►
From YouTube: Mesos Developer Community Meeting (July 27th 2017)
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
A
D
A
A
D
G
Also,
can
you
toss
on
the
agenda
it'd
be
good
to
how
we
fit
what
the
frequency
knew.
We
wanted
to
keep
doing
with
every
tune
once.
G
G
Agenda
up
you're
really
well,
I,
don't
know
what
I'm
gonna
be
the
Sterling
kind
of
house.
It's
been
coloring,
the
last
yeah,
so
I
just
thought
like.
Maybe
if
we
do
it
less
frequently,
and
we
could
even
just
a
sign
out
forms
working
routes,
gonna
present
it
one
innovation
workers
can
present
it.
One
we're
gonna
film,
some
other
agenda
stuff,
it
might
help
help
got
cuz
I.
We
know
they're
out
there.
F
G
G
So
you
see,
as
I
mentioned,
there
was
probably
like
ten
eleven
folks
showed
up,
and
we
talked
about
some
of
the
current
performance
improvement
under
beam,
and
then
we
talked
about
some
of
the
future
ones.
We'd
like
to
do
and
tried
to
sort
of
slot
in
who
might
be
working
on
what
things
the
hour
went
by
super
fast.
Actually,
we
barely
I
mean
we
could
have
kept
talking
to
do
it
if
it's
talking
for
another
two
hours,
probably
about
stuff,
is
happening.
G
So
an
update
on
some
of
the
current
work.
That's
being
done
so
basically
I
would
say
the
work
is
sort
of
separated
into
two
I
guess:
I
call
it
two
main
things:
performance
improvements
in
the
process
itself
and
then
performance
improvements
being
made
sauce
and
then
each
of
those
kind
of
subdivided
as
well.
So
in
the
live
cross.
G
Like
the
actual
interfaces
into
the
process
like
I'm
like-
and
you
can
even
talk
a
little
bit
more
about
this,
but
basically
a
lot
of
our
dispatchers
and
defers
and
then
some
actors
are
on
futures
and
all
these
things
we
have
to
copy
a
lot
of
objects
around,
and
so
people
are
looking
into
how
we
can
move
things
using
C++
move
semantics
or
maybe
even
replace
them
with
something
like
a
share.
The
shared
pointer
pointers
then
third
place
is
in
the
general
HTTP
path.
So
we've
kind
of
been
optimizing.
G
Some
of
the
message
passing
paths
but
the
general
my
sheep.
They
should
keep
that
so
the
cool
thing
for
the
general
message
passing
optimizations
is
so
far
with
about
three
or
four,
like
major
optimizations,
there's
a
tremendously
huge
performance
improvements
so
for
one
of
the
benchmarks
that
M
that
we're
running
it
used
to
take
about
a
hundred
and
forty-one
seconds
to
run
the
benchmark.
G
Now
it
takes
them
on
our
max
about
four
seconds
three
to
four
seconds
and
on
the
Linux
box,
with
a
bunch
of
cores
like
seconds
or
so,
and
that's
just
like
a
ping-pong
and
that's
a
passing
thing,
which
is
that
it's
a
really
basic
ping
pong
create
a
bunch
of
actors.
Have
we
send
messages
back
and
forth?
So
that's
great,
you
know,
I
think
it's
some
low-hanging
fruit
performance,
optimization
perspective
and
that's
gonna
be
great.
G
Then
the
future
work
that
the
forest
working
group
started
talking
about
as
I
mentioned,
was
beyond
the
process
also
in
maysa
itself,
specifically
around
HDPE
handling
and
some
the
HDPE
stuff
and
Mase
us
and
I
forget
the
list
of
all
those
things,
but
there's
a
bunch
of
other
stuff
we're
going
to
look
at
and
do
they're
gonna
be
creating
benchmarks.
So
we
can
drive
the
optimizations,
which
I
think
is
great
and
I
think
in
a
future
one
we
can
present
what
those
benchmarks
are
and
talk
about
the
firm's
crude.
G
We
get
current
only
way,
I'm
the
only
host
we're,
definitely
I,
write
blog
posts,
yeah
some
of
the
performance
optimizations
we've
done
thus
far
on
the
processor,
are
pretty
interesting
as
well,
stressing
you
know
even
stressing
Linux,
semaphores
and
other
things
like
that
and
seeing
how
we
can
do
those
even
more
efficiently.
So
it's
been
cool
stuff
any
questions
again,
sir
I
said
it
was
super
high
level.
So
if
people
have
like
I'm
having
answered
94
questions
that
be
some
of
the
process
performance,
optimization
stuff
people
are
curious,
but
that's
your
high
level,
man.
G
Process
stuff
some
of
its
already
landed,
so
you
can
actually
turn
on.
You
can
actually
turn
on
like
right
now.
If
you
were
to
go,
take
Mesa,
so
you
could
compile
it
with
a
lock
free,
run,
cool
and
then
probably
the
end
of
the
week.
You'll
be
able
to
compile
it
with
a
lock
free
event,
queue
as
well,
and
then
you'll
also
be
able
to
compile
it
with
a
user's
face.
G
You
just
a
semaphore,
which
is
the
last
row,
optimization
everything
that
that
we're
doing,
because
we're
introducing
like
some
pretty
serious
stuff
or
already
reducing
all
of
it
behind
configuration
flags.
So
all
the
existing
code
paths
for
the
most
part
there's
a
couple
of
code
paths
that
are
slightly
different
because
they're
choosing
some
easy
optimizations
that
we
could
do
like
replacing
maps
with
have
perhaps
you
know
looking
things
up
at
it
in
the
hash
map
and
stuff
like
doing
a
count
and
then
doing
brackets,
which
are
basically
devil
workups
in
fast
paths.
G
This
was
really
add
up,
adding
a
few
double
check,
locking
idioms
certain
places.
So
all
that
stuff,
you
know
all
that
stuff
change,
but
for
the
most
part,
all
these
things
are
configured
behind,
compile
flags
so
that
over
time
we
hope
to
make
them
defaults.
But
right
now
we
can
let
people
keep
really
comfortable
about
once
heard
about
the
correctness.
J
When
there's
a
lot
of
events
and
well
I,
don't
know
exactly
what
that
is,
but
like
sometimes
you
had
some
performance
issues
and
also
probably
guys
also
there's
some
benchmarking
for
the
process
and
they
used
to
do
for
some
numbers.
Yeah.
G
J
C
G
G
G
That's
what
we
yeah
yeah
we're
gonna
you
eat
through
all
that
stuff,
yeah
yeah
the
end
of
message
passing
benchmark
the
hundred.
So
when
you
go
from
one
hundred
and
forty-one
seconds
down
to
a
couple
seconds,
you
basically
go
from
a
hundred
thousand
messages
per
second
to
about
three
or
four
million
messages
per
second,
so
significant
amount
of
throughput
increase
that
you
get
make
those
changes
right,
yeah
and
again.
D
I
J
Yeah
I
think
attendance
is
pretty
good
I
wrong,
like
10
people
every
time
yeah,
so
we
have
like
Apple
Twitter
folks
today,
I
you
might
like
Microsoft
guys,
because
we
want
to
discuss
something
about
the
ownership
stuff,
bony
ownership,
that's
kind
of
relevant
to
the
backers
of
the
guys.
So
you
might
the
Microsoft
the
guys
to
the
meeting
as
well
today
so
I.
So
we
have
a
listen
notes.
I
think
we
have
a
planning
spreadsheet.
That's
why
that's
what
we
do!
So
that's
pretty
much
like
all
the
things
that
we
want
doing
that
area.
J
J
The
Devonian
is
your
stuff
right
now,
like
a
kind
of
legacy
issue
where
I
we
always
change
the
owner
of
the
flan
to
be
the
same
as
a
container
user
as
things
gross
like,
for
example,
when
I
support,
docker
containers
contain
yourself,
can
define
user
which
might
be
different
than
the
host
user,
and
also,
we
add
more
type
of
volumes
like
super
volume,
sandbox
volume,
dr.
longing
and
some
other
volumes,
and
we
and
also
Apple
guys
at
this
like
share
along
and
support
those.
J
I
think
we
have
two
discussions
so
far
in
the
working
group
and
we're
winning
for
Microsoft
the
guys
to
get
back
to
us
from
Windows
perspective
on
this,
so
that
we
can
have
a
design
I
think
if
we
go
to
them
notes
I
think
we
kept
our
design
dog
here
so
far.
So
this
one
this
one,
it's
kind
of
not
a
like
exactly
sign
dog
but
like
it's
kind
of
like
not
dumped,
we
go
to
this
one.
J
We
try
to
capture
the
problem
and,
and
the
user
stories
want
to
support
and
try
to
figure
out
a
solution
that
can
satisfy
all
these
user
stories
for
that
yeah.
So
why
ownership
stuff
is
what
we
discussed
recently
and
I
can
previously
discussed
the
CPU,
pinging
I
think
the
first
few
working
groups-
and
we
also
just
got
lucky
post
hook,
I-
think
that's
led
by
a
poll
that
they
want
to
do
this,
an
isolator
that
can
inject
arbitrary
commands
before
container
start
and
after
container
stop.
J
They
want
to
inject
random
commands
and
they
don't
want
to
build
an
isolated
I
want
a
generic.
Actually
that
can
just
allow
people
to
specify
shell
command.
Looks
like
that
and
I
think
Jamie
speeches
demo,
the
the
in
head
capability
support
in
one
of
the
meetings.
I
think
the
recent
things
we
get
on
made
his
capability
is.
We
can
first
of
all,
we
support
him
beyond
capabilities
right
now,
which
means
you
can't
have
an
unruly
users.
Have
a
certain
capability
like
glass.
This
is
the
main
you
can.
J
J
The
host
for
isolation
is
like,
before
those
container
on
the
post
that
Network
each
each
I
wants
to
the
SAS
or
Coast
port
right
now
there
is
no
isolation
between
them,
which
means
that
container
that
not
being
a
sign.
He
given
port
can
actually
this
along
that
port,
because
Colonel
don't
provide
any
isolation
for
the
host
network.
J
J
Do
plan
to
yeah
I
do
plan
to
so
we
have
come
on
board,
you
know.
Do
we
have
a
common
work
for
four
I
probably
want
to
do
that
to
properly
label,
have
a
special
label
somehow
to
to
to
make
sure
those
tickets
show
up
on
the
comment
board
rather
than
using
us
I
mean
it's
just
easier
for
me
to
use
a
spreadsheet
at
the
moment.
I
J
It's
pretty
much
like
the
people
who
asked
for
this
feature
this
side
of
urgency.
Pretty
much
like
a
users
say:
hey
look
at
the
definition
here,
so
T
zero
means
I
need
that
in
production.
Right
now
without
that
affects
my
customers,
she
wants
that.
The
feature
has
been
planned
for
the
next
six
months.
We
plan
to
roll
out
in
production.
T2
is
nice
to
have
I.
Don't
have
any
plans
to
use
that
feature.
So
that's
how
I
categorize
those
features,
so
TZ
is
something
that
either
a
bug
or
some
cool
name
clapping.
J
Yeah,
we
also
discuss
a
bunch
of
stuff
I
think
the
agenda
either.
Next
time
is
for
the
McGill,
burries
and
I.
Think
we
discussed
one
time
by
three.
We
should
continue
discussion.
One
thing
that
we
want
to
do
this
Oh
see
I
support,
given
that
OSI
is
one
point
all
right
now
last
week.
So
that's
the
thing
we
should
do.
Asap
to
Jimmy
and
one
thing
I
really
wanted
to
do
is
I,
don't
wanna,
just
be
okay
under
the
OCI
store
there.
J
J
We
have
cache,
therefore,
that
your
artifacts
I
do
want
to
leverage
that
especially
OC
I
defined
this
very
nice
like
ass,
can't
content
base,
address
content
and
dress
were
content,
addressable
storage
things,
so
so
we
do
want
to
see
if
we
can
unify
all
the
artifacts
1920
some
users
using
that
as
a
kind
of
foundation
to
build
up
I
would
see
on
support.
So
that's
kind
of
discussion
went
into
next
time.
J
I
don't
know
immediately,
but
people
are
pretty
excited
about.
The
OCR
OSI
is
supposed
I
mean
initial.
The
OCI
spec
right
now
is
very
compatible
with
stalker,
so
I
have
a
feeling
that
doctor
not
just
Sapporo,
see
I
in
dr.
Akaka
relative,
so
I
was
they
did
that
then
I
can
see
the
reason
that
people
will
move
to
that
model.
Ocaml
right
now.
The
fact
that
he
knows
the
images
back
and
doctrine
is
madness.
J
One
single
pretty
much
just
one
single
media
type,
the
BND
like
Oh
CSV,
Amiga
OC
eye
doctor
to
be
ng
da
darker.
So
that's
the
only
difference
pretty
much
so
I
can
see,
and
also
there's
some
like
publicly
supporting
the
registry
already
cause.
Yes,
so
how
see
like
people
trying
to
meet
you
that
pretty
soon
and
doctor
will
support
at.
J
Well,
I
mean
I,
mean
I,
wanted
average
this
chance
to
to
support
us
unify
artifacts
and
which
I
think
is
super
valuable
for
us
now.
It
can
simplify
the
GC,
because
right
now
would
have
different
GC
for
Thatcher
and
different
GC
for
it.
Unfortunately,
I
would
have
to
do
that
twice.
I
other
times
on
OCI
store,
there's
urgency
to
be.
D
D
J
J
I
think
we
did
a
lot
of
discussion
that
there,
because
this
is
something
not
new
people
have
been
asking.
You
have
me
asking
this
feature
for
a
long
time
and
actually
like.
We
also
took
a
look
at
what
kubernetes
does
in
terms
of
that.
I'm
sewing
an
interesting
thread
there,
where
I
got
got
shut
down
by
four
people
initially
work,
so
Google
has
some
poor
folks
that
has
those
experience
I
think
they
mentioned
I
port
used
to
have
what
used
to
do.
J
One
thing
that
they
so
one
way
to
do
that:
I
support,
CPU,
Keaney
and
gonna
kick
for
given
box
and
reserve
some
things
for
dedicate
course,
and
the
rest
you
like
in
your
fair
sharing
or
first
of
all,
like
GFS
and
the
futures.
But
one
comment
from
that
discussion
thread
is
like
poor.
People
are
very
regret
on
that.
J
The
desperate
wants
to
revert
that
change
to
a
stay
where
there's
no
like
a
partition
on
the
know
like
what
part
of
that
aesthetically
first
part
of
dies
like
Cola
and
they're
sharing
stories,
but
they
don't
mention
the
reason.
Why
is
that?
But
the
best
pre
regret
that,
and
they
want
to
revert
that
changed
using.
So
we
discussed
like
kind
of
like
that,
the
rationale
like
it
was
Croton
accounts
on
that
special
type
a
week.
J
J
Yes,
yes,
if
you
look
at
a
dog
request,
yeah,
if
you
look
at
dog,
it's
an
abstraction
layer,
I
think
what
they're
suggesting
like
proposing
is
I
have
a
affinity
info
inside
container
info.
Basically,
the
info
saying
that
Oh
for
this
container
I
want
all
the
profits
of
that
container
to
be
honest
in
core
or
on
the
same
socket
on
the
same,
even
though,
if
you
can
see
the
mean
up
here
so
basically
like
you
specify
which
CPU
you
want
to
use,
you
specify
was
affinity
requiring
your
container.
Oh
yes,
so.
D
J
J
J
That's
that's
true,
and
by
Intel
yeah
I,
don't
I,
don't
think!
That's
something
that
I'm
kubernetes
is
gonna
stuff
like
accept
shorter.
M
J
Like
that,
the
feedback
game
is
like
I,
don't
think
the
talk
has
sufficient
background
information
for
us
to
make
a
decision
like
that,
I
think
the
action
now
that
means
captioning
and
has
anyone
actually
run
red
marks.
Yeah
I
think
they
have
some
benchmark
to
show
how
much
performance
influence
we
can
get
by
running
a
few
line
program.
But
it's
like
a
micro
benchmark.
It's
not
very
that's
how
many
little
misses
I
think
he
captured
the
using
the
perfect
campaign
to
thread
on
the
same
I.
Think
that's
a
good
one.
J
J
Yeah
I
think
we
were
right
now,
focusing
on
the
tier
0
things
at
the
moment,
and
once
we
address
those
we're
gonna
move
to
you
on
yeah
and
one
thing
I
think
I
definitely
help
is
documentation,
I'm
mark
as
t0
I.
Think
the
right
on
documentation
is
pretty
weak
on
missus
container
risers,
so
it's
pretty
bad.
J
So
in
the
back
I
mean
the
easiest
thing
you
can
do
for
the
projects
we
improve
documentation
and
so
far
I
think
we
don't
have
an
owner
because
I
don't
think
we
have
time
for
that,
but
I
think
the
going
I
mean.
Maybe
one
way
is
like,
while
we're
cleaning
up
stuff
like
the
longing
stuff
and
graduate
included
documentation
for
that
part,
so
that
we
can
hopefully
in
the
future,
that
we
can
complete
documentation
for
mrs.
green
lights.
Now
that
I
can
see
issue.
J
N
N
O
A
Both
just
share
whatever
we
so
the
thing
that
I've
been
working
on
with
Jay
in
New
York
is
starting
up
a
regular
mate
system
newsletter.
We
can
publish
just
so.
We
have
like
a
steady
stream
of
content
going
out
on
the
mates
goes
blog,
so
I
sent
out
a
few
emails
and
we
all
sent
out
a
few
emails
miniver's
to
try
to
gather
content
and
New
York
was
going
to
assemble
the
first
issue
that
focused
on
storage
work,
so
we're
trying
to
gather
contributions
from
folks
who
have
been
working
on
storage
stuff.
A
A
A
A
I
D
P
D
So
these
are
all
the
different
distros
that
we
considered
central
outside
of
the
clinics,
well,
the
door
to
Debian
and
the
art
sites,
so
the
red
ones
here
are
past
their
end-of-life
support
from
wood,
so
I'm
considering
them
to
not
be
worth
supporting
polo
for
a
1404
which
are
super
old,
but
they're
LPS's,
so
they
still
work
all
of
them.
All
of
these
have
these
commands
that
I'm,
like
manually,
run
on
rocker
images
and
actually
built
NATO's
answers
with
deep
motive.
So
the
green
here
is
not
just
this.
Compiler
is
available.
D
D
No
was
reach
back
yet,
but
I
reached
out
to
Cody
and
know
that
I
could
actually
just
create
watch
instances
on
AWS,
so
I
think
I'll
just
do
that.
The
test
press,
those
everything
else
works
just
fine.
There
didn't
even
goes
like
to
be
the
density,
which
is
actually
either
support
or
insect
Hudson
problems
currently
with
Apple
play.
I
want
to
get
these
records
to
whether
people
would
be
okay
with
installing,
like
a
newer,
client
or
actually
stock.
D
Is
just
essentially
is
all
attitude
has
been
pushed
back
in
the
past,
because
people
thought
that
issues
worked
out
of
the
box,
but
look
at
the
for
saying
that
all
most
of
these
other
distros,
we
need
to
fetch
some
problem
in
I.
Don't
really
see
why
voicex
pencil
tomb
in
there
so
why
preference
would
be
to
say
just
install
your
I,
don't
work
on
crime,
I.
N
D
G
D
Think
I
think
for
that
stuff,
winching
I
thought
this
can
teach
about
it.
I
think
the
best
better
way
to
do
that
would
be
to
just
have
a
fight
for
my
my
area,
that's
separate,
which
is
fixed
for
everyone,
as
opposed
to
everyone
having
to
catch
a
specific
version
of
high
quality,
but
he
makes
me
say:
3:9,
like
you
know,
all
these
different
distro
is
finding
a
specific
version
of
client
would
be
pretty
difficult
so
for
a
quick
format.
D
I
think
we'll
have
a
project
specific
version
that
we
ship,
so
so
it's
over
yeah
I
thought
that
it
really
in
relation
to
the
funds,
all
that
stuff
I
think
it's
not
really
big,
but
it
would
definitely
be
much
easier
if
we
just
saying
we
all
support
humble
right.
So
that's
what
we're
leaning,
I'm!
Okay,
though
out
of.
D
D
D
D
H
D
Just
how
would
we
think
about
backwards?
Compatibility
is
like
you
know.
If
we
have
language
changes
that
are
coming
up,
and
then
you
have
first
to
actually
investigate
the
language.
Changes
are
actually
smaller
than
other
books
that
are
about
to
be
there
like
that
cycle,
cable,
all
like
putting
all
that
foreign
language,
survivability
is
and
or
that
level,
because
people
have
to
reevaluate
reevaluate
their
code
bases
and
stuff
anyway,
as
far
as
the
community,
this
opposes
community
from
the
surveys
I've
seen,
most
people
are
still
running
on
G,
60
or
80
or
90.
D
D
D
M
D
What
was
the
file
system
library,
so
the
file
system,
library
from
boost
has
been
essentially
fixes,
but
right
now
we
do
a
lot
of
file
system
stuff
in
stone
and
windows,
compatibility
and
all
that
stuff
and
andrew
has
been
wanting
to
use
boost
file
system
for
a
long
time,
but
because
of
the
binary
dependency,
because
it's
kind
of
arsenic
president.
No,
it's.
F
D
G
G
I
was
tryna,
don't
work
about
this
earlier
today
as
well.
You
have
dr.
images
for
a
bunch
of
different
things.
We
build
on
right.
Can
we
click
those
docker
images
and
even
the
darker
files
someplace
in
public,
because
right
now
and
you
go
to
like
build
meso.
So
when
you
go
to
the
docs
page,
it's
like,
oh,
and
if
you're
on
this
thing
run
these
seven
commands
and
you're.
Can
this
thing
run
these
eight
commands
and
this
thing
on
these
nine?
G
M
E
I
J
B
J
G
Hutton
a
so
suppose,
so
are
we
good
just
put
those
up
there,
as
people
could
at
least
like
pull
down
and
try
to
build
in
flavors?
We
can
even
have
ones
where
race
is
already
built
inside
people
could
just
play.
Obviously
there's
things
out
to
do,
but
it
has
obviously
sink.
We
can't
see,
but
if
that's
the
cases
were,
let's
do
the
same
with
VMs.