►
From YouTube: SIG - Storage 2023-07-17
Description
Meeting Notes:
https://docs.google.com/document/d/1mqJMjzT1biCpImEvi76DCMZxv-DwxGYLiPRLcR6CWpE/edit#
A
A
A
A
A
A
C
A
A
Okay,
I
mean
we've
seen
a
few
issues
of
this
kind
of
lately
with
different
different
scenarios.
Add
a
method
for
updating
the
starts.
I
wonder:
is
it
something?
Would
it
be
easy
to
just
potentially
extend
the
the
validity
of
the
certificate
by
default?
Do
you
think,
instead
of
adding
a
configuration,
I
guess
so.
E
All
this
is
configurable
there
are.
There
are
parameters
to
configure
our
certificate
rotation
intervals,
so
they
they
could
do
that,
but
the
search
that
we
generate
for
upload
are
really
only
meant
to
be
used
internally
for
communication
between,
like
the
openshift
router
and
our
service,
or
between
an
Ingress
controller
and
our
service.
E
Because,
like
the
the
DNS
names
aren't
gonna,
you
know
work
for
the
search
that
you
know,
even
if
the
duration
is
longer
they're
gonna.
You
know
that
your
laptop
doesn't
have
the
same.
You
know
DNS
as
our
server
in
the
cluster,
so
the
names
will
be
off
so
they
have
to
do
stuff
with,
like
you
know,
Etsy
hosts
you.
C
A
F
A
F
F
Yeah,
so
basically
what
he
is
trying
to
do.
He
would
like
to
use
sun
in
order
to
share
basically
to
implement,
read,
write
many
in
order
to
partition
the
storage
and
he
did
some
measurement,
and
it
seems
that
using
qcal
format
and
the
snapshotting
from
from
qmu
it
seems
to
be
the
fastest
on
the
most
performance.
F
F
A
Okay,
so
we
can
kind
of
treat
this
as
an
FYI
in
this
meeting
for
people
to
dig
in
it's
a
cool
seems
like
a
cool
project.
I
did
encounter
and
right
now
the
name
of
it
is
escaping
me,
but
there
was
another
team
at
Red
Hat
somewhere,
which
was
also
looking
at
doing
some
kind
of
a
meta
provisioner.
That
would
actually
it
was
for
local
storage
and
it
would
support
essentially
multiple
back
end
storage.
So
you
could
use
nvme
devices.
You
could
use
standard
local
block
devices.
A
You
could
use
local
LVN
any
number
of
things
like
that,
and
then
it
would
present
a
similar
kind
of
a
unified
PVC
provisioning,
so
that
also
seems
to
carry
with
it
a
bit
of
overlap
and
with
now
three
different
groups,
seemingly
attempting
to
do
something
similar.
It
really
seems
like
there's
potential
for
something
here.
F
A
Yeah
I'll
see
you
after
the
call
I'll
see
if
I
can
dig
up
that
email
thread
and
put
a
link
to
that
project
here
as
well.
So
we
kind
of
have
all
three
in
one
place
and
it
seems
like
yeah.
It
would
be
kind
of
neat
to
kind
of
converge
the
requirements
into
something
and
see
what
what
could
be
done.
Yeah.
F
A
Okay,
all
right.
That
sounds
good
thanks
for
adding
the
agenda
item
and
Vivek.
If
you
wanted
to
jump
back
in
later
on
with
more
details
when
your
audio
is
resolved,
feel
free
or
you
could
add
some
more
context
into
the
agenda
for
everyone,
great
okay.
So,
let's
move
on,
we
do
have
a
bunch
of
things
here.
So
let's
go
to
Alex
your
topic
about
the
cube
root,
storage,
Lane
issues.
A
We
touched
on
this
one,
maybe
before
you
joined
I've,
added
a
comment
in
there
to
request
some
more
details
on
the
exact
use
case,
because
Michael
was
kind
of
talking
about
how
the
certificates
are
generally
for
internal
communication
and
so
shouldn't.
There
shouldn't
usually
be
an
expiration
problem,
so
we
kind
of
want
to
understand
a
little
bit
more
about
what's
going
on
there
before
and
that
it's
also
is
configurable.
So.
G
G
So
Brian
from
the
CI
team
noticed
this
there's,
maybe
like
a
few
one
male
thread
and
there's
PRS
from
Alexander,
but
today
earlier
today
we
were
looking
at
the
memory
consumption
and
we
believe
that
that's
the
issue.
Basically,
we
we
get
like
30
gigs
per
CI
job
and
we
were
exceeding
that
with
new
CDI,
so
I
think
the
the
question
for
this
meeting
is,
if,
if
it
makes
sense,
maybe
it's
hard
to
answer,
but
does
this
make
sense?
G
There
was
a
three
gigabyte
increase,
sounds
a
bit
much
for
just
you
know
a
few
popular
controllers
that
just
Loop
over
state
but
combine
the
CDI
bump
with
at
CDN
memory,
which
is
being
used
in
in
the
CI
lanes,
and
we
get
basically
we
get
like
leader.
Elections
happening
a
lot.
We
have
just
just
straight
up
dead,
API
servers
yeah,
so
that
makes
the
tests
perform
worse,
so
I
mean
we'll
we'll
get
we'll
get
to
a
point
where
the
lanes
are
green,
but
what's
the?
E
Yeah,
that's
a
good
question.
Yeah,
you
think
you
know
we
added
these
new
controllers
and
I
think
what
we
can
look
at
is
I
guess.
Well,
you
know
they
use
the
informers,
so
they're
cash
and
stuff,
but
it
should
be
a
shared
patch
with
everything
else
and
maybe
we're
caching,
some
new
resources,
but
every
gig
still
seems
like
a
lot.
E
G
G
A
D
E
Yeah
I
I
was
I
can
help
out
looking
in
both
of
those
directions.
I
tried
to
make
some
changes
to
the
way
the
tests
were
structured,
but
that
didn't
help
things
so,
but
what
Alex
is
saying
is
seems
to
be
consistent
with
some
of
the
logs
I've
seen
from
tests
and
whatnot,
where
weird
things
are
happening.
C
A
Yeah
I
mean
there's
definitely
some
while
we're
kind
of
maintaining
multiple
ways
of
achieving
the
same
thing
for
the,
since
we
have
the
compatibility
going
on
there's
an
expected
that
we
would
be
using
some
more
resources,
but
that
seems
like
a
lot:
okay,
cool
all
right,
so
I
think
we
have
at
least
a
rough
plan
on
that
one
in
terms
of
looking
into
it.
Let's
see,
if
there's
some
updates.
B
Hey
so
it's
something
that
I
thought
I'd
raise
as
an
option
or
a
Time
feeling
for
this
meetings
to
add
to
this
meeting
lightning
talks,
meaning
5
to
10
15
minutes
of
talking
on
any
subject
that
you
want.
You
don't
have
to
be
a
member
in
the
team.
It's
anyone
from
the
community
can
also
raise
a
topic.
He
thinks
or
she
thinks
is
interesting
for
the
for
the
entire
meeting
and
people
in
the
meeting
you
can
also
I
will
add
an
option
of
requests.
B
B
I,
don't
know
that's.
Basically
it
and
people
will
just
volunteer
to
do
the
lighting
talks
or
by
requests.
Do
the
lighting
talks
I
thought
about
doing
a
lightning
talk
on
volume
populators
as
a
start,
but
I
did
demo
it.
We
did
demo
it
in
the
last
demo
session,
so
I'm
not
sure.
If
there's
anyone
in
this
current
meeting
that
haven't
heard
it.
B
A
B
Yeah
I
thought
it's
possible
to
either.
You
know,
send
me
ideas
or
add
to
this
document
like
a
bullet
point
of
requests
and
a
link,
maybe
to
a
doc
that
does
a
list
of
ideas
for
lightning
talks
and
people
can
volunteer
to
do
them
and
I
can
raise
it
at
the
end
of
each
and
meeting
either
requests
and
someone
volunteers
to
take
it
or
someone
volunteering
to
do
something
on
his
own
accord.
B
C
A
B
A
Sounds
good
I'll?
Let
you
revamp
this
then
to
you
know
to
be
something
that's
kind
of
useful,
but
it
seems
that
we
can
just
keep
it
right
here
in
the
dock.
It'd
be
a
nice
useful
place
for
it.
I'm
excited
about
this
I
think
it'll
be
cool
to
have
some
some
presentations
in
here
from
different
folks,
so
that'll
be
cool.
B
C
B
A
E
E
G
E
E
E
D
E
A
Cool
all
right
sounds
good,
yeah,
I
think
a
little
more
context
would
be
needed
all
right,
so
I'm,
going
to
jump
back
into
the
issues
was
that
okay,
so
we
were
seems
like
there's
a
I'm
just
gonna
see
how
far
down
that
was
oh
yeah.
So
this
appears
to
be
almost
the
like
the
third
one.
But
okay,
let's
go
up
to
metric
names,
failed
from
winter.
A
A
G
D
A
Do
we
have
a
do?
We
have
an
issue
with
so
yes,
I
guess,
with
the
Clone
progress
and
stuff
I'm,
just
kind
of
wondering
is
that
going
to
affect
something
like
when
we're
listing
data
volumes
on
the
command
line
or
something
with
vert
CTL
or
anything
like
I'm,
just
kind
of
curious?
If
who
might
be
depending
on
those
names
as
they
are
today,.
C
A
A
A
E
C
E
C
E
It
may
be
fixed
I,
I,
just
I
I
forget
the
current
state.
I
think
it's
yeah.
It
may
be
fixed
for
everything,
though,
because
I
think
what
we
do
now
is
the
web
hook
creates
a
short-term
token
on
the
data
volume
and
then
I
think
pretty
much
right
away.
I'll
have
to
check
the
latest,
but
I
think
we
add
a
longer
term
token
to
the
data
volume
and
then,
when
we
create
the
target
PPC,
there
is
a
longer
term
token
added
to
that.
A
E
A
C
E
D
A
D
D
A
E
Yeah
yeah
I
mean
I
I
think
it
looks
like
you
know.
The
the
source
pod
is
posting
data
to
the
upload
server.
It
would
be
I
wish
we
had
the
upload,
but
a
lot
of
what's
happening
on
the
upload
server
side.
Maybe
we
do
in
a
previous
comment,
but
yeah
I'm
not
sure
what
we
can
do
here.
I
mean
clusters
are
sized
differently.
You
know
if
it
takes
well,
although
we
should
be
yeah
I,
don't
know
not
sure
why
you
know
the
upload
server
pod.
A
E
They
are
yeah,
that's
a
good
question
too
yeah
I
mean
you
know
we
could
have
some
like.
You
know.
Global
configuration
like
you
know
only
X
many
concurrent
host
clones,
but
that
would
be
like
a
new
feature
and
it
would
really
only
be
useful
I
think,
for
you
know,
constrained
clusters
or
people
that
wanted
to
do,
but
yeah.
E
A
C
A
D
D
C
A
A
D
A
A
G
A
Yeah
or
if
there's
a
yeah
I,
don't
know
that's
interesting
Unless.
Somehow
this
code
code
is
didn't
being
called
before
the
like.
If
it's
copying
over
from
scratch
or
something,
although
it
says
we're
deleting
it
here
by
now
so
yeah-
that's
really
interesting,
but
we've
asked
for
the
debug
info,
so
we
we're
kind
of
working
towards
resolving
that.
So
nothing
else
to
do
here,
I'd,
say
and
then
this
is
interesting.
Cdi
import
from
URL
is
massively
slower
than
a
manual
wget
or
CTL
image.
Upload.
G
G
So
there's
this
nvd
kit
filter,
that's
just
playing
nice.
With
these
type
of
mirrors.
It's
called
the
read
ahead:
filter,
okay
and
there
was
somebody
making
like
somebody
external
to
our
team,
made
the
pr
and
they
mentioned
that
on
on
their
machine.
It
reduced
the
import
from
90
minutes
to
one
minute.