►
From YouTube: 2019-07-23 :: Crimson SeaStor OSD Weekly 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).
A
Okay,
let's
start
and
last
week,
I
will
testing
predict
an
exchange
on
CBT
which
allows
us
to
be
too
wrong.
On
a
single
note,
and
but
today,
I
I
was
running
to
the
dirted
failure
caused
by
the
mission,
support
of
Duty
connection
and
then
I.
How
were
the
reviewing
chambers
change
it
to
to
put
the
full
story
into
alien
world
but
I?
A
In
long
terms,
so
I
I
tried
to
try
to
port
rocks
db2
sISTAR
because
the
sISTAR
and
offers
the
estate
of
a
thread
which
allows
try
to
rob
to
to
it
cute
applicable
coil
and
when
it
caused
the
probe
block
the
core,
it
will
switch
out
using
a
new
context,
and
that
says
I
think
that's
exactly
what
we
need
and
I'm
working
on
it
at
this
moment.
So.
B
A
B
That
this
is
a
way
bigger
deal
like
it
has
expectations,
but
how
threads
work
that
are
actually
common
between
Windows
and
Linux
and
has
actually
patient
github
blocking
syscalls
work
that
are
genuinely
common
between
Linux
and
Windows.
What
you're
talking
about
is
totally
new
it.
The
idea
to
try
to
get
this
merged
upstream
I
think
so.
So
why
isn't?
Why
is
that
better
than
simply
doing
that
up
at
the
object
store
interface,
boundary,
creating
a
layer
right
above
object
store
where
anytime,
you
try
to
make
a
blocking
call,
it's
simply.
A
A
B
A
B
B
You
spit
up
a
second
that
actually
runs
the
blue
store,
binary
wrapped
with
some
kind
of
shared
memory.
I/O
cue,
it's
not
different
from
what
you
do
with
two
threads
performance,
wise.
Potentially,
it
might
be
a
little
slower,
I
guess
what
I'm
saying
is
we
had
a
choice
between
investing
engineering
work
in
making
blue
store
work
well
and
Caesar,
or
we
could
choose
to
invest
it
in
a
new
store
that
would
actually
work
well
with
you.
B
C
A
B
B
B
B
C
B
It's
it's
still
like
engineering
work,
but
at
least
it's
not.
You
know
shared
for
this
right.
It's
just
a
new
little
everything.
It
took
its
overhead
like
we're,
not
gonna,
get
honest
performance
numbers
receipts
are
out
of
Oedipus
because
it
will
be
slower
that
that
communication
overhead
will
matter,
but
it
was
a
test
on
a
real
object
or
a
limitation
before
we
have
people
just
correct
the
subject.
My.
B
A
C
B
C
B
Blue
sure
is
not
gonna.
Get
us
encore
that
colors.
Those
are
like
installations,
where
your
goal
is
just
a
lot
of
hard
disks
right.
So
you're
not
you're,
not
going
to
dust
in
the
court
for
OSD.
That
would
be
very
expensive,
but
also
in
those
contexts
you
don't
really
care
about.
I
forbid
right.
You
can't
drive
it
through
the
sibyl's
anyway,
so
I
guess.
The
question
is:
how
hard
would
it
be
to
build
something
like
that?
And
if
the
answer
is
hard,
then
it
makes
sense
to
do
other
stuff.
A
B
Mean
like
later
on,
if
we
and
it's
also
possible
that
ceased
or
might
not
have
trouble
supporting
disks
or
spinning
disks
we
made,
it
may
turn
out
that
actually
it
works
fine
on
the,
but
only
minor
tuning
changes
right,
we're
not
going
to
design
for
it.
But
if
it
happens
to
work,
then
we
don't
need
rooster
at
all.
A
B
It's
not
it's
an
interface.
It's
it's
a
it's
a
layout
thing
right.
We
could
make
choices
with
the
metadata
layouts
that
simply
don't
work
for
hard
disks,
in
which
case
it
might
not
make
sense
to
try
to
add
a
second
vote,
or
it
could
turn
out
that
our
choices
actually
don't
have
to
lie
far
though.
Okay,
so
Walkley
yeah.
B
D
A
B
A
B
Because
they'll
totally
grant
you
that
contact
switching
between
processes
is
more
expensive
than
process
in
contact
switching
between
threads
but
context.
Switching
between
threads
is
already
very,
very
expensive.
The
marginal
cost
of
doing
it
between
processes
is
not
that
much
higher
and
that's
not
likely
to
do
the
dominating
cost
case.
But
the
real
argument
is
more
that,
even
if
we
completely
converted
blue
store
to
work
and
cease
or
it
well,
it's
not
obvious
to
me
at
least
that
that's
what
we
want.
There
might
be
some
devices
where
it
was
appropriate,
but
persistent
memory.
B
A
B
A
A
A
A
A
A
A
C
B
And
so
I
haven't
looked
at
the
interface,
but
is
the
end
of
the
interface
in
rocks
to
be
those
are
in
terms
of
blocking
halts
right?
Yes,
so
can
we
implement
an
end
that
does
what
you're
describing
under
rocks
to
me
so
that
rocks
we
never
needs
to
know
that
doesn't
require
change
trucks
to
be.
That
would
be
great
right.
No.
A
A
C
C
Huge
issue:
if
those
new
taxes
are
not
contended,
maybe
we
could
live
with
them.
Well,
if
a
mutex
is
just
taken
and
it's
an
containers,
all
it's
done
is
basically
a
technical
operation.
There
is
no
blocking.
There
is
no
fuel
access
call.
The
question
is
whether
those
mean
that
are
really
continental,
that,
if
not,
maybe
maybe
in
an
optimistic
case,
we
could
live
with
them
or.
A
B
A
C
B
A
B
A
B
But
if
they
accept
it
upstream
and
they
might
I
mean
there,
I
I
think
there
have
been
a
couple
of
efforts
to
run,
rocks
TV
in
C
star
that
were
like
research
type
type
deals.
It
sit
there
like
a
level
TV
implementation
in
the
C
star
tree,
or
is
that
something
else
you.
C
A
A
B
A
B
B
B
The
real
question
is:
can
we
avoid
a
copy?
That's
that's
the
part.
That's
really
going
to
hit
us
like
if
we're
accepting
a
right,
and
we
have
to
do
a
second
memory
copy
over
into
a
shared
memory
region
that
will
actually
hurt.
So
the
trick
will
be
that
when
we
read
off
the
wire
we
read
into
the
shared
memory
space
to
begin
with,
however,
we
will
likely
want
to
do
that
for
several
different
kinds
of
I/o
bagans.
B
C
B
C
C
B
C
B
B
I'm
not
super
worried
about
it,
because
we're
not
that
far
along
to
designing
Feaster,
I,
guess
I'm,
just
trying
to
game
out
the
obvious
costs
of
running
in
a
second
process,
and
let's
put
it
another
way,
if
the
obvious
cost
can
be
solved
by
doing
a
thing
we
want
to
do
anyway,
then
I,
actually
like
it
more
at
the
target,
because
mitigating
those
effects
would
have
been
things
we
wanted
to
do
anyway.
Yes,
however,
it
leads
us
into
doing
development
on
things
that
don't
help
us
in
the
future.
Then
it's
less
attractive.
B
B
Is
whatever,
because
we're
not
we're
not
going
to
use
boost
or
in
in
the
long
term,
for
those
sorts
of
latency
constrained
workloads
anyway?
And
it's
going
to
be
lost
in
the
noise
I
think
or
it
won't
be
internal
sure
not
to
have
been
a
bad
idea.
But
it
little
bit
a
bad
idea
that
one
of
the
mad
expensive
to
explore.
B
B
A
B
B
B
B
B
B
B
A
B
B
B
B
A
Know
what
I
mean
what
I'm
talking
about
is
the
shared
share
coda
base
in
the
final
in
the
infinite
eight
on
my
approach,
we
only
need
to
switch
boost
or
to
sit
all
right,
blue
store
and
the
we
don't
need
to
touch
the
supporting
color
base
or
two
two,
two
four
to
add:
if
defying
alien
a
temple
store
and
have
three
copy
of
done,
freak
three
condition
of
them
in
the
long
term.
If,
if
we
call
with
my
approach,
we
only
have
two
of
them
like
we
see
star
on
without
system.
B
A
B
B
A
In
both
cases,
we
work
through
the
coda
based
over
the
store,
but
but
the
difference
is
that
with.
If
we
could
go
into
a
new
world,
we
need
to
change
everything
change
either.
For
example,
we
need
to
change
the
safe
context,
support
two
tablets.
Three
cases.
Why
is
the
classical
team?
Oh
yes,
when
you
don't
have
a
three,
those
three
of
them
to
avoided
the
timber
confliction
and
the
tool
to
allow
the
castle
tricycle
of
T
and
sorry
crimson,
Latif
and
other
a
lot
of
demons
to
cause
a
stir
in
the
sting
ripple.
B
B
A
B
B
D
C
D
D
These
two
old
existing
together,
for
example,
proof
can't
puff
counter
staff
contacts
configure
proxy.
They
are
different,
but
you
have
to
build
into
the
one
binary,
so
we
couldn't
use
just
a
155,
we
start
to
add
we
find
the
code.
We
have
to
use
true
macro
definition,
because
all
the
chromaticity
compared
and
they're
the
waste
system
and
help
tell
the
code
if
it
is
filled
at
the
blue,
stop
for
the
crimson
OS.
You
have
to
check
another
macro
definition.
D
D
D
D
D
B
B
D
B
Basically,
it
shouldn't
be
the
case
that
there's
anything
in
common
between
crimson,
common
and
stuff
Commons.
You
should
have
a
third
thing:
that's
like
agnostic,
common
or
something
we
got
away
with
it.
We
got
away
with
the
crimson
common
because
we
weren't
trying
to
link
back
to
anything
else,
but
but.
B
B
A
B
B
D
C
A
A
A
B
A
B
We
stick
in
a
different
concept
that
makes
this
whole
conversation
to
propose.
You
build
an
entirely
separate
binary
that
finds
its
own
stuff.
There's
no
conflict,
no
problem
right.
We
don't
have
to
fix
any
of
those.
We
can
put
it
off
if
we
want
to
be
in
the
same
namespace,
though
we
probably
shouldn't
spread
it.
These,
like
conditional
namespacing
throughout
this
code
base.
That's
gonna
be
weird
right.
C
A
So
I
think
we
need
to
determine
some
intermediate
conclusions,
so
we
have
a
like
to
discuss.
Problem
is
currently
y22.
Co
is
the
Le
right,
Pluto
or
not,
and
properly
how
to
reach
resolve
the
problem.
If
we
go
with
Illinois
sr-22
how
to
isolate
the
different
symbols
just
regarding
to
the
first
one,
we
we
have
a
an
idea
to
separate
the
classical
talk
into
a
different
process,
but
we
need
to
care
about
to
be
cared
about
it.
A
Technology
introduced
some
unacceptable
performance
equation
and
I
and
I
will
push
the
a
change
to
github
to
show
you,
so
you
can
have
idea
what
they
were
close
to
portal.
Sorry,
Doc's,
db2,
sitar,
patella,
Durga,
I.
Think
the
motive
kind
of
thing
is
to
talk
to
convince
the
upstream
to
get
is
merged,
it
could
take
a
longer
and
then
another
problem
is
is:
if
we
want
go
with
the
who
is
the
push
to
to
analyze
the
posture
we
need
to?
A
We
have
two
ways
why
to
to
introduce
the
decent
namespace
to
to
protect
the
classic,
the
the
share
infrastructure
used,
pipe
criminality
and
the
cross
eulogy.
Another
way
is
it
properly
to
to
find
a
way
to
to
hide
the
classical
symbols,
to
use
a
PI
by
L
in
either
boost
or
when
you
do
something
static
investigation
into
that
right.
D
B
B
A
C
A
B
A
B
A
D
A
But
another
way
is
that
way:
we
have
a
different
but
convenient
convenient,
convenient
library,
for
example.
It
could
be
named
analyzed
pool
tour
which
only
exposes
necessary
symbols
consumed
by
by
chromosomes
d,
but
it
won't
expose
any
it
will
linked
against
safe
comment,
but
it
won't
expose
any
symbols
exposed
by
ship
common,
unlike
other,
unlike
other
part
of
the
system-
pockets.
The.
B
B
Any
class
that
gets
passed
between
crimson
and
will
store
has
to
be
in
a
wheel,
honest
to
god,
common
library
that
the
hey
it's
right.
That's
the
only
tricky
part
but
I
think
it's
just
zap
context,
and
that's
it
right
that
in
the
config
stuff,
something
something
has
to
take
the
config
out
of
crimson
and
shove
it
into
blue
swore,
whatever
that
it
has
to
know
about
those.
B
D
B
Okay,
as
long
as
crimson
only
ever
talks
about
its
own
sub
context,
blue
store
only
talks
about
its
own
stuff
context
and
the
thing
that's
responsible
for
passing
the
one
to
the
other
knows
about
both.
That's
that's
the
trick.
So
I
think
you
would
have
to
separate
the
safe
context
machinery
out
into
its
own
object
with.
E
B
B
E
B
C
A
D
Implemented
the
datastore
API
used
currently
and
I
found
that
in
the
size
OSD
and
they
don't
they
have
for
each
each
object
store.
They
have
its
own
collection
definition,
so
they
don't
ex
post
collection
to
the
OST.
The
gesture
exposed
something
that
collection
handle.
It's
the
pointer
is
other
the
parent
class
for
the
collection,
so
also
each
object
store,
has
its
own
collection,
implementation,
so
kind
of
a
is
where
you
to
see
see
a
style
collection.
It's
their
problem,
so
I
have
to
add
the
back
at
the
wrapper
for
the
collection.
A
C
Dbt
I
finished
the
Python
work
it
sent
in
addition
to
the
one
period,
the
local
mode
for
CBT
Center
last
week,
I
also
issued
another
pairs.
One
is
with
the
perf
stat
integration
and
three
part
and
rework
of
the
monitoring
infrastructure
infrastructure
in
CBT.
Another
one
is
very
simple
example
of
a
CPT's
configuration
file.
It's
already
matched
I'm
I'm.
Switching
back
to
the
introductory,
however,
I
would
also
meet
a
new
meal
topics.
C
A
A
A
C
E
C
A
C
Okay,
so
I
don't
know
well
well,
you
lost
me.
I
have
a
networking
issue
you
what
I'm
saying
that
what
we
decided
to
do
this
week
is
work
on
the
messenger
back
off
and
I'm
doing
that,
while
waiting
and
trying
the
handling
suggestions
that
we
talked
about
I
hope
to
have
enough
code
today
to
show
the
radical
work
with
him
on
it.
That's
the
main
idea.