►
From YouTube: Ceph Crimson/SeaStore 2021-06-09
Description
A
Let's
start
last
week
I've
been
reading
the
implementation
of
spdk
and
f2fs
for
first
time
expiration
on
on
how
they
do.
A
That's
category
and
next
time
next
step
would
be
to
investigate
on
the
the
lba
implementation
and
and
and
segment
cleaner
to
to
figure
out
what
what
what
we
need
to
do
to
to
collect
the
necessary
information
for
for
performing
more
effective
computer
collection.
One
thing
I
can
think
of
is
to
to
add
a
sequence
id
which
is
a
monolithically
increasing
number.
Whenever
we
have
a,
we
have
a
right
and
the
track
is
tracked
on
the
global
on
the
device
world
and
also
check
it
on
the
on
the
pork
wide
or
segment
wide.
A
C
B
Position
in
the
journal
stream
tells
you
its
age,
because
we
always
write
to
the
head
of
the
journal.
Also,
it's
possible
you're,
getting
very
ahead
of
where
c-store
is.
B
C
B
B
I
mean
we
know
why
it's
crashing.
It's
not
like.
That's
a
mystery.
It's
because
the
v
again
handling
isn't
particularly
clean,
which
is
why
I'm
adding
interruptible
futures
to
see
historically.
B
C
A
B
Has
been
able
to
get
it
to
as
long
as
you're,
not
trying
to
run
the
osd
itself
and
you're
using
the
fio
tool
that
just
exercises
transaction
manager,
that's
able
that's
reasonably
stable,
so
you
might
be
able
to
get
interesting
information
that
way,
but
that's
where
I
think
you
should
start.
I
think
you
should
start
with
empirical
observation
of
how
c-store
actually
behaves
and
then
start
formulating
strategies
for
what
you
think
the
segment
cleaner
should
do
differently.
B
A
B
In
the
sense
that
you
can
age
segregated
extents,
yes,
so
we're
adding
that
machine
or
zohan's,
adding
that
machinery
right
right
right
now.
D
D
All
know
the
trick
get
on
get
all
know
the
goal
to
know
entry.
You
give
us
solution,
but
I
still
met
that
problem
still
is
sometimes
and
I'm
I'm
still
tracing
on
it,
and
sometimes
the
system
start
up
and
I
can
create
the
created
pool
and
when
I
run
the
fl
testing,
the
reactor
report
is
stalled
there.
So
the
the
passing
cannot
continue.
So
just
the
the
the
output
reported
reactors
reactors
start
there.
D
So
when
I
fix
the
the
oh,
no,
the
tray,
no
entry,
I
will
continue
on
the
reactor
start
issue.
That's
all
the.
D
D
Many
numbers,
but
I
didn't
pay
attention
to
that
numbers.
So
many
some
some
number
yeah.
A
A
D
C
C
E
F
Hi
everybody
scrubbing
working
on
my
no
problems
with
the
next
step,
which
is
the
factoring
of
the
square
back
end,
almost
ready
for
up
to
four
trying
to
making
it
into
pr
and
thanks
to
sam
this
morning,
helped
me
understand
something
I
have
a
fix
to
a
previous
pr.
That's
it
for
me.
C
C
B
Bugging
through
the
lba
implementation,
converting
it
to
the
future,
I'm
just
hitting
a
bunch
of
stuff
in
interruptible
aerator
that
isn't
really
implemented.
So
I'm
adding
unit
tests
as
I
go
and
backfilling
stuff.
That's
it.
G
I
was
still
debugging
the
extent
placement
manager
I
run
into
a
gcc
bug.
I
guess-
and
it
took
me
some
time
to
overcome
it.
Actually
I
didn't
overcome
it.
I
I
I
downgraded
my
my
gcc
version
from
ten
to
nine
and
it
will
it
works
now.
I
think
I
can.
G
I
can
push
the
pr
this
week.
That's
all
for
me.