►
From YouTube: Ceph Crimson/SeaStor OSD Meeting 2020-08-05
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
last
week
I
created
a
cup
a
couple
of
fix
to
enable
this
little
run
still
based
treasure
test
with
with
crimson
and
the
latest
test
result.
I'm
pasting
the
latest
result
in
the
chat
window,
so
it's
blocked
by
some
failures
in
in
bluestock,
but
I
don't
think
it's
a
it's
a
really
relatable
related
to
brewster.
I'm
I'm
going
to
look
into
it
this
afternoon
and
apart
from
it,
I
was
leaving
sam's
apr
to
to
add
the
pink
machinery
to
the
airbn3,
and
that's
me
to
me.
B
Oh
okay,
so
I'm
I
updated
the
pr
according
to
the
comments,
and
so
please
review
it.
If
it
is,
I
assume
that
the
the
extend
map
tree
is
started
in
the
old
node.
Oh
no,
the
block,
so
I
will
not
store
it
in
extern
map
tree.
So
when
the
all
know
the
cached
extent
will
be
duplicated
or
something
and
it
will
updated
the
external
map3.
B
So
each
transaction
can
use
this
correct.
Exchange
map
tree
root
information,
and
I
I
next
task.
I
rebased
my
code
to
sam's
rba,
pin
patch
and
found
some
problem
and
sam
said
it
is
decreased
reference.
Reference
is
missing,
so
wait
for
them
to
update
it
later
and
then
I
can
reveal
rebase
pin
patch.
C
To
bother
just
rebase
on
the
pin
patch
grab
just
comment
out
the
deck
that
that
graph
falls.
I
checked
your
unit
test
passes
under
those
conditions
your
current
patch
doesn't
build
for
base,
though
so
do
that
stuff.
First.
B
B
B
C
B
A
Covered
and
stem.
C
Well,
actually,
you
had
a
bunch
of
of
unaddressed
comments
on
your
pr
from
me
still.
B
B
Oh,
that's
because
the
kb,
a
fixed
kv,
node
layout,
needed
some
constant,
constant
operation,
so
cipher
le64
doesn't
provide
that
to
one.
So
I
provided
the
error.
B
Yeah
I
changed
it
to
the
change
modified
for
the
ending.
C
B
So,
do
you
mean
I
need
to
use
the
set
64
seth
aries
six-fold.
C
B
Address,
let
just
use
the
self
air
e64
originally.
B
B
B
B
C
B
If
the
error
address
in
the
value
part,
so
when
they
set
value
in
fixed
kv,
node
layout
to
set
the
value,
there
is
problem,
they
need.
The
input
parameter
is
the
constant
it's.
If
the
address
in
the
k,
k
key
is
no
problem.
C
C
So
I'm
not
quite
sure
what
problem
you
ran
into
if
you
want
to,
you,
can
send
an
email
with
a
question,
but
at
least
this
version
I
can't
seem
to
get
to
build.
C
Oh
maybe
whip
c
store
extent,
map.
B
A
C
Yep,
I'm
adding
support
for
that
graph
properly
so
that
it
will
correctly
retire
the
extent
and
you
know
get
rid
of
it
out
of
the
cache.
I
have
what
I
think
is
a
working
version
and
then
I'm
back
to
working
on
a.
A
Okay,
let's
see
I'm
regarding
to
your
european,
I
think
it's
I
got
your
point
and
what's
missing
is
just
to
add
more
a
little
bit
more
comment
regarding
to
why
we
need
to
ping.
Probably
it's
a
all.
C
A
C
C
A
A
C
C
It's
extremely
annoying
to
retrofit
the
data
structures
to
do
this,
so
I'm
doing
it
now,
like
all
the
internal
and
memory
data
structures
need
to
be
able
to
do
this
efficiently
because
you
have
to
mess
with
it
all
the
time,
so
no,
I
actually
do
want
to
do
it
now.
I
understand
that,
strictly
speaking,
I'm
not
super
worried
about
performance
now,
but
we
will
need
this
for
sure.
A
I
understand
but
sam
my
concern
is
you
make
this
this
amount
mandatory
to
to
add
the
to
ping,
all
the
parent
of
a
note
in
in
cache,
but
but
we
we
don't
have
a
way
to
to
to
evict
a
a
note.
A
parent,
for
example,
out
of
the
cache.
C
C
C
So,
in
other
words,
cash
pressure
applies
to
the
things
we
actually
want
to
cash.
We
don't
actually
want
to
cash.
The
the
lba
notes
at
all
the
internal
nodes.
We
only
we
only
keep
in
cash
if
there's
a
child
in
cash
under
them.
There's
no
other
reason
to
do
it
and
the
reason
we
keep
them
in
cash
is
that
you'll
notice,
this
implementation
doesn't
have
a
second
logical
address
lookup,
that
is
the
latter
cache.
D
Last
week
I
was
on
developing
the
interrupt
interruptible
narrative
future
during,
during
which
I
I
found
that
maybe
a
series
of
error-rated
version
of
future
facilities
like
parallel
for
each
or
when
all
succeed,
that
needs
to
be
implemented.
So
I'm
now
working
on
that
part
and
that's
all
for
me.
A
D
Yeah,
I
need
to
add,
I
need
to
add
those
those
primitives
so
that
I
can
replace
errorator
future
will
replace
c
star
future
with
aero
radio
feature
on
the.
I
o
execution
path.
A
E
Yeah,
I
think
I'm
I'm
working
on
a
similar
implementation
like
sam,
currently
is
working
on,
so
I'm
working
on
the
to
I'm
working
on
to
implement
a
tracking
facility
so
that
if
a
tree
look
up
in
all
node,
if,
if
someone
else
hold
the
return
cursor,
the
node
will
be
available,
kept
available
and
kept
checked.
E
C
E
C
E
I
think
there's
another
reason
that,
if
in
the
in
that
transaction,
if
you
need
to
access
that
node
in
the
in
the
extent
after
an
asynchronous
operation,
and
if
you
hold
the
cursor
of
that
oh
note,
you
that
all
node
pointer
will
be
kept
updated
with
less
effort.
Because
if
you
want
to
access
that
all
node
again
in
another
continuation,
yeah.
A
D
C
E
It's
it's
it's!
It's
just
to
keep
track
of
the
the
cursor.
To
the
oh
note,
in
the
leaf
extent,
if
you
need
to
keep
track
omnids
in
across
the
continuations,
because.
A
E
C
C
Yeah,
why
a
cursor,
though,
is
it
that
you
expect
to
re-look
up
the
same?
Oh
note
or
you
just
you
just
need
to
keep
a
pointer
at
the
page.
It
was
it
wasn't.
That's
all.
C
C
E
That's
that's
that's
interesting
because
the
reference
is
the
reverse,
so
so
it's
a
cursor
holding
the
leaf
node,
the
reference
of
the
leaf
node
and
the
leaf
node
holding
the
reference
of
the
internal
node
until
the
root
node.
So
if
we
right
that's
that's
that
the
entire
structure
will
be
freed,
I'm.
C
Not
worried
about
it,
creating
no!
It's
not
that
it's
that
so
so,
let's
say
your
leaf.
Node
has
a
reference
to
extent
e
right.
Yes,
oh,
but
yeah
yeah,
but
in
in
in
memory.
So
the
trouble
is
that
if
I
relocate
e
because
I
did
a
compaction
or
something
then
the
in-memory
structure
you're
pointing
at
is
now
invalid.