►
From YouTube: OCI Weekly Discussion - 2022-01-20
Description
agenda/notes: https://hackmd.io/El8Dd2xrTlCaCG59ns5cwg?view#January-20-2022
A
Maybe
we
can
give
an
update
on
the
working
group
stuff
like
most
people
there,
but.
C
Hey
josh,
hey:
are
you
able
to
give
the
update,
I'm
just
dropping
my
kids
to
school,
so
it's
I'm
getting.
A
C
A
C
A
I
guess
here
so
there's
no
agenda
this
week,
so
we
talked
about
a
minute
ago
talking
about
what's
going
on
with
the
reference
group,
the
reference
types
working
group
and
I
will
post
the
notes
to
this.
A
A
Tuesday,
thank
you
trying
to
get
together
the
mission
statement.
That
lockheed
is
put
a
pr
against
in
github
and
there's
a
little
bit
of
discussion
about
that.
A
I
think,
though,
the
general
idea
was
to
simplify
and
distill
what
it
is
that
we're
trying
to
achieve
there
and
then
for
the
next
steps,
I'm
seeing
something
about
garbage
collection
in
here
I
I
was
a
little
lost
on
that.
If
someone
wants
to
talk
about
that,
a
bit.
C
I
think
the
the
agreed
next
steps
were
distilling
the
proposals
into
a
set
of
requirements
and
then
getting
agreement
on
the
requirements
and
then
sorry
I
live
at
the
top
of
a
very
big
hill,
so
I'm
puffed,
the
other
one
was
you
were
going
to
pull
out,
create
a
straw
person
example
that
we
could
use.
A
Yeah
and
to
elaborate
on
that
last
piece.
Basically,
my
point
was
so
that
we
don't
get
bogged
down
in
the
specifics
of
the
tooling
and
projects
that
each
of
us
and
our
companies
are
trying
to
promote
that.
Perhaps
we
could
do
something
friendly
like
that,
would
have
parallels
with
s
bombs
and
signatures
and
other
types
of
things
that
we
might
want
to
link
together
in
a
registry,
and
if
you
check
out
the
there's
a
new
channel
in
slack
wg
reference
types.
A
Where
there's
a
discussion
on
this,
I
brought
up
ice
cream.
Vanessa
talked
about
cookies.
I
I
don't
know
if,
at
this
point
it's
getting
more
distracting
than
than
that,
but
that's
pretty
much
it
so
that
that
meeting
is
tuesdays
at
11,
pacific.
A
D
There
was
a
comment
about
doing
some
pr
pruning
was
our
interest
in
doing
that,
or
do
we
just
not
have
the
right
people
here
for
that.
A
I
see
john,
I
see
anon,
I
see
samuel
what
what
specifically
was
there
something
specific
I
see
mike.
I.
D
I
didn't
have
any
on
my
list,
but
that
was
amy's
suggestion.
A
I
I'd
be
happy
to
to
share
screen
and
do
some
of
that
is
there.
Did
anybody
have
any
issues
that
they
had
proposed
or
wanted
to
look
at
specifically.
E
E
D
A
A
B
F
I
Okay,
what
were
the
hot
topics?
Did
we
did
we
get
the?
What
is
it
the
extensions
thing
like
like?
Did
we
get
that
worked
out
or
we
still
kind
of.
I
G
I
Oh
okay,
oh
it
looks
like
people
like
module
over
operation.
That's
that's
good.
I
All
right
well
good
to
see
you
is
that
sam
and
mike
and
tian
all
right.
Well,
I
guess
I'll
hop
off
then
or
if
you
guys
have
any
questions
or
anything
like
that.
I
I
Well,
like
gogo,
went
away
right,
and
then
I
guess
this
is
like
completely
irrelevant
for
oci,
but
like
gogo
has
pretty
much
gone
away
and
then
we're
kind
of
we
like
there's
a
lot
of
usage
of
gogo
across
the
community.
But
now
there's
this
new
thing
called
source
relative
and
the
paths
that
kind
of
helps
generate
things.
I
I
Something
it's
so
the
hardest
part
about
getting
proto
f
generation
to
work
out
of
with
modules
was
that
you
had
to
kind
of
use
the
go
path
to
set
the
output
path
to
like
get
it
to
go
next
to
your
files.
I
So
all
source
relative
does
is
get
rid
of
that,
so
that
when
you're,
when
you
set
your
output,
it'll
go
next
to
the
protobuf
file,
which
is
kind
of
what
you
want
and
they
so
that
fixes
a
lot
of
problems
with
the
generation
just
wondering
if
anybody
knew
anything
about
that.
G
I
think
you're
alone
and
caring.
I
That's
a
tough
two
meetings.
That's
a
tough
meeting
time
for
me
to
make
yeah.
G
G
I
I
Maybe
I
should
show
up
to
the
to
that
group
and
do
it
and
at
least
say
hey,
are
we
interested
in
this
and
like
looking
at
it
as
a
as
a
possibility,
and
hopefully
I
don't
get
tomatoes
thrown
at
me.
G
C
I
I
So
they're,
not
like
hammering
your
database
with
crazy
queries
like
you're,
serving
them
with
like
a
formed
designed
query,
but
at
the
same
time
they're
able
to
like
like
say
they
need
to
add
a
new
field
or
they
want
to
like
join
it
with
some
other
object,
or
something
like
that,
and
do
that
without
your
support,
which
is
which
is
good
but
you're
still
having
your
query.
Space
is
still
limited
to
what
you've
targeted
with
the
performance
of
your
application.
G
I
No,
no,
no
I'm
saying
you
kind
of
know
the
queries
that
are
going
to
come
in.
You
know
the
relationships
are
going
to
get
queried
up
front,
and
so
you
don't
need
to
serve
arbitrary
queries.
G
I
I
Yeah
yeah
yeah
well
gremlin's
the
project
is
it?
Is
it
gql?
I
hope
it's
called
gql,
but
the
no
it's
not
that
this
is
like
more
about
it's
like
a
view.
Language
would
be
kind
of
the
way
to
put
it.
I
F
I
Yeah
yeah
right
and
you
found
you
found
two
instances:
yeah
right
off
the
bat
first
two
results
yeah.
I
looked
at
rejoiner
at
some
time
because
I
was
just
like
oh
cool.
They
have
a
java
thing
for
it.
It
looks
actually
more
advanced
than
what's
available
in
the
go
community.
That
was
the
only
problem
we
had
with
graphql
is
the
go.
I
But
the
javascript
stuff
is,
is
impressive,
like
you
can
give
it
like.
A
database
schema
it'll
like
automatically
generate
a
schema
for
you
and
you
can
and
it
will
serve
arbitrary
queries
pretty
much,
but
the
ghost
stuff
you
end
up.
Writing
these,
like
resolver
things,
and
you
can
really
control
how
you
craft
queries
against
your
your
database
back
in
and
make
it
easily
kind
of
work
across
different
data
sources
and
whatnot.
I
So
it
might
be
interesting,
but
I
mean
like
we
might
find
with
with
registries
the
that
you
know.
The
set
of
queries
that
you
really
want
to
have
are
are
pretty
small
right.
You
might
be
able
to
define
those
relationships
and
give
them
proper
names.
I
I
G
I
G
I
Yeah
yeah
and
then
we
oh,
we
also
have
go
generics
to
deal
with
too.
G
A
I
G
I
Yeah
yeah
yeah,
I
I
think
it'll
take
a
little
bit
to
figure
out
like
okay.
Am
I
making
a
mistake
with
this
design?
I
I
played
with
it
a
little
bit
and
you
can
definitely
work
yourself
into
a
corner,
but
it's
pretty
powerful
and
it
like
it
does
solve
some
problems.
I've
had
before
the
I
I
I
hope
it
could
do
something
about
like
kubernetes
client
go,
for
example,
like
the
parameterization
of
interfaces
would
be
great
over
types.
G
I
haven't
touched
it
in
a
month,
but
an
attempt
to
rewrite
clanko
using
generics.
I
Yeah,
that
would
be
that
would
fixed
like
90
percent
of
the
problems
and
just
the
amount
of
code.
There
would
also
go
down
and
then
also
you
would
reduce
the
code
generation
required
for
crts
to
probably
just
the
types
themselves.
I
Yeah
yeah,
I
I
I
have
a
project
that
uses
crds
and
I
haven't
updated
it
in
a
while.
I'm
I'm
dreading
that
moment,
there's
usually
because
the
with
that
code
generation
usually
there's
some
sort
of
internal
dependency
that
you
end
up
with
as
well.
That
breaks
at
least
that's
what
I
saw
on
the
kubernetes
stuff.
G
When
we
started
there's
no
story
for
like
upgrading
like
up
bumping
versions
and
by
the
time
I
stopped
carrying,
it
still
wasn't
a
bad
state.
I
think
just
like
how
do
we
service
these
versions?
We
don't
okay,
cool.
I
guess,
like
the
set
of
rules
you
needed
to
follow,
to
like
be
able
to
change
a
crd
to
a
new
version
without
breaking
stuff
was
totally
incomprehensible
to
me.
I
Yeah,
I
think
that's
a
difficult
problem.
Yeah,
I
think
you
just
have
to
have
you
just
support
them
both.
G
I
G
I
G
I
Yeah
versioning
yeah
versioning
is
not
a
solved
problem.
I
think
you,
I
think,
there's
some
belief.
There
is
because
they'll
be
like
oh
somber,
just
use
that
and
everything
works
out,
but
there's
a
lot
of
like
subjective
things
around
it
that
require
case
specific
mitigations.
I
This
was
like
around
1
13,
maybe
114
time
frame
there
it
changed.
There
was
like
a
change
in
the
way
that
it
handled
certain
kinds
of
timestamp
formats,
and
so
it
would
basically
layers
that
were
generated
on
one
version
of
docker
would.
I
I
I
And
then
it
would
so
that
it
would
generate
something
that
would
actually
hash
different
yeah.
This
was
the
like
pax
and
gnu
stuff
and
then,
when
you
like
start
looking
at
tar,
because
everybody's,
like
oh
yeah,
tar's,
been
around
forever,
the
format's
got
to
be
stable,
there's
actually
like
three
different
formats
of
varying
different
adoptions
and
behaviors.
I
G
I
Yeah
yeah
he
got
that
one
right
nailed
it
I
mean,
because
I
mean
there's,
there's.
I
G
I
And
you
also
have
to
ask:
is
this
important
enough
to
the
disposition
of
the
content
to
break
a
hash
if
it
changes
right
and
every
time
you
go?
No,
it's
not
important
enough,
but
then
there's
a
use
case
that
relies
on
that
and
then
you
add
it
back
in
you
cause
churn
and
that
and
the
hashes.
So
now
you
have
two
things
that
would
have
hatched
equal
before
are
now
different,
and
so
you
have
to
like
each
step.
You
take.
I
You
have
to
make
these
decisions,
whereas,
like
you
know
what
you
can
say,
if
you
just
use
tar,
you
know
you
can.
I
It
looks
fine
until
it
breaks
and
it's
mostly
not
breaking,
because
the
tool
the
underlying
tooling
is
stable.
C
C
I
G
Yeah
star
gc,
the
I
think
the
e
part
mostly
is
the
reshuffling
of
like
access
order
and
also
they
win
actually
added
it
to
container
d.
G
I
think
star
gz
is
very
clever.
I
wish
I
had
thought
of
it.
G
It
it
basically
chunks
the
gziptar
into
a
concatenated
series
of
g-zip
tars.
I
G
Yeah,
which
clients
interpret
just
fine
and
then
at
the
end
depends
a
index.
I
Yeah,
okay,
because
we
okay-
because
we
talked
about
things
like
this,
but
we
were
always
like.
Oh,
how
do
you
get
a
tar
file?
That'll
be
valid,
two
things
that
don't
know
about
it,
but
then
yeah.
So
this
yeah.
This
is
pretty
clever
to
actually
make
it
work.
G
I
I
I
Just
because
it
would,
it
would
create
the
insecure
situation
where
people
might
try
to
pull
apart,
that
metadata
file
before
they
like
verify
the
hash,
to
get
some
information
about
how
to
like
unpack
it
or
something
like
that.
G
G
The
these
three
defects
like
propose
this
to
the
image
spec.
I
don't
really
know
how
to
respond.
I
G
G
Yeah,
I
think
it
looks
for
this
annotation
and
it
does
a
range
request
on
the
end
of
the
layer
and
there's
also
an
annotation
that
describes
like
the
size
of
the
table
of
contents
and
so
or
maybe
they
wanted
one.
But
you
should
be
able
to
make
a
range
request
to
get
the
entire
table
of
contents
and
then
from
there,
just
lazily
access
stuff.
G
Like
each
entry
has
the
like,
which
chunk
it's
in,
and
so
you
can
do
another
range
request
to
just
request
that
gzipped
tarball
chunk.
I
Okay
and
we
lose
some,
do
we
lose
some
compression
efficiency.
B
G
I
think
star
gz
is
that
uses
the
star
gz
format.
I
think
the
e
star
gz
bit.
It
involves
prioritizing
files.
I
So
I'm
gonna
give
this
like
a
idea
lgtm
and
then
we
need
to
have
all
the
maintainers
read
it
in
detail.
G
G
I
Yeah,
but
if
it
might
be
easier
to
have
a
separate
target,
z,
spec
and
then
have
the
annotation
and
distribution
reference
it
and
be
like
hey:
here's
how
to
detect
targey
z
layers
and
your
etarjeezy
layers,
and
if
you
encounter
them.
Here's
what
to
do
on
the
client.
I
Okay,
all
right!
Well,
I
have
to
drop
off
so,
but
it
was
good
to
go
through
this
stuff
with
y'all.
I
don't
know
if
we
can
reach
out.
I
I
made
a
comment
on
the
etard
gz1.
I
The
might
be
good,
I
don't
know
if
we
can
reach
out
to
this
person.
I
don't
know
who
this
person
is,
let's
see
if
they
can
present
it.