►
From YouTube: Velero Office Hours - Oct 8, 2020
Description
See agenda and notes at https://hackmd.io/I3u1x0u9T46KhuYZN4LX-A
A
So
the
mainly
the
thing,
so
what
I
wanted
to
ask
is
all
about
so
say:
for
example,
if
you
go
with
any
operator
based
approach,
or
else
if
you're
gonna
provide
your
service
on
top
of
kubernetes
platform,
so
mainly
the
crd
based
approach,
if
you
look
at
it
right,
so
the
life
cycle
of
all
those
resources
or
stick
with
the
same
cluster
until
unless
you
don't
have
solutions
like
veliro
right
so
valuero
basically
helps
you
to
take
the
backup
of
all
the
kubernetes
resources
and,
let's
say
because
of
the
disaster.
A
If
you
wanted
to
migrate
from
one
cluster
to
another
cluster,
so
valyrio
basically
helps
you
to
restore
to
the
totally
new
cluster
and
then
all
the
resources
will
be
backed
up
and
then
everything
works
well,
but
main
problem
is
the
restoration
time.
Basically,
it
depends
on
the
size
of
the
cluster
and
the
size
of
the
resources.
What
is
running
on
the
cluster
so
in
so
that
basically
takes
based
on
the
size
it
may
take
one
hour.
A
A
We
have
believe
you
so
that
you
know.
Basically,
you
can
avoid
the
restoration
time
so,
let's
say
two
to
three
hours:
downtime
based
on
the
size.
So
what
do
you
think
about
this
kind
of
an
use
case?
Yeah?
So
that's
what
I
mainly
wanted
to
ask.
B
So
I'll
jump
in
so
you're
asking
specifically
in
the
case,
so
you
want
to
kind
of
have
two
clusters
running
and
keep
them
in
sync.
So
you've
got
one
that
is
running
backups
and
another
one.
That's
continually
restoring
from
the
main
cluster.
That's
running.
B
So
I
don't
as
far
as
I'm
aware
at
the
moment,
we
don't
have
any
support
within
valero.
To
do
that.
I
think
that's
you
clearly
correct
me
if
I'm
wrong
here,
but
I
imagine
you
could
have
something
that
you
that
you
write
around
it
so
that
you
could
have
something
another
cluster
which
is
able
to
access
the
the
backups
from
cluster,
a
for
example,
and
cluster
b
could
if
it
could
access
those
backups,
you
could
have
some
kind
of
other.
B
I
imagine
you
you'd
have
to
write
your
own
system
around
that
to
be
able
to
to
continually
run
those
restores.
I
think
I
did
see
your
issue
and
I
did
see
nolan
commented
on
it
that
it
would
be
a
great
feature
to
have,
but
it's
it's
not
on
the
road
map
at
the
moment.
So
I
think
it's
something
that
we
can
definitely
raise
to
your
product
team
and
have
a
discussion
and
try
and
get
a
sense
of
like
the
priorities,
because
obviously
we
have
to
balance
other
things.
A
B
I
think
it
would
be
a
great
feature
to
have,
but
I
have
to
raise
it
with
with
the
rest
of
the
team.
A
Yeah,
so
definitely
so
because
this
in
general,
so
anyone
require
this
kind
of
a
feature
so
because
of
course,
that
there
are
many
things
also
available
outside
also,
you
can
also
build
your
own
one
to
take
the
backup
and
restore,
but
we
need
to
kind
of
bring
up
something
like,
even
if
you
are
doing
the
restore
there.
Actually,
you
will
run
into
a
sort
of
a
small
down
time
yeah.
So
if
your
cluster
is
very
big,
actually
the
downtime
will
be
more.
So
when
you
do
the
restoration.
A
So
somehow
we
need
to
address
the
problem
of
so
how
you
can
minimize
the
downtime
of
a
system
because
of
a
disaster
yeah.
So
so
that's
what
the
main
point
actually
what
I
wanted
to
address.
So
let's
say
I
was
figuring
out
a
tool
so
which
can
help
in
this
aspect
so
that
so
we
can
make
use
of
it,
and
also
this
will
be
beneficial
for
anyone
outside
in
general.
A
So
everybody
are
looking
for
a
tool,
so
of
course
they
require
a
backup
and
restore
and
then
even
if
it
is
doing
the
restoration,
so
they
want
to
minimize
the
the
downtime
in
general.
So
that's
where
we
need
to
bring
up
some
kind
of
I
mean
this
is
another
one
concept,
so
what
I
think
of,
but
there
will
be
also
the
other
concepts.
A
So
as
you
experts,
actually
you
might
know
it
of
there
will
be
another
possibilities,
also
how
you
can
design
the
system
yeah,
so
this
just
wanted
to
put
across.
So
what
valerio
basically
thinks
about
in
this
use
case.
B
So
I'm
not
so,
I
will
admit
I'm
pretty
new
to
the
valero
team,
so
I'm
not
actually
familiar
with
kind
of
the
standard
times
that
it
usually
takes
for
restores
crazy
kind
of
set.
Ask
is
like
two
to
three
hours.
Is
that,
like
a
typical
amount
of
time,
that's
usually
required
for
doing
a
restore
on
the
cluster.
B
I'm
just
wondering
like
if
there's
some
interaction
here
with
maybe
the
the
particular
provider
that's
been
used,
or
whether
you're
using
brassic
or
something
like
that,
whether
that
might
cause
the
restores
to
take
longer,
I'm
not
sure.
B
C
I'm
reading
through
this,
I
was
just
able
to
open
this
slack
to
understand
what
this.
So,
I
guess
what
you
we,
what
you're
suggesting
is
that
we
have
a
way
to
to
restore
from
a
cluster
without
doing
the
step
of
packaging,
the
cluster
in
and
packaging
the
cluster,
so
just
go
directly.
A
A
No,
it's
it's
not
the
speed
of
the
restore,
so
the
mainly
I
wanted
to
my
main
target
is
to
bring
down
the
downtime
yeah.
So
if,
if,
if
you
are
doing
the
restore
so
in
typical
use
case
how
everybody
they
run,
the
system
is
that
term
that
will
be
a
one
cluster
that
will
be
their
productive
cluster.
They
will
take
a
backup
continuously,
so
in
case
of
a
disaster,
they
will
basically
spin
up
a
new
cluster
and
then
they
will
restore
it
from
the
the
old
cluster.
So
then
everything
is
backed
up.
A
So
then
your
system
is
without
any
data
loss,
so
your
system
is
back
and
then
everything
will
be
up
and
running,
but
this
restoration
during
the
restoration
you
are.
The
solution
will
be
there
in
a
downtime,
so
it
may
take
half
an
hour.
It
may
take
one
hour.
It
depends
on
the
size
of
your
cluster
and
then
how
many
applications
are
running.
Services
are
running.
It's
completely
depends
on
that
one.
A
So
now
the
my
main
target
is
to
bring
down
the
down
time
during
the
restoration,
so
how
you
can
basically
address
the
problem
to
bring
down
the
the
downtime
yeah.
So
I
don't
want
to
have
longer
downtime,
so
let's
say
with
the
single
cluster,
I'm
running
it
and
then
restoring
totally
new
cluster.
That
is
taking
example
one
hour,
so
I
need
to
address.
I
wanted
to
reduce
down
that
one
hour
into
example:
some
one
minute
or
five
minutes
maximum,
so
how
I
can
do
that.
So
that's
what
the
problem
I'm
trying
to
resolve.
C
Okay,
I
understand
now
so
so
to
answer
the
the
question
I
I
would
need
some
numbers,
but
even
even
if
I
had
the
numbers
I
wouldn't
have,
I
don't
know
the
answer
off
the
top
of
my
head.
So,
like
you
already
said,
it
depends.
How
long
about
a
restore
take
depends
on
the
size
of
the
cluster.
Is
that
that's
being
backed
up.
A
C
Think
I
think
the
the
size
limits
that
you
have
set
for
your
clusters
also
would
make
a
difference,
of
course,
the
less
memory
you
have,
the
less
efficient,
any
the
the
the
slower
any
operation
would
be
so
maybe
increasing
the
the
the
memory
would
make
a
difference.
I'm
not
sure
I'm
just
thinking
through
this.
As
far
as
the
proposed
solution
for
making
restore
faster,
which
of
course
we
are
that's
a
very
much
in
our
interest
to
to
go
in
that
direction.
C
We
would
have
to
investigate,
because
if
a
restore
takes
is
taking
one
hour,
let's
say,
but
the
actual
downloading,
from
from
the
storage
and
and
unpacking
is
not
overtaking
that
time.
If,
let's
say
it's
just
a
minute,
then
the
solution
is
not
what
the
solution
is
is
not
is
addressing
the
wrong
problem,
so
the
problem
might
be
somewhere
else.
Does
it
make
sense?
What
I'm
saying.
C
So
so
so
maybe
what
we?
What
we're
talking
about
here
with
this
issue,
is
removing
the
middleman,
removing
the
the
the
storage,
so
you,
instead
of
fetching
from
storage
refresh
directly
from
the
cluster,
but
if
fashion
from
the
storage
is,
is
really
not
over
overtaking
the
total
amount
of
time.
For
this
restore
I'm
not
sure
that
that
would
be
the
solution,
but
also
if
your
restore
is
taking
a
ridiculous
a
long
time
then,
regardless
of
having
to
fetch
the
restore
from
the
s3
storage,
maybe
there
is
something
else
there
that
needs
to
be
looked
at.
A
B
Yes,
one
thing
I
was
just
wondering
is:
are
you
whenever
you
take
a
backup
and
perform
the
restore?
Are
you
backing
up
the
entire
cluster?
Maybe
there
are
only
subsets
of
the
applications
on
there
that
you
really
need
to
have
restored.
So
if
you
do
have
clusters
running
in
parallel
cluster
a
and
cluster
b-
and
you
want
to
like
fail
over
to
cluster
b-
maybe
you
could
already
have
certain
services
running
in
there
and
then
you
only
need
to
back
up
a
subset
of
cluster
a
to
get
it
back
and
functional.
A
So
basically,
I
wanted
to
keep
both
the
clusters
in
sync.
So
any
point
in
time
there
will
be
one
cluster
is
active,
so
say.
For
example,
I
have
cluster
a
and
cluster
b
cluster.
A
is
my
productive
one,
so
I
always
keep
my
cluster
b
sync
with
cluster
a,
but
that
will
be
there
in
a
passive
mode.
So
that
is
not
an
active
one.
So
my
cluster
a
is
basically
active
one.
So
let's
say
due
to
some
situation,
my
cluster
is
down
so,
which
means
that
I
no
need
to
take
a
resto.
A
I
need.
I
don't
need
to
do
the
restoration,
because
my
cluster
be
already
in
sync
with
cluster
a.
I
just
need
to
bring
the
cluster
b
to
active1,
so
my
downtime
is
reduced,
so
my
cluster
b
is
already
active.
So
again
the
application
is
works
as
it
is
yeah,
so
without
much
downtime.
So
so
this
is
the
approach.
Basically,
I
was
looking
into
with
respect
to
veliro,
or
else
there
will
be
other
possibilities.
Also
what
khalsa
was
mentioning
about
yeah.
So
maybe
there
are
so
many
possibilities,
so
we
need.
A
Maybe
if
you
can
suggest
us
some
the
best
one,
so
that
will
be
really
great
yeah,
so.
A
A
C
D
What
you're
talking
about
is
more
replication
than
backup
they're
two
different
concepts,
so
there
was
focused
on
backup
and
making
sure
that
we
have
a
disaster
recovery
in
case
of
anything
else,
not
working
for
you.
You
can
have,
depending
on
your
setup,
you
can
have
replication
going
to
one
place.
You
can
have
backups
going
to
a
third
place,
so
that
way,
your
you're
absolutely
confident
that
you
can
survive
a
disaster
but
yeah
what
what
you
describe
is
more
more
replication
than
back,
especially
continuously
replicating
things.
That's
replication.
D
B
Thanks
for
using
the
issue
like
I
said
well,
we
can
discuss
this
and
some
of
the
other
points
that
have
come
up
today.
At
the
add
those
to
the
issue,
like
I
said
reason
with
our
product
team,
maybe
to
investigate
some
of
the
other
things
to
do
with
general
performance
and
reducing
the
restore.
B
A
So
also
yeah,
so
so
I
also
heard
from
so
one
of
my
teammates
is
that
so
they
actually
using
valerio
and
in
a
productive
environment.
Actually
their
system
is
very
complex
one
and
they
actually
mentioned
about
some
issues
with
the
valerio
to
handling
the
secret.
Let's
say
if
you
talk
about
the
distributed
systems
like
keema,
I'm
not
sure
you
know
about
keema,
so
it's
it's
an
extension
platform
on
top
of,
let's
say,
k
native
yeah.
A
So
so
it's
a
serverless
system,
so
they
actually
mentioned
that
they
have
issues
with
valerio
while
taking
an
backup
and
then
during
the
restoration,
because
the
system
is
pretty
distributed.
A
It's
it's
a
micro
services
architecture,
it's
pretty
distributed.
So
when
they're
restoring
it
back
so
they're
kind
of
missing
the
secrets
and
then
the
connections
again
it
was
not
working
as
expected
after
the
restoration,
I'm
not
having
much
details
right
now.
So
maybe
I
can
also
get
the
details
and
then
I
can
also
open
issue
in
the
the
valerio,
the
github
yeah.
So
I
just
wanted
to
understand
from
you
is
that
have
you
also
ran
into,
or
you
already
know,
some
sort
of
issues?
A
B
So
I
don't
know
specifically,
so
what's
sorry
please
clarify,
were
they
saying
that
they
were
having
specifically
issues
with
restoring
secrets
or
was
it
the
applicant
or
the
app
that
they
were
using
for
managing
the
secrets
within
their
cluster?
You
mentioned
keem.
I
sorry
I
missed
the
name
of
the
thing
that
they
were
deploying
yeah.
I'm
I'm
not
aware
of
like
any
particular
issues
with
with
large-scale
restores.
A
Yeah,
okay,
so
that's
what
I
don't
have
the
exact
details
right
now,
but
I
got
to
know
about
a
few
things
yeah.
So
the
one
thing
what
they
talked
about
is
mainly
the
secrets
yeah,
so
in
and
so
other
things
here.
I
need
to
get
more
details
from
them,
and
so
mainly,
I
asked
so
currently
believe
you
have
any
known
limitation
for
this
complicated
or
complex,
distributed
applications
yeah.
So
during
the
history,
after
the
restoration,
it
has
any
issues,
let's
say
after
the
restoration.
A
So
what
we
expect
is
that
everything
should
work
well
so
before
how
it
was
working
yeah.
So
so
that's
what
I
just
wanted
to
ask
so
mainly,
but
I
can
get
the
details.
You
know.
C
So
if
they
are
in
in
the
actual
backup,
that's
a
good
sign
and
then
we
just
have
to
focus
okay,
why
it
was
not
being
restored,
but
if
they
were
not
even
backed
up,
then
you
have
to
see
okay.
How
is
the
backup
being
created?
That's
excluding
the
secrets,
because
there
are
a
way
to
exclude
things,
and
there
are
ways
to
include
everything
that
you
want.
So
that's
one
step
that
I
have.
A
B
Basically,
well
thanks
for
thanks
for
joining
and
bringing
us
your
questions,
yeah
we'll
definitely
follow
up
on
the
the
issues
and
things
like
that
and
we'll
we'll
look
out
for
the
issue.
The
second
issue
that
you
mentioned
once
you
get
more
information
from
your
teammates.
A
B
30
yeah,
you
should
go
and
get
some
rest,
but
thanks
for
joining
us
and
yeah,
we'll
we'll
chat
on
github.