►
From YouTube: Ceph Crimson/SeaStore 2021-08-11
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
Let's
start
last
week
I've
been
adding
the
actually.
I
I
had
a
follow-up
over
the
last
disc
of
one
of
the
discussion
we
had
in
zinka's
meeting
to
add
the
to
add
the
c
lamp
build
to
to
our
ci
to,
namely
to.
A
A
I
think
in
general
the
cylinder
compiler
is
more
much
less
tolerance
than
than
gcc.
So
I
think
we
we
won't
use
gcc
in
future
and
we
will
stick
with
clan
in
hope
that
every
the
the
the
if
the
build
pass
on
zealand
it
should
pass
on
gcc.
That's
my
assumption.
C
A
So
I'm
I'm
trying
to
fix
it,
and
one
other
issue
is
to
to
is
in
the
fio
because
it
failed
to
build
with
either
with
a
cylinder,
10
or
sealant
11.
In
my
case,
so
I'm
trying
to
to
the
the
the
root
cause
is
that
we
are
we
having
the
statement
of
using
namespace
lcd
one
over
our
head
fire,
which
is
very
popular
and
practically
it's
been
included
by
like
hundreds
of
the
source
file,
so
I'm
trying
to
remove
it,
and
I
cr
I've
created
pr
to
address
it.
A
Another
issue
is
in
fio
itself,
where
it
it
failed
to
to
compare
with
the
with
the
latest
version
of
it,
as
fio
failed
to
compile,
with
with
a
plus
plus
compiler
of
a
gct
or
with
clan,
I'm
creating
patch
to
address
it
by
using
the
atomic
offered
piston
library
to
implement
the
sequential
lock.
D
Last
week
I
finished
the
changing
the
allocated,
the
hint
by
using
shadow
id
poor
id
and
hashi
prefixed
and
still
working
on
the
comments,
and
I
have
discussed
with
insing
and
will
update
the
the
pr
according
to
in
his
comments,
and
currently
I
found
the
fio
testing
without
debug
information
output.
Put
it
easy
to
fail,
so
maybe
the
I'll
be
a
tree
currently
and
easily
get
a
dirt.
D
Before
I
can
doing
the
test
with
a
small
concurrency,
for
example,
the
ldap
set
to
two
and
number
drop
is
to
one,
but
currently
I
can't
finish
finish
it
so
so
it's
easily
to
to
field.
Currently,
that's
all.
C
E
D
E
D
Yeah
I
before
I
can
doing
the
fl
test
with
the
small
currency.
Maybe
that's
true,
given.
E
C
F
E
G
Suggestion
and
I've
just
really
pushed
push
the
pr,
and
now
I'm
debugging
the
or
I
think
it
is
almost.
G
E
C
A
Engine,
could
you
elaborate
a
little
bit
on
the
on
conflict,
moves
from
unknown
to
lb
part?
I
I
don't
quite
get
it.
B
Yeah
and
go
to
the
section
diagram
to
compare
this
is
to
compare
the
before
scenario
and
after
scenario.
B
At
one,
the
the
first
comparison
you
can
see
actually
the
invalidating,
invalidate
transactions
are
intense.
The
previous
rate
is
about
1.5,
the
ratio
is
about
1.5,
but
after.
A
B
If
you
look
at
the
scenario
before
the
scenario
before
is,
is
when
oh
no
tree
hasn't
been
integrated
with
interrupted
future
and
hasn't
been
fully
integrated
with
interrupted
future
and
after
scenario
is
that
our
system
has
been
fully
integrated
so
after
integrating
fully
integrated
the
the
interrupted
future
since
transaction
invalidation
is
more
intense
than
before.
E
Before
we
would
tend
to
detect
the
problem
in
the
o
node
layer,
whereas
now
the
very
first
extent
that
you
touch
that
causes
an
invalidate
will
cause
it.
B
C
C
B
E
E
So
for
now
I
think
the
interesting
thing
is
that
we're
like
we're.
With
the
latter
changes,
we
see
conflicts
more
in
the
ono
trees
than
we
do
in
the
latter
trees.
That's
actually
kind
of
a
neat
change.
I
think
that
would
happen
once
it's
no
longer
crashing
the
lba
tree,
which
maybe
this
week
or
next
month,
whatever
once
we
wrap
up
those
bugs
we'll
start
testing
on
larger
workloads
and
then
we'll
have
to
work
out
what
we
want
to
do
about
conflicts.
C
A
I'm
imagining
if
we
could
have
a
storage
attached
to
our
jenkins
thinking,
though
probably
we
could.
We
could
integrate
this
this
tool
into
our
ci.
E
This
kind
of
sensitive
performance
testing
is
really
difficult
to
create
binary
tests,
for
especially
with
c-store,
where
the
gross
behavior
of
it's
going
to
change
so
radically
over
over
time.
Maybe
in
about
18
months,
eastern's
behavior
will
be
stable
enough.
That
it'll
make
sense
to
compare
it
month
to
month,
but
I
expect
this
to
change
radically
soon,
so
I
don't
think
it's
worth
putting
time
into.
I
also
think
it's
something
that
everyone
developing
c-store
will
be
paying
close
attention
to
kind
of
all
the
time
because,
like
once,
it's
no
longer
crashing.
A
E
The
but
like
I,
I
didn't
debug
chad
may's
crash
specifically
because
I
know
of
one
really
severe
problem
with
the
existing
lba
tree.
The
way
it
finds
the
available
logical
address
space
when
it's
doing
an
allocation
is
buggy.
It
will
do
the
wrong
thing
if
a
split
happens
during
the
insertion.
So
I
just
know
that
that's
broken
so
not
especially
concerned.
E
E
B
E
D
D
D
Already
will
ready-
maybe
it's
going
too
soon
ready
to
provide
appear.
C
E
That
would
be
great
if
you
could
also
include
a
link
to
that
branch
in
your
email.
That
would
help.
C
H
I
apologize
for
being
late.
Let
me
provide
updates
since
last
week.
First
of
all,
the
sequencing
appears
sent
it's
got
approved,
thank
some
and
special
one
for
the
for
the
comments.
I'm
rebasing
it
to
refresh
process
master,
and
I
planned
a
follow-up
that
minimize
that
minimize
the
footprint
of
cl
on
client
request
the
up
sequencer
impose
a
little
bit.
I
want
to
drop
the
pref
op
that
was
introduced
to
client
request
totally
for
the
sake
of
op
sequencer.
H
I
think
the
same
information
is
basically
available
on
the
on
the
operation
in
the
operation
registry.
H
Add
the
cost
of
doing
some
some
extra
filtering,
but
I
think
this
filtering
would
happen
on
cold
path
on
call
paths,
so
it
will
be
basically
a
tradeoff
between
complexity,
footprint
and
and
overhead,
but
on
hot,
on
cold
path.
Trying
to
trying
to
do
to
do
that.
H
That's
one
thing:
another
is
about
another
stories
about
cicd,
I'm
working
on
on
bringing
the
make
perf
check
about
back.
Last
week
I
updated
I
upgraded
one
of
the
notes
to
focal
it
turned.
It
was
actually
a
bumpy
road.
A
lot
of
out
of
issues.
I
had
to
learn
to
learn
about
netplan
and
the
big
changes
in
in
networking
configuration
focal
growth.
H
B
H
F
H
Yeah,
but
this
this
job
is
particular
it's
it's.
It
hanked
because
of
run
cbt
because
run
cbt,
pawn
crimson
using
vstart
passing
dash
dash
mamster
instead
of
dash
dash
theonster,
so
we
basically
executed.
We
basically
run
crimson
with
alinstar
adapting
memster
instead
of
using
the
answer.
That's
wrong.
That's
that's!
Actually,
a
that's
a
fallout
from
the
pr
that
generalize
alienster
to
other
to
other
objects
or
object
stores.
That's.