►
From YouTube: 2019-04-25 :: Ceph Performance meeting
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,
so
there's
been
a
lot
of
movement
this
week
on
pull
requests
partially
because
our
stale
BOTS
now
automatically
closing
stuff.
So
whole
bunch
of
old
PRS
got
closed.
I,
don't
know
if
that
anyone
here
had
anything
closed
right
now,
Braddock
had
a
bunch
of
stuff
closed.
It
looks
like
Adam
I,
think
you're
survived
it
because
you've
been
periodically
updating
yours,
so
I
think
you're
in
good
shape.
A
B
Well,
originally,
we
didn't
issue
in
very
long
enumeration
of
mops
under
some
conditions.
I
implemented
blue
FS
read
ahead
on
my
own,
and
then
we
realized
that
Europe's
DB
embedded
in
Nautilus
has
gone
ahead
and
I
tried
to
well
I
verified
how
it
works,
and
it
appeared
that
actually
no
read
ahead
is
happening
and
I
did
some
investigation,
which
showed
that
we
need
to
implement
the
fetch
method
in
blue
ox
environment
and
hence
I
made
that
well.
But
no
I
have
no
much
difference.
A
A
B
B
A
Automatic
iterator
read
ahead
is
available,
starting
at
five
point:
12
for
buffered,
I/o
and
5.15
for
direct
I/o.
So
is
this?
Is
this
different
than
what
you're
talking
about,
or
is
this
same
thing
the
same
thing
yeah
same
thing:
okay,
okay,
so
at
least
for
older
I'm,
pretty
positive
for
luminous
and
mimic
at
least
we'll
need
to
upgrade
rocks
TB
to
get
this
I,
don't
know
if
I
remember
for
Nautilus,
if
we
upgraded
actually
be
sufficiently
or
not.
B
A
Yeah,
alright
cool,
alright,
so
there's
a
clean
up
PR
here
from
ma,
Jinping
I,
don't
actually
know
how
much
that
affects
performance,
but
it
was
listed
as
with
a
performance
tag,
so
I
included
it
there's
some
ongoing
discussion
here
about
introducing
execution
stage
in
crimson
OSD.
Damn
I
saw
you
commented
on
that
red
X
night
here,
but
did
you
have
any
any
thoughts
on
the
discussion.
A
A
A
A
A
A
A
D
D
E
D
It's
part
of
the
test
merge,
but
I'd
rather
I
mean
it's
been
tested.
That's
a
bug
fixes
to
it,
but
since
I've
already
had
to
make
a
couple
tweaks
just
to
merge
it
with
rgw
so
far,
I'd
rather
just
make
sure
I
have
a
working
rgw
first,
basically,
so
we
have
basically
I'd
rather
hat
makes
for
the
interface
has
a
has
a
that.
Someone
consuming
the
interfaces
in
the
codebase
before
we
merge
the
interface.
E
D
F
D
I'm
expecting
that
I'm
going
to
get
done
with
the
rgw
part
pretty
soon
once
we
get
that
through
testing,
that
should
you
know
make
every
be
able
to
make
everything
you
want
in
a
big
chunk
mean
doesn't
matter
to
me.
It
was
a
matter
of
in
general
of
them
if
it
merges
in
general,
as
I've
mentioned
the
former
I
plan,
to
make
some
changes
to
the
interface
anyway,
but
I
can
always
make
changes
to
it
after
it
gets
merged
in,
like
once,
Jason
Dillman's
committed
well.
F
First
of
all
you
just
that
is
before
you
said
earlier
that
it
should
there
should
be
a
consumer
that
works,
whether
it's
in
the
codebase
or
not.
I,
think
this
is
sound,
so
it
shouldn't
really
merge
and
tell
there
until
you're
finished
debugging.
The
one
client
that
currently
exists
so
that
when
it
goes
in
something
communal,
could
immediately
in
theory,
use
it
and
it's
legit
yeah.
A
So
only
other
updated
PR
I've
got
here
is
that
sage
really
doesn't
want
to
lose
the
work
that
Adam,
the
other
Adam
had
done
on
age.
Testing
of
the
different
blue
store
allocators,
though
I
think
he's
hoping
that
maybe
Adam
can
take
some
of
that
work
and
turn
it
into
like
a
unit
test
just
that
we
can
go
back
and
retest
that
stuff
easily
again
in
the
future.
A
Let's
see
lots
of
no
movement
but
many
fewer
than
then
there
were
previously,
though
one
I'm
still
very
interested
in
is
the
IOU
ring
PR.
That's
here,
I
think
I
may
try
to
get
a
5:1
Colonel
installed.
The
one
of
the
insert
notes.
I
can
try
this
because
the
results
look
pretty
pretty
inspiring
I
did
happen.
It
I'll
warn
everyone
that
yen's
put
out
an
April
Fool's
joke
where
he
said
that
he
was
gonna.
Remove
Li
Bao
from
the
colonel
in
favor
of
you
ring
in
five
one
and
I.
A
A
Ya
various
other
things
in
here,
not
nothing,
I,
think
that
has
a
whole
lot
of
movement
on
it
right
now,
alright,
so
the
the
only
thing
I
really
have
to
mention
is
that
Adam
has
been
continuing
to
work
on
looking
at
tea,
rocks,
dB
and
he's
still
not
necessarily
getting
like
super
high
levels
of
performance
or
anything
out
of
it.
I
think
it
was
pretty
similar
to
what
we're
seeing
with
rocks
DB,
but
he
has
verified
that
the
write
amplification
is
way
way
lower.
A
So
we're
writing
much
less
data
out
to
the
the
DB
devices,
which
is
really
good,
I.
Think
because
I've
been
seeing
more
and
more
hardware
configurations
coming
out
where
people
are
sticking,
the
DB
on
on
nvme
drives
or
SSDs
that
have
very
low
drive
or
drive
writes
per
day
durability,
though
he
did
mention
that
it's
hitting
the
block
device
that
were
not
keeping
all
the
data
on
the
the
the
DB
partition
in
in
a
way
that
he
would
have
expected
so
he's.
A
Looking
into
that
to
try
to
figure
out
why
I
don't
know
if
it's
just
a
temporary
expansion
of
space
space,
amplification,
that's
needed
or
or
what
is
going
on
exactly
but
he's
trying
to
figure
it
out.
So
then
that
might
actually
improve
performance
down
the
road
we'll
see,
and
that's
that's
basically
it
that's
all
I've
got.
Does
anyone
have
anything
they'd
like
to
talk
about
this
week.
A
A
compiling-
and
there
are
some
kind
of
minimal
but
important
changes
that
are
made
in
the
build
process,
though
it's
not
100
percent
drop
in
with
existing
rocks
DB.
My
hope
is,
if
this
pans
out-
and
it's
really
really
useful
for
us-
maybe
we
can
go
back
to
the
rock
ski,
be
guys
into
the
toshiba
guys
and
say
you
know,
let's,
let's
get
this
in
rocks
to
be
itself.
You
know
this
is
this
is
crazy
having
a
totally
separate
distribution
for
this?