►
From YouTube: 2019-10-01 :: Crimson SeaStor OSD Weekly 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
Hello
I'll
have
to
look
at
a
good
I
forgot.
What
my
comment
was.
B
D
A
C
Carry
me
yep,
oh
okay,
so
just
like
I
didn't
make
matching
visible
progressing
constant
as
I
then
takes
the
carry
sport
to
when
the
siege
introduced
to
Tommy
on
stick
start
and
I.
Seen
as
a
nice
step,
step,
2,
oh
by
the
way
I
made
some
progress
on
I'll,
see
that
I,
the
rock
TB
its
I
turned
out
to
the
progression
is
was
introduced
by
the
so
being
so
out
of
hours.
It
out
a
threat.
C
I
think
the
next
definitely
to
to
do
more
more
testing
to
it
to
take
out
if
we
can
to
improve
the
situation
without
increasing
the
right
file
paper.
But
if,
if
I
increase
the
prayer
part
right
right,
metaball
file
path
or
the
performance
can
can
empower.
Waster,
quick,
classic
rocks,
DP
I.
Think
as
a
long-term,
non
long-term
long-term
plan
is
to
use
to
use
the
theta
rider,
Roxy,
P,
Alton
and
Illinois's
Brewster
will
be
sometime,
and
if
the,
if
the
today
rider
doctor,
not
not
work,
we
need
to
use
T.
C
A
I
would
like
to
point
out
that
the
performance
concerns
for
blue
store
are
strictly
short-term
for
the
medium
and
long
term.
We
don't
care
because
we're
writing
a
different
store
for
high
performance
devices.
Blue
store
is
only
going
to
be
relevant
for
hard
disks
and
slower
SSDs.
So
in
those
time
skills,
even
if
it
did
add
a
little
bit
of
overhead,
we
still
rather
use
it
if
it
greatly
improved
our
development
over
overhead.
That
is,
if
aliens,
if
they
only
a
nice
blue
store,
is
a
pain
in
the
ass.
C
C
A
So,
in
the
long
term,
in
soap
like
if
blue
store
is
our
plan
for
supporting
slower,
hard
disks
and
SSDs,
we
know
we
hear
about
performance
a
little.
We
don't
want
to
be
tremendously
wasteful,
but
if
we're
only
talking
about
a
few
percentage
points,
we
want
that
the
easier
to
death
path.
That
way
we
can
spend
more
effort
on
the
on
the
real
back
end.
C
C
A
E
C
A
E
Hearing
problem
so
that
thought
that
in
the
Peary
adventure
there
is
a
scheduled
event,
each
the
raddest
parameter
and
in
the
context,
when
the
finishes
rider
doing
the
uncommitted,
the
uncommitted
function,
I
used
sister
Eileen
submitted
two
two
seven
eight
function
executor
in
sepia:
zero,
whether
that
aiding
submit
to
doesn't
have
work.
It's
not.
C
E
E
A
E
E
A
E
E
A
Yeah,
that's
that's
right!
Yeah!
That's
that's
right!
So
you
have
to
wrap
as
callbacks
I
think
so
every
time
transaction
gets
passed
in,
you
need
to
look
at
all
of
the
contexts
that
are
supposed
to
be
fired
and
wrap
them
with
they
call
back
that
will
instead
schedule
the
call
back
on
the
main
core.
Is
that
what
you
were
doing.
E
E
A
Yeah,
that's
one
way
to
do
it.
You
just
make
ups
what
the
callback
fulfills
the
promise.
That's
fine.
E
C
C
E
B
B
B
B
B
D
But
I
did
manage
to
do
some
work.
I
started
to
start
understand
how
how
to
implement
socket
on
system,
and
the
first
thing
I
noticed
is.
There
is
a
small
data
structure
which
maintains
the
other
registrations,
mostly
mostly
data
structure.
Now
I
was
trying
to
think
how
to
write
this
one
in
the
system.
I
thought
about
using
Gaussian,
then
I
read
into
I.
Think
it
was
you
and
a
few
other
people
discussing
I'll,
see
you
in
the
system
two
years
ago,
a
year
and
a
half
ago
and
I
think
the.
D
D
A
D
A
C
A
D
A
A
We
need
like
it's
not
just
an
RC.
U
thing!
We
now
have
to
work
out,
okay,
which
sub
Damon
is
needs
to
handle
this
like.
If
it's
a
PG
command,
it's
not
just
a
simple
RC.
U
thing
the
PG
will
be
specific
to
a
core
and
not
necessarily
the
one
that
received
the
Edmonds
second
command.
So
it's
not
a
trivial.
A
A
Yeah
we'd
have
to
do
some
actual
work
and
it
would
depend
on
on
the
command.
We
would
have
two
bikes,
it
doesn't
think
okay,
so
they
fall
broadly
into
a
few
different
categories:
they're
broadcast
ones
where
everyone
gets
the
same
thing
and
then
we
aggregate
the
results.
Okay,
seven
are
specific.
To
course
we
have
to
route
it
to
that
core
and
then
get
a
result
back
or
maybe
they're
all
broadcast
I
don't
know,
but
like
we
don't
we
don't.
We
don't
have
to
solve.
It
now
is
I
think
what
I'm
saying
for.
D
A
D
A
B
A
A
B
C
B
A
B
A
Okay,
so
I
have
a
few
things
so
working
at
the
recovery
backfill
thing,
but
it
occurred
to
me
that
there's
no
easy
way
to
refactor
the
existing
classical
C
code
without
figuring
out
how
I
want
crimson
to
do
background,
recovers,
scheduling,
throttling
and
scheduling
so
I'm
implementing
those
parts
in
crimson
right
now,
I've
pulled
the
EM
clock
code
over
sort
of
just
the
shim
layer,
whatever
no
big
deal
and
now
I'm
implementing
the
part
that
every
one
of
these
sort
of
op
concepts
that
get
started
currently
just
starts
running
as
a
B
star
plus
word
func
continuation,
whatever
fiber,
whatever
yeah
inside
their
reactor.
A
Let's
run
right,
but
if
there
are
already
too
many
things
running
you'll
get
stopped
until
there's
a
slot,
mostly
so
that
we
can
start
things
like
background
recovery
operations
and
expect
the
bill
to
sit
there
until
there's
slack
capacity
available
for
them.
This
will
accommodate
the
this
will
accomplish
the
following
things:
in
classic
OSD.
It
will
do
the
thing
that
the
sleeps
currently
do.
It
will
do
the
things
that
M,
the
prioritized
queue
currently
does,
and
it
will
do
the
thing
that
the
background
queues
for
we
do
Oh.
C
A
Won't
have
to
work
out
what
that
actually
for
today
for
now
yeah
the
longer-term
goal
will
be
we're
writing
a
scheduler,
so
we
need
to
know
from
C
star
what
how
much
time
you're
spending
idle
so
that
we
can
increase
the
cap
if
we
need
to-
and
we
may
want
to
make
it
smarter
about
the
cap-
it
may
not
be
as
strict
number
running
it
might
be.
A
am
I
approaching.
A
Am
I
running
out
of
idle
time
when
we're
running
out
of
Io
idle
time,
all
of
the
usual
things
os
scheduler
does
to
decide
what
to
run
a
thread
so
we'll
eventually
have
to
grow
into
a
lot
of
that
stuff.
But
this
is
a
start
and
then
so.
This
was
a
I.
Think
the
third
category
on
your
roadmap
coordinating
background
I/o.
So
that's
I
think
that's
actually
a
prerequisite
for
that
goal.
Fortunately,
I
don't
think
it's
two
months
worth
of
work.
It's
a
relatively
small
piece
of
that
goal.
C
A
C
C
A
Right
now
we
have
no
back
crimson
as
it's
currently
written
has
zero
background
tasks
at
all.
All
it
does
is
do
straight
lines.
Synchronous
operations,
I
mean
not
real,
it's
not
truly
saying
for
those,
but
for
a
client
operation.
You
receive
a
message.
You
do
the
work
associated
with
it.
You
get
rid
of
the
message.
It
doesn't
generate
its
own
work,
so
backfill
can
get
its
own
work.
So
it's
time
to
solve
the
problem.
A
I
think
other
things
I
wanted
to
talk
about
were
oh
and
as
part
of
that,
I'll
also
have
to
implement
a
object,
read
and
write,
locking,
which
means
object.
State
needs
to
grow
to
be
what
object
context
currently
is
because
there
four
different
ways
to
stop
I/o
on
an
object
depending
on
which
head
on
which
background
operations
are
pending,
because
the
back,
like
the
state
machine
for
when
an
I/o
is
allowed
to
proceed
in
the
face
of
backfill,
is
like
just
a
little
bit
more
complicated
than
you'd.
Think
so,
I'll.
A
Add
that
step
as
well
same
deal,
though,
if
there's
no
actual
reason
to
block
it's,
not
gonna
block,
it's
just
an
additional
check,
but
it
does
mean
that
we
need
something
that
I
like
either
it
gets
crammed
into
object,
state
or
object.
State
needs
a
wrapper
either
way,
I'll
have
to
change
those
interfaces.
So
a
little
bit.
D
A
I
didn't
mean
the
config
system
and
that
literally
config
option,
currently
all
the
ones
I
found
or
actually
also
OSD,
config
options.
Things
like
data
that
kind
of
stuff,
but
I'm
probably
gonna,
want
some
for
the
scheduler,
because.