►
From YouTube: Ceph Crimson/SeaStor OSD 2020-06-24
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
Hello,
hi
tommy
hi.
So
let's
start
as
always,
I'm
pasting
the
url
to
the
edit
pad
in
in
chat
window.
A
Last
week,
I
I
was
still
working
on
the
async
another
tree
and
was
reviewing
some
of
the
very
nice
blue
storylines
to
peer
by
by
ritek
and
in
this
week
I
will
try
to
wrap
up
the
review
of
heartbeat
and
there
will
be
a
replicas
by
place
by
engineer,
stem
and
that's
it
to
me.
B
Okay,
so
I
have
put
my
or
property
chat
code
here
and
I'm
going
to
move
to
the
new
sam's
latest
version
and
because
there
are
some
updated
for
the
data
apply
and
code
so
and
then
we'll
write
in
the
test
case
to
do
the
test.
B
So
that's
all
cool.
C
Okay,
I
am
working
on
so
in
order
to
flush
dirty
pages
or
generally
to
rewrite
a
page
that
currently
exists.
You
need
to
reload
the
page
containing
any
incoming
pointers
to
that
page.
So,
for
any
logical
extent,
that
would
be
the
leaf
lba
tree
node
and
for
any
any
inner
lba
tree
nodes.
That
would
be
the
parent
node
for
that,
because
you
need
to
rewrite
the
physical
location
of
it
and
in
general
we
want
to
keep
those
in
memory
anyway
because
of
temporal
locality.
C
So
I'm
adding
support
to
the
cache
to
track
those
dependencies
and
ensure
that
all
the
parent
pin
parent
extents,
are
pinned
what
this
means.
Logically,
is
that
any
logical
extent
such
as
one
of
your
extent
tree
or
oh
no
tree
extents
or
an
actual.
You
know
data
containing
extent
if
it's
in
memory,
all
of
its
mapping
pages
in
the
lba
tree
will
be
in
memory
as
well.
C
C
It's
useful
for
ensuring
that
flushing,
a
dirty
page,
doesn't
require
us
to
do
reads
because
that
can
make
the
cache
memory
grow
during
a
time
when
you
need
it
to
shrink,
get
any
of
that.
I'm
adding
support
for
that.
That's
the
prereq
for
both
flushing
dirty
pages
and
cleaning
segments.
D
A
D
A
C
A
C
Okay,
in
other
words,
I
don't
need
the
parents-
I
I
don't
need
the
oh,
no
tree
parents
for
that,
one
to
rewrite
it,
but
I
do
need
the
physical
address
pages.
C
D
This
week
I
was
trying
to
investigate
some
zero
copy
data,
serialization
serialization
methodology
like
the
flat
buffers,
in
the
hope
that
maybe
we
can
tackle
the
issue
that
current
the
current
encode
and
the
deep
and
decode
method
burns
too
much
cpu
and
that's
what
we
do.
A
C
D
C
Monkeying
about
with
the
pathology
code
itself,
I
expect
and
then
fixing
any
bugs
that
come
up.
I
suspect
it
will
be
more
work
than
it
seems
like
it
will
be.
So
I
think
that's
where
we
need
to
focus
our
efforts
now.
C
A
And
also
use,
I
think
we
can
use
a
standalone
test
for
some
use
cases.
C
That's
fine,
but
if
recovery
and
backfill
not
work,
that's
what
we
need
to
be
doing.
That's
that's
the
point!
So
if,
if
we're
at
that
stage,
then
we
need
to
start
making
the
thrasher
test
work,
because
that
should
work
now
right.
That
was
the
point
of
getting
blue
store,
recovery
and
backfill
working.
Those
three
together
should
give
us
the
ability
to
run
the
conventional
thrashing
test.
C
D
A
D
E
Last
week,
I
I
continued
implementation
of
the
split
and
currently
the
the
right
side.
Node
can
be
correctly
formed
using
up
using
current
tests,
and
that's
all
for
me.
A
Next
any
comments,
questions.
E
So
coffee,
if
you
want
to
you,
want
my
help
to
help
you
to
go
through
the
heartbeat
code.
I
can
you
just
please
pee
me.
E
A
Convenient
for
you,
thank
you
anything
else.
Nope.
C
Do
we
want
to
talk
a
bit
about
the
topology
work,
so
I
think
we
are
getting
close,
so
I
think
it's
worth
working.
Have
you
guys
looked
at
all
at
the
code
that
makes
pathology
work?
Do
we
have
any
volunteers
to
work
on
it.
D
I
don't
have
access
to
the
to
the
set
library.
C
C
C
A
We
have
a
dedicated
web
page
for
for
process
for
for
applying
for
all
account
on
cplab
and
the
account
for
the
open,
vpn
access
to
it.
Oh.
C
Yeah,
so
that's
that's
step
one
that
gets
you
the
ability
to
run
them
at
all.
So
at
that
point
I'm
going
to
email,
you
like
the
suite.
I
think
we
want
to
focus
on
getting
to
work
first
and
where
the
python,
for
it
is
a
good
first
step,
would
be
to
just
write
it
on
the
regular
sef
classic
and
then
have
a
look
at
the
code
and
measure
the
logs
and
see
you
know
how
it
works
and.
A
C
If
we
have
to,
if
I
mean
if
stiff
idm
is
another
month
of
work,
then
yeah,
I
think
we
should
just
do
our
own
thing
and
switch
back
to
the
set
adm
good
when
it's
available.
If
that's
hard,
then
we
can
wait,
but
I
I
in
other
words
the
longer
we
wait
to
do
this.
The
longer
we're
writing
code
that
we
can't
test
and
that's
dangerous.