►
From YouTube: Ceph Crimson/SeaStore 2021-03-17
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
So,
let's
start
and
last
week
I've
been
reviewing
engines
pr
and
here's
the
pr2
to
improve
the
performance
of
an
illinois
store,
and
I
will
continue
on
my
my
unit
test
for
the
client
request
sequencer.
B
Hello,
everyone.
The
past
week,
I've
been
working
on
refactoring
our
send
in
con
in
connection
kefu,
told
me
that
you
and
him
found
a
a
small
bug
in
crimson,
where
we're
using
intrusive
ptr,
where
we're
sending
the
messages
instead
of
a
unique
ptr.
A
A
B
B
Yeah,
so
I've
been
I've.
I've
been
working
on
on
changing
that.
That's
actually
it's
actually
a
pretty
big
change
and
I
still
need
to.
I
mean
also
locate
all
the
places
that
this
change
needs
to
take
place.
It's
kind
of
hard
to
to
grab
for
that.
You
know,
for
example,
or
you
can
grab
for
a
make
message,
because,
usually
usually
we
call
send
right
after
we
call
make
message.
B
I
guess
it
is
possible
that
it
is
possible
that
sometimes
we
actually
call
make
a
message
from
some
function
and
then
call
send
from
a
nested
function,
so
that
that
is
also
something
that
I
need
to
work
on
yeah.
So
right
now
I
just
added
the
new
a
new
make
a
new
make
message
under
the
crimson
namespace
that
returns
a
unique
ptr
instead
of
an
intrusive
ptr,
and
it
also
had.
B
I
also
needed
to
add
it
as
a
friend
in
all
of
the
classes
under
mess
under
src
messages,
because
the
current
the
current
the
current
make
messages
had
to
be
defined
as
a
friend
in
all
of
them.
So.
A
Everyone,
just
a
quick
thought:
have
you
considered
instead
of
using
a
choose,
a
smart
pointer?
We
could
just
move
semantic
of
the
other
message.
The
other
time
we
pass
the
message.
We
we,
instead
of
passing
a
pointer.
We
just
move
a
has
a
moved
instance
of
the
message.
Does
this
work
or
it
cannot
be
avoided
because
we
need
to
invoke
the
move
move
construct
every
time.
B
This
was
actually
after
I
me
and
radic
had
a
couple
meetings
about
it
and
at
the
first
meeting
already
he
suggested
using
a
unique
ptr.
I
haven't
tried
of
thinking
how
to
implement
it
without
it
actually.
B
Yes,
yeah,
so
I'm
just
working
working
on
that
right
now.
C
C
A
C
A
Thank
you
come
here.
D
I
try
to
test
the
nvd
release
build,
but
the
skill
has
problem
after
applying
sam's
patch,
when
I
change
the
file
configuration
to
the
different
io
depths
different
job
number
and
the
the
the
time
every
different,
every
every
configurator
has
made
different
issues.
So
I
will
wait
sam
to
finish
the
modification
of
gc
and
then
redo
the
test,
so
so
no
features
developer.
Currently,
if
you
have
other
tasks-
and
I
I'm
I'm
glad
to
take
the
another
task
and
I'm
reading
the
code
currently
that's
all.
D
Yeah
but
every
every
configuration
account
to
finish
the
test.
There
are
many
different
issues
and
I
have
already
replied
to
the
sam.
He
has
no
it
and
I,
if
you
know,
if
you
have
other
features
needed
to
develop
engine,
I
don't
know
what
what
can
happen
to
to
them
to
do
the
io
path,
because
it
cannot.
The
test
cannot
be
divided
into
small
pieces.
So.
A
D
I'm
sure
yeah
I
will
take
some
time
and
consider
it
and
reveal
the
design.
I
think
I
have
discussed
with
some
about
m2
and
you
get
some
his
idea.
D
C
D
Yes,
we
can
send
the
message
right.
We
can
send
the
data
through
the
through
the
development,
okay,
the
communication
right
you,
you
can
send
the
data
to
to
the
different
course,
but
you
still
can
send
a
mac,
a
pointer
right.
The
the
other
cars
can
use
that
more
important
point
to
the
data.
D
C
D
Yeah
yeah.
Okay,
that's
all
so.
Some
skill
has
problem
on
the
in
the
gc
and
the
way
to
thumbs
finish
the
his
modification.
A
I
think
what
you
should
want
to
emphasize
is
that
we,
you
need
to
to
take
evidence,
efforts
into
consideration,
because
everyone
wants
to
you
use
a
unique
pointer
to
to
as
an
interface
when
sending
a
message.
But
in
your
case
you
wanted
to
to
use
a
foreign
pointer
for
for
passing
the
message
between
to
to
sending
a
message
to
a
given
osd,
even
though
that
ocd
could
that
connection
messenger
could
live
on
a
different
core.
A
B
D
A
A
D
E
E
Last
week
I
I
adjusted
my
some
of
my
pr's
according
to
kofu's
suggestion,
and
I
am
still
reading
the
source
code
of
c
store
and
trying
to
comprehend
the
detailed
design
I've.
I
noticed
that.
E
Some
future
store,
futurized
store
apis,
have
not
been
implemented
in
c
store.
I
intended
to
do
some
do
some
work
in
that
area,
but
I
don't
know
if
it
is
feasible
right
now
I
mean
I
don't
know
if
that
implementation
is
blocked
by
some
other
work.
That
must
be
done.
First.
A
A
By
the
way
we
already
merged
the
the
patch
from
from
shahan
to
fix
the
recovery
and
backfill,
so
hopefully,
in
next
regularly
technology
wrong.
We
could
find
a
green
test,
a
green
test
result
of
the
threshold
test.
Currently
we
have
like
nine
ish
tests,
which
are
performed
by
our
current
job.
One
of
them
is
a
is
a
crash,
a
threshold
search
test.
It
had
been
failing
for
for
four
months.
Hopefully
we
could
have
all
of
them
green.
A
So
shohan's
change
has
been
merged
like
seven
years
seven
days
ago,
but
the
last
batch
was
performed
10
10
days
ago.
So
that's
why
we
still
have
a
test
failures
in
the
stretch
test,
but
hopefully
in
the
next
next
week.
Also
we
could
have
a
or
green
test
result
in
performed
by
a
technology
current
job.