►
From YouTube: Monitoring Cloudflare's IPFS Gateways - Thibault Meunier
Description
This talk was given at IPFS Camp 2022 in Lisbon, Portugal.
A
Real
goal
is
like
to
share
how
we
have
set
up
our
monitoring,
how
we
use
it
and
definitely
happy
to
see,
especially
because,
like
there's
a
lot
of
analysis
that
has
been
done
and
like
a
lot
of
data
that
could
be
available
so
see
how
like
these
logs,
could
help
you
down
the
line
and
just
to
inform
how
this
is
going
so
yeah
I'm
a
research
engineer
at
cloudflare
I
work
on
the
various
distributed
web
project,
ipfs
ethereum,
also
about
split
in,
like
all
the
standards
like
DNS
and
I,
like
croissants,
so
I
know
we're
in
Portugal
and
like
it's
mostly
pastelinata.
A
They
also
have
very
weird
brioche
croissant,
which
I'm
not
trying
to
qualify
as
croissant,
but
yeah.
That's
the
case.
So
we'll
see
if
there
are
some
available
later
today
today,
the
presentation
so
I'm
going
to
present
quickly
an
overview
of
like
cloudflare
system
what
it
means,
then
what
is
clever
monitoring
the
various
systems
we're
using
and
then
where
to
go
next
things
we
have
in
the
pipeline
for
like
Improvement
for
this,
so
yeah,
first
class
architecture.
A
So
most
of
the
you
in
the
room,
I
think
know
what
cloud
like
a
Gateway
is
it's
an
HTTP
interface
to
the
ipfs
network.
A
The
main
URL
for
cloudflare
Gateway
is
cloudflapfs.com,
even
though
we
also
do
provide
the
capability
for
people
to
have
like
X5
and
on
certificate
on
their
own
domain,
and
so
they
can
serve
whatever
CID
they
want.
As
mentioned
before,
the
Gateway
is
closely
tied
with
like
standard
HTTP
cache
which
sits
close
to
like
end
users,
and
so
that
allows
us
to
like
Leverage
standard
web
infrastructure
to
serve
ipfs
content
to
retrieve
ipfs
content.
A
Overall,
that's
the
architecture
that
we
have
in
terms
of
data
Generation,
it's
not
limited
to
that,
but
we
these
are
the
main
systems
that
we're
using.
So
we
have
a
monitor
that
is
just
checking
the
health
of
our
Gateway
via
various
active
probes,
and
then
we
have
more
like
data
that,
like
is
sent
by
the
Gateway.
So
we
have
like
Prometheus
data.
A
A
What
I
didn't
mention
is
like
we're
not
using
row
Cubo
nodes
we're
using
like
Kubo
as
a
library
which
I
think
is
like
kind
of
a
plan
of
having
lib
ipfs
at
some
point,
but
like
it's
kind
of
like
half
in-house
half
open
source
implementation
for
this
foreign.
What
is
kind
of
tricky
that
we've
seen
operating
like
an
HTTP
Gateway
is
for
ipfs
the
kind
of
like
base
layers.
You
have
a
CID
and
you
look
at
Which.
A
Cid
is
being
requested
how
it
is
accessed,
like
you
have
bit,
swap
you
have
like
the
content
routing
for
HTTP.
You
add
like
one
a
on
top
and
like
the
main
layer
we
see,
we
see
is
like
the
naming
so
like
you
can
have
like
a
row,
a
raw
domain,
so
like
ipf
as
a
cloudflare.com,
which
would
act
as
a
Content.
A
But
then
you
might
want
to
have
like
I,
don't
know
a
cache
on
top
and
then
you
have
like
another
customer
on
top
and
then
you
have
like
another
C
names
and
so
being
able
to
have
a
view
through,
like
all
these
different
steps,
it's
kind
of
tricky,
and
so
that's
why
yeah
trying
to
figure
out
what
to
make
sense
of
this
mess
leads
to
a
lot
of
like
debugging
and
headaches
so
to
dive
a
bit
more
in
these
systems.
A
What
actually
is
Cloudland
monitoring
and
so
like
these
very
systems
that
I've
mentioned
before,
and
some
of
you
also
mentioned
in
like
previous
presentations.
So
first
of
all,
we
have
Prometheus,
which
is
very
useful,
like
it's
a
Time
database
working
on
like
low
cardinality
when
there's
like
high
continuity.
A
Just
explode,
so
we
have
like
limited
number
of
metrics
per
cardinality,
but
it's
very
good
for
like
live
monitoring
so
like,
for
instance,
if
you
want,
like
the
number
of
like
respond
that
you
have
like
on
your
back
end
or
like,
if
you
want
to
know
like
I
think
Leo
mentioned
about
like.
Is
it
like
from
a
Gateway
or
not
Gateway
like
a
very
small
flippers,
is
very
good
for
that.
It's
also
connected
to
our
internal
auditing
systems,
based
on
like
the
threshold
rates.
A
Another
system
that
we're
using
is
Sentry
and
so
century
is
like
different
than
Prometheus.
In
a
way,
it's
not
that
much
of
a
database,
but
it's
more
like
an
event-based
and
like
event
based
and
like
focus
on
like
the
errors
that
our
system
could
be
producing.
So
the
good
thing
is
like
it
provides
context
about
the
areas
that
we
see.
So,
for
instance,
like
you,
have
the
host
name.
A
And
finally,
we
have
clickhouse.
So
clickhouse
is
just
cloudflare
way
to
have
a
lot
of
data.
I
was
like
arbitrary
data
with
like
High
cardinality,
so
it's
a
column
based
database
cloudflare,
also
samples
the
the
logs
and
like
data
that,
like
we
go
to
clickhouse,
so
that
we
just
don't
store
everything
that
happens.
We
can
store
some
Aggregates
and
it's
far
more
useful
for
long
term.
A
Finally,
something
we've
open
sourced
earlier
this
year
is
the
ipfs
monitor.
So
it's
like
a
black
box
monitoring
tool
for
the
ipfs
Gateway
contestant,
like
multiple
scenarios
or
like
update
of
the
DNS
link
caches
new
content
created
on
like
their
spinning
service,
and
it
also
provide
Prometheus
exports
so
that
we
can
like
have
alerts
when
something
is
going
bad.
A
Like
I,
don't
know
the
maximum
is
like
half
a
second
which
is
good
and
when
the
CID
isn't
available,
that's
really,
depending
on
like
the
node
and
the
strategy,
so
we've
tuned
it
a
bit
since
since
then,
but
mostly
it
just
takes
time
to
figure.
Cid
is
not
there,
which
is
not
that
good,
because
you
have
like
this
loading
screen
just
waiting
for
nothing
from
that.
Where
do
we
go
next?
A
A
Also
the
like
on
content,
reading
and
content,
providing
having
like
Gateway
plus
indexes
Plus,
some
more
like
interactive
participation
with
like
data
providers
like
important
because,
like
I,
don't
know
like
a
minute
to
like
retrieve
a
Content,
it's
just
too
much
on
on
certain
providers,
then
having
like
some
standard
around
like
Gateway
names
and
resolution,
is
I,
feel
so
important
and
that's
mainly
I
think
going
through
like
the
ipns
land,
we're
not
seeing
that
much
usage
for
ipns,
but
definitely
that's
compared
to
its
usage.
A
There's
a
lot
of
requests
coming
in
in
terms
of
like
how
to
tune
and
how
to
define
so
there's
been
a
new
ipip
coming
into
Define
ipns,
but
yeah.
Definitely
more
involvement
through
here.
It
could
be
interesting.
A
A
Are
people
actually
building
dags
that
are
efficient
for
caching,
or
is
it
just
like
using
the
like
normal
Kubo
like
DAC
building,
and
that's
it
I
know
like
some
projects
like
have
been
using
efficient
content
layout
should
be
like
for
Docker
npm
or
like
some
videos
that
we've
seen,
but
overall,
it's
kind
of
like
an
explored
area
that
could
be
interesting
to
reduce
the
amount
of
storage
similar.
A
One
thing
that
is
coming
and
like
definitely
more
for
like
one
additional
sources
like
distributed
tracing
so
on
a
similar
note
to
Sentry
like
this
sort
of
tracing,
would
allow
to
just
better
understand
through,
like
the
whole
Pipeline
and
various
systems
that
we
have
were
actually
is
something
taking
time.
So,
if
we
see
the
time
for
discovering
content
is
just
too
long
that
maybe
like
help
us
prioritize
the
things
that
like
we
want
to,
have
and
want
to
see
and
finally,
integration
with
Enterprise
tool.
A
It's
more
like
a
Gateway
saying,
like
generally
I,
feel
there's
been
especially
the
way
like
the
those
packages
and
like
ipfs
nodes
have
been
built.
It's
sometimes
tricky
to
like
integrate
one
new
tool
without
just
rebuilding
this
thing.
So
yeah,
that's
more
General
comment.