►
From YouTube: 2020-05-14 :: 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
until
well,
while
they're
still
deliberating
in
the
court
meeting,
we
can
just
get
started.
Let's
see
so
the
only
new
PR
I
saw
this
week.
That
appeared
to
be
performance
related
is
that
we
are
looking
at
upgrading
to
newest
rocks
DB
there's
a
lot
of
fixes,
including
some
data,
corruption,
fixes
and
crashes
and
the
start
fixes.
So
all
of
that
looks
really
good.
A
There's
a
number
of
performance
improvements
that
are
listed
in
various
different
ways,
try
to
remember
some
of
the
things
that
were
in
there
from
the
last
release
that
we
were
using,
but
just
various
things,
I
think
some
iterator
speed,
ups
other
cache
improvements
and
just
other
general
performance
enhancements.
So
theoretically
it
looks
good,
we'll
see,
I
think
keep.
You
is
going
to
try
to
run
it
through
his
new
Jenkins
PvP
test
framework.
A
A
A
This
parallel
crush
calculation
for
the
balancer
PR
that
was
closed
without
comment.
I,
don't
know
if
that's
because
it
it's
just
kind
of
sitting
there
for
a
while
and
I'd
been
rearview
sage
had
previously
been
reviewing
that,
but
since
he's
on
leave
now
for
a
little-
maybe
maybe
that
was
just
close,
because
no
one
else
had
looked
at
it.
Yet
I
don't
know
so
anyway
that
closed
in
terms
of
updated
PRS.
A
B
I'm
not
sure
there
was
a
particular
assert
that
Patrick
wanted
me
to
fix,
which
I
did
well
I.
Think
I
did
but
it's
difficult
to
tell
with
the
fest
tests,
not
working.
Okay,
well,
I
think
I'm
just
going
to
make
sure
that
I'm
no
worse
than
what
master
is
doing
sure
and
if
I'm
no
worse
than
masters,
do
then
master.
Then
we
should
be
able
to
go
ahead.
I
think
cool.
A
Yeah,
it
I,
don't
know
how
long
we'll
take
this
of
the
fest
test
working
again,
but
personally
it
seems
like
if
we're
talking
master
we've
got
quite
a
bit
of
time.
Let's
just
get
it
hit.
This
is
my
thought,
but
that's
just
me
all
right.
Let's
see
what
else
updated
OS
blue
store
up
was
a
lot
was
to
reprocess
all
this
one.
So
the
the
deal
here
is
that
I
think
everyone's
a
little
scared,
because
the
blue
store,
locking
is
complicated
and,
and
previously
it
was
as
easy
to
say
no
because
it
was
breaking
they.
A
Oh
is
it
was
not
passing
technology
tests,
so
that
was
you
know,
an
easy
answer,
but
now
they've
they've
claimed
that
they
were
able
to
fix
the
problem
from
before
and
it's
passing
now
so
I
guess.
Our
first
job
is
just
to
rerun
it
through
Keith
who's,
whip
testing,
suite
and
if
it
passes,
then
we
have
to
really
the
more
difficult
job
of
actually
going
through
it,
and
one
of
us
needs
to
figure
out
how
whether
or
not
this
is
actually
safe,
which
is
gonna,
be
harder.
A
So
I
don't
know,
I
guess,
ki,
foo
and
I
will
both
maybe
try
to
muddle
our
way
through
and
look
at
it.
But
if,
if
it
is
safe,
it
looked
like
it
was
a
nice
performance
improvement,
it's
just
and
yeah.
We,
he
probably
probably
should
do
that
anyway.
Probably
one
of
us
needs
to
actually
become
familiarize
ourselves
with
the
locking
here,
but
anyway
that's
on
us.
So
let's
see
what
else
lots
of
no
movement
stuff.
A
That
Adam
there's
your
other
one
here,
the
FIFO
data
log,
one
and
then
Sam's
ceased
or
initial
PR.
That's
getting
a
lot
of
reviews,
but
no
no
specific
movement.
This
week,
my
MDS
expected
files
PR
that
got
some
early
review
from
young,
but
hasn't
seen
much
movement
lately,
I'm
working
on
building
on
top
of
that
to
try
and
pre
export
directories
that
had
fragments
created
as
part
of
the
expected
files
PR
here.
A
So
the
idea
there
is
that
you,
both
pre
split
the
directories
into
fragments
and
then
now
that
you've
got
a
bunch
of
fragments
in
advance
instead
of
using
the
balancer
to
try
to
migrate
those
around
when
they're
hot,
which
is
kind
of
when
it's
hard
to
do
so,
because
they've
got
files
and
the
vssr
hot,
and
so
it's
taking
a
long
time,
there's
all
kinds
of
contention
already,
if
you've
already
pre
fragmented
them.
What
I'll?
A
Maybe
if
you
just
do
a
round-robin
or
a
hash
based
distribution,
the
fragrance
for
a
directory,
that's
specifically
expecting
to
have
lots
of
files-
and
maybe
these
are-
is
also
mark
that
they
expect
to
be
hot.
So
we'll
see
see
if
the
helps
I
suspect
it'll
be
really
fast,
but
then
it's
you
know
more
work
for
the
user
to
make
those
front
those
hints
so
anyway,
what
else
this
reduce
memory
footprint
one
from
Igor.
We
need
to
get
that
in
soon
I.
Think
Oh
keepers
get
that
would
keep
you
testing
so
hopefully
that
past.