►
From YouTube: Velero Community Meeting/Open Discussion - June 23, 2020
Description
June 23, 2020
Status Updates
nrb
Lots of reviews, but sorry if I didn’t get to yours
Lots of internal work too
carlisia
On community support this week
kubebuilder PR still getting reviews: https://github.com/vmware-tanzu/velero/pull/2561
PR reviews
Adding tests to: Add BSL controller - https://github.com/vmware-tanzu/velero/pull/2617
ashish is out
Worked on backing up volumes with restic by default
Discussion Topics
nrb Restore hooks - let’s have a seperate call for this to address some concerns about using init containers.
Main use case seems to be quiescing/unquiescing databases
Will need to use some sort of tool to send this to a backup PV, and a tool to take it from the backup PV to a runtime PV
nrb/carlisia - Public/private APIs
Let us know what you rely on!
This will be phased, not all at once
In 2.0, the current Go client for Velero will be replaced by a controller-runtime client. The informers/listers will be replaced by a client manager.
Vishnu (Enable restic by default - https://github.com/vmware-tanzu/velero/pull/2611) and (Progress on backup/restore by plugins - https://github.com/vmware-tanzu/velero/pull/2543)
Contributor Shoutouts
@adamrushuk: Update basic-install and release-instructions documentation
@JessicaGreben: add a supported provider: Storj object storage
@adamrushuk: Update basic-install.md: Add windows cli installation option via chocolatey
@tbatard: Update Jekyll to 4.1.0
A
Hello,
everyone
and
welcome
to
another
episode
of
the
Balearic
community
meeting,
slash
open
discussion
today
is
June
30th
23rd.
There
we
go
2020
and
per
usual.
We
got
some
status
updates
and
a
bunch
of
discussion
topics.
So
let's
get
started.
I'm
gonna
start
my
share
my
screen
here
and
if
you
have
any
discussion
topics,
please
add
them
to
the
meeting
notes
there
we
go
right.
First
up
we
have
Nolan.
B
Hey
everybody
not
a
whole
lot
of
coding.
Last
week,
more
reviews
and
meetings,
helping
people
will
figure
out
what
to
work
on
so
yeah
lots
of
reviews.
I'm.
Sorry,
if
I
did
not
get
to
your
review-
and
there
was
also
a
lot
of
internal
work
on
some
VMware
projects
related
to
Valero,
so
not
a
whole
lot
to
add
other
than
that.
B
B
There's
a
call
for
the
other
maintainer
x'
to
take
a
look
at
that,
and
then
we
have
a
discussion
this
week
about
the
restore
hooks
PR
that
I
put
here.
We're
gonna
have
a
separate
discussion
because
it's
probably
gonna
be
longer
than
we
can
fit
into
this
call
and
all
and
still
be
able
to
fit
other
topics
in
but
yeah
mostly
this
week.
I
would
really
like
to
get
these
design
things
in
that's
the
main
goal.
D
D
Any
question
anyone
has
questions
about
this.
Alright
can
go
back
to
that
video,
so
I,
still
going
through
peer
reviews
and
adding
test
that
new
BSL
controller
and
I
also
have
to
add
I.
Think
I'm
going
to
look
into
Eddie
web
hooks
for
that
to
vet
to
validate
to
the
do
validation
on
BSL
creation
and
update,
create
and
update.
F
A
B
B
So
this
will
hopefully
alleviate
a
lot
of
confusion
around
rustic,
because
we
in
the
past
have
had
the
requirement
to
have
people
go
in
and
annotate
each
volume
that
they
wanted
to
have
targeted
by
rustic,
which
is
admittedly
annoying
so
hopefully
with
1.5.
This
will
be
a
lot
less
irritating
and
a
much
smoother
experience
so
yeah.
That
was
something
sheesh
got,
got
merge
in
this
past
week.
B
G
B
B
Thanks
so
ho,
so
the
the
the
proposal
has
is
tries
to
run
the
containers
and
have
them
get
adopted
by
I'm.
Sorry
pods
tries
to
have
them
get
adopted
by
whatever
owns
them,
so
whether
that
be
deployment
Damon
set
replicas
are
stateful
said.
The
problem
that
could
come
up
is
depending
on
what
application
that
is.
B
They
may
be
starting
up
with
empty
data,
and
these
restore
hooks
would
then
run
and
try
to
insert
data
in
the
middle
of
the
applications
execution,
so
she
isn't
on,
but
he
was
asking
about
instead
of
the
current
design
I've
proposed.
What?
If
we
do
this
with
in
hit
containers
and
maybe
modify
the
containers
and
put
in
it?
I'm
sorry
I,
keep
saying
containers,
modify
the
pods
and
put
omit
containers
on
them
that
run
the
hooks,
whatever
those
may
be,
or.
B
Some
other
form
that
makes
sure
we
run
and
populate
the
data
before
we
start
the
application.
So
I
think
that
is
a
summary
of
the
current.
The
current
state
of
the
proposal,
I
know
mark
mentioned
late
last
week
that
in
his
project
that
uses
Valero,
they
actually
used
in
it.
Containers
I,
don't
know
if
anyone
else
has
anything
that
they
would
want
to
jump
in
and
say
something
on.
H
B
The
data
source
field
right
now
I
actually
talked
to
shinging
about
this.
The
cig
lead
for
cig
storage.
I
talked
to
her
about
this
yesterday
that
data
source
field
is
only
used
for
CSI
snapshots
right
now.
It
there
are
plans
to
use
that
for
a
generic,
what's
being
called
a
generic
data,
populate
er,
but
that's
not
even
in
alpha
yet
upstream,
so
whenever
that
would
be
ready.
B
H
E
Sorry
I
have
a
quick
question
regarding
that
need
a
Papa
later
Phil,
if
you
guys
don't
mind
so
as
a
SD
summary
that
Nolan
mentioned
states,
we
really
want
a
way
for
the
applications
to
run
some
sort
of
executable
arm
during
restore
right
now.
The
end
goal
of
that
executable
is
to
generate
or
populate
data
for
the
applications
to
run,
but
in
some
cases
can
it
be
different,
and
in
that
case
came
this
data
populate
er,
be
not
the
right
thing
for
us.
B
It
let's,
let's
the
application:
no,
it
can't
even
get
the
PVC
or
PVE
until
the
data
is
there
with
the
restore
hooks
or
with
anything
out
of
band
like
Valero.
We're
kind
of
we
have
to
rely
on
something
like
in
it
hooks
or
something
like
that
to
basically
get
a
ready,
PVC
jump
in,
do
something
and
then
start
out
the
application.
G
G
The
one
of
the
use
cases
that
we're
using
backup
hooks
and
rustic
and
restore
books
floor
is
for
Postgres,
and
so
we
have
a
backup
hook
in
two
volumes
and
emptied
their
volume
and
it
runs
PG
dump
over
there
and
then
it
has
that
and
then
rest.
It
concludes
that
in
the
back
up
in
the
restore
hook,
so
we
we
would
create
an
init
container
inside
the
post
res
during
restore
to
do
the
restore.
G
But
the
problem
is
you
can't
run
PG
restoring
in
a
NIC
container,
because
the
second
container
now,
which
is
Postgres
it's
not
running
so
PG,
restore,
can't
talk
to
a
postgrads
because
there's
actually
no
instance
of
postgrads
running.
So
we've
resolved
this
problem
like
or
like
and
are
an
Arg
example.
G
We
solve
this
problem
by
putting
the
restore
hook
in
a
different
pod,
like
in
a
different
total
different
workload,
write
an
API,
and
so
this
API
is
not
going
to
report
healthy,
but
it
allows
postcards
to
come
up
and
then
it
can
do
the
PG
restore.
But
it
definitely
is
just
just
another
thing
to
think
about
when
we're
thinking
about
our
restore
hooks
should
work.
B
G
As
it
turns
out
postgrads
in
my
sequel
in
a
lot
of
these,
have
these
in
entry
point
in
it
directories,
and
so
you
can
actually
just
dump
the
directory
there
in
blood.
Postgrads
will
start
up
from
it.
So
there's
a
there's,
a
potential
workaround,
but
that's
why,
like
I'm,
not
sure
whether
this
is
something
that
is
very
application
specific
or
whether
it's
a
bolero
like
consideration
and
how
every
store
book
should
be
implemented.
G
B
And
I
think
I
think
Valero's
goal
is
to
provide
building
blocks
that
like
we
can.
We
can
get
most
of
the
way
there
and,
like
you
said,
like
Postgres
has
a
way
around
it.
So
in
an
it
container
could
put
stuff
into
like
a
well-known
location
and,
and
one
thing
in
sheesh
and
I
talked
about
I
think
Andrew
Reid
had
brought
up
was,
for
example,
if,
if
you've
got,
if
you
have
to
talk
to
a
database,
let's
say
you
don't
have
a
well-known
location,
you
can
put
information
in.
B
G
Yeah,
the
way
that
we
have
you
know
like
a
suitable,
restore
hook
thing
implemented
to
solve
our
problem.
We
basically
we've
always
created
this
init
container
in
our
API
pod
here,
and
it
has
a
simple
script
in
it.
That
looks
for
the
postcards
dump
file
and
if
it
exists,
it
runs
the
PG,
restore
and
then
deletes
the
file.
B
G
I
mean
so
the
application
that
we're
delivering
is
is
a
tool
that
allows
other
ISPs
to
write
include
their
application,
but
we
actually
so
we're
using
Valero
in
two
different
ways.
We
use
it
to
backup
our
own
state,
which
does
include
a
post
cards
database
and
a
mini.
Oh
that's
s3,
you
know,
object
store,
but
then
an
ISV
who's
packaging,
our
code
inside
our
application
can
also
use
it
and
it
turns
out
that
they
often
have
a
Postgres
database
inside
most
applications
and
so
they're.
Looking
at
our
example
as
a
way
that
there.
B
G
Run
Postgres
myself
yeah,
which
was
an
interesting
you
know
as
we
wanted
to
move
the
restore
over
to
a
separate,
Poppa
separate.
You
know
workload
in
order
to
like
allow
postcards
to
spin
up
it
actually
because
of
the
way
the
bolero
works.
We
actually
had
to
move
the
backup
over
to
that
mother
pod
also.
Otherwise,
the
would,
during
the
restore
time
like
the
volumes
attached
to
the
wrong
workload,
and
we
couldn't
actually
do
the
restore
right
right.
B
E
I
had
some
thoughts
on
that.
He
stating
the
problem
so
correct
me
if
I'm
wrong,
but
the
the
problem
that
we
are
trying
to
solve
the
restore
hook
after
this
discussion.
It
looks
like
more
of
a
policy
issue
that
some
applications
want
synchronous,
behavior,
isn't
they
want
their
leased
or
hooks
to
run
before
the
application
start,
whereas
other
applications
might
want
a
synchronous
behavior.
That
is
they
want
that
he
stood
who
chewed
on
it
after
the
application
starts
on.
B
G
If
we
don't
have
synchronous
and
asynchronous
or
pre
and
post
restore
hooks,
you
don't
actually
you
ultimately
don't
have
a
one-to-one
mapping,
and
so
there's
probably
holes
in
there
right,
like
the
reason
that
a
post
back
of
hook
exists,
probably
can't
like
whatever
reason:
I'm
not
using
those
right
now,
but
I.
Imagine
you're
going
to
need
a
post
restore
hook
in
order
to
like
provide
the
inverse
of
that.
B
And-
and
the
overall
goal
is
Valero,
just
has
no
way
of
there's
no
restore,
there's
no
restore
action.
Right
now,
like
we
have
backup
hooks.
There
are
no
restore
hooks
so
like
we
need
something
to
unpack
that
data
that
was
was
backed
up
so
we're
trying
to
get
this
in,
but
we
want
to
make
sure
we
get
to
restore
hooks,
implement
it
in
such
a
way
that
there's
not
really
a
possibility
for
data
corruption
and.
F
C
B
That
would
be.
That
would
be
great
I.
Think
that's
a
longer-term
thing
we
want
to
do
it
I
think
we
want
to
get
to
that
point.
We've
we've
got
surveys
out
there
and
got
ideas
on
making
higher
level
constructs
to
back
up
based
on
like
the
application,
whether
that's
there
there's
been
talk
of
like
doing
it
based
on
helm,
charts
or
the
cig,
apps
application
construct,
or
some
excuse
me
some
way
of
using
something
that
tells
us
hey.
B
This
is
what
my
app
is
made
out
of
it's
made
out
of
these
pods
and
it's
not
just
pods
it's.
This
pod
has
to
go
first,
but
I.
Think
right
now
we're
we're
in
the
building
blocks
phase,
so
I
agree
that
that
would
be.
That
would
be
really
nice
and
we
want
to
get
there,
but
I
think
we're
also
not
quite
there
yet.
F
Actually,
in
the
building
block
phrase,
maybe
you
want
to
take
advantage
like
have
hooks
based
on
the
on
the
kubernetes
board
events,
not
just
on
on
talero
pv,
restore
events
or
resource
with
students
in
general.
I
think
that
our
use
case
I'm
from
the
excavator
team,
I'm
the
PM
of
this
team.
I
think
that
our
use
case
and
marks
use
case
seems
very
similar
if
not
identical.
We.
F
B
B
B
C
No
worries
Nolan
I
was
just
saying
thanks
everyone
for
this
really
great
discussion
and
for
some
of
the
use
cases
that
were
brought
up
here.
We
are.
We
have
comments
running
on
the
PR
on
get
and
are
working
on,
scheduling,
a
session
that
we
will
record
and
will
post
on
the
community
channel
to
dive
deep
on
this
particular
PR,
and
so
we're
hoping
we'll
get
that
information
out
to
you
all
and
we're
hoping
that
you
all
will
join
us
for
that
as
well.
A
B
B
We
are
going
to
try
to
move
api's
go
api's
that
are
not
in
this
set
into
an
internal
package
so
that
we
can
have
a
more
Delaney
ated
go
API.
Now
this
is
not
going
to
be
immediate.
This
is
gonna
be
kind
of
over
as
we
find
things
if
you
rely
on
anything
not
in
this
list.
Please
let
us
know
through
this
issue
26:54
and
we
will
try
to
accommodate
you.
B
Will
eventually,
yeah
client
and
generated
will
eventually
change,
but
for
now
this
is
where
this
is.
Where,
like
the
Valero
client
would
go.
If
you
want
to
call
CR
DS
the
client
for
CDs
I
know
some
of
the
plugins
are
using
these
ok
and
as
well
as
generated,
but
these
these
are
our
purposefully
plug
public
api's.
So.
B
B
B
B
Yeah
and
and
I
did
a
search
to
github
I
think
this
covers
most
code.
I
I
don't
think
I
could
be
completely
exhaustive
and
I.
Don't
know
that
all
like
just
because
I
search
on
github
doesn't
mean
that
God
code,
some
people
could
have
code,
that's
not
on
github,
so
we're
putting
that
call
out
there
and.
B
Yeah,
let
us
know,
like
Carly
CI,
said
we're.
Probably
we're
gonna
be
moving
towards
a
q
builder
and
controller
runtime
structure
in
2.0,
and
so
the
client
in
generated
code
will
be
different
and
you
can
use
the
cube
builder
client
to
register
the
velaro
scheme,
so
the
client
can
generated
packages
will
be
gone
by
then,
but
for
now
this
these
are
the
intentionally
public
API.
The
rest
is
kind
of
a
byproduct
of
how
we
structured
things,
and
we
haven't
been
great
about.
B
About
deprecations
and
not
moving
things
around
so
if
you've
been
relying
on
that,
we've
just
been
kind
of
moving
when
we
needed
to
move
stuff
around,
we
did.
But
it's
come
up
that
hey.
Maybe
we
should
deprecated
some
of
this
stuff
when
we
do
that,
so
we've
decided
how
about
we
instead
try
to
move
that
into
the
internal
directory
structure
so
that
we're
not
we're
not
tell
it
we're,
not
exposing
stuff.
We
didn't
intend
to.
E
Have
a
question:
so
it's
in
connection
with
the
D,
client
and
degenerated
packets.
That's
gonna
be
for
it
in
2.00.
Wait
controllers
a
time
client!
So
is
there
any
place
where
folks
can
find
out
how
the
controller's
runtime
generated
client
will
be
different
from
the
actual
client
that's
generated
by
client
go
or
maybe
I
might
be
missing
the
exact
specific
but
the
current
client
package,
and
if
consumers
are
actually
using
decline
package,
would
they
have
some
sort
of
upgrade
path
from
that
client
package
dependency
to
the
controller
runtime
time
package
stuff
like
that?
B
B
The
my
understanding
and
currently
CEO
you'll
have
to
help
me
here.
My
understanding
is
with
the
cube
builder
client.
You
get
a
scheme
for
an
API
group
and
you
register
that
with
the
cube
builder
client,
so
you
would
import
the
valero
API
group
scheme
give
that
to
a
cube
builder,
client
and
the
cube
builder
client
could
then
look
up
any
any.
F
B
B
My
currently
see
has
dealt
with
this
a
lot
more
than
I
have,
but
my
understanding
is
using
controller
runtime.
You
won't
have
to
generate
these
specifically
typed
clients
like
we.
We
do
for
the
Valero
client
anymore.
You'll,
just
use
you'll
generate
a
scheme
which
knows
about
all
of
the
registered
CR
DS
and
the
go
types,
and
then
you
put
that
scheme
into
the
client.
That's
a
controller
one-time
client,
yeah.
E
That
makes
sense.
I
was
just
curious,
but
because
I
know
that
for
the
current
clients
we
can
have
informers
on
the
generated
Valero
API
consumers
can
use
those
informers
to
react
on
Valero
changes.
I
was
not
sure
how
those
informers
knew
change
when
we
move
to
controller
and
time
I
probably
need
to
do
some
baking.
I
can.
D
Get
a
very
quick
rundown
on
that.
So
so
do
you
so
with
the
controller
runtime,
if
you,
if
you
use
the,
if
you
get
a
client
through
the
manager,
because
you
can
also
get
a
cry
straight
up,
clients
without
using
the
manager
hope
it
makes
sense.
But
if
you
get,
if
you
using
the
client
that
gets
generated
through
the
manager,
if
you
stand,
she
ate
a
manager
and
then
you
get
used
the
client
from
there
whenever
it
does
a
get
or
a
list.
Its
Fasching
fetching
from
the
cache.
E
Yeah
that
makes
sense,
but
so
I'm,
sorry
when
I
say
informers,
I
mean
folks
can
use
those
informers
to
write
like
control
loops.
One
changes
to
quell
their
OCR's
and
stuff,
like
that,
I'm,
not
sure.
If
controller
Trenton
manager
exposes
informers
to
I
mean
I
might
be
taking
this
discussion
in
wrong
direction.
I
would
have
to
look
up
and
get
back
yeah.
D
So
the
I
would
suggest,
if
you're
going
to
look
in
so
I
start
with
the
manager,
because
the
manager
provides
the
sinking
the
watching
and
he
does
that
you
we
don't
have
to
manage
that.
We
can
tweak
a
if
we
need
we
need
to,
but
you
straight
up,
you
create
a
controller,
you
pass
a
manager
and
that
is
going
to
to
keep
watching
and
if
you
want
to
get
any
form
in
specific.
There
is
also
a
way
to
do
that.
If
you
want
to
check,
if
something
has
things.
E
J
Yes,
I
have
one
kitchen
so
as
of
now
under
this
package
back
up
and
packages
too,
we
have
some
files
so
which
are
having
some
generic
API
like
creating
some
delete,
requests
or
using
some
resource
for
the
store
or
backup.
So
are
we
planning
to
create
new
package
so
that,
through
new
package,
we
can
expose
those
API,
so
they
can
use
it
or
we
are
going
with
this
existing
package.
Only
I.
B
B
There
was
also
like
there
there's
an
issue
somewhere:
I'll
have
to
find
it,
but
there
was
there
was
an
issue
to
at
some
point,
make
make
package
backup
impact
of
package,
restore
more
like
a
library
and
I.
Think.
The
big
things
we
want
to
put
in
internal
are
like
the
controllers,
but
some
elements
of
package,
backup
and
package
restore,
may
get
pulled
into
internal,
but
we
can
definitely
discuss
what
parts
of
package
backup
and
package
restore
our
public
but
yeah.
B
D
That's
exactly
what
we
need
to
know.
We,
we
don't
want
to
break
anybody's
code,
but
we
we
find
ourselves
right
now
where
we
have
to
be
I
mean
if
we
are
going
to
be
careful
about
not
bringing
anybody's
cold,
we
have
to
maintain
backwards,
compatibility
with
everything.
Well,
it
will
right
and
that
just
really
restraining
for
us.
There
are
in
their
practices
that
we,
we
know
for
sure
people,
but
now
using
you
should
not
be
using
in
in.
We
have
to
answer
the
question:
do
we
have
to
worry
about
maintaining
backwards
comparability?
D
Well,
where
does
if
we
know
people
are
not
using,
we
put
an
intern
that
we
don't
have
to
worry
about
that.
So
we
just
want
to
make
sure
we
focus
on
what
we
do
have
to
worry
about
and
make
sure
we
don't
break
anything
for
anybody,
but
at
the
same
time
we
can
move
forward
without
having
to
worry
about
breaking
things,
people
we
totally.
In
other
words,
if
we
know
people
are
using
it,
we
are
going
to
accommodate
you
one
way
or
another.
A
B
I
want
to
clarify
this
does
not
or
should
not
turn
on
rustic
by
default
when
you
install
velaro.
This
is
when
you
pass
the
CLI
flag.
It's
the
third
bullet
point
when
you
turn
on
the
server,
if
you,
if
you
tell
the
valero
server
to
default
the
volumes
to
rustic,
it's
if
you
have
velaro
start
up
now
without
adding
that,
then
Valera
is
going
to
behave
as
as
it
does
today.
I.
D
B
H
And
this
thing
like
for
user,
will,
let
me
make
more
sense.
Like
I
mean
I
understand
how
we
caught
with
back
up
volume
excludes
the
because
it
is
Ernie's.
Users
are
supposed
to
providing
the
volume
see
how
to
back
up
Eucharistic.
So
when
we
like
to
reduce
the
kind
of
lower
or
like
the
the
manual
steps
to
perform.
So
it's
really
helpful
in
that
way
listing
like
gamma
mu
step
is
we
labeled
helps
the
plugins
are
not
interested,
then
those
will
be
done
by
default.
H
B
H
H
H
H
B
H
I
H
B
B
H
B
H
H
G
B
B
I've
got
two
for
Adam
Rush.
He
they're
actually
related.
He
added
a
Windows
CLI
installation
option
via
the
chocolaty
package
manager
and
give
us
some
instructions
for
how
to
get
that
updated.
So,
thank
you
very
much.
Adam!
That's
really
cool
Jessica
graben
added.
She
developed
an
added
documentation
for
the
store,
Jay
and
tardigrade
object,
storage,
plugin
and
that's
pretty
cool
I.
Think
I'm.
Sorry
I
forget
his
name.
Was
it
Kevin,
Kevin,
Kevin,
I,
wanted
to
say
Jeff,
but
I
knew
that
wasn't
right.