►
Description
Marine-js: How to run marine wasm services on JS peer: architecture overview and how to try it
Our goal is to make the Fluence JS peer as capable as the Fluence Rust peer. Previously there was no way to run wasm services on the JS peer. Now this feature is available and it's now easier to treat both types of peers as the same
If you want to run services both on Rust and JS peers, this feature will surely help you. Marine-JS unlocks the ability to make mind-blowing things with aqua and deploy services dynamically on any type of peers
Useful Links
https://fluence.network/
https://twitter.com/fluence_project
https://doc.fluence.dev/docs/
https://doc.fluence.dev/aqua-book/
https://github.com/fluencelabs/aqua
https://github.com/fluencelabs/aquavm
https://fluencenetwork.notion.site/fl...
https://github.com/fluencelabs/marine...
https://t.me/fluencedev
https://fluence.chat
A
All
right,
everybody
thanks
for
being
here,
is
another
fluency
community
call,
and
actually
it's
gotta
be
a
really
good
one.
We
have
a.
We
have
a
lot
going
on
today
we
have
tech
teams
update
from
the
court
team,
and
then
we
have
a
community
developer
presentation
from
one
of
the
previous
hackathons
very
cool,
and
then
we
get
a
couple:
educational
talks
from
the
core
dev
team,
including
marine
js,
and
how
to
run
web
assembly
on
jsp,
which
is
super
super
cool.
B
So
speaking
about
the
updates,
we
are
focused
on
stabilization
over
the
last
month
since
the
last
community
call.
So
we
have
actually
a
lot
of
changes.
We
will
make
a
lot
of
releases
tomorrow.
We
decided
to
shift
it
one
day
to
to
have
just
more
time
for
the
team
to
prepare
for
for
the
talks,
but
the
main
updates
will
come
from
pavel
valeri
and
alexey
later
later,
during
the
call.
So
let's
just
proceed.
A
D
D
I
start
we
both
started
off
our
first
experience
with
the
internet
was
web,
one
which
was
all
about
democratization
of
information.
So
it
was
a
very
pro-social
approach
to
the
internet.
It
was
very
idealistic,
and
that
was
what
got
us
excited
about
it.
Then
of
course,
web
2
came
along
and
we
spent
a
few
years
turning
rich
people
even
richer
and
making
a
little
bit
of
money
ourselves,
but
it
wasn't
nearly
as
satisfying
well.
D
A
few
months
ago
I
read
an
article
on
andreessen
horowitz
that
was
about
dows
and
how
they
are
bringing
gig
economy
and
decentralization
back
to
the
internet
in
in
web
3,
and
that
was
really
exciting
to
me
and
literally
about
two
weeks
later.
I
put
in
my
resignation
at
amazon
and
about
a
week
and
a
half
after
that
we
were
working
on
the
showing
point
hackathon.
So
it's
been
a
it's
been
a
whirlwind
tour:
we're
really
enjoying
ourselves.
E
It's
it's
better
than
finding
a
needle
in
a
haystack,
because
in
a
needle
in
a
haystack
you
have
all
these
unrelated
straw
pieces.
You
know,
there's
no,
no
relationship
with
it
in
in
the
data
world,
we
can
selectively
find
data
that
is,
is
related
roughly
towards
the
objective
we're
shooting
for
pick
through
it
pick
through
massive
amounts
of
data.
Let's
take
an
example
like
github,
there's,
200,
plus
million
different
repositories
and
github
contributed
by
70
million
different
users.
E
That's
a
lot
of
stuff
can't
go
through
it
very
quickly,
but
if
we
look
through
it
and
find
out
specific
items
within
there,
we
can
sort
through
and
find
out
a
lot
of
information.
For
example,
looking
through
commits
and
that
commit
logs.
You
can
see
what
files
they've
put
in
what
types
of
files,
what
the
programming
languages,
how
many
lines
of
code?
Libraries,
all
sorts
of
things
that
we
can
find
from
that.
So
I
can
sort
through
very
quickly
and
find
little
bits
and
pieces
and
feed
it
into
one
consolidated
block
of
data.
E
So
from
that
here's
one
one
thing
where
fluence
can
can
play
a
part:
we
can
divide
and
conquer
these
vast
amounts
of
data
across
different
nodes
have
them
sort
through
the
data.
It's
very
data
intensive,
but
it's
not
terribly
cpu
intensive.
So
we
can
feed
through
large
amounts
of
data,
find
little
bits
and
pieces
and
put
them
back
together.
E
D
So
yeah
it
makes
it
it's,
it's
really
nice
creating
the
whole
pipeline
and
having
both
sides
working
together.
So
the
data
engineering
that
matt
does
creates
this
really
dense
highly
concentrated
data
that
works
really
nicely
in
machine
learning
models
which
we
can
then
apply
to
try
to
find
software
engineers
who
are
similar
to
each
other
find
software
engineers
that
are
that
are
particularly
capable.
D
Their
skills
are
particularly
applicable
to
an
individual
project
and
the
goal
there
really
is
to
make
it
easier
for
web3
hackers
to
jump
from
project
to
project,
to
explore
new
directions
with
their
careers
to
have
a
little
bit
more
freedom
in
in
what
they
do
with
their
time.
Less
of
the
you
know,
kind
of
golden
handcuffs
that
we
see
working
for
giant
corporations
and
web
too.
D
So
we
want
to
try
to
make
that
a
little
bit
more
standard
way
of
of
doing
things,
and
you
know
not
have
to
be
beholden
to
the
big
corporations
and
again
it
doing
the
data
science
side.
It's
it's
really
nice
doing
that
stuff
on
on
fluents.
You
guys
are
using
marine,
which
is
a
nice
high
or
I'm
sorry,
are
using
rust,
which
is
a
nice
high
performance
language,
putting
it
into
wasm
modules,
which
are
really
high
performance,
but
can
run
on
any
platform.
You
don't
need
specialized
hardware.
D
I
think
there's
a
really
good
opportunity
for
us
to
deploy
this
code
out.
You
can
spin
up
lots
of
nodes
very
quickly
when
you
need
a
lot
of
compute
all
at
once
and
then
drop
them
back
down
so
that
you're
not
paying
for
them
when
you're,
not
using
them.
It's
a
really
nice
combination,
doing
data
engineering
and
data
science
on
the
fluence
platform.
I
think
it's
going
to
be
really
exciting
for
us.
A
I
do
so.
It
was
not
very
long
ago
when
you
partook
in
the
hackathon,
basically
neuter
us
new
to
fluence
need
to
react.
If
I
remember,
I
mean
new
to
everything,
I
think
the
google
sign
up
form
was
probably
the
only
thing
you
recognized
when
you
entered
the
hackathon
and
which,
which
is
great
and
now
you're,
basically
working
on
building
distributed
decentralized
data
pipelines
and
just
I'm
just
personally
interested.
How
do
you
I
mean
you
accelerated
your
growth
in
this
space
very,
very
quickly.
D
D
Yeah
so
so
the
the
work
itself
wasn't
new
to
us.
It
was
just
the
languages
and-
and
you
know,
we've
been
learning
languages
for
for
30
years,
so
learning
another
new
language
isn't
isn't
that
big
a
deal?
The
idea
that
the
process
flow,
the
distribu
decentralized
process
flow,
that
fluence
does
was
really
the
biggest
learning
experience
for
us
and
honestly
at
first,
when
I
first
looked
at
it,
I
I
I
didn't
get
it.
D
I
wasn't
totally
convinced,
but
then
I
started
looking
through
the
you
know
the
the
language
for
for
decent
for
distributing
the
the
sections
of
processing
and
it
started
to
make
a
lot
of
sense
to
me,
and
so
we
were
just
going
to
do
it
to
do
a
hackathon
to
to
just
kind
of
start
exploring
the
space,
but
it
turned
out.
We
really
like
what
you're
doing
and
we're
going
to
try
to
integrate
it
into.
You
know
into
our
long-term
vision
of
doing
data,
engineering
and
data
science
pipelines.
A
Yeah
this,
this
is
great.
This
is,
as
I
said,
like
we're
super
interested
in
data
pipelines
and
I
think
a
couple
of
the
announcements
that
are
coming
up
right
after
you
now
are
very
very
pertinent
to
what
you're
doing
and
definitely
want
to
obviously
stay
in
touch
with
you
and
support
your
in
your
in
your
journey
there.
G
F
F
F
Also,
we
improved
error
handling
and
every
call
return
streams
stream
of
the
errors
and
you
can
easily
debug.
Now
all
the
also
your
flow
with
registry,
so
latest
version
for
latest
stable
version
for
not
registry
franchise
in
aqua.
On
the
slide,
please
update
it
if
you
want
to
use
registry
in
the
stable
condition
and
thank
you
and
check
out
our
documentation
in
our
book
for
with
stuff
and
try
out
our
examples.
F
B
B
B
For
example,
imagine
that
you
have
a
user
and
this
user
has
many
devices
like
the
phone,
the
notebook.
I
don't
know
what
and
wants
to
get
some
updates
from
the
network
like
in
a
messaging
application.
For
example,
you
can
assume
that
there
is
some
label
for
the
resource
and
you
can
try
to
root
to
find
all
the
roots,
all
the
providers
of
this
resource,
and
in
this
case
the
resource
itself
is
just
the
presence
that
this
client
device
is
being
connected
to
the
network
and
you
can
find
the
route
to
this
device.
B
Another
use
case
is
when
you,
for
example,
want
to
deploy
some
some
data
holding
service
to
the
network,
and
then
you
want
to
find
this
service
after
you,
I
don't
know
rebooted
your
computer
or
find
the
resource
for
another
another
beer
like
if
you
create
your,
I
don't
know
avatar
service
for
just
avatar.
You
can
register
the
providers
like
fluence
nodes
with
ipfs
services,
with
this
avatar
image,
for
example,
to
the
registry
and
then
find
it
resolve,
get
the
image
another
example.
B
The
third
example
is
building
clusters
in
the
network,
like
replica
says,
for
example,
or
something
like
that
for
doing
even
load
balancing
in
some
cases.
In
this
case,
you
treat
the
resource
as
the
cluster
id
or
the
whole
cluster
is
the
resource.
The
resource
id
identifies
these
distributed
resources
and
you
can
find
that
all
the
providers
can
find
each
other
using
this
mechanics
mechanism
and
you
can
find
which
peers
are
currently
online
and
serving
this
cluster
for
example
so
hand.
These
three
examples
in
mind.
B
G
So
I
need
to
spare
your
data
yeah.
I
will
good,
do
you
see
it
the
presentation
or
no?
Yes,
we
do.
Yes,
we
do
yeah,
so
I
will
present
the
marine.js
milestone.
2,
as
we
are
going
through
this
story
in
this
another
zone.
I
will
we
will
support
the
arbitrary
single
module
services
in
fringe
jazz,
not
arbitrary,
but
with
like
fewer
models.
You
know
without
body.
G
But
we
have
some
problems
when
we
trying
to
do
it,
we
we
previously
could
we
could
write
only
javascript
services
for
french
js
or
we
could
wrap
a
marine
service
into
binderon
business
engine
for
us
and
compile
it
with
bioengine,
make
a
javascript
cracker
and
pass
it
to
finish.js
to
run
it.
So
it's
painful,
and
if
you
don't
have
some
service
that
needs
private
data
or
something
then
the
only
way
to
use
it
from
browser
is
to
pass
this
private
data
to
the
relay,
and
it
is
not
good.
G
G
G
It
runs
this
called,
whilst
in
investment
it
uses
interface
types
to
have
a
good
interface
like
to
have
ability
to
pass
structs
inside
button
and
receive
start
from
watson,
strings,
arrays
or
whatever,
and
also
it
provides
binaries
interface
for
binaries
like
you,
can
have
a
wrapper
for
a
curl
function
that
marine
will
create
for
you
and
provide
it
for
the
wassens.
So
you
can
call
the
binary
ecosystem
from
your
business
so
and
it
is
good
for
using
as
part
of
the
sun
software
like
trust
node,
which
uses
marine
to
run
services.
G
G
So
we
decided
to
reuse
four.
We
think
that
it
is
simpler
to
it's
faster
to
implement
and
we
could
improve
performance
in
future,
and
so
there
there
was
the
problem
we
couldn't
run
wasn't
in
wasmer
in
a
browser,
so
we
should
use
the
browser
native
web
assembly
on
time
which
is
accessible
from
javascript
webassembly
module.
G
G
G
G
G
G
We
need
that
module
name,
function,
name
and
arguments
in
the
sound
form,
so
we
also
have
an
example
of
how
it
looks
like
how
to
use
the
marine
js
inside.
We
can
have
a
service
that
we
have.
Some
structs
have
some
function,
that
gets
arguments
returns
some
values
so
and
how
we
call
it
in
many
using
maningers,
we
just
provide
module
name
service
function,
I'll
build
a
json
that
represents
the
arguments
and
get
get
the
result
check
the
error,
if
needed,
or
just
check
the
result,
if
we
sure
that
it
does
not
fail.
G
What's
interesting
that
it's
not
about
that
stupid
deceleration,
correlation,
digitalization,
transistor,
the
marine,
actually
checks
types
x
if
the
function,
if
the
types
match
the
function,
signature
and
return,
meaning
whatever
meaningful
errors
for
everything,
and
what
is
also
good
that
as
the
we
have
only
this
one
string,
we
can
dynamically
decide
what
service.
With
what
arguments
we
need
to
call
what
function
it
unlocks
many
options
for
the
ngs
implementation.
G
So
next
in
the
diagram
of
how
that
all
works,
so
fluency
receives
a
request
called
a
service
from
aqua.
Then
he
dispatches
the
service
to
friendship
service
which
finds
the
service
by
its
id
of
the
module
in
rangers
and
just
passes.
The
arcs
for
two
interface
types
to
check
the
signature
and
the
stuff
and
interface
types
work
with
a.
G
Js
was
interface,
the
wrapper
we
made
that
works
in
memory
organizes
the
puts
all
the
things
in
the
right
way
called
splashing
function
which
can
work
with
it
with
this
state
and
also
interface,
types
need
to
get
the
result,
and
it
can
also
also
contain
some
dynamically
allocated
things
or
complex
structures,
and
then
we
just
return
the
results.
All
the
way
back
and
the
fluencys
can
pass
that
service
result
to
aqua.
G
G
The
next
update,
which
which
we
will
announce
tomorrow,
is
running
arbitrary,
arbitrary,
pure
models
in
princess,
and
it
was
an
interface
for
it
that
I
showed
you
also.
The
future
work
is
supporting
widely
and
some
host
imports.
Maybe
imports
of
javascript
functions
that
are
user
needs
to
be
able
to
run,
and
then
the
last
is
full
support
of
multimodal
services.
So
basically
it's
the
most
that
javascript
can
have
and
it
will
make
javascript
appear
as
close
to
raspberry
as
possible.
C
Just
a
little
like
a
little
technical
detail
on
the
javascript
side
between
fluence.js
and
fluence
app
service,
I
wanted
to
know
that
actually,
the
app
service
is
riding
in
the
background,
so
this
call
dispatching
is
happening
to
either
web
workers
in
the
browser
on
worker
threats
in
node.js.
So
previously
we
had
performance
issues
with
aquavium
and
we
decided
to
run
all
awesome
stuff
in
background.
So
it
is
also
cool
that
you
can
run
all
the
services
in
background
without
sacrificing
the
performance
of
your
application
inside
browser.
G
If
the
connection
is
lies
outside
this
diagram,
when
aqua
vm
executes
on
some
node,
the
interpreter
runs
the
script.
It
goes
gets
the
result
that
shows
you
you.
I
need
to
have
results
from
this
service.
Called
this
service
called
this
service
call,
then
friendsjets
calls
the
services
gets
results
and
return
returns.
These
results
to
the
aqua
interpreter.
G
This
interpreter
executes
code
with
these
results
and
when
it
decides
to
that
all
the
work
on
this
peer
is
done.
It
returns
the
result
without
call
request
and
transgress
sends
the
particle
to
the
next
year.
That
will
be
executing
the
script
that
the
next
sphere
that
is
needed.
So
everything
on
the
diagram
is
on
your
computer
cool.
Thank
you.
A
G
C
So
at
the
moment
we
can
only
load
the
single
wasn't
files
and
register
them
as
services.
But
we
are
working
hard
on
implementing
the
full
potential
of
marine
on
gspr.
So
in
future
we
will
be
able
to
register
services
like
the
same
way,
for
example
using
aqua
or
aqua
cli.
So
it
would
be
possible
to,
for
example,
load
some
wasa
modules
to
ipfs
and
then
create
a
blueprint
from
them
on,
for
example,
an
old
gsp
and
that
instantiate
the
service
there.
So
it
would
be
possible
in
the
future.
B
I
may
add
something
something
like
the
the
use
case
for
this
milestone
and
like
what
we
will
probably
do
with.
It
is
a
scenario
like
the
following:
alexey
just
presented
the
updates
in
in
the
registry,
and
there
is
a
security
model
in
the
registry
when
you
need
to
make
some
bytes
from
the
structure
to
sign
these
bytes
and
the
logic
to
to
make
this
white
array
to
provide
the
signature
and
hence
to
to
check
the
signature
currently
is
implemented
on
rust
and
is
available
only
on
the
rust
nodes.
B
B
I
go
to
the
network
and
say:
okay,
I
want
to
use
it
and
inside
this
script
I
will
download
the
module
that
makes
these
bytes
locally.
It
will
be
a
pure
function.
I
will
deploy
it.
Maybe
I
will
use
the
custom
api
not
the
same
as
we
have
on
the
rust
node,
but
it's
not
to
be
used
from
aqua,
cli
or
something
like
that,
because
javascript
are
still
not
fully
featured
compared
to
the
rust
nodes.
B
C
C
C
C
C
C
C
So
it
is
pretty
much
these
lines
so
these
ones.
So
we
can
see
that
our
service
was
written
in
rust
and
but
then
we
deployed
it
to
fluence.js,
run
it
inside
the
node
and
have
it
working
okay.
Now,
let's
do
the
same
thing
but
inside
browser.
So
let's
go
with
another
example,
and
just
just
to
show
you
that
the
code
is
pretty
much
the
same,
we're
doing
the
same
thing,
but
we
are
running
it
in
the
browser.
C
C
C
C
A
C
Yeah:
okay,
okay,
that's
totally
true!
You
can
deploy
this
service
created
from
this,
wasn't
file
to
node
to
rust,
node
and
it
will
output
the
same
thing.
But
in
order
to
do
that,
I
have
to
create
first
a
blueprint
and
well
first
module
definition.
If
I
remember
correctly,
then
the
blueprint
and
then
I
have
to
deploy
it
to
some
node,
so
I
won't
be
able
to
do
it
right
now.
A
B
C
C
C
A
G
So
I
want
to
add
that
we
are
not
really
limited
to
rust
code.
We
use
sqlite
that
is
written
on
your
support
path.
I
don't
remember,
and
we
have
written
interface
types
adapter
ourselves
like
the.
It
is
a
simple
script
like
like
the
webassembly
commands,
and
we
embed
it
into
wasn't
using
our
marine
tool
and
users
use
this
as
collide
in
webassembly
in
works
in
js,
and
it
works
in
cross
nodes.