►
From YouTube: Ceph Crimson/SeaStore 2021-10-07
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
All
right,
let's
get
started,
let's
see,
I
spent
a
lot
of
time
reviewing
the
devices
pr
from
schwein.
You
saw
that
I
pushed
some
commits
cleaning
up
some
stuff,
I'm
still
reviewing
the
last
two
and
a
half
or
the
last
commit
and
a
half
or
so
from
your
patch.
So
I'll,
let
you
know
I'm
about
to
re-push
in
the
next
20
or
30
minutes
a
version
of
the
comments
I
pushed
that
actually
works
instead
of
crashing
immediately.
A
Yep
and
you'll
want
to
take
a
look
over
them.
I
haven't
tested
them
very
thoroughly,
so
you
should
do
whatever
testing
you've
been
doing.
Okay,.
A
The
main
goal
here
is
to
make
it
so
that,
as
many
things
as
possible
are
compilers,
for
instance,
exposing
segment
directly
means
that
it
was
easy
to
convert
the
compound
4-bit
28-bit
combo
into
just
like
a
number
which
would
make
it
really
easy
to
compare
it
to
segments
where
it's
not
appropriate.
So
I
want
to
make
that
kind
of
thing
private.
B
Okay,
all
right
barth,
hello,.
B
A
B
A
C
I
just
tried
to
bring
up
the
cluster
with
the
upstream
build
17
17th
version.
Exact
version
is
like
17.0.0
iphone
7998,
so
I'm
facing
the
few
issues
to
add
the
osds.
So,
first
of
all
I
would
like
to
know
which
versions
I
can
try.
Current
master.
A
Second,
I
haven't
tried
to
bring
up
a
cluster
the
way
you're
doing
it.
So
again,
I
suggest
that
you
create
a
bug
with
the
steps
you
took
and
then
or
I
will
take
a
look
at
it,
there's
a
good
chance
that
it
just
actually
doesn't
work.
There's
a
bug,
okay
or
that
there's
some
alternate
thing
you
have
to
do
or
some
config
you
have
to
fill
in.
I
don't
know.
B
A
C
B
A
I
guess,
if
you're
trying
to
get
you
could
get
it
out
of
damon.
I
think
not
actually
sure
to
do
that.
I
always
do
my
own
build
locally.
A
B
D
Multiple
device
pr
for
three
hand
and
updating
the
journal
patching
pr,
but
after
testing
it
seems
not
improvement
at
the
latency.
It's
it
it's
getting
worse
for
the
latency,
and
I
try
to
try
to
calculate
the
the
functions
enter
time
and
exist
time
for
the
for
the
right
recorder.
D
But
when
I
use
the
perf
to
a
probe
the
function-
and
it
always
tell
me
the
the
function
is
in
line
and
not
no
return.
A
point,
no
matter
how
I
set
the
function
out
in
line
with
the
c
plus
c
flags
and.
D
A
I'm
not
sure
about
that.
Sorry
you'll
have
to
dig
into
how
people
usually
use
perf.
For
that,
can
you
guess
so?
If
journal
batching
is
working,
you
should
see
larger
ios
submitted
on
average.
So
you
might
look
at
the
disk
statistics
see
if
the
total
number
of
rights.
D
Most
is
one
in
the
queue
in
the
queue.
Sometimes
it
is
zero
changes.
A
A
Anybody,
okay!
Well,
that's
what
I
would
look
into.
It
means
that
the
osd
probably
isn't
saturated.
So
what
you're
observing
is
an
increase
in
latency
on
acute
depth,
one
workload,
which
means
you
need
to
improve
your
implementation.
It's
adding
overhead.
D
But
I
don't
know
how
to
employ
it
already
write
the
several
ios
in
in
just
once.
That's
right.
D
No,
no
in
most
time,
I
think
I
used
io
testing
so
most
time.
At
that
time,
only
one
io
requests
are
coming,
and
I
also
modified
notice
that
there
are
two
three
ios
in
the
queue
and
sometimes
10
00
10,
in
the
queue
and
sometimes
in
over
100,
in
the
queue.
D
D
A
A
It
might
need
double
buffering,
then
this
might
be
something
to
put
on
the
back
burner
for
a
while.
It
means
we
may
need
something
more
complicated
if
this
is
a
problem
at
all.
D
A
Is
possible,
in
which
case
it
would
be
better
to
focus
on
something
else.
Instead,.
A
D
E
Family,
so
I
had
to
spend
a
couple
days
doing
some
other
stuff
for
stretch,
but
I
am
looking
through
how
ciladb
does
messaging
because
they,
you
know,
wrote
c
star,
and
I
want
to
see
if
there
are
any
obvious
traps
that
they've
identified,
that
we
need
to
work
around
in
doing
in
a
process
communication
to
route
messages
to
different
cores
for
stuff,
like
moving
pg
operations
to
the
right
place
before
I
start
doing
that
kind
of
multi-reactor
work,
and
that
is
going
on
this
week.
B
A
B
B
This
week
I
I
made
notifications
as
sam
suggested
and
for
the
for
the
multi-device
pr
I
I
was
on.
I
was
on
the
national
day
holiday
this
week,
so
I
I
didn't
make
make
much
progress
on
other
stuff
I'll
try
to
get
the
get
the
sprite,
lfs,
gc
and
the
extent
placement
strategy
work
in
sistor
next
week.
A
Okay,
cool
chen
may
you're.
In
luck,
there
was
recently
a
couple
of
youtube
videos
created
for
this
I'll,
send
them
to
you
in
an
email.