►
From YouTube: Ceph Crimson/SeaStor OSD 2020-08-19
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
I
was
still
working
on
the
technology-based
treasure
test
and
I
run
into
couple
issues
which
is
not
right
related
to
the
back
backfield
recovery,
which
is
the
which
is
which
was
in,
which
was
due
to
the
missing
operation
which
which
I
support,
are
supposed
to
be
the
building
block
of
the
of
the
effect
view
and
recovery.
A
I
added
the
the
a
support
of
them
and
I
realized
that
sometimes
the
object
doesn't
necessarily
have
the
map.
So
that's
why
my
last
run.
I
was
over
threatened
field
because
of
that,
so
I'm
adding
the
as
an
aerator
support
to
the
newly
added
operations
to
ensure
that
the
the
at
least
the
error
message
and
the
filter
are
unless
less
in
less
surprising.
Let's
put
it
this
way,
and
I
will
I
plan
to
advertise
more
of
the
operations
along
the
way.
B
Hi
for
me,
two
things
last
week,
the
first
one
was:
I
continued
working
on
the
eio
handling
case,
I'm
trying
to
figure
out
currently
where,
where
in
our
code
should
the
handling
exactly
should
be
done,
it
seems
like
there
are
two
cases
where,
where
we
want
to
to
do
that-
and
that
is
one
is
when
we
have
a
crc
mismatch
and
the
second
one
where
the
object
store
is
returning
eio
traditionally
in
classical
osd,
it
was
just
done
by
the
function,
a
rep
repair
primary
object
and
it
was
called
from
do
read
and
do
sparse
read.
B
So
the
equivalence
of
the
equivalent
of
these
two
functions
in
crimson
is
pg
back
and
read,
and
pg
back
and
sparse
read,
but
it
does
it
so
so
there
it
seems
like
it
handles
the
crc
mismatch
case.
I
mean
it
that
it
still
doesn't
have
the
light
we
need
to.
We
can
handle
it
there,
but
replicated
back-end
is
also
calling
object,
store
read,
so
we
need
to
handle
that
too,
and
I'm
not
sure
at
which
level
would
that
be
correct
to
handle
and
the
second
correctly.
B
Yeah,
it's
the
same
exactly
so
it's
the
same
behavior
for
both
of
them.
The
reason
I'm
not
just
doing
it
in
pg
backend
is
because
it
seems
like
there
are.
There
are
other
parts
that
we
need
to
to
implement
this
handling
for
every
place,
essentially
that
that
calls
objects
to
read
so
that
that's
what
I'm
trying
to
figure
out
right
now,
and
the
second
thing
is
wait
any
more
questions
on
that.
A
A
Regarding
to
the
put.
B
Okay,
yeah
thanks.
That
would
be
great
and
this
the
second
thing
is.
Last
week
I
came
across
a
paper
that
was
presented
in
in
in
fast
20
the
conference
like
six
months
ago.
It
introduces
a
sort
of
extension
to
the
to
crush
algorithm.
B
It
basically
talks
about
the
possible
problem
you
have
with
crush
with
data
migration
happening
well
when
there
are
non-trivial
cluster
expansions
and
it
introduces
like
a
an
expression
to
it
that
overcomes
this
issue.
I
brought
it
up
in
last
week's
self
performance
weekly
with
with
mark
and
josh,
and
we
we
that
was
the
main
topic
of
discussion
last
week,
and
we
are
going
to
continue
talking
about
it
this
week.
B
I'm
going
to
link
the
the
paper
here
in
the
chat
just
so
everybody
have
an
idea
what
I'm
talking
about,
and
you
can
look
at
it
at
your
free
time
currently
in
my
drive,
because
I've
also
edited
the
pdf
a
little
bit.
I
added
like
color
coding
and
marking
of
a
particular
points
that
I
found
critical
or
informative,
so
that
that
should
make
the
reading
experience
just
easier.
So
you
can
see
like
what
are
the
the
main
points
and
you
can
also.
B
Yeah:
okay,
I'll
also
link
the
original
paper.
D
E
A
F
Ritak
hello
last
week,
not
not
much
for
crimson
just
reviews
working
modes
was
working
mostly
for
blue
fest
and
now
we've
gotten
a
segment
in
rgw
around
offloading
the
omap
stuff.
However,
I'm
addressing
a
bunch
of
comments,
I
mentioned
reviews
reviews
comments
on
the
unit.
F
Testing
for
backfield
code
found
actually
a
bug
there
in
a
in
bathroom
implementation
working
to
to
solve
it,
but
I
put
a
question
on
the
right,
which
is
which
probably
related
to
a
broader
broadband
to
the
recovery
stuff
and
what
is
about?
What
is
our
plan
for
pg
log
based
recovery?
A
At
least
the
fact
sorry,
the
recovery
is
not
quite
a
financial
automation,
machine
right.
C
A
A
C
We
should
start
triaging
out
bugs,
as
you
identify
things,
may
put
them
into
a
document
or
create
bugs
and
then
we'll
just
start
picking
them
up.
Radik
what's
happening
is
turning
up
actual
shortcomings
in
crimson
that
make
it
unusable
until
those
are
fixed.
It's
not
interesting
to
like
change
the
structure
of
existing
code.
We
have
to
fix
it
first.
F
F
What
what
I'm
trying
to
extrapolate
here
is
that
this
is
the
same
approach
would
be
viable
for
for
the
pg
log
based
recovery,
and
I
started
thinking
how
much
it
would
it,
how
much
time
it
would
consume
to
actually
restructure
the
current
the
current
code,
how
much
it
cost,
how
time
will
be
needed
to
bring
it
to
the
penn
state
machine
concept.
If
it
is
under
one
week,
then
we
could
start
real
testing.
F
C
A
F
Okay,
I'm
fine
with
that
one,
but
not
on
that
till
we
don't
have
the
testing
the
only
testing
thrashing
testing
whatever.
I
would
be
far
away
from
climbing
that,
let's
say
backfill
backfield
is
ready.
A
Yes,
by
the
way
I
I
I
was
also
updating
the
has
rashid
prepared
by
josh.
To
summarize
the
to
my
students
we
have
like
like
for
for
each
working
item.
We
have
a
time
a
milestone,
connecting
to
the
the
release
time,
for
example,
which
which
feature
will
be
ready
in
pacific
and
which
picture
will
be
in
the
queue
release.
So
I'm
trying
to
cross
out
some
items,
and
I
I
realized
that
we
haven't
finished
any
of
them
with
the
concrete
proof.
D
A
I
I
I
just
posted
the
the
url
to
the
to
the
search
basis
suite
I
she
come
whipped
up
and
later
on.
I
updated
a
little
bit
and
I
am
tracking
the
videos
we
have
and
I
will
add
a
I
will
put
some
transfer
transfer
most
some
of
these
failures
to
a
a
ticket,
so
any
one
of
us
could
get
a
chance
to
describe
them
and
try
to
fix
them.
We
can
call
it
later
on
on
it.
C
Can
you
send
out
like
an
email
with
what
tag
we're
putting
these
under
and
how
to
search
for
them?
That
way,
as
things
get
added,
we
can
find
them
and
we
know
what
tag
to
put
them
under
as
we
find
them
as
well.
It's
crucial
by
the
way
when
you
guys,
at
least
when
we
were
stabilizing
classic
oc
the
first
time
when
you,
if
you
bother
looking
through
a
pathology,
run,
please
take
the
time
to
write
a
bug
up
for
each
crash.
C
Even
if
you
don't
have
time
to
investigate
yours
yourself
like
take,
take
the
time
to
look
and
see
if
it
matches
an
existing
bug
and
if
not
create
a
new
one,
you
don't
have
to
assign
it
to
yourself.
That's
not
the
point.
The
point
is
simply
to
make
sure
that
we
have
a
if
you've
like
it
can
take
20
minutes
to
an
hour
just
to
identify
what
the
crash
was
so
once
you've
invested
that
time
make
a
bug.
A
Sure
we'll
do
that
we
we
already
have
a
category
dedicated
to
to
crimson,
so
we
don't
need
a
tag.
I
would.
I
would.
G
Everybody
and
I'm
scrubbing
till
now
it's
now
the
code
is
as
good
as
the
original
classic
code,
I'm
cleaning
it
up
and
preparing
it
for
first
of
june
cleaning.
That
means
the
apart
from
removing
extra
code,
I'm
finding
my
no
options
for
failure
which
I'm
fixing,
which
takes
time.
So
it
will
take
me
for
some
time
more
to
complete
this
test,
and
I
will
open
like
discussed.
Like
we
discussed
yesterday,
I
will
open
a
package
for
the
problem,
I'm
seeing
in
the
original
code
with
one
of
the
standalone
test.
G
Yes,
it's
clean
up,
because
now
it
has
hundreds
of
log
messages
inside
and
any
comments
and
cleaning
it
up
and
preparing
it.
Yes,.
A
C
Okay,
I'm
debugging
dirty
block
right
out.
I
have
the
code
written
basically
and
I
ended
up
flushing
out
a
lot
of
the
machinery
needed
to
do
generalized,
garbage
collection
as
as
well
so
that'll
be.
I
would
do
that
as
a
server
pr,
but
I've
got
a
lot
of
that
done
as
well.
So.
D
This
week
I
just
finished
coding
for
embedding
interrupt
interruption
mechanism
into
already
future,
but
right
now
there
are
still
some
serious
bugs
in
it
and
I'm
trying
to
deal
with
it
and
that's
all
for
me.
D
D
A
A
A
The
test
for
the
unit
test
for
for
recovery,
because,
because
redex
working
on
the
framework
for
for
exercising
factsview,
I
think
most
of
them
could
be
reviewed
for
testing
recovery.
D
Okay,
I
think
I
can
I
I
think
I
can
start
working
on
a
unit,
a
unit
test
for
recovery.
After
I
finish,
the
interrupt
interruption,
work.
A
Sure,
probably
you
could
take
a
look
at
the
the
pr
by
direct
attack
to
to
see
to
see
what
what
what
do
you
think
about
it.
And
so
we
can
edit
the
missing
pieces,
which
could
could
be
reused
by
recovery.
A
A
If
nothing
else
have
a
good
one
and
talk
to
you
later.