►
From YouTube: Velero Community Meeting / Open Discussion - Oct 6, 2020
Description
Oct 6, 2020
Status Updates
dsmithuchida
Working on e2e testing and ginkgo framework
bridget
Couple of outstanding PRs that I’d like reviewed (#2951, #2969)
Been working on some bugs:
Backup deletion fails if cloud resources already deleted #2980
This is a regression in behaviour from v1.4.2 due to the introduction of DeleteItemActions. We can introduce a fix, but how should we handle the case where there are DeleteItemActions to perform but the resources are deleted?
Unable to install Velero using go get #2966
We can fix this by upgrading the version of CSI external-snapshotter but it requires updating the version of Kubernetes modules we’re using to 0.19.0. What is the impact of upgrading these libraries?
nrb
Not a whole lot of updates; was on semi-PTO last week.
Intending to do a “support-a-thon”/bug smash session when the whole team is back
Discussion Topics
Waiting for Additional Items to restore
Issue #2980 - deleteitemactions regression
Happens all the time, but what do we do when there’s DeleteItemActions but no backup tarball?
Since there’s no way to really know what to match the plugins against, log that there was no tarball and continue, skipping the plugins.
Community Contributions
@sseago: restore proper lowercase/plural CRD resource
@svcua: [velero] Bump Velero version to v1.5.1
@xUnholy: [velero] fix bug using workload identity issue #161
@nirav-chotai: [Velero] Fixing restic daemonset yaml
A
B
A
C
Hi
everyone
so
I've
just
posted
I've
got
a
couple
of
pr's
which
are
pretty
small
but
things
that
I'd
like
to
get
reviewed,
and
one
of
them
is
an
improvement
to
the
tour
release
scripts.
So
I'm
just
thinking
that
we
might
have
a
1.5.2
release
coming
up,
and
so
I'd
like
to
get
that
review
and
get
that
in
the
other
thing
is
I've.
C
Just
it
was
a
bit
of
a
quiet
week
last
week,
so
I've
just
been
working
on
some
some
bugs
one
of
them
was
to
do
with.
If
you
try
to
delete
a
backup
where
the
associated
cloud
resources
have
already
been
deleted,
the
backup
deletion
fails
and
then
the
the
customer
resource
gets
stuck
in
the
deleting
phase
and
then
doesn't-
and
this
is
due
to
during
the
processing
of
delete
item
actions,
it
tries
to
download
the
tarball.
So
if
the
tarball
is
missing,
it
can't
get
stuck
there.
C
So
this
because
this
is
a
new
feature
in
1.5.
This
is
been
a
regression
of
sorts
in
terms
of
like
behavior.
So
it's
just
kind
of
I
I've
taken
a
look
at
storage
work
on
a
fix
for
it,
but
then
kind
of
got
stuck
where
I
realized
that,
in
the
case
where
the
cloud
resources
are
missing,
what
should
the
behavior
be
if,
thirdly,
item
actions
for
that
plugin
to
be
processed?
And
how
do
we
communicate
that
to
users?
C
C
C
D
D
D
D
D
Yeah,
I
think
that
makes
sense.
I
I
think
we're
gonna
want
to
keep
pretty
close
to
what
kubernetes
is
doing.
I
know
there's
there's
a
fix
for
some
crd
marshaling
stuff.
That
would
be
really
nice
that
we've
got
a
bunch
of
workarounds
spread
throughout
the
valero
code
base
that
will
be
in
kubernetes
120..
D
D
D
So
it's
it's
just
when
you
go
through
go
get
right!
Now:
okay,
okay,
thanks
yeah
yeah
this!
This
worked
for
our
our
releases,
it's
just
if
you,
if
you
try
to
install
it
through
goget,
I
think
there's
a
different
code
path
that
it
uses
to
read
the
go
modules
or
something
I'm
not
exactly
sure
what
logic
is
different
between
the
two,
but
I
I
can
definitely
reproduce
this
and
it's
basically.
If
we
follow
the
chain,
we
have
to
update
the
external
snapshotter
that
we
depend
on,
but
they
use
kubernetes
0-19.
A
A
D
D
This
is
not
a
commitment,
but
just
a
guess
that
I
think
that
will
probably
be
a
mix
of
1.5,
1.5.2
and
1.6.0,
where
those
fixes
will
land,
but
I
don't
know
yet,
and
we
also
owe
the
1.6.0
milestones
or
the
the
list
that
we
were
going
to
commit
to.
But
the
number
one
thing
were
were
our
team
is
intending
to
deliver?
Is
the
end-to-end
testing
so
that
we
have
solid
testing
before
we're
doing
big
changes
in
galero?
D
I
think
this
week
I'm
going
to
be
catching
up
on
reviews
that
I
I
didn't
really
get
to
last
week,
so
I
don't
think
I
need
a
whole
lot.
The
the
main
ask
I
have
is
the
the
discussion
topic,
which
I
don't
think
I
see
scott
on
the
call
but
yeah.
I
think
the
main
thing
this
week
will
be
catching
up
on
reviews
and
and
support
issues.
A
A
D
We
can
wait
until
scott's
on
this
is
basically
for
for
folks
that
don't
know
this
is
a
design
topic
that
scott
brought
to
us.
Jack.
We've
actually
been
in
discussion
on
this
for
a
few
months,
and
this
is
a
change
to
the
restore
item,
action
plugins
that
allows
the
plugins
to
wait
until
the
additional
items
are
actually
restored
and
then
proceeds.
So
the
the
reasoning
here
was
in
openshift.
D
D
So
for
maintainers.
I
would
ask
that
you
read
over
this
and
make
sure
it
makes
sense
and
it
fits-
and
I
guess,
for
the
rest
of
the
folks-
do
the
same
and
and
see
I
I
know,
there's
a
couple.
I
see
dylan
from
red
hat.
I
I
know
this
is.
This
is
a
big
thing
for
openshift,
but
I
can
see
it
generally
fitting
in
with
the
rest
of
valero.
E
So
if
you
have
openshift
installed-
and
you
have
a
docker
registry
running
on
that
on
the
cluster
then
before
we
can
actually
go
and
recreate
all
of
the
image
stream
resources
which
again
they're
openshift
custom
resources,
but
they
depend
heavily
on
these
images
already
being
pre-populated
in
the
registry.
So
that's
kind
of
the
driving
force
for
this,
but
I'm
unsure
if
scott
thought
of
a
few
other
use
cases
or
not
in
his
head,
I've
only
been
briefed
slightly
on
this
design.
So
yeah,
it's
kind
of
directly
myself.
D
Yeah
and
there's
there's
a
link
to
this
out
to
an
issue
where
scott
and
I
had
a
back
and
forth
for
a
while
it
was.
It
was
kind
of
asynchronous.
I
think
so.
I
think
if
I
remember
it
stemmed
out
of
there,
there
was
a
race
condition
in
valero,
where
we
submitted
to
create
a
custom
resource
definition,
and
then
it
wasn't.
D
Sometimes
it
wouldn't
quite
be
ready
to
consume
until
kubernetes
did
some
validation
and
then
we
could
get
into
a
state
where
that
validation
wasn't
done,
but
valero
was
submitting
the
custom
resources,
and
so
they
wouldn't
actually
get
restored
because
kubernetes
didn't
know
it
was
a
valid
definition.
Yet
so
we
put
in
some
weight
logic
and
kind
of
this.
This
grew
out
of
that
because
I
think
I
think
the
intent
was
hey.
E
D
D
D
C
Yeah,
so
if
the
yeah,
so
I
think,
if
the
backup
char
ball
is
non-existent
or
if
the
entire
directory
associated
with
the
backup
has
been
removed,
and
so
what
happens
is
it
tries?
There's
a
download
tries
to
download
that
in
order
to
perform
back
the
delete
item
actions,
even
if
there
are
no
associated
delete
item
actions
for
that
particular
backup.
So
I
I
thought:
oh,
we
can
check
for
there
are
well.
C
We
can
check
for
the
case
where
there
aren't
any
delete
item
actions
associated
with
this
backup
and
then
skip
the
download
in
that
case.
But
then
I
kind
of
raised
another
question
in
my
head.
What
was
reason
the
question
of
what
the
behavior
should
be
in
the
case
where
there
aren't
there
are
associated
deletion
actions,
but
their
resources
have
been
removed.
D
C
C
C
B
So
we
put
this
in
for
the
vsphere
plug-in
and
I
think
orphaned
objects
are.
You
know,
acceptable.
It's
like
in
the
current
setup
we
have
the
was
the
snapshots.json
file.
The
volume
snapshot
json
file.
If
that
gets
lost,
we
would
not
call
the
the
volume
snap
shadows
to
delete
correct
right.
That's
basically
the
same
behavior.
D
Yeah
like
this
to
me,
I
I
need
to
re,
read
the
issue
more
generally,
but
it
sounds.
It
sounds
to
me
like
someone
went
around
valero
and
cleaned
out
the
object
store
and
then
tried
to
do
a
valero
delete,
because
if,
if
valero
was
managing
the
object
storage
itself,
it
should
clean
up
everything
in
an
orderly
fashion.
D
D
D
Yeah
yeah,
okay,
so
it
sounds,
it
sounds
like
we
just
need
better
error,
checking
and
logging
all
right.
That
makes
sense.
I
wasn't
sure
if,
if
this
was,
if
I
was
just
misunderstanding,
or
if
there
were,
there
were
other
resources,
but
it
sound.
It
sounds
like
these
folks.
Have
these
folks
had
delete
item
action,
plugins
installed
and
we're
missing
the
tarball.
B
C
D
Right
right,
I
remember
putting
that
in
but
okay,
but
you
were
just
saying
you
were
thinking
about
the
failure
cases.
Yes,
okay
got
it
yep
because
I
remember
trying
to
fetch
it,
regardless
of
whether
they
were
there
or
not,
but
because
because
that
I
think
that
was
the
original
behavior
but
okay,
I
I
get
yeah
yeah.
It
makes
sense
that
we
would
move
on.
If
there
are
none,
just
go
back
to
the
old
behavior,
and
if
there
are
some
we
log
it
and
say
hey.
D
D
A
D
D
We
introduced
some
regression
in
the
crd
weighting
that
we
mentioned
before,
and
scott
tracked
this
down
and
fixed
it.
So
this
fix
will
be
coming
out
in
one
five
two.
This
has
been
a
little
bit
of
a
whack-a-mole
in
in
getting
the
proper
crd
resource
names
in
the
code.
So
thank
you
to
scott
for
tracking
this
down
and
making
sure
we
had
the
correct
strings.