►
From YouTube: State of the Cephalopod 2022
Description
Join us November 3-16th for different Ceph presentation topics! https://ceph.io/en/community/events/2022/ceph-virtual/
A
A
B
I'll
speak
to
the
the
community
quickly,
so
we
wanted
to
say
a
bit
about
the
the
Seth
Foundation,
because
maybe
people
aren't
aware
of
the
different
companies
that
are
supporting
Upstream
stuff,
so
we
have
cueta.
Over
the
last
year,
we've
maintained
a
healthy
self
Foundation
membership.
We've
got
10,
Premier
Members
all
shown
as
on
this
slide
there
and
there's
also
13
General
and
11
associate
members
the
set
Foundation.
You
know
it's
responsible
for
for
coordinating,
kicking,
kicking
off
and
and
coordinating
the
events,
but
it's
also
funding
labs
for
Upstream
testing.
B
It's
funding,
a
documentation,
writer
I,
see
Zach
is
on
the
call,
and
it's
also
coordinating
some
and
funding
some
marketing,
some
kind
of
generic
self
marketing
the
board,
which
is
the
staff
Foundation
board
which
directs
this
this
all
of
this
resources
is
composed
of
those
10
Premier
members
and
then
also
one
elected
General,
member
and
one
elected
associate
member
and
that's
meeting
monthly
now,
and
it's
it's
quite
taking
over
quite
well
now.
Another
initiative
recently
from
from
the
last
few
months
is
there's
also
a
user
Dev
monthly
virtual
event.
B
That's
nicely
coordinated
by
Laura.
That's
that's
on
the
call
I've
seen
so
that's
you
know.
I
I
hope
that
many
of
you
have
have
attended
that
already,
but
it's
a
it's
a
forum
where
you
can
connect
and
interact
directly
with
the
Upstream
stuff
developers
post.
Some
questions
hear
about
new
features
participate
in
in
testing
new
features.
Things
like
that.
B
There
was
a
there
was
a
kind
of
hackathon
Grace
Hopper
open
source
day,
where
we
had
had
a
few
developers
participating
on
some
kind
of
paper,
cuts,
nice,
low-hanging,
fruit
issues,
I
hope,
I,
hope
people
are
also
paying
attention
to
the
regular
Tech
talks
that
are
announced
on
the
mailing
list.
Those
are
quite
useful
if
we
want
to
Deep
dive
into
certain
topics,
and
here
we
are
with
cephalicon
virtual
2022.
C
Thanks
Don
all
right,
so
in
terms
of
the
release
process,
like
you
all
heard,
there
have
been
a
lot
of
improvements,
but
in
terms
of
the
release
schedule,
we
stick
to
maintaining
two
releases.
At
a
time
we
are
currently
working
on
the
reef
release.
You
can
see
that
on
the
shot
there
we're
here
and
next
we
are
going
to
start
working
on
the
S
release
and
very
interestingly,
we
started
discussing
what
the
S3
should
be
named
yesterday.
C
C
C
C
C
So
we
decided
to
create
this,
give
up
stream
sepia
lab
cluster,
which
is
essentially
a
cluster
of
thousand
osds,
but
it
should
be
noted
that
it's
logical
that
scale-
it's
not
real
world
setup,
but
even
even
then,
it
allows
us
to
think
test
things
at
a
scale
and
for
the
Quincy
release.
We
made
sure
that
once
we
had
our
debt
freeze,
we
were
continuously
testing
bills
in
this
particular
cluster
and
there
were
a
lot
of
issues
that
got
identified
and
fixed
before
we
actually
did
the
first
couple
of
RCs.
C
So
this
is
an
ongoing
process.
It's
not
that
the
it
was
just
on
frequency.
We
are
also
using
this
givea
cluster
to
test
Point
releases
or
for
Quincy
at
this
point,
and
it
must
be
also
noted
that
this
is
on
top
of
the
L
lrc
that
we
already
have
the
long
running
cluster
is
the
real
word
cluster.
It's
not
a
scale
of
thousand
osds,
so
we
at
least
have
two
Upstream
clusters
that
we
are
actively
testing
Point
releases
on.
Beyond
that
there
were
other
clusters.
C
There
was
a
red
hat
scale
lab
cluster.
That
was
also
emulating
some
of
the
work
that
you
were
doing
in
Posse
and
also
validating
the
fixes
that
we
made
for
it,
and
this
went
up
to
8,
000
osds
of
logical
scale.
A
lot
of
other
boundaries
were
pushed
and
issues
were
also
highlighted
and
fixed
for
the
reef
release.
And
finally,
the
there
was
a
lot
of
effort
from
Mark
Nelson
and
the
Upstream
developers
to
do
high
performance,
classic
testing
and
I.
C
Think
we'll
talk
about
the
details
in
the
later
slides
for
those
interested
there's,
also
a
Blog
series
of
blog
posts
that
have
been
done
on
the
performance
workbook
that
we've
done
in
Quincy
so
feel
free
to
search
for
those
all
right
with
that.
Let's
talk
about
what's
new
in
different
components
and
stuff,
starting
with
rados.
C
So
in
Quincy,
the
most
exciting
thing
I
would
say
that
came
in
was
qos
in
the
OSD
by
default,
which
means
that
there
is
no
longer
a
need
for
users
to
play
with
sleep
settings
and
manual
configuration
options
to
be
able
to
prioritize
client
versus
background
operations.
Background
operations
here
means
recovery,
scrub,
snap
trim,
video
deletion,
kind
of
things.
So
that's
pretty
cool
and
there
was
a
lot
of
effort
that
it
was
put.
C
So
some
of
you
may
be
aware
that
the
code
definitely
existed
in
Pacific,
but
it
was
not
on
by
default,
so
the
Quincy
cycle
was
used,
stabilized
bug,
fix
and
performance.
You
know
test
and
clock
the
the
algorithm
that
provides
us
and
safe
and
with
enough
confidence
we
were
able
to
make
it
default
and
that's
support.
The
second
Point
covers
the
extended
testing
that
we
did
across
workloads
and
across
different
kinds
of
scale
in
Blue
Store.
A
lot
of
exciting
things,
I've
listed
a
few.
C
A
few
of
them
that
might
be
interesting,
is
that
we
removed
allocation
metadata
from
Rocks
TV,
and
this
essentially
significantly
improved
small
right
performance,
and
this
shows
up
in
our
performance
slides
that
we
have
also
shared
earlier.
We
also
made
blue
FS
improvements
and
the
final
grain
locking
piece
is
the
one
that
we
are
very
excited
about
this
other
than
core
cleanup.
This
also
helps
us
nail
down
a
lot
of
bugs
that
we've
been
seeing
in
this
area
in
blue
FS,
so
this
has
been
a
pretty
major
effort
in
Quincy.
C
Other
than
that,
we
also
have
cache
pinning
that
has
been
implemented
and
there
are
other
improvements,
also
like
zero
block
detection.
That's
just
for
large
scale,
testing
that
we
are
using.
The
idea
is
that
we
should
be
able
to
write
zeros
to
Blue
Store
for
large
scale
testing
without
filling
up
the
actual
devices.
That's
another
ongoing
effort.
That's
going
on,
but
moving
on.
We've
also
got
messenger
improvements,
so
we
now
have
on
the
wire
compression.
So
we
already
had
encryption
so
with
Quincy.
We
have
on
the
wire
compression
for
osd2
SD
communication.
C
This
is
not
on
by
default,
but
we
would
love
to
hear
some
feedback
about
this
there's
there
is
extensive
documentation
about
it
as
well.
There's
more
cons,
concise
and
meaningful,
explore
Ops
reporting
in
the
cluster
log.
Now
we've
added
Health
warnings
yeah.
This
must
be
noted
that
in
Quincy
we
added
warnings
for
file
store
deprecation
in
The
Reef
release.
We
will
be
removing
support
for
file
store.
C
So
if
you
are
currently
on
Milestones,
maybe
of
importance
to
you,
there's
also
another
warning
about
the
required
release
flag
because
sometimes
users
don't
set
it
and
end
up
shooting
themselves.
In
the
foot,
there's
also
I
will
talk
about
the
Telemetry
stuff
in
the
later
sites,
but
there's
an
improved
opt-inflow
and
Telemetry,
starting
with
Quincy
next
slide.
Please.
C
All
right
so
with
the
reef.
What's
coming,
this
client
versus
client,
GEOS,
initial
implementation
and
this
initial
implementation
is
in
liberators.
The
support
for
client
versus
client
and
liberators
will
be
added
in
in
the
reef
release,
and
there
is
a
lot
of
testing
that
needs
to
be
done
so
with.
It
must
be
noted
that
it's
a
complex
piece
of
work.
C
So
we
are
working
on
a
mechanism
for
M
clock
to
also
do
that
in
the
reef
release
in
terms
of
blue
store,
lots
of
exciting
and
high
performance
work
coming
in
there's
going
to
be
a
custom
wall.
Support
for
Rocks
TV,
so
we
are
getting
rid
like
the
idea,
is
to
use
the
custom
wall
instead
of
of
roxdbs,
and
that
comes
with
a
little
plethora
of
performance
improvements
with
with
that
there's
other
yes,
there's
a
new
one
tracker
per
object
mode
that
will
replace
the
shared
blob
logic
here.
C
The
ideas
that
we've
identified
on
a
bunch
of
bottlenecks
in
in
the
current
code,
which
show
up
in
especially
things
like
testing
of
RBD,
mirroring
and
when
the
workloads
are
snapshot
streaming
extensive.
So
this
is
going
to
give
us
performance
improvements
in
those
areas.
And,
finally,
this
4K
allocation
unit
for
blue
FS
been
seeing
a
lot
of
discussion
in
the
user
list
around
this,
where
users
are
running
into
blue
FS.
You
know
space
in
in
shared
devices
where
both
your
DB
and
data
are
on
the
same
device.
C
So
this
is
going
to
be
an
improvement
in
that
area.
Still
looking
forward
to
this
as
well
beyond
that,
we've
got
balancer
improvements
so
as
we
all
know
that
the
balancer
is
on
by
default,
but
it
does
not.
It
currently
just
takes
into
account
the
number
of
pgs.
The
idea
here
is
to
add
another
mod
in
the
balancer,
which
optimizes
reads
in
the
sense
that
the
primer,
the
number
of
primaries,
that
particular
OSD
is
hosting.
This
will
not
be
on
by
default.
C
It's
going
to
be
added
as
a
support
in
the
USD
map
tool.
First
and
based
on
the
feedback
we
get
we're
gonna
at
yeah.
Add
it
to
the
default
code
base
beyond
that
some
other
improvements,
so
there's
going
to
be
work
in
the
area
of
capturing
data
availability
score.
This
will
have
an
initial
implementation
for
this
and
reef,
because
the
idea
is
that,
apparently,
we
have
snapshots
like
set
status,
but
we
don't
have
anything
that
tells
us
availability
over
time.
C
C
All
right
now,
let's
talk
about
the
Crimson
project.
I
know
a
lot
of
folks
are
very
excited
about
this
and
waiting
for
this.
So
for
those
who
are
not
aware,
the
Crimson
project
is
basically
a
high
performance.
Readout
of
the
USD
currently
supports
RBD
workloads
and
replicated
pools
with
the
blue
store.
Backend.
C
We
don't
backboard
things
for
for
Crimson,
it's
really
cutting
edge
and
there
are
lots
of
improvements.
Getting
added
as
we
speak.
So
what's
coming
in
reef
is
the
initial
multi-reactor
support.
The
S
release
will
add
further
improvements
to
the
messenger.
In
this
area
there
is
going
to
be
usability
improvements,
so
we'll
have
Flags
such
as
Crimson
pool
type,
which
will
ensure
that
Crimson
osds
don't
boot
and
unless
these
flags
are
set.
C
So
this
allows
users
to
not
show
themselves
in
the
foot
again,
there's
going
to
be
initial
support
for
snapshot,
Snapchat
main,
which
is
very
exciting,
so
that
fits
in
very
well
with
the
RBD
workloads.
There's
going
to
be
improvements
in
c-store,
there's
Next
Generation,
Object,
Store
implementation,
going
on
that's
going
to
be
focused
on
efficiency
and
touring
and
in
terms
of
what's
currently
planned
for
the
S
release.
Is
the
c-store
multi-reactor
support
cheering
support?
C
C
All
right,
like
I,
said
I
wanted
to
touch
upon
Telemetry
a
little
bit
and
there's
going
to
be
a
full-fledged
talk
about
it
later.
So
what
came
into
Quincy
was
an
enhanced
opt-in
flow.
Why
is
this
important,
and
why
is
the
school
is
because
this
allows
developers
to
collect
data
based
on
new
features
that
that
we
think
are
important,
so
it
provides
a
lot
of
flexibility
to
add
metrics
that
we
want
to
collect
in
terms
of.
C
Why
is
it
important
to
users
is
that
they
can
keep
sending
only
what
data
they
opted
to
send
in
in
the
first
place,
whenever
there
is
any
new
data
to
be
collected,
we
don't
force
opt-in,
so
so
you
can
re-opt
it.
If
you
are
interested
in
sending
that
particular
data
explicitly
we
make
users
aware
of
the
data
sharing
license
as
well.
C
Telemetry
has
had
a
bunch
of
channels
like
we
know
basic
crash
device
indent,
but
in
Quincy
we
added
a
new
channel,
which
is
the
both
Channel.
It
is
off
by
default,
but
this
is
a
a
huge
Improvement
in
terms
of
what
we
can
achieve
with
Telemetry
is
that
it
helps
us
gather
performance.
Metrics
that
developers
can
go
and
look
at
come
up
with
optimizations
look
at
bottlenecks
and
do
things
like
that,
so
I
think
that's
pretty
cool
there's
been
focus
on
crash
report
analysis.
C
So
a
lot
of
you
who
use
the
safe
tracker
may
already
be
looking
at
this
Telemetry
creates
automatic
tracker
reports
with
the
crashes
that
are
getting
reported
from
user
clusters.
So
there
is
the
integration
with
our
redmi
bug
tracker.
We
have
daily
reports
of
top
crashes
that
are
happening.
This
is
very
important
when
we
especially
do
new
Point
releases.
C
Telemetry
helps
us
understand
if
there
is
a
regression
that
we
accidentally
introduced
and
didn't
get
caught
in
our
integration
testing.
There
is
a
fancy
internal
dashboard
that
some
of
the
users
don't
have
access
to
directly,
but
developers
do
but
for
for
users
they
always
can
look
at
the
public
dashboard.
That
has
also
been
improved
a
bunch
in
the
Quincy
release
and
there's
also
extensive
device
dashboards.
So
you
can
see
which
HDD
or
SSD
models
of
safe
users
are
deploying
next
slide.
Please.
C
All
right
so,
what's
coming
in
Reef
the
this
continued
work
on
the
backend
analysis
of
the
Telemetry
data,
so
all
the
performance
data
that
we
are
collecting.
We
aim
to
analyze
it
in
ways
that
are
meaningful
to
us.
So
there's
work
going
on
in
that
direction.
There
are
tools
for
developers
to
use
crash
reports
and
identify
and
prioritize
bugs.
C
One
of
the
most
interesting
things
that
we
added
was
also
the
ability
to
identify
which
orchestrator
module
is
being
used
in
safe
cluster,
and
with
that
we
now
have
the
ability
to
understand
how
many
Rook
clusters
are
there.
How
many
safety
DM
clusters
are
there,
and
you
know
what
other
exists
that
we
don't
know
about
beyond
that.
C
We
will
also
be
expanding.
The
data
set
via
safe
collector
and
other
mechanisms
like
Standalone,
collector
and
other
data
resources,
so
overall
lots
of
exciting
stuff
coming
into
Telemetry.
So
if
you
still
haven't
opted
in
this
is
the
time
to
do
it.
I
would
say
with
that
I
think
I'm.
At
the
end
of
my
my
slides.
Let's
move
on
to
the
next
and.
A
Yeah
thanks
Leah
yeah,
so
I'll
go
over
the
other
components
and
stuff
pretty
briefly
at
a
high
level.
There's
more
detail
on
the
slides
that
you
can
look
into
later.
I'll
share
these
on
the
mailing
list.
After
the
cut
talk,
so
in
the
dashboard
realm,
there
are
a
number
of
improvements
in
Quincy
a
lot,
a
lot
related
to
some
new
Services.
A
You
can
deploy
like
SNMP
I'm,
trying
to
make
it
easier
to
do
certain
things
like
expanding
your
cluster
and
it's
playing
NFS
and
there's
now
an
integrated
way
to
report
bugs
with
stuff
directly
through
the
dashboard.
So
you
can
you
don't
even
have
to
log
into
redbine
or
anything
or
create
an
account.
You
can
just
directly
if
you
see
an
as
you
report
it
through
the
dashboard
in
a
very
simple
way.
A
You
can
actually
even
do
this
in
the
CLI
as
well,
if
you're
not
even
using
the
dashboard,
so
it
makes
it
a
little
bit
easier
to
report
ebooks
you
find
with
for
at
least
going
and
going
forward
in
the
dashboard.
That's
more
focused
on
kind
of
more
operational
improvements
for
for
day,
two
sorts
of,
or
the
three
like
after
your
initial
install.
What
can
you
do
with
the
cluster?
A
A
A
A
A
A
With
if
this
version,
Rook
also
adds
that
true
plugin,
which
makes
it
easier
to
troubleshoot
and
to
maintain
and
debug
anything
going
wrong
with
osds
or
monitors,
so
you
can
easily
start
up
the
AE
container
with
the
same
versions
and
the
same
access
to
disks
of
a
given
monitor,
OST
demon
has
and
repair
it
if
needed.
That
sort
of
thing
I
also
had
a
new
support
for
NFS
snapshots.
Restoring
cloning
and
resizing.
A
One
is
around
for
around
newer
interfaces
so
that
the
there
a
few
there's
existing
ways
to
interact
with
RBD
main
ones
being
the
kernel
driver
and
the
user
space
Library,
which
is
often
exposed
through
qmu
super
and
VMS,
a
third
one
weighs
also
by
scuzzy.
This
is
not
really
being
focused
on
it
so
much
these
days
and
said
that
we're
focusing
more
on
in
nvme
of
Gateway,
which
is
a
bit
of
a
collaboration
between
IBM,
Intel
careers
and
red
hats.
A
A
lot
of
interesting
research
going
into
a
new
long,
structured
data
format
beheaded
by
the
Massachusetts
open,
Cloud
folks,
particularly
Peter
destroyers
from
Northeastern
University
he's
got
a
getter
repo
available
here
and
it
just
very
promising
initial
performance
numbers
and
it
looks
like
a
very
interesting
potential
path
in
the
future.
But
is
it
at
this
point?
It's
a
kind
of
early
days
going
from
a
research
prototype
and
starting
to
become
something
that
is
more
more
usable
in
the
real
world.
Hopefully,
in
the
next
couple
releases.
A
For
rgw
the
readers
Gateway,
there
are
additional
S3
apis
and
enhancements
to
those,
including
server
for
server-side
encryption
and
for
SV
select,
which
is
a
a
way
to
kind
of
query
data
and
to
only
get
back
the
bits
and
pieces
that
you
actually
need,
instead
of
making
an
entire
kubernetes
objects.
This
has
a
lot
of
integration
into
a
number
of
different
kinds
of
analytics
tools
and
the
addition
of
the
parquet
object
format
allows
even
more
things
to
work
with
it
quite
seamlessly.
A
There
also
is
a
lot
of
focus
on
multi-site
improvements,
including
enabling
you
to
recharge
buckets
that
are
being
replicated
and
optimizing.
The
way
that
the
data
is
stored,
as
well
as
improving
the
load
balancing
of
application
so
that
it
all
narrative
views
can
participate
in
not
just
a
few
of
them.
A
We
also
a
lot
of
work
going
into
the
rgw
zipper
projects,
which
is
essentially
a
abstraction
layer
that
allows
I
VW
to
run
on
different
backends,
there's
a
kind
of
a
prototype
backend
based
on
a
SQL
database,
Brands
equator,
other
sorts
of
SQL
databases
and
maybe
maybe
quite
useful
for
disconnected
the
workloads
where
or
perhaps
very
small
deployments.
But
you
don't
want
to
run
a
full-blown
stuff
cluster
underneath
something
also
it's
quite
exciting,
to
see
the
distributed.
A
And
finally,
it's
ffs
there's
an
again
a
number
of
amount
of
large
Focus
around
performance
and
stability,
but
initially
that
there's
a
few
newer
features
being
at
work
done
so
first
ffs
top
and
Quincy.
There
are
a
number
of
more
useful
metrics
being
added,
so
you
can
track
what
clients
are
doing
more
precisely.