►
From YouTube: 2020-02-18 :: Ceph Crimson 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
C
C
They
start
and
last
week,
I
was
reviewing
any
things
play
out:
dude
ping
nut
this
ping,
the
messenger
to
us
given
CPU
core
and
also
reviewing
she
has
supported
to
pass
the
PT
look
down
done,
come
to
the
PT
back-end
and
meanwhile
I'm
walking.
On
the
own
note,
although
the
between
condition,
based
on
their
transaction
interface,.
C
C
C
D
It's
exactly
the
same,
the
constraint
circuits
are
done
the
same
way.
The
only
difference
is
whether
you
do
the
splits
before
you
consider
the
children
or
after,
and
it
slightly
changes
the
constraints,
but
why
one
way
or
if
you're,
referring
to
the
fact
that
you
change
references
from
the
parents,
that's
true
either
way.
C
I
coulda
shoot
share
you
with
with
a
working
progress
branch,
so
we
can
well.
D
Actually,
if
you're
drawing
a
diagram
might
actually
be
easier.
During
that
point
of
view,
I
mean
we're
still
so
again
we're
so
early
in
the
interface
level
that
I
have
so
little
faith
in
my
current
transaction
manager
that
H
that
I'm,
not
as
worried
about
that,
but
drawing
a
diagram
of
how
you're
planning
to
do
it
might
might
clarify
it.
Okay,.
D
Provided
that
it's
possible
to
implement
yes
from
the
other
side,
anyway,
okay,
I'm,
just
saying
drawing
a
diagram,
might
make
it
easier.
My
suspicion
is
that
you're
slightly
off
about
whether
or
not
so
changes
always
propagate
all
the
way
to
the
root
right,
unless
it's
possible
to
mutate
one
of
the
earlier
blocks
and
above
the
transaction
manager
that
H
layer,
you
actually
can
you're
allowed
to
mutate
one
of
those
blocks.
That's
that's
permitted!
D
C
D
D
D
D
To
be
clear,
I
actually
think
it's
easier
to
implement
the
top-down
version.
It's
just
cleaner,
but
that's
just
my
opinion
and
it
doesn't
change
in
this
since
you're
not
gonna,
do
explicitly
I,
don't
think
and
again.
This
is
this.
Is
the
part
I'm
worried
about
this
thing?
I'm
just
writing
to
you
right
now:
the
subsystem
where
the
transactions
are
rejected
if
they
conflict,
that's
the
part
that
may
end
up
being
above
the
transaction
manager,
layer
I
think
it's
easier
to
do
below
it.
D
D
D
E
C
F
Won't
work
this
won't
work
because
well,
this
would
work
in
case
of
perfectly
lined
ability,
but
compiler
could
decided
some
parts
of
the
buffer
library.
Implementation
are
not
in
line.
It
are
too
late
into
symbols
into
symbols
in
same
in
same
life,
statical
library
and
crimson
and
alien
eyes.
Blister
at
the
moment
makes
some
implicit
assumptions
that
direct
on
some
on
its
dependencies
that
they
are
the
same
in
both
crimson,
plain
water,
let's
say,
and
the
ally,
crimson
alien
world
and
okay.
F
D
D
You're
gonna
have
to
namespace
bufferless.
Sorry,
that's
the
way
it
goes.
It's
part
of
the
public
every
one
weeks
and
you're
gonna
have
to
actually
convert.
When
you
move
between
a
list
root.
Two
blue,
two
alien
blue
store,
you're
gonna,
have
to
actually
change
the
type.
If
you're
clever,
you
could
make
it
so
you
it,
but
that
will
repel
your
code.
Okay,
fine.
F
D
A
E
F
D
F
D
B
D
D
F
D
D
F
F
D
D
A
D
F
F
F
D
D
D
D
E
D
F
F
F
F
F
D
The
crimson
code
is
pretty
careful
about
not
copying
or
false,
except
what
it
literally
wants
to
make
a
copy
and
with
replication
you
do
actually
have
to
make
a
copy
right.
You
copy
it
in
two
different
messages
and
also
to
the
transaction,
so
yeah
I
can
see
that
it
wouldn't
surprise
me
about
that.
Much
I'm,
a
little
surprised.
We
did
that
good,
a
job
of
squashing
all
copies,
but
you
know
not
that
charge.
F
Last
week,
I
was
focused
mostly
on
error
on
pack-hunting
in
classical
OSD,
especially
around
always
the
the
map
issue,
and
also
in
messenger
from
that
I
was
where
I
was
reworking.
The
ex
could
watch
not
if
I
think
it
got
merged.
Pranks
this
week,
I
was
doing.
I
was
reviewing
couple
of
things,
especially
the
the
ionized,
the
alien
adapted
for
Earth,
blue
star
and
now
I'm
sitting
to
to
the
recovery
stuff.
That
means.
B
Everybody
and
what
did
most
of
the
week
I
limited
the
time
doing
prison,
but
what
I
did
was
going
over
a
code
for
scrubbing,
related
to
scrubbing,
trying
to
understand
and
writing
comments
inside
the
inside
the
code.
I,
probably
won't
I
won't
push
those
comments,
but
maybe
later
on,
I'll
ask
you
to
review
if
I
got
things
right
and
more
or
less
that
most
of
what
I
did
I
did
had.
I
did
add
a
minor
patch,
too
sister
was
accepted
related
to
clunk,
but
other
than
that
I
don't
see
anything
else
and
they
send.
A
B
I
really
I
will
I'm
I'm
doing
it.
Flu
writing
reading
the
code,
understanding
it
using
videos,
they
were
crews
or
we
will
have
more
questions.
We
publish
it
to
the
whole
team
to
most
comments.
One
is
that
I,
don't
know
if
you
not
his
sister
and
he
had
a
lot
of
patches
regarding
the
very
low
level,
a
code,
the
future
regarding
the
implementation
of
futures
and
some
utilities
and
regarding
the
destruction
I.
A
D
D
D
D
A
D
D
The
entries
in
the
first
place
then
doing
something
less
wildly:
costly,
I,
don't
know
what
the
answer
will
be:
I
suspect
it
will
be
hard
and
I
don't
want
to
hold
up
this
patch
on
it
not
hard,
but
is
it
well
I,
don't
think
it's
heavy
heart
I
think
it's
gonna
require
changes
to
the
ratos
protocol.
I
mean
literally
the
wire
probe
protocol
for
sending
the
messages
which
will
cause
a
compatibility.
C
C
D
A
problem
with
how
we're
encoding
the
log
entries
when
we
put
them
into
the
message,
so
it's
I,
think
enough
work
that
I
don't
want
to
hold
up
this
patch.
If
we
want
to,
we
could
add
a
config
option
to
disable
it
just
so
that
we
can
make
it
easy
to
figure
out
whether
when
we
do
fix
it,
we
have
fixed
it
well.
That'll
give
us
a
baseline
to
test
against
to
figure
out
how
much
CPU
were
wasting
on
this
operation,
but
even
that
I
think
we
can
wait
on.
D
D
Mean
this
needs
to
be
fixed
in
classic
OST.
To
be
perfectly
honest,
like
this,
this
exact
code
also
runs
the
classic
OST
and
it's
costing
the
same
amount
of
CPU
there.
So
we
we
do
need
to
fix
this
I.
Just
don't
think
we
should
block
the
patch
all
up.
Is
that
what
you
meant
run
and
this
you're
right
this.
C
D
C
D
D
The
wire
protocol
so
there's
the
messenger
internal
protocol
that
describes
how
the
how
the
handshakes
work
and
how
the
sessions
are
set
up.
That's
the
messenger
layer.
This
is
actually
the
message.
D
Is
the
binary
contents
of
the
mos
Diop
message?
It's
of
the
mos
I'm
sorry,
M
OSD
wrap
up
the
change
names
at
some
point,
I
can't
report.
It's
called
it's
the
it's
the
wrap-up
message,
the
exact
way
we're
choosing
to
encode
the
log
entries
is
inefficient.
It
actually
were
encoding.
The
log
entries
the
same
way
when
we
write
it
to
the
object
store
so
we're
wasting
it
twice.
F
C
F
C
D
C
G
Last
week,
I
was
most
mostly
notifying
my
Clemson,
please
log
PR
and
I
did
some
profiling
of
the
Crimson,
OST
and
I.
Think
the
maid
in
the
major
performance
bottleneck
is
out
of
the
pairing
states,
app
analog
method
and
I
tend
to
agree
with
agree
with
you
guys
about
that
home.
This
might
need
some
fundamental
change
about
the
way
we
do
data
or
serialization.
G
D
G
D
A
sort
of
a
well
contained
problem
that
we
can
solve
and
if
there
are
other
ways
to
do
it,
we
can
make
us
at
the
PG
log
structure
itself
can
be
directly
encoded.
Did
a
PG
log
structure
is
incredibly
complicated
because
it
supports
a
bunch
of
fields
that
are
utterly
irrelevant
from
normal,
writes,
yeah.
A
D
F
C
D
F
D
Improve
our
problem
mark
Nelson,
storms
much
much
more
general
with
blue
store.
It's
really
really
expensive.
Writing
log
entries,
regardless
of
how
you
encode
them
because
of
how
rocks
VP
deals
with
them.
So
that's!
Actually
it's
related,
certainly,
but
even
just
even
just
fixing
the
encoding
of
a
classic
would
be
a
huge
one.
A
And
I
just
share
a
link
to
summarize
my
current
findings
about
and
that
our
network
stack
I
think
it
would
be
good
to
reason
why
we
need
why
me
we
might
need
post-tax,
the
POSIX
and
the
native,
and
we
need
to
be
aware
of
issues
if
we
want
to
support
both
of
them
and
I
hope
we
we
are
in
the
same
page
on
this
and
I'm
still
working
on
the
performance
evaluation.
Part
I
think
it
will
be
close.
I
will
publish
data
and
finish
this
document
as
soon
as
possible.
That's
tough
for
me!