►
From YouTube: Ceph Performance Meeting 2022-05-05
Description
Join us weekly for the Ceph Performance meeting: https://ceph.io/en/community/meetups
Ceph website: https://ceph.io
Ceph blog: https://ceph.io/en/news/blog/
Contribute to Ceph: https://ceph.io/en/developers/contribute/
What is Ceph: https://ceph.io/en/discover/
A
Let's
see
so,
okay,
today
a
couple
of
closed
pr's
that
I
saw
this
week,
the
the
first
is
the
one
we
were
just
talking
about
a
little
bit
ago:
adam's
excellent,
excellent
pr
that
figured
out
how
to
fix
the
issues
that
cropped
up
in
quarry's
pr,
basically
just
not
doing
bounding
the
whole
space
iterator,
which
it
turns
out.
We
can't
safely
do
so.
A
The
tests
that
were
failing
before
are
now
fixed
and
the
cory's
pr
was
reinstated
into
pacific,
and
I
don't
remember
if
we've
reverted
it
in
master
now
we
might
have,
but
in
the
event,
both
that
pr
and
atom
six
are
going
into
1628
and
going
into
master,
and
it's
looking
looking
good,
so
that
was
excellent
work.
We're
we're
still
getting
it
in
for
the
1628
release.
I'm
I'm
very
very
happy
about
that.
A
Okay,
next,
my
fix
for
argument,
parsing
c-star
made
it
in
there
was
some
really
good
feedback.
I
had
gotten
from
kefu
and
and
radic
and
that
that
trunk,
the
the
code
beyond
both
what
it
was
previously
and
then
what
my
original
version
of
it
was.
So
that
was
really
good
and
yeah.
That's
in
now.
A
So
we
are
again
correctly
passing
the
star
command
line
options
to
to
c-star,
which
we
hadn't
been
for
a
little
while
a
couple
of
updated
prs,
my
pr
for
enabling
tc
malik
in
when
using
cstar.
A
So
it
turns
out
that
if
you
compile
ceph
with
a
c
star
support,
neither
crimson
nor
the
classic
osd
will
be
using
alec
they'll,
both
be
back
to
using
lipsy
and
and
we're
we're
even
worse
than
we
used
to
be
with
it
it
for,
for,
like
a
three
gigabyte,
blue
store
cache
size,
we're
using
like
33
gigs
of
ram.
So
it's
it's
really
really
nasty.
A
So
this
basically
just
changes
now
just
changes
the
the
conditional
and
the
perfectly
logic
that
that
basically
causes
us
to
to
not
link
with
tc
malloc
when
it
when
we
use
with
c-star.
Now.
The
problem
that
we're
hitting
right
now
is
that
we're
seeing
some
invalid
pointer
asserts
that
spir
appear
to
be
coming
from
address.
A
Sanitizer
matan's
been
been
looking
at
it
and,
and
he
thinks
that
is
actually
the
ones
that
he's
he's
tracked
down
anyway,
have
been
due
to
a
known
behavior
in
tc
malik,
and
that
this
is
a
little
bit
of
a
it
was
not
something
on
our
end
is
I
don't
know
if
you
wanna,
say
it's
a
false
positive,
it's
not
exactly,
but
but
it's
not
not.
On
our
end,
having
said
that,
some
of
the
tests
that
I
ran,
I
saw
sink
faults
in
and
that's
that's.
A
You
know
a
bigger
problem
and
I
don't
know
if
it's
due
to
this
or
due
to
something
else
going
on,
but
we
probably
can't
merge
this
pr
until
we
we
track
all
this
down,
so
that
pr
right
now
is
just
kind
of
stuck
in
that
state.
A
My
pr
for
switching
to
a
time-based
near-fit
algorithm
needs
to
be
rebased
and
updated,
but
otherwise
I
think
it
looks
good.
The
the
feedback
I've
got
right
now
is
to
use
actually
use
the
the
old
version
of
the
config
options,
because
it
has
less
impact
on
on
hot
path,
performance,
which
I
I
agree
with
100.
A
I
thought
that
the
cash
thing
that
we
introduced
also
was
a
way
to
fix
that,
but
we
we
don't
seem
to
have
a
real
strong.
We
haven't
really
decided
to
guess
if
we're
going
to
continue
using
legacy
options
or
try
to
move
everything
to
the
new
new
way
of
doing
it.
So
I
don't
know.
I
guess
I
don't
care
that
much,
but
whatever
people
want
to
be
is
fine
with
me
as
long
as
this
doesn't
have
a
runtime
impact.
A
Okay
and
then
last
here,
rgw
deus
cell
implementation.
I
saw
that
there
might
have
been
just
a
little
bit
more
review
on
that
and
it
needs
a
rebase
now.
So
I
guess
that's
back
in
the
author's
court,
and
that
was
what
I
saw
for
updated
prs
this
week.
A
All
right:
well
then,
this
first
discussion
topic
for
roxy
b,
iterator
bounding
work.
If
adam
was
here,
I
just
wanted
to
give
him
the
opportunity
to
talk
a
little
bit
about
what
he
saw
and
if
he
has
any
future
plans
for
looking
at
any
of
this
beyond
what
was
there.
I
know
he
has
said
that
for
some
of
the
feedback
that
he
got
is
that
we
need
better
test
cases
surrounding
this
and
I
think
he
does
plan
to
implement
them,
but
didn't
want
to
slow
down
the
roll
out
of
this
fix.
A
So
I
think
probably
the
next
steps
are
getting
better
test
coverage
around
how
some
of
our
iterators
work,
but
not
there
yet
so
move
on.
Then
I
guess
I
already
kind
of
talked
about
the
resident's
economic
work.
There's
not
a
whole
lot
more
to
discuss
there.
A
I
think
we
just
need
to
figure
out
the
aborts
and
sig
faults.
The
imports
maybe
are
taken
care
of.
At
this
point,
I
will
say
that
this
is
kind
of
required
for
us
to
be
able
to
do
more,
not
just
performance
testing
but
actually,
like
stressed
stress,
testing
of
crimson.
We
we
we
need
this
working
better,
be
able
to
proceed
with
that,
so
this
is
kind
of
high
priority.
A
The
other
thing
I'll
mention
is
that,
even
when
using
tc
malloc,
which
seems
to
work
with
our
tests,
our
performance
tests,
we
I'm
not
seeing
it
sig
fault
or
anything.
A
We
do
stall
out
eventually
on
rights,
and
I
don't
think
it's
actually
related
to
this.
I
talked
to
sam
a
little
bit
about
it
and
he
said
that
he
can
get
crimson
into
a
state
right
now,
where
he
does
he
stalls.
A
He
said
that
the
the
ones
he's
looking
at
are
related
to
peering,
so
it's
not
any
of
this
stuff,
but
but
that
will
probably
be
the
next
step
after
we
get
this
dc
malik
vr
merged
is
then
to
start
tracking
down
these
stalls
and
trying
to
help
sam,
if
you
know
figure
out
if
this
is
the
same
ones,
that
he's
seeing
or
not,
okay
and
then
moving
on
the
the
last
topic
I
had
was
that
I'm
starting
to
look
at
igor's
work
where
he's
implementing
a
blue
store
wall
rather
than
using
the
roxdb1.
A
I
just
started
that
yesterday
I
get
it
to
compile
and
run
fine,
but
it
does
not
appear
to
be
using
yours.
New
new
machinery
in
in
blue
store,
so
I
just
need
to
read
the
code
and
figure
out
why
it
doesn't
appear
to
be
a
config
option.
I
don't
see
any
with
that
code,
so
maybe
there's
something
else.
I
need
to
do
there
to
to
make
it
work
in
this
kind
of
you
know
not
even
pr
state
yet.
A
So
I'm
still
working
on
that,
but
hopefully
next
week
I'll
have
more
to
share
and
that's
that's
all
I've
got
guys,
you
know
short
meeting
any
anything.
Anyone
else
wants
to
bring
up
this
week.
A
All
right
well,
then,
we're
wrapping
up
very
early
today
have
a
great
week.
Everyone
and
we'll
see
you
next
week.