►
From YouTube: Ceph Performance Meeting 2022-08-18
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
A
That
was
a
new
PR.
This
weekend
was
also
merged
right
away.
So
if
they
get
nice
fast
turnover
times,
they
get
to
work
on
much
more
development
oriented
code
right
now,
so
that
one
is
in
and
then
the
other
closed
PR
was
in
rgw.
There
was
this
deos
back
end
blue
code
that
was
submitted,
which
is
kind
of
interesting.
A
It
would
potentially
let
rgw
were
on
top
of
deos,
which
is
kind
of
like
a
the
successor
to
luster
in
a
way,
I,
don't
exactly
know
what
their
relationship
is
like
now,
since
Intel
I
think
sold
off
most
of
their
their
luster
assets
to
ddn
and
but
they
kept
deos.
So
I
don't
know
what
their
relationship
is
like
now,
but
in
any
event,
this
PR
was
going
to
do
that,
and
it
sounds
like
the
author.
A
Now
has
closed
it
in
favor
of
a
new
PR
that
I
assume
uses
a
a
better
library
or
a
newer
version
of
the
library
for
for
deos.
So
we'll
we'll
see
when
and
if
a
new
PR
comes
out
for
that
I
saw
three
updated
PRS
this
week.
This
first
one
is
adding
primary
balance
scoring
to
the
balancer
from
Josh
Solomon
that
that
showed
up
I
think
last
week
and
there's
just
been
some
review
and
discussion.
I
think,
radic
and
and
Laura
have
been
reviewing
it.
A
A
I
think
that
now,
basically
is
ready
to
move
forward.
I,
don't
I,
don't
think
there
I
didn't
see
any
additional
concerns
on
it
and
then
finally
Igor
your
your
PR
for
getting
rid
of
status
updates
on
each
transaction.
It
looks
like
that's
been
added
to
whip
Theory
testing,
so
hopefully,
hopefully
we
can
merge
it
soon.
A
And-
and
that
is
what
I
saw
for
for
PRS
this
week-
I
didn't
quite
make
it
through
the
end
of
the
list,
but
I
I
suspect
there
probably
wasn't
anything
in
the
last
like
10
years
that
that
changed
this
week.
So
if,
if
there
was
probably
only
close
by
the
bot,
so
did
I
miss
anything
this
week
from
anybody.
A
Okay,
well,
then,
moving
on
the
only
discussion
topic
I
have
this
week
is
that
on
the
the
tombstone
roxdb
Tombstone
removal
project,
Adam
had
made
a
suggestion
when
I
spoke
to
him
last
week
that
maybe
we
should
try
to
see
if
we
can
modify
rocksdb
itself
to
issue
mem
table
flushes.
When
we
see
a
certain
number
of
tombstones
during
iteration
like
they
do
with
compaction,
so
I
went
and
looked
through
some
of
the
Rock's
DB
code.
A
That
does
this
and
kind
of
legitimate
is
I
I,
don't
think
it
would
be
necessarily
super
straightforward
and
easy
to
do.
It
might
not
be
bad,
but
it's
it's.
It's
definitely
not
like
an
easy,
easy
thing
to
implement
right
now,
there's
a
couple
of
things:
this
code
that
they
currently
have
only
works
for
elite
tombstones
and
not
the
elite
range
tombstones.
A
So
that's
one
kind
of
strike
against,
even
even
using
this
for
compaction,
which
is
you
know,
was
for
at
the
moment,
but
then
also
what
the
code
does
is
basically
just
sets
a
flag
that
is
read
by
some
kind
of
background
compaction
thing
to
know
that
a
compaction
should
happen.
So
it's
not
like
you're
doing
a
direct
triggering
of
compaction
when
you
hit
a
certain
number
of
tombstones
during
iteration.
A
It's
just
that
you
you
set
a
flight
to
asynchronously
later
on,
go
do
compaction
whenever,
whenever
that
other
part
of
the
code
gets
around
to
looking
at
it,
the
the
gist
of
it
is
that
most
of
the
work
is
actually
being
done
in
the
compaction
part
of
the
code.
This
thing
that
manages
it
so
I
didn't
see
evidence
that
there's
quite
the
same
infrastructure
on
the
mem
table.
A
Flush
side
of
this
there
might
be
maybe
I
just
missed
it,
but
but
we'd
have
to
do
something
similar
or
we'd,
set
a
flag
that
that
we
should
flush.
A
All
that
combined
I'm,
basically
back
to
thinking,
maybe
the
right
place
to
do.
This
is
in
our
own
KV
layer,
where
we
can
just
track,
deletes
and
delete
ranges
and
and
manually
issue
the
compaction
or
the
flush
ourselves.
When
we
see
that
we've
issued
a
certain
number
of
deletes
and
that
code
I
believe
would
be
a
lot
simpler
and
easier
to
deal
with
than
trying
to
do
like
modify
rocksdp
and
kind
of
slip
in
our
own
stuff
there.
A
So
anyway,
that's
that's
where
I'm
at
on
that
I'll
probably
have
to
Adam
more
about
it
on
Monday
and
see
what
he
thinks
that
here
today,
but
that
that's
basically
all
I've
got
any
thoughts
or
questions
or
comments
on
that
stuff.
A
A
A
All
right
well,
then,
thank
you
all
for
coming
and
enjoy
the
rest
of
your
day
and
enjoy
your
week
thanks
guys.