►
From YouTube: SONiC DASH Workgroup Community Meeting May 3 2023
Description
PR289 - Still needs volunteer - Dockerfile.saithrift-client has a static tag in the FROM instruction
SmartNIC Summit June 13 - 15 in San Jose, CA
PR336 - [ACL]: Optimize DASH ACL by introducing cluster tag
Continue “acl tag” convo
Q&A
PR346 - Change SAI submodule URL to opencomputeproject/master - pending
A
Community
called
today
today
is
our
67th
Dash
meeting
and
what
I'm
showing
on
the
screen
here
are
the
notes
from
last
time
and
so
people
who
attended
last
time
in
our
notes
for
anyone
new
on
the
call,
and
so
we
talked
about
the
quick
recap
of
the
P4
workshop
on
the
Intel
campus,
and
we
mentioned
that
the
smart
Nick
Summit
is
coming
in
June
in
San
Jose,
so
I
believe
I'm
planning
to
be
there.
Gerald,
I'm,
I,
know
I,
think
Chris
or
someone
had
planned
to
be
there.
A
So
we
could
talk
about
that
a
little
bit
and
then
I
know
Clark
from
Alibaba
wanted
to
talk
about
his
Nat
PR,
and
we
talked
a
little
bit
about
changes
to
psi
apis
and
some
ideas
around
that,
and
so
that's
what
happened
last
week
and
then
for
anyone
new
I,
usually
post
the
YouTube
videos
here
in
these
links,
just
the
community
call
the
high
availability
call
and
the
behavioral
model
call
if
we
have
those.
So
that's
what
was
happening
and
if
anyone
has
a
PR
or
something
they
want
to
discuss.
A
A
A
C
B
B
B
B
So
last
week,
sorry,
when
we
talked
about-
let's
say
you
add
an
IP
to
three
tags,
so
you
give
it
a
list
of
three
tags
and
let's
say
you
wanna
one,
two,
three
and
let's
say
you
want
to
remove
from
two
I
think
the
ID
was
well.
You
would
give
it
a
set
of
the
remaining
two
tags
that
IP
belongs
to
I.
Believe
that's
that's
what
was
proposed
last
week
to
make
this
work
to
make
a
remove
work.
B
However,
the
problem
is
the
set
taking
a
un32
as
a
tag
doesn't
allow
to
give
a
list
of
attributes
the
API,
the
Dashi,
the
PSI
API.
You
can
do
a
set
on
an
attribute
and
that's
it.
There's
no
attribute
count.
So
how
does
that
work?
It's
kind
of
similar
to
also
the
set
on
ECR
rules
where
we
cannot
give
a
list
of
port
or
anything
like
that.
B
So
to
me,
I'm,
not
sure
how
we're
going
to
do
the
set.
Unless,
like
is
your
rules,
we
basically
well,
we
will
see
our
rules,
we
would
create
a
new
set
of
rules
and
then
swap
on
The
View
on
the
eni
and
then
and
then
delete
the
old
stuff.
Well
here
it's
the
same
IP
as
a
key,
so
I'm
not
sure
how
we're
going
to
do
that
unless
we
remove
the
prefix
from
all
the
tags,
and
then
we
add
them,
we
add
that
prefix
to
all
the
exit,
the
remaining
tags.
B
A
D
B
D
B
D
B
E
E
B
D
D
B
B
B
D
F
D
D
B
C
F
D
E
D
B
D
B
B
If
there
are
any,
a
new
group
will
be
created
and
then
the
new
group
will
basically
be
set
on
the
eni.
Does
that
apply
to
tag
as
well
I
like
to
think
so,
meaning
that,
as
long
as
a
group
is
not
applied
to
any
ni,
if
a
rule
references
a
set
of
tags,
I
suppose
because
it's
a
mask
then
so
as
long
as
the
group
is
not
applied
to
the
dni,
we
can
add
prefixes
to
tag.
We
can
expand
the
tag.
B
D
B
B
B
B
D
B
D
B
I
think
the
actuality
different
needs
to
be
updated
with
the
the
fact
that
a
tag
while
being
in
the
program
in
a
rule
that
is
active,
you
know
the
group
is
applied
to
an
E
and
I
that
can
be
expanded
on
the
Fly
I.
Think
that's
an
important
factor,
because
that
changes
from
the
whole
ACL
not
being
modified
while
in
use
algorithm
right.
B
G
Yeah
I
had
a
couple
of
questions
and
Uncle
tags
too.
This
is
okay,
the
first
one
just
to
make
sure
that
everybody
is
on
the
same
page,
because
it's
a
bitmap
you've
done
is
that
understanding
that
the
rule
is
the
match.
If
any
one
tag
in
the
role
matches,
we
need
one
tag
in
the
prefix
derived
from
the
prefix.
D
E
D
B
B
B
I'm
trying
to
understand,
what's
now
it's
32-bit
prefix,
to
tag
bitmap
up
to
32
tags.
It
is
the
same
32-bit
map
program
in
DCR
rule.
Is
that
correct?
Okay?
So,
basically,
when
we
go
to
a
higher
scale
than
32,
both
of
those
not
only
the
tag,
programming
on
the
prefix
and
the
SEO
rule
we
have
to
change
is.
G
G
Yeah
I
guess
the
answer
at
that
time
was
that
it
should
I
mean
so
there
were.
There
were
two
different
tags
coming
from
one
from
ten
one,
slash,
eight
and
one
from
ten
one
slash
sixteen
yeah.
So
the
packet
has
packaged
and
matched
both
and
it
should
get
both
tags
or
something
was
the
response
I
believe,
but
my
understanding
was
that
it
should
be
an
LPM
based
lookup,
which
means
it
should
only
match
one
of
those
prefixes
right.
E
G
G
G
D
G
G
F
D
G
D
J
J
D
My
point
is,
there
will
be
a
rule,
one
with
let's
say:
Tag
2
match
and
some
action
rule
two
with
let's
say
tag
one
match
and
some
action,
but
your
packet
when
it
comes
I,
think
I
mean
if
I
I'm,
just
thinking
a
hypothetical
scenario.
But
the
point
is
the
the
packet
coming
and
and
all
the
tags
Associated
must
be
derived
for
any
rule
to
to
match,
because
if
you
only
rely
on
one
super
super
tag,
I
don't
know
how
the
rules
can
can
be
made
to
match
there.
D
D
K
D
G
J
H
See
I
think
there
should
be
a
configuration
example
that
makes
this
very,
very
absurdly
clear
in
a
case
that
that
and
and
this
should
be
a
pseudocode
description-
not
a
data
plane,
implementation
requirement-
this
is
a
behavioral
requirement
of
how
matching
works.
It's
not
how
you
have
to
implement
it
right.
H
If
a
data
plane
wants
to
do
a
single
longest,
previous
match,
look
up
and
get
17
tags
for
the
single
packet
and
look
up
all
17
tags
and
some
turnary
look
up
that's
up
to
it,
but
probably
most
won't
do
that,
but
that's
effectively
how
you're
describing
what
the
control
plane
wants
to
happen.
Logically,
right
behaviorally,
it
should
be
as
if
you
did
that
yeah.
G
H
Are
many
many
ways
that
all
give
the
same
answer?
There
are
many
ways
to
do
this
in
data
plane
that
all
get
the
same
answer.
I,
don't
think
the
dash
is
going
to
require
a
data
plane
to
implement
it
one
particular
way
as
long
as
it
does
a
correct
way,
but
they
need
to
clearly
Define
what
they.
The
correct
answer
is
for
matching.
K
H
B
H
Difference
yeah
an
example
with
overlapping
prefixes,
each
assigning,
different
tags
and
rules
matching
on
you
know
different
ones
of
those
tags
and
showing
subtle
cases
that
these
kind
of
would
answer.
These
kind
of
questions
would
be
ideal,
I,
don't
I,
don't
know
if
anybody
feels
comfortable
creating
such
an
example
it's,
but
it.
H
A
E
G
A
G
D
G
J
A
G
H
H
A
A
A
A
A
A
F
All
right,
but
that
breakage
of
the
170
1755,
is
being
looked
into
to
get
to
get
it
fixed
so
that
yeah
that
the
problem
is
that
the
dash
related
changes
in
the
video
went
in
and
then
this
PR
came
in
between
and
the
one
on
top
is
the
vxlan
fix
that
we
need
for
us
to
move
to
the
pointer,
to
move
to
that
instance
of
sign
as
a
sub
module
within
Dash.
So
we're
unable
to
move
to
the
vxlan
fix
that
we
absolutely
need
for
overlay
without
the
1755
picks.
F
F
A
D
F
D
E
D
E
F
F
D
F
F
Right
sure
this
is
not
urgent,
but
you
know
to
think
about
it
at
some
time.
The
same
question
for
a
a
similar
question
for
metering
metering:
we
don't
have
the
PSI
apis
right
now,
so
the
implementation
cannot
be
completed
without
testing,
at
least
in
the
dash
test
hardness.
We
cannot
add
metering
part
without
the
sign
apis.
D
K
F
A
F
J
A
A
J
J
A
A
A
D
A
E
D
So
Sanjay
the
the
sonic
gnmi
container
has
the
capability
to
program
the
app
DBS
right
like
directly
like
so
you
can.
You
can
basically
use
gnmi
to
set
or
create
any
entries
in
in
the
app
DB
and-
and
you
just
need
to
use
that
framework
to
program
those
Dash
entries
right
so
because
there
is
no
Yang
model,
validation
or
those
things
for
for
dash.
Yet
so
it
can
just
bypass
those
things.
Yeah
I.
D
A
A
A
D
F
D
F
A
B
D
D
B
When
I
tried
to
integrate
to
sing
D
that
didn't
exist,
so
a
pull
side,
radius
and
updated
enough
of
thing
D
to
basically
be
able
to
yeah
the
size.
Reserve
yeah
the
side
changer,
to
write
the
app
Hardware
to
write
the
radius
database
so
that
it
can
be
fed
into
the
CD
and
call
my
style
layer.
So
that
worked.
E
D
C
C
B
A
A
I
think
if
I
had
anything
else
again
we'll
be
down
in
California
for
the
smart
Nick
Summit,
so
you
guys
reach
out
to
me
if
I
apparently
I
have
an
exhibit
table
I'm.
Looking
for
like
a
banner
to
go
behind
the
table,
it's
a
short
notice
that
kind
of
scrambling
a
little
bit.
But
if
you'd
like
to
chat
with
me
offline
feel
free,
you
know
my
email
address
and
my
phone
number
I'll
put
them
in
the
chat
window.
So
anyone
have
anything
else.