►
From YouTube: Ceph Crimson/SeaStore 2021-09-15
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
see
last
week
I
pushed
a
pr
with
some
epm
simplifications
and
fixes,
and
I'm
working
on
getting
a
toothology
job
that
will
actually
run
c
store
working,
so
there
will
be
further
fixes
for
that.
I
expect,
should
we
how's
it
going.
B
Last
week
I
finished
some
small
pr
binary
search
for
lb3
and
omap3
for
the
literature,
and
I
did
the
magics
forget
the
alpha
extent,
and
literally
next
so
later
next
is
it's
now
bigger
than
I
look
extense
the
banded
surgeon
not
to
provide
a
much
improvement.
It
can't
see
implemented
when
using
the
perf
analysis
so
that
the
confliction
still
caused
lba
tree
allocate
exchange
to
waste
more
cpu
cycles.
A
Cool,
so
the
next
investigation,
therefore,
is:
why
are
the
lba
addresses
so
close
together?
Can
you
generate
some
kind
of
a
histogram
of
where
they're
concentrated.
C
B
D
B
And
only
I
will
output
print
out
and
maybe
ten
times
I
missed
two
times
about
the
sure
yeah.
A
Aggressively
either
way
you
guys
may
want
to
do
that
and
try
to
reproduce
it
with
debugging
and
see
if
you
can
work
out
what
it's
doing.
It's
probably
like
the
last
few
times.
The
last
instances
of
this
that
I
fixed
were
cases
where
we
were
somehow
adding
the
same
extent
twice
or
other
similar
bugs
like
the.
A
The
epm
behavior,
where
we
reuse
the
physical
address
of
the
segment
we
were
cleaning,
would
cause
it
because
it
caused
undefined
behavior
in
the
intrusive
set
we're
using
to
track
things.
So
there
are
a
lot
of
ways
to
cause
it.
That's
what
I'm
saying,
if
you
guys
can
reproduce
it
with
debugging
and
try
to
narrow
it
down.
That
would
help
I'll
also
be
trying
to
do
that.
Sure.
C
C
C
C
Okay
last
week,
I
I
submitted
several
small
papers
and
mainly
was
working
on
the
device
support
and
I
think
it's
I
think
it's
almost
there.
I
I
hope
I
can
send
me
the
pr
in
the
next
one
or
two
days.
That's
all
right.
D
This
in
invariant
is
is
too
strong
because
it's
transitive
from
knee
transitive
from
leaf
node
throughout
to
the
root
node.
So
I
think
it
will
increase
the
rate
of
conflict
yeah,
but
changing
yeah,
but
changing
this
environment
is
not
not
very
easy,
because
it
leads
to
major
refactors
to
do
split,
merge
and
look
up,
but
I
think
it's
still
worth
the
effort,
as
for
conflict,
is
one
of
the
major
issues
from
providing.
So
I
I
think
I
will
start
this
cleanups
this
week.
A
Yeah,
the
lba
treat
does
not
use
that
variant
in
its
implementation
yeah.
So
that's
an
unusual
one.
The
other
thing
I
will
point
out
as
far
as
conflicts
go
is.
I
am
concerned
about
the
the
ratio
of
false
to
next.
In
allocate
extent
that
means
that
the
lba
addresses
aren't
well
distributed.
D
D
A
All
right,
one
more
shameless
plug
every
week,
either
before
this
version
of
this
meeting
or
the
morning
version
of
this
meeting,
I
do
a
45
minute,
rados
q,
a
so
if
anyone's
interested
and
wants
a
sort
of
a
longer
form
way
to
just
ask
whatever
kind
of
question.
That
is
a
good
time
to
do
it.
It
shows
up
on
the
ceph
community
calendar
as
rados
q.
A
all
right
have
a
good
day,
all
right,
bye.