►
From YouTube: 2019-12-24 :: 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
A
A
A
B
A
C
A
It's
a
it's.
A
very
large
change,
I
mean
a
large
addition
to
crimson
subtree.
One
thing
I
come
into
my
mind,
is
that
we
are
using.
We
actually
have
a
very
large
refectory
in
this
area
in
you
know,
in
classical
tea
actually,
which
tried
to
unify
the
two
commands
and
a
sock.
So
the
tail
command
can
write
it
right
through
to
the
service
fight,
Oh
Steve,
bypassing
the.
A
City
new
fire
to
commend
us
come
over
the
map.
Let's
put
this
way,
and
the
same
applies
to
the
a
sock
approach,
though
I'm
trying
to
figure
out
how
we
can
have
this
at
this
part,
building
two
crimson
as
well,
because,
ideally,
we
need
to
be
compatible
always
with
the
class
ability
and
I.
Think
it's
a
very
beautiful
feature,
because
one
is
able
to
pass
the
petal
a
single
command
to
a
given
Oh
Steve,
without
talking,
without
the
need
to
talk
to
the
underlying
basic
right,
very
useful.
B
B
To
accept
this
appeal
and
then
move
the
command
interface
to
use
the
same
mechanism
and
try
to
combine
it
to
push
it
or
to
merge
it
somehow,
with
the
classic.
Is
the
I'm
not
sure
it's
possible
because
there's
a
lot
of
mr.
code
just
a
way
of
holding
there,
but
maybe
copy,
like
I
started
to
do
copy
the
idea
of
insulating
everything
into
this
into
these
commands?.
B
B
A
B
If
you
saw
one
of
the
new
comments
I
had
in
the
new
version
of
the
ER
I
heard
a
problem
when
trying
to
abort
the
OSD
when
trying
to
use,
for
example,
a
SIGINT
on
the
OSD
and
get
it
and
get
it
to
terminate
nothing
gay
wasn't
working
right,
I
tracked
it.
What
I
did
this
week
was
to
track
it
down
to
mishandling
of
acceptable
by
sister,
at
least
when
equal
is
they
used
and
I?
B
B
B
I
think
it's
not
good
I'm,
not
sure,
yet
that
if
it
is
compatible
or
not,
and
what
every
date
is
combined
is
we
do
the
UNIX
domain,
sockets
and
and
every
other
socket
and
the
unifying
the
handling
the
way
I
didn't
want
to
do
originally
so
the
code
is
now
is
not
make
a
change
in
a
major
way.
Maybe
it's
still
I'll
test
it
and
I'll
check.
If
my
if
my
change
is
still
applicable,
if
so,
I
try
to
push
it.
If
not,
it
will
be
a
local
language.
A
Kind
of
logo
change
on
top
father's
aunt
/
old,
oh
the
branch
of
the
DBZ
star.
Otherwise
we
need
to
have
our
fork.
We
will
have
some
local
commit,
but
that's
not
relate
to
the
core
code,
others
till
the
DB
or
sister
on
its
own,
really
to
how
we
handle
the,
for
example.
The
adjacent
sanitizer
change
in
the
same
excrete,
but
we
don't
touch
the
the
total
edge
or
towards
CC,
is
led
B,
so
I
would
suggest
we
avoid
that
kind
of
a
logo
change
that
will
be
the
start
of
for
versatility.
Be
that
something.
B
B
A
But
but
I
think
we
are
already
picking
up
realities.
These
are
and
a
very
recent
two
sisters
commit
recently,
so
you
will
have
a
to
adapt
to
it.
That's
my
guess.
Actually
you
might
want
to
hang
out
here.
This
hour
see
started
model
to
commit.
Probably
you
will
have
it
to
up
to
Europe
here
to
adapter
change.
Okay,.
A
A
A
A
C
D
C
D
Objects
because
you
know
production
environment
two-week
time.
We
kind
of
hope
that
we
can
use
a
relatively
big
vehicle.
Big
object
to
store
data,
for
example
a
10
10,
10
petabytes
sized
object,
which
means
the
object.
Number
is
relatively
small
and
weak,
keep
all
of
the
object
metadata
in
the
OST
memory
to
avoid
extra
extra
I/o
operation
when
doing
data
read
and
write.
D
D
D
D
D
A
D
D
D
D
D
We
think
that,
in
this
is
a
store
is
not
considering
implementing
the
rate
as
level
replication
mechanism
in
it
inside
it.
Then,
if
the
system
does
not
take
that
into
consideration,
then
after
the
implementation
of
say
store,
we
may
find
when
they
find
it
hard
to
implement
an
efficient
to
replication
mechanism
up
on
top
of
it.
So
we
think
that,
for
example,
we
if
we
can
implement
a
much
more
efficient
snapshot
measurement
and
then
that
should
be
useful
to
an
efficient
replication
mechanism.
A
D
D
Snapshot,
differences
to
another
cluster
believe
the
corrector
snapshot.
Creation
operation
itself
will
involve
much
overhead.
Then
we
cannot
take
the
very
efficient
who
cannot
make
a
very
efficient,
very
at
a
very
high
rate
of
snapshot
creation,
which
means
this.
The
interval
between
snapshots
may
be
too
long,
and
when,
when
disaster
happens,
we
we
may
lose
more
data.
That's
the
sort
of
thought
I.
A
Said
I've
done
to
the
use
case,
servers
a
finer-grained,
a
snapshot,
which
means
we
can
have
a
Leicester
lost
when
wisdom.
That
happens,
but
that's
basically
a
requirement
in
kettle
how,
regarding
to
how
we
connect
to
to
address
that
kind
of
a
requirement
and
how
it's
really
to
639
at
least
and
this
face.
Oh.
A
A
The
line
online
via
me
device
as
a
prop
device.
We
need
to
use
some
some
low-level
primitives
to
take
advantage
of
the
hardware
q
regarding
to
the
Ahmad,
multiple
channeling.
Yes,
I
also
agree
with
you
that
if
we
wanted
to
we,
if
you
will
be
using
multiple
channeling,
it
will
be
translated
to
rhythm
right
in
in
the
FTL
enabled
SSD.
In
that
case,
I
think
we
will
be
degraded
to
a
single
journaling
and
let's
adhere
to
it
to
its
job.
A
C
A
A
A
C
So
the
second
I'm,
looking
at
the
issue
of
of
ACN
I
just
run
the
unit
has
messenger
intestines.
It
doesn't
happen
with
this
unit
has
so
maybe
I
will
need
to
start
V,
star
cluster
and
try
to
I.
Guess
that's
that's
happened
when
we
stopped
a
caster
right.
It
is,
it's
writing
to
closed
afraid,
socket
right.
The
OSD
is
writing
to
a
closed
pocket.
Oh.
A
C
C
Test
but
oh
but
I
just
round
it
several
times,
I
didn't
see
any
errors.
Funny
saying
is
it
just
happen,
they're,
real
or
they're.
Frequently
each
there's.