►
From YouTube: Ceph Crimson/SeaStor OSD Weekly 2020-06-03
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
Then,
let's
start
on
last
week,
I
updated
my
only
trip
here
on
top
of
a
SAM
spear
and
you're
helping
murder,
so
I
can
continue
with
the
I
think
they
stink
version
of
my
PR
I.
Think
I
will
have
it
as
a
follow-up
up
here.
The
first
first
step
is
to
have
a
synchronized
version,
getting
kind
of
synchronize
a
very
merger.
It
says
on
the
on
the
review.
The
next
thing
is
to
try
to
try
to
review
and
I,
read
ex
recovery
broker
here
and
get
these
PR
merged.
Oh
by
the
way.
A
C
Discussion
with
with
stamp
and
I
hope,
I
think
it's
better
stop
the
extracted
common
code
currently
because
I,
don't
understand,
they're,
worried
detail
of
the
implementation
for
extended
map
tree
and
the
IRB
a
tree
there
is.
There
are
some
difference
in
the
details
for
the
beech
tree
implementation,
so
I
started
to
implement
them.
It's
Ted
the
tree
itself.
Currently
so
I'm
writing
the
extended
tree
retreat.
You
know
that
in
position
not.
A
D
C
C
E
E
E
F
D
All
right
push
to
the
right.
I
pushed
the
changes
for
the
second
PR
up
and
it
looks
like
they
merge,
so
that's
good
other
than
that.
I
worked
on
some
stuff
unrelated
to
crimson,
but
I
am
now
switching
back
to
working
on
basically
a
segment
registry
right
now.
It
just
blindly
assumes
that
I
can
just
increment
the
segment
ID.
We
need
a
way
to
track
which
segments
are
in
use
and
which
ones
are
freed,
so
that'll
be
another
physically
addressed
structure
hanging
off
of
the
route
block.
That
should
be
relatively
simple.
A
D
A
D
D
D
A
D
Interface
is
already
there
transaction
manager,
so
journal
has
an
interface
where
calls
back
into
transaction
manager
where
it
asks
for
the
next
segment
and
where
it
releases
them
so
like
when
it
opens
a
segment,
it
asks
whatever
gets
passed
in,
give
me
the
next
segment
ID
you
want
me
to
use.
That
starts.
Writing
to
that
and
when
it
closes
it,
it
says:
okay,
I
finished
with
the
segment
ID
do
with
that
information.
What
you
will
come.
F
D
A
A
B
Last
week,
I
was
mainly
debunking.
The
block
recovery
code-
I've
got
a
question
here.
I
want
I,
intend
to
add
some
testing
code
for
the
recovery,
the
mechanism,
but
since
this
involves
a
lots
of
shutting
down
and
putting
up
OSD,
so
I
wonder
if
there
is
any
facility
within
the
current
of
tooth
ology
or
QA.
A
B
D
A
D
D
D
The
problems
that
it's
not
a
correctness
test
like
there's
a
lot
that
it
won't
test
it'll,
just
basically
test
to
see
whether
the
I
was
he
literally
crashed.
That's
all
it'll
notice.
It
won't
notice,
for
instance,
if
the,
if
the
objects
are
the
wrong
version
or
if
we
didn't
recover
correctly
or
if
they're
like
it
won't,
do
any
validation
and
their
technology
test
that,
due
to
all
of
that
stuff
so--but,
if
you
do
wanna
get
that
working.
That
would
be
fine
too,
like
we
need
that
eventually,
because
we're
gonna
need
we're.
D
Gonna
need
to
be
able
to
do
that
kind
of
benchmarking
to
find
out
what
the
impact
of
recovery
on
foreground
IO
is
from,
but
I
think
we
should
tackle
technology
as
the
next
task.
That's
that's!
Why
we're
tackling
log
based
recovery
and
back
flow
right
now?
It's
so
that
we
can
then
integrate
crimson
and
technology
and
start
it
and
get
it
running
at
our
nightly
to
us.
I.
A
A
D
D
Qa
/
stand
alone,
they're,
just
basically
bash
girls,
though
that
could
work
too
I
think
what
I'm,
what
I'm
saying
is
I,
don't
wanna
spend
a
lot
of
time
on
an
intermediate
solution.
I
want
to
get
like
basic
testing
of
this
and
then
get
the
technology
thing
doc.
That's
where
we'll
really
know.
A
A
A
F
Okay,
okay,
so
there's
a
delay,
switch
on
the
mic
so
last
week,
I
added
a
test
for
node
insertion
and
the
lookup
and
fix
some
bugs
in
the
code
and
the
node
content
can
be
dumped
it.
So
that
helps
me
to
write
gist
in
order
to
illustrate
how
how
it
looks
like
logically,
and
why
I'm
going
in
this
way
without
need
to
dig
into
tedious
implementation
details,
though
I
will
paste
here.
C
A
C
F
C
I,
try
to
enlarge
the
enlarge
the
memory,
but
I
still
have
other
issues
so
also
I
asked
her
the
sister
I
started.
The
readiness
Roddick
had
not
used
the
sister
allocator,
so
it
it
used
should
use
another
sister
patch
to
fix
that
easy,
batter,
batter
patch
is
not
merged
in
the
C
star.
So
for
us
the
simpler
a
truth
we
saw
it
adjust
yields
that
you
fall
down.
Peter.
A
A
C
C
A
E
We
just
can
even
correctly
when
we
made
that,
because
of
because
of
performance,
there
was
some
issue
in
the
past:
around
object
around
audience,
sterile
implementation
and
to
not
hurt
not
hit
that
back
the
master
base
test
we
have
enabled
we
switched
the
alligator
from
sister
to
BBC,
just
for
default,
vids
and
now
what
well?
What
would
the
rationale
for
the
switching
also
degrees
degrees
builds?
Sorry,
I
haven't
gotten
the
initial.
The
initial
sentences.