►
From YouTube: Ceph Crimson/SeaStore 2021-09-08
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,
so
looking
over
the
segment
cleaner,
I
think
it
doesn't
have
that
problem,
but
I'm
going
to
need
to
audit
more
ordered
it
more
carefully
tomorrow.
When
I'm,
you
know
awake.
A
Interestingly,
I
with
that
fix
to
the
handle
split.
I
can't
reproduce
the
assert
anymore.
At
least
I
haven't
been
able
to
for
the
last
three
or
four
hours.
A
Well,
let's
give
it
a
couple
of
minutes:
it
just
crossed
10
o'clock.
Oh
well,
it
just
crossed
sort
of
the
meeting.
We
should
at
least
get
greg.
C
So
this
listing
is,
I
use
the
curve
to
do.
The
performance,
profiling
and
airbnoc
extent
seems
occupied
about
20
of
the
cpu
cycles.
D
A
Yes,
like
there's
a
there's,
a
trivial
improvement
there,
where
we
should
be
doing
a
binary
search
when
we're
looking
at
the
individual
nodes
that
would
speed
things
up
substantially.
Do.
B
A
A
So
the
first
thing
to
do,
I
think,
would
be
to
work
out
whether
we're
spending
our
time
primarily
in
that
initial
tree
traversal
or
if
what's
happening,
is
we're
spending
a
ton
of
time.
Looking
through
adjacent
keys,
where
we
can't
find
an
insertion
point,
if
it's
the
former,
then
the
binary
lookup
thing
will
help.
If
it's
the
latter,
then
something's
not
working
correctly
with
hinting,
we
shouldn't
generally
be
allocating
in
a
portion
of
the
lba
tree.
That
has
a
lot
in
it.
Does
that
make
sense.
A
C
A
C
Calling
exchange
the
iteration
next,
it's
called
homicides.
A
Yeah
compared
with
how
many
times
alec
extent
is
called
at
all
okay,
that
ratio
should
give
us
a
hint
as
to
what's
going
on
there.
Okay
anyway,
let's
do
the
meeting.
A
C
I
have
pasted
the
link
in
the
chat
window,
so
everybody
can
check
it
and
the
initial
result
is
the
draft
area
result
is
found
that
rb
allocate
exchange
occupied
about
twenty
percent
of
this
total
cpu
cycles.
C
The
perf
collected
the
cpu
cycles
and
only
on
the
cpu
one,
so
the
current
osd
running
on
the
cpu
one,
and
so
so
that's
the
result,
and
just
now
an
exam
suggested.
We
use
the
proof
counter
to
calculate
how
many
iterations
next
is
called
versus
the
allocate
extended
counters
right.
B
A
Okay,
greg.
E
I'm
just
working
on
assimilating,
more
stuff
and
going
through
the
seastar
tutorial.
I
noticed
a
few
random
things
that
I'll
send
in
prs
for
to
make
just
working
with
vstart
a
little
nicer,
and
I
have
to
take
a
little
bit
of
time
to
do
some
unrelated
stuff
for
stretch
clusters
sounds
good.
Gotta,
do
some
backwards
for
downstream
and
add
a
command.
B
Oh,
I
was
fixing
bugs
within
the
extent
placement
manager
pr-
and
I
think
this
this
week
I
will
return
to
the
multi-device
support
work,
I'll
I'll,
get
it
done
as
soon
as
possible.
That's
all.
A
A
Okay,
yeah,
okay,
okay,
all
right
I'll
run
this.
So
when
I
ran
this
earlier
today,
I
almost
immediately
got
an
I
o.
Hang
I
didn't
track
it
down,
though,
and
I
wasn't
getting
any
useful
debugging.
Have
you
seen
that
yourself
or.
A
B
D
Yeah
last
week
I
have
implemented
the
hints
in
the
ono
g
and
I
I
added
clients
into
independent
way
to
collect
and
pass
metrics.
So
I
think
it
should
also
works
for
mbd
tour
with
c
store
back
hand.
D
So
it's
it's
more
generic,
and
now
I
added
eight
eight
more
graph
with
with
io
iops
and
this
throughput
and
some
c
star
magic.
I
think
there
are
there's
one
unique
matrix
here,
which
is
reactor
utilization.
D
D
So
next
I
think
I
will
help
and
here
have
you
guys,
insist
on
and
you
have
some
style
metrics.
I
don't
understand
it.
D
A
I'm
looking
at
your
cpu
utilized
ratio,
oh
get
started.
That
would
be
the
portion
of
the
time
we
spend
waiting
right.
D
D
Sketch
schedule
is
the
reactor
queues,
the
main
means
the
name
of
the
cube
is
another
key
name
named
exit
by
default
and
if
we
add
more
scheduling
group,
there
will
be
more,
I
think
around
metrics.
So
so,
if
we
need
to
see
the
total
utilization,
which
is
the
reactor
utilization.