been
really
successful
for
people
in
the
cloud
native
space,
because
all
the
tools
in
place
well
with
demo
a
there's,
no
beautiful,
config
format
right
fluent
DXs
is
XML.
B
A
Is
a
yaml
engineer
these
days,
so
yeah
yeah,
but
but
what
you
said
about
the
making
it
parallel
processing
the
multi-core.
That
sounds
definitely
like
something
that
would
give
a
superior
performance.
That
sounds
sounds
exciting
and-
and
you
mentioned
the
before
the
thing
about
kubernetes
I-
think
you
did
some
interesting
things
around
kubernetes
events
right.
B
B
But
people
who
knows
has
more
experience
think
about
how
fast
this
car
can
go,
plus
how
much
gas
I
need
for
that
right
and
in
Computing
is
the
same
right.
You
can
have
a
very
really
fast,
a
tool
or
agent,
but
also
a
how
much
power
you
need
to
process
that
oh
yeah
I
need
five
servers:
okay,
that's
wrong,
and,
and
talking
about
the
private
product
around
and
at
least
with
flow
embed.
B
We
make
sure
that
we
can
have
a
very
high
throughput,
a
real,
so
use
the
resources
that
we
have
in
a
very
smart
way
that
can
give
us
more
processing
power
in
less
time,
but
even
using
less
CPU
that
different
type
of
architectures
and
now
going
back.
This
has
performed
the
performance
a
perspective.
So
in
the
second
point,
was
there.
B
Sorry
about
that,
it's
a
lady
so
from
they
I
think
that
from
year
2018
we
support
the
kubernetes
filter
inside
flow
embed.
That
means
that
when
we
collect
the
logs,
we
can
reach
the
logs
with
labels
and
annotations,
and
now
recently
we
get
this
need,
like
a
my
kubernetes,
a
cluster
generate
events
that
also
I
need
to
analyze
for
security
purposes
or
general
analysis.
B
So
we
just
ship
a
new
plugin
that
also
is
called
a
kubernetes
events
in
the
input
side
that
just
connects
to
the
API
server
and
is
able
to
retrieve
all
the
events
by
default,
or
you
can
select
certain
namespaces
and
ship
them
as
normal
logs
over
the
over
the
pipeline
and
yeah
we're
getting
a
lot
a
ton
of
a
good
feedback
about
a
lot
of
good
feedback
about
that.
I
think
that
is
becoming
the
next
default.
B
A
Wow,
that's
impressive,
so
most
of
the
plugins
just
understand
the
architecture.
Are
these
done
by
the
core
team,
or
is
that
something
that
the
external
contributors
can
write?
Plugins
just
understand
the
Dynamics
and
how
it's
different
or
same
as
friendly
in
terms
of
the
activity
of
the
community
in
creating
generating
plugins.
B
They,
for
example,
the
stackdriver
teams,
contributes
and
have
Engineers
work
in
a
stackdriver
plug-in
inside
flowing
bit,
which
is
part
of
the
core,
so
they
just
work
directly
on
Upstream,
for
example,
we
have
the
data
plugin
that
was
contributed
by
data,
though,
and
different
companies
contribute
to
their
own
connector
that
either
was
missing
or
we
start
we
just
created
and
they
so
there's
a
ton
of
value
because
hey
if
we
have
thousands
of
users
using
this
agent,
we
might
have
a
connector
I.
Think
Nightfall
is
another
company.
B
I
think
explained,
contributed
some
time
to
their
own
spline,
connector
and
yeah.
It's
fun,
because
the
fun
part
here
is
like
people
contribute
by
buying
it
right.
They
need
to
solve
a
problem
or
they
need
to
solve
a
business
problem
and
they
find
a
good
value
by
investing
time
and
hours
in
influence.
I
think
we
don't
have
like
a
synthesis,
a
probe
based
in
C
language.
We
don't
have
a
thousand
contributors
in
the
code.
B
I
think
that
we
are
a
bit
over
300
in
total,
but
now
yeah,
because
most
of
our
people
solve
the
problem
with
processing
with
Lua
now
well,
with
version
two
and
also
a
couple
of
months
ago,
we
improved
the
the
wasam
interface,
so
you
also
can
create
plugins
in
Watson
for
the
input
of
filters.
So
if
you're
using
the
awesome
layer,
you
can
use
Python,
you
can
use
rust
or
anything
or
any
other
fancy
language
that
you
like.
B
A
Nice
and
I
think
you're
you're
planning
on
announcing.
We
have
we're
now
at
August
and
in
a
couple
of
months
time
we're
going
to
have
cubecon,
North,
America
I.
Think
the
big
announcements
is
going
to
be
around
the
version
2.2.
If
I'm
not
mistaken,
they
want
to
tell
us
what
give
us
a
teaser
about.
What's
what's
up
and
coming.
B
Yeah
well,
I,
don't
want
to
lose
spoilers
here,
but
okay,
two,
the
two:
it's
coming
for
kubecon.
That
will
be
a
well,
maybe
end
of
October
the
release.
It
will
be
ready.
A
one
of
the
paintballs
that
we
found
with
the
community
was
around
monitoring
of
the
agent
right,
so
I
I
any
kind
of
project
where
you
get
the
problems
at
different
layers
right.
We
used
to
have
a
connectivity
functionality
and
now
are
more
around
usability
or
how
to
make
easier
to
create
these
pipelines.
B
So
one
of
them
is
well
it's
about
the
monitoring
of
the
agent,
we're
chiping,
a
new
endpoint
with
new
grafana
dashboard,
so
people
can
get
more
insights
and
we
are
preparing
another
secret
project,
which
is
a
terminal
UI
based
spray
for
flu
embed
right
that
you
can
monitor.
You
can
create
this
this
stuff
and
get
some
charts
in
your
terminal
is
pretty
cool
and
so
yeah
I'm
gonna
share
some
information
around
October.
It's
it's
looking
pretty
pretty
cool
and
the
other
is
around
well
performance
improvements
and
I
always
mention
this
and
people
say.
B
Oh,
why
you're
talking
always
about
performance
because
there's
more
Demand
right?
There's
always
more
demand,
oh,
if
you're
now
able
to
process
I,
don't
know
100
000
messages
per
second.
Now
we
need
120..
Okay.
What's
the
next
bottleneck?
How
we
can
improve
it,
how
we
can
adjust
architecture,
so
performance
will
always
there,
and
we
are
now
working
closely
now
with
a
person,
who's
writing
the
flu
and
bit
book
it
doesn't
start
from
us
it's
starting
from
Phil
Wilkins
who
created
the
flu
in
the
book
she
started
reading
now
the
fluent
big
book.
B
So-
and
we
say
you
know,
we
should
work
together,
at
least
as
maintainers
we're
going
to
to
contribute
in
the
radiation
Edition
in
the
book,
because
one
thing
that
we
notice
with
a
technical
books.
It
happened
to
one
person
that
I
know
that
wrote
the
kubernetes
on
the
H
book.
The
problem
is
that
he
chipped
the
book.
After
writing.
A
book
disclaimer
takes
12
months.
B
It
don't
work
the
moment
that
you
ship
the
book,
maybe
some
of
commands
or
configuration
that
used
to
be
supported
at
that
time,
are
no
longer
available
or
they
are
deprecated.
This
happened
to
him
right
chapter,
one
and
two
is
broken
because
they
deprecated
a
bunch
of
them
k3s
so
and
it's
all
failed.
I
don't
want
that.
This
happened
to
you,
so
we
want
to
make
sure
that
whatever
we
ship
in
the
next
version
either
flew
in
bit
three.
B
We
were
thinking
at
that
time
or
2.2
yeah
we'll
be
working
for
your
version
and,
if
we're
moving
to
yaml,
if
we're
moving
to
more
collection
of
Matrix
and
I,
don't
know
open
tilling
to
payloads
type.
Let's
make
sure
that
you
have
the
right
support
from
us
because
yeah
this
is
kind
of
new
has
been
there
for
more
than
one
year,
but
I
want
to
make
sure
that
your
readers
and
people
who
use
flambit
and
consume
this
book
can
get
a
good
experience
now.
Windows
improvements
are
coming
more
and
more.
B
We
keep
shipping
more
metrics
for
Windows.
I
would
say
that
it's
not
like
really
big
new
future
right,
because
we're
still
there's
a
lot
of
things
to
tackle
even
still
in
logs
in
metrics
and
traces
in
processing.
A
yaml
will
improve
too
right.
The
way
that
we
support
yaml
today
works.
It's
fine
but
I
think
we
need
to
do
some
improvements
for
the
tooling,
who
just
does
a
general
strict
mode,
which
is
more.
You
have
to
use
lowercase
applicants
yeah
the
case
right.
We
just
use
underscore
like
in
the
old
way,
the
classic
Mode.
B
We
need
to
fix
that
and
yeah.
We
will
have
more
news
around
traces
too,
how
to
select
traces,
you
know
and
how
to
not
to
correlate
all
of
them
and
not
to
create
all
the
Matrix.
Sorry,
all
the
the
data
type,
the
Telemetry
data
types
when
they
go
through
the
pipeline,
because
that
is
aggregation,
but
we're
doing
doing
some
selection
through
processors,
so
certain
traces
that
are
not
useful.
You
can
discard
them
right
away.
You
can
write
the
run.
B
The
Lua
filters
on
top
of
them
I
think
that
that
part
is
really
interesting
and
one
of
the
challenges
that
we
found
with
people
doing
tracing.
Maybe
you
can
elaborate
on
that.
You
have
more
experience
than
the
mean
and
that
it's
like
yeah
I'm
receiving
all
the
traces
in
in
my
collect,
for
example,
but
I
don't
need
all
that
data
and
and
The
Collector
also
is
presents.
So
much
data
that
maybe
we
can
discard
them
right
from
the
edge
right
where
the
SDK
is
sending
this
information.
There's.
A
B
B
So
there's
a
lot
of
experimentation
and
news
around
connecting
all
these
type
of
use,
cases
class
and
we're
working
hard
with
the
community
to
create
trainings
and
right
now
we
are
running,
for
example,
the
new
flu
and
bit
webinar
series
summer
series
that
are
being
out.
We
have
one
today,
I
think
it's
running
at
this
same
time,
I.
B
Anorak
and
Thiago
is
running
that
one,
so
please
Community
can
it
will
be
really
good
if
you've
enjoy
and
yeah
so
I
think
yeah
performance,
GMO,
Windows
processors
for
traces,
a
more
processor
for
Matrix,
so
a
lot
of
things
that
are
being
really
useful
for
people
and
yeah
we're
preparing
for
flu
and
bit
version
3,
also,
which
is
the
next
year.
We
plan
to
launch
it
on
Paris
in
kyocon,
2024.,
exciting.
A
Great,
so
definitely
for
audience
check
out
the
flu
and
bit
team
at
open
observability
day
at
the
pre,
the
the
pre-cubecon
event
that
week
and
then
obviously
at
kubecon
itself
in
Chicago
in
November,
and
until
then
they
want
to
share
with
the
audience
where
they
can
reach
out.
I
know
that
there
is
a
very
active
fluent
slack
but
Beyond
to
share
where,
where
to
find
you
where
to
reach
out
to
the
community.
B
Yeah
well,
there's
a
couple
of
channels.
Asynchronous
communication,
it's
like
is
always
good
and
if
you
go
to
fluency
or
fluent
bit
website,
you
will
get
the
slack
link
is
fluent
Dash
all
but
yeah
you
get
a
shortcut
in
there
and
the
other
is
at
Twitter
for
more
random
communication.
Actually,
sometimes
we
get
asked
for
help
in
Twitter.
B
Github
is
also
not
a
resource,
but
we
have
the
flu
and
bit
Community
meetings
happening
every
two
weeks
right
that
have
been
very
useful,
so
people
from
different
companies
join
us,
they
share
about
their
use
cases
and
we
are
there
to
just
try
to
help
as
much
as
we
can
with
suggestion
or
listening
about
the
struggles
that
they
might
have
with
fluency
or
fluent
bid,
and
try
to
see
how
we
can
at
least
for
people
who
are
from
different
companies.
In
that
call,
we
can
prioritize
in
our
roadmaps.
B
A
B
Yeah,
so
before
how
to
reach
out
to
me,
one
big
ask:
we
are
running
the
flu
and
bit
survey
this
year
right.
One
of
the
challenges
that
we
have.
Another
challenge
we
have
is
like:
we
have
ton
of
users
right
and
we
get
a
huge
Community,
but
sometimes
it's
hard
to
understand
where
the
struggle
are
or
how
we
can
do
a
better
job
for
them
for
the
community.
So
we're
running
this,
this
survey
that
is
published
in
the
fluent
bed
and
fluency
website,
so
you
go
to
the
main
site.
B
A
That's
great
that's
great,
and
thank
you
very
much
and
do
stay
with
me
for
the
last
part
of
the
show,
which
is,
as
always,
the
breaking
news,
and
today,
I
have
quite
a
few
updates
that
I
wanted
to
share
with
the
Community
First
of
all
is
Prometheus
that
now
supports
otop.
A
recent
enhancement
enables
Prometheus
to
ingest
open,
Telemetry
metrics
using
a
new
otlp,
compatible
ingestion
endpoint.
A
Obviously
it's
it's
still
experimental,
so
it's
under
a
feature
flag,
so
be
aware
of
that.
But
if
you
want
to
to
hear
more
about
that,
I'm
posting
on
the
comments
here
for
the
live
stream,
audience
and
I
will
also
post
for
the
obviously
for
the
for
the
listeners
and
the
podcast
apps.
A
short
brief
on
that,
which
is
also
very
similar
to
what
you
said.
A
You're
recognizing
that
Prometheus
is
is
is
a
de
facto
way
of
processing
but,
on
the
other
hand,
otlp
or
open
telem3
and
octopus.
The
protocol
is
becoming
more
and
more
common
and
people
are
expecting
to
to
be
able
to
ship
that
and
be
able
to
ingest
that,
from
the
other
end,
so
definitely
aligned
with
what
you
shared
with
us
as
well,
and
then
we're
talking
about
Prometheus
prom
con
we'll
take
place
in
Berlin
this
year.
The
dates
have
been
published.
A
It's
at
the
end
of
September,
28th
and
29th.
It
will
be
the
eighth
annual
conference
of
Prometheus
and
all
the
open
source
ecosystem
around
it,
so
cortex
tunnels
and
so
on.
It's
a
good
place
to
hear
use
cases,
case,
studies,
best
practices,
new
features
and
Integrations,
and
so
on.
Also,
the
integration
with
open
television
I
just
mentioned
so
do
check
it
out
if
you're
into
the
Prometheus
ecosystem
and
by
the
way,
there's
also
an
open
call
for
proposal.
A
B
A
A
So
we're
going
to
get
all
the
goodness
from
the
elastic
elasticsearch
community
and
the
elk
star
community
and
the
knowledge
about
log
semantic
conventions
into
the
open
Telemetry.
Now
that
it's
separate,
we
can
get
the
maturity
of
the
specification
while
still
carrying
on
with
the
semantic
convention.
So
this
is
a
definitely
a
an
interesting
one
to
check
out
I
actually
posted
about
it
just
just
today.
So
it's
really
really
fresh
hot
off
the
press.
Eduardo.
B
Yeah
I
have
mixed
opinions
about
it
in
general,
not
because,
and
not
from
the
angle,
looks
good
or
bad.
It's
like
from
years
ago,
there's
always
looking
for
to
have
these
standards
in
general,
and
we
always
get
some
struggles
right
and
we
always
ended
up
with
different
heterogeneous
systems
that
speaks
in
different
languages
right
and
what
I'm
all
about
in
is
about
to
support
every
single
standard.
B
Every
single
schema
that
exists
around
because
from
my
angle,
it's
not
to
to
create
a
standard
but
be
able
to
connect
this
different
standard,
these
different
schemas
and
always
thinking
in
the
user,
that
if
they
are
using
XYZ
tomorrow,
they
can
switch
vendor.
They
can
switch
protocols
and
things
will
run
smoothly.
So
I'm
really
happy
to
see
this.
Also
in
the
film
Community,
we
have
again
a
lot
of
these
requests
for
elastic
common
schema
and
other
schemas.
It
is
part
of
the
news
for
the
next
release.
B
We
are
having
some
layers
around
that
too,
but
as
far
as
we
are
adding
value
to
the
user,
I
think
that
I'm
I'm,
all
in
I,
don't
want
to
say
yes,
I
believe
this
is
a
huge
standard.
This
this
will
fix
everything,
because
time
sometimes
demonstrate
that
it's
not
one
thing
that
helps
is
multiple
of
them,
but
happy
to
to
learn
more
about
it
and
yeah
and
talk
to
the
community,
because
one
thing,
as
even
us
right
releasing
software
pitching.
This
is
what
you
should
use.
When
you
go
talk
to
the
users.
B
A
Yeah
sounds
good.
Another
update
on
the
latest
release
of
kubernetes
1.1.27
on
the
observability
side,
there's
a
very
important
feature.
There's
now
it
introduced
a
new
feature
called
node
log
query
that
allows
viewing
logs
of
the
services
running
on
the
Node,
so
you
can
actually
use
the
cubelet
API
to
just
query
and
understand
the
logs.
This
is
still
Alpha
release
of
the
feature
and
it's
behind
the
feature
gate.
A
It's
essentially
a
project
for
a
global
tracing
for
kubernetes
control
plane,
and
we
all
know
it's
kubernetes
is
async.
It's
event
driven.
It's
declarative,
so
it's
it's
interesting
in
the
context
of
tracing,
because
the
typical
tracing
approach,
where
the
spans
or
the
operations
ordered
by
which
operation
in
Fox,
which
doesn't
really
fit
well
this
model.
So
it
was
fascinating
to
see
that
kilometry
took
a
different
approach.
A
Maybe
a
bit
similar
to
case
span
for
those
who
are
familiar
with
this
open
source
where
span
is
created
for
the
actually
not
for
the
operation,
but
for
the
kubernetes
object
itself
and
each
event
that
happens
on
that
object.
Is
a
child
spam
and
kilometry
Collective
data
from
the
API
server
the
audit
logs
and
and
uses
that.
So
it's
a
very
interesting
project.
I'm
sharing
here,
an
interesting
blog
post
that
was
posted
on
the
cncf
blog.
Have
you
run
into
that
that
one
by
any
chance,
Eduardo.
B
A
I
don't
follow
the
kubernetes
news
like
in
every
single
release,
but
I
always
try
to
come
up
with
what
is
new
in
the
observability
space
but,
for
example,
these
giveaway
things
is
really
interesting
right
and,
for
example,
in
the
past
we
use
everybody
used
to
saturate
the
API
server
to
grab.
You
know
the
metadata
and
but,
for
example,
the
equivalent
has
the
same
information
was
never
exposed
and
well.
Alaska
was
doing
some
work
in
fluently
with
flu
embed
on
that
area.
B
A
Amazing,
so
thank
you
very
much
I'd
like
to
thank
you
again
for
joining
me
and
for
joining
us
and
sharing
all
the
latest
and
greatest
with
fluent
bit
and
fluent
D,
and
can't
wait
to
hear
the
announcements
around
the
qcon.
You
also
had
a
very
good
developer
week
right
they
a
week
ago
or
so
right.
So
the
community
has
also
got
face
time
together.
Yeah.
B
We
we
stand,
we
we
were.
We
were
in
Houston
like
a
couple
of
weeks
ago,
so
we
had
a
private
meeting
with
some
some
developers
from
different
companies.
So
it
was
a
really
good
experience
to
understand,
pains
what
we
can
solve
challenges
and
you
know
be
more
work
on
opinion
on
this
type
of
instances
as
a
roadmap.
So.
A
A
So
you
can
find
all
the
details
on
on
the
Twitter
handle
at
open,
observe
where
you
can
hear
when
we're
live
streaming
and
to
also
share
your
comments.
Suggestions,
news
bits
if
there's
anything
interesting
that
you
want
me
to
include,
and
with
that
I'd
like
to
thank
you,
everyone
for
listening,
I'm,
Dotan,
horvitz
and
see
you
at
next
month's
episode.