►
From YouTube: Ceph Crimson/SeaStor OSD Weekly 2020-09-16
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
Okay,
so
let's
start
last
week,
I
I
was
still
working
on
the
thresh,
a
totology
basis,
threat
test
and
books
will
show
how
to
address
some
some
issues.
So
the
last
latest
issue
with
the
stretch
test,
was
a
a
segment
thought
in
blue
store.
B
Hi,
so
some
updates
on
the
dio
errors.
Pr
looks
like
david
zaffman
reviewed,
left
some
reviews
on
it.
I'm
gonna
link
it
just
you
guys
want
to
take
a
look
and
looks
like
he
approved
them,
but
there
are
still
some
some
concerns
that
that
he
raised
that
I'm
gonna
address.
B
He
left
the
comment
six
hours
ago,
so
I've
only
just
just
seen
it
but
looks
he
approved,
looks
like
he
approved
the
pr-
and
I
still
I'm
still
just
reading
through
code
for
for
snapshots,
there's
a
lot
to
a
lot
to
learn
there.
So
I'm
just
really
still
just
learning
that,
yes
and
that's
it.
A
I
just
have
a
quick
question:
do
you
think
it's
the
the
we
can
do
something
different
than
we
did
before
and
because
we
we
now
have
a.
We
are
using
c
store.
B
Yeah,
so
I
I
saw
that
I
I
saw
the
email
you
you
sent
me
kefu
with
trying
to
think
if
we
can
have
in
support
for
for
snapshots
inside
c
store.
A
C
A
I
mean
probably
the
for
tracking
the
ot,
sorry,
the
the
objects
extent,
for
example,
using.
D
D
A
E
I'm
working
on
the
test
case
for
the
tree
and
debug
currently
working
on
this
blade
operation
and
then
doing
the
diagonal
and
the
kafu.
Can
I
use
the
reader's
bench
to
reproduce
that
issue
brother
or
blue
rooftop.
A
Yes,
two
probably
it
could
be
difficult
to
reproduce
this.
The
segment
thought
we
found
in
slash
test,
because
we
cannot
run
the
stretch
test
without
the
help
of
testology.
A
If
you
really
want
to
reproduce,
I
I'd
suggest
to
take
a
look
at
the
the
new
introduced
future
to
run
technology
on
the
laptop.
But
I
didn't
get
a
chance
to
to
try
it
out.
E
Oh,
it
need
a
cluster.
Yes,
probably.
A
F
Everybody
scrubbing
that's
what
I
did
yesterday.
I
still
have
one
problem.
One
test
case
in
the
stand
alone
that
doesn't.
F
D
I
am
debugging
the
gc
stuff.
Hopefully
this
won't
take
more
than
a
couple
of
days
after
this
I
will
package
it
into
a
pr
and
that
will
be
essentially
the
end
of
the
transaction
manager
layer.
What's
up
your
is
that
okay.
D
Yeah,
so
once
this
is
ready,
so
what
that
will
mean
is
that
from
the
transaction
layer
down,
add
a
coarse
layer
that,
as
many
features,
won't
exist,
some
of
the
crc
stuff
isn't
in
place
yet,
but
you
will
be
able
to
write
for
an
arbitrary
period
of
time
without
running
out
of
space,
because
gc
will
run
in
parallel
and
do
the
maintenance
work
that
needs
to
be
done.
So
that
means
that
we
can
benchmark
it,
which
is
what
I'm
gonna
do.
D
Next,
I'm
gonna
rig
up
an
extremely
simple
test,
harness
that
operates
at
the
transaction
manager
layer
so
that
I
don't
need
stuff
above
it
now
that
will
compare
its
performance
to
a
raw
disk
and
also
this
other
back
end.
I
want
to
test
just
because
it
should
be
a
convenient
thing
to
do.
At
the
same
time,
I
really
really
doubt
that
the
performance
will
be
good,
so
that
will
be
more
a
jumping
off
point
for
the
next
sequence
of
things.
To
do.
D
D
D
A
D
On
it,
I
haven't
looked
at
it
closely.
I
figured
I
wouldn't
bother
looking
at
it
in
detail,
unless,
until
I
get
a
chance
to
benchmark
it
and
find
out
what
cpu
overhead
is,
its
transactions
are
actually
more
powerful
than
we
need
them
to
be
so
it's
I
would
give
it
a
30
likelihood
that
it
just
kills
itself
immediately
at
that
level,
but
I'm
curious
what
your
thoughts
are.
A
You
haven't
got
a
chance
to
to
performance
it
or
to
play
around
it,
but
the
takeaway
of
the
looking
at
this
design
that
there
are
two
two
components
of
this
hr
sc,
one,
the
mpo
another,
is
actually
itself
and
I'm
pro
exposed
to
two
different
apis.
One
is
based
on
block.
Another
is
based
on
log.
The
log
is
basically
append
only
as
data
structures,
yeah.
D
A
D
A
C
Last
week
I
have
decoupled
the
three
node
implementation
out
of
the
node
layout
implementation
and
I
will
continue
to
do
the
to-do
list
in
the
pr.
C
C
I
think
I
think
it
is
worse
to
look
at
the
descriptions
and
the
award
hatchet,
maybe
not
for
the
detailed
okay.
A
Thank
you.
I
will
try
to
pop
up
some
some
questions
by
the
end
of
this
fight.
Back
at
the
end
of
the
next
week
possible,
we
can
show
discussion
on
your
design
this
time.