►
From YouTube: SONiC DASH Workgroup Community Meeting Oct 5 2022
Description
PR225 for community to try for Chris, possibly merge by Friday
OCP: Taras presenting SAI testing catch-up
Scale #'s updated in documentation
A
A
B
I'll
just
talk
a
little
bit
about
progress,
working
on
an
issue
that
Hannah
filed
when
he
was
trying
to
write
an
IPv6
test.
He
was
adding
IPv6
the
seal
support
and
now
providing
a
unit
test
for
that
and
the
test
gets
stuck
when
it
goes
from
test
one.
The
tested,
because
I've
been
working
on
that
for
several
hours-
and
you
know
I,
don't
have
resolution
yet,
but
I'm
working
on
it
and
I'll
try
to
help
more
debugging
details
to
the
issue.
You
know
as
I
make
progress.
Yeah.
B
It
it's
nothing
you're
doing
wrong.
Believe
me,
it's
in
fact
you
can
take
the
existing
ipv4
test
and
just
duplicate
it
in
another
file
with
a
different
name
and
the
same
bug
happens
when
you
switch
from
test
one
to
test
two.
It
just
hangs
and
it
has
something
to
do
with
the
server
Communications
between
Thrift
and
grpc.
You
know
it's.
It's
I'm
having
to
drill
in
pretty
deep.
B
There's
nothing
to
present
yeah
okay,
I
will
I
will
upload,
let's
say
some
debugging
snapshots
to
the
issue.
Hopefully
you
know
this
morning.
If
I
get
time
for
people
to
help
like,
for
example,
Marion
having
to
do
a
P4,
runtime
I,
don't
know
yet
I'm
making
some
progress.
Okay,.
C
Right
and
as
you
know,
Chris
you
know
I
validated
the
V6
code
by
essentially
removing
the
V4,
you
know
file,
and
then
you
know
running
it
and
then,
since
that
V6
was
the
only
test
in
the
in
the
directory.
It
ran
fine
and
you
know,
but
however,
I
could
not
really
check
in
or
commit
to
it,
because
the
fact
that
it
would
have
basically
broken
everything
else.
You
know
had
it
I've
done
it
so
on.
C
B
That's
that's
why
I'm
working
on
it
right
now,
so
you've
done
your
part.
Now
other
people
have
to
pitch
in
and
figure
out
where
it's
stuck
so
yeah
I've
run
your
test
and
iStation
and
the
link
the
the
branch
Who
provided
helps.
B
It's
may
just
have
to
do
with
the
PTF
framework
and
the
way
it's
being
used
right
now
may
have
nothing
to
do
with
how
it's
nothing
to
do
with
the
test
itself.
So
I'll
provide
more
details
and
ask
for
help
from
the
appropriate
experts.
D
Debugging
this
issue,
but
yeah
I,
don't
seem
to
have
the
full
conclusion.
I
don't
know
if
he
has
joined
today,
then
we
can
talk
about
it,
but.
B
D
B
I
would
just
mention
that
I
know
I,
don't
know
if
anyone
else
has
tried,
but
I
had
the
pull
request
where
I
use
Azure
ACR
for
for
Dockers.
Instead
of
my
Docker
hub
and
I
wonder
if
anyone
else
has
tried
to
run
that
branch
and
verify
it
works
for
them,
because
it's
it's
been
in
the
queue
for
a
couple
weeks
now
and
I
would
eventually
like
to
get
it
merged
in.
B
C
B
Containers
right
so
yeah
that
was
that
wasn't
that
was
definitely
an
interim,
so
I
guess
what
I'll
say
is
unless
there's
any
feedback
to
the
contrary.
I
would
like
to
merge
this.
Let's
say
Friday
unless
there's
any
dissension,
because
I
don't
want
to
think
the
language
there's
other
things,
that'll
have
to
change
once
that's
put
in
so
I
want
to
kind
of
unblock
the
queue.
B
Okay,
anyone
is
is
Keen
interest
in
trying.
All
you
have
to
do
is
clone
it
and
do
make
all
and
try
it.
That
would
be
enough,
and
you
know
in
running
it's
I've
run
it
dozens
of
times.
It's
always
nice
to
have
an
independent
person.
Try
it
just
to
make
sure
there's
not
an
environment
dependency
that
I
I
assumed.
C
Sounds
good
yeah
one
thing
that
Chris,
you
know
you
know.
In
our
conversation
we
talked
about.
There
were
two
workflows
right.
The
second
one
was
somebody
else
were
to
make
changes
to
the
container
and
want
to
check
in
and
so
forth
that
I
have
not
given
it
a
shot.
So
I'm
not
sure
if,
if
you
want
somebody
to
take
a
look
at
it
or
you
think
that
you
know
there's
already
enough,
testing
has
been
done
to
that.
You
know.
You
know
that
it's
kosher.
B
Yeah
thanks
there
are
a
documented
three
different
workflows
in
my
readme
and
it's
mainly
a
matter
of
someone
trying
it
out
so
that
I'm,
not
the
only
person
who
maintains
the
docker
files
would
be
good
for
someone
to
to
go
through
that
yeah.
That
would
be
there's
a
readme
in
there.
B
B
B
Yeah
towards
the
bottom,
the
very
end.
There's
all
these
workflows
right.
It's
kind
of
crazy
looking,
but
if
anyone
wants
to
try
and
try
to
let's
say
just
make
a
test,
Docker
file
and
practice
these
workflows,
they
have
to
work
with
them.
Otherwise,
the
first
time
someone
actually
wants
to
do
it,
we'll
just
we'll
just
do
it
then
there's
enough
I'm
feel
confident.
This
is
95
correct.
B
B
So
yeah,
oh
there's
no
urgency
really
in
someone
trying
all
those
things
I
think
it's
just
good
practice,
but
people
are
busy,
so
it
can
always
be
done
on
demand,
for
example,
whoever's
working
on
the
bmb2
like
Marian
or
his
colleagues
or
other
people.
If
they're
working
on
the
new
p4c
compiler
changes
they're
going
to
eventually
have
to
run
into
this
and
we
can
work
together,
then
there's
no.
E
B
E
E
No,
it
just
was
the
name
of
the
topic
that
I've
talked
on
before
the
meeting
was
started.
So
this
is
what
where
we
are
going
to
discuss
the
site
testing
and
the
dash
testing
as
a
subset
of
it
on
ocp.
A
Okay,
okay
got
it.
Yeah
I'll
have
to
watch
that
session
for
sure.
A
And
did
you
guys
have
any
other
PRS?
You
wanted
to
look
at
me.
A
Okay-
and
that
was
features
needed
in
bmv2.
That
kind
of
thing
that
we
talk
about
in
behavioral
model.
F
Can
I
look
at
the
update
for
the
skill
number?
Please?
F
E
F
Yeah
I
have
nothing
to
say,
but
just
everyone
to
be
aware
of
these
scale
numbers
things
are
hero
numbers
or
like
a
year
old,
so
those
got
synchronized
with
some
other
page
I
think
it
was
the
hld
Dash
hld
page
that
already
had
this
number,
so
the
Hero
Test
Got,
updated
too.
C
Sorry,
what
are
these
numbers
Christina.
A
When
we
had
the
original
documentation
we
had
in
the
transforms
document,
we
had
some
scale
numbers,
some
initial
swing
at
the
scale
numbers,
and
they
were
located
here
inside
of
the
packet
transforms
document
here
at
the
very
top
one
of
the
initial
documents
we
were
doing
scale
per
dpu
here
and
then
we
replicated
these
numbers
over
into
the
this
document,
which
was
the
testing
the
Hero
Test,
and
so
this
is
what
we
updated.
Was
these
two.
G
A
C
F
Present
in
I
think
Sonic
Dash
hld.md
file-
it's
just
as
a
hero
test,
did
not
have
these
numbers
in
it.
So
basically,
if
the
numbers
got
synchronized
across
the
pages,
so
there
is
no
one
page
says
the
number
and
the
other
Pages.
The
other
number
right
there
already
they're
present
for
a
couple
of
months
now.
G
D
That
seems
like
a
lot
of
routes.
It'll
be
good
to
have
like
a
little
bit
of
breakdown
on
IPv6
and
ipv4
as
well.
Yeah
I
think
that
will
help
a
lot.
G
That's
actually
a
good
idea.
Good
thing
you
take
that
down.
That
is
something
that
we
should
be
working
with.
Michael
I
think
that
not
a
lot
of
people
use
IPv6
on
the
inner,
but
we
should
set
the
expectations
as
if
they
do,
we
have
a
couple
of
customers
that
use
that
large
Enterprise
customers,
so
maybe
Michael
could
help
us
quantify
that
a
little
bit
better
Christina,
that's
a
good
good
point,
just
the
IPv6-
and
this
is
the
overlay.
G
The
external
you
know,
doesn't
matter
that
much
like
the
other,
but
the
overlay
being
IPv6
will
get
a
different
number
and
she's
right
about
that.
So
we
will
add
that.
B
G
In
general,
most
of
the
overlays
ipv4,
but
we'll
we'll
we'll
look
at
that,
let's
figure
out
what
the
number
needs
to
be
first
and
then
we'll.
Maybe
I
don't
want
to
make
it
so
complex.
But
maybe
we
need
a
matrix.
G
A
We've
had
meetings,
we
have
these
here.
B
E
G
And
then,
eventually,
Chris
I
think
that
we're
going
to
end
up
with
like
the
hero
test
for,
like
you,
know,
performance
and
then
we're
going
to
have
functional
testing,
which
is
you're
what
you're
working
on
with
your
entire
test
apparatus
and
then
we're
going
to
have
in
Cloud
testing
and
that
that
is
becoming
really
really
interesting,
especially
with
what
I've
seen
from
last
week
with
the
site
perk
and
what
it
can
do,
and
we
are
also
you
know
internally,
we
are
developing
tests
with
T-Rex,
so
T-Rex
is
open
source.
G
So
we
like
that,
that's
good
we're.
We
are
actually
developing
tests
that
can
be
replicated
by
anybody
in
any
cloud
and
pointing
at
devices
that
look
like
this.
So
that's
a
whole
area
that
over
time
will
get
developed,
but
my
I
kind
of
envisioned.
You
know
hero
tests,
you
know
that
runs
in.
G
We
look
for
regressions,
so
the
stress,
I
I,
see
functional
tests
for
regression
of
functionality
and
then
I
see
that
we
will
share
our
tests
that
we
developed
for
the
cloud
that
can,
you
know,
run
in
any
Cloud
essentially
to
test.
You
know
different
applications
with
different
loads.
G
As
a
third
way
of
testing
and
and
that
will
be
driven
by
T-Rex,
which
we're
writing
ourselves
and
then
potentially
cyber
where
people
are
interested
in
like
using
a
getting
a
lot
of
different
types
of
of
traffic,
pointing
at
devices
like
this
and
seeing
if
we
can
some
somehow
break
the
device
and
so
I
see
a
lot
of
work,
there
we're
already
doing
a
lot
of
work
with
the
T-Rex
just
to
get
it
going.
G
You
know
we
have
been
able
to
get
millions
of
connections
per
second
already
in
a
real
cloud
and
and
we're
working
on
that.
But
this
is
a
whole
in
very
interesting
area
that
will
so
I
I
said
from
the
beginning.
I
want
Dash
to
be
very,
very
test.
G
Oriented
quality,
oriented
and
I
think
that
that
just
means
that
we
will
be
attacking
this
thing
from
multiple
angles
and
and
trying
to
create
as
many
different
types
of
traffic
patterns
as
possible
and
make
sure
that
when
these
things
go
out
that
they're
going
out
with
quality
but
we'll
share,
you
know
as
we
go
along
and
we
build
an
automated
test,
we'll
be
sharing
that
at
least
on
how
you
can
build
it.
B
We
make
Also
regarding
T-Rex,
we
we
did
do
over
over
a
year
ago,
a
year
ago,
last
summer
we
created
a
Snappy
client
library
for
T-Rex,
but
it
was
only
at
that
time
implementing
L2
and
L3
streams,
not
statefulness
right,
but
we
do
have
a
goal
of
adding
stateful
connections
to
Snappy
over
time
and
one
of
the
goals
is,
you
know
you
can
use
a
common
API,
regardless
of
the
underlying
traffic
generator
engine,
whether
it's
hcsc
or
T-rex
or
Hardware,
based
so
I
pasted
a
link
in
there
Gerald
that
whichever
team
is
using
T-Rex,
they
might
want
to
take
a
look
at
that.
B
G
Yeah
we
we
definitely
can.
This
has
to
be
like
I
I,
think
this
benefits
everybody,
whether
building
Hardware
or
whether
they're
working
on
the
simulator
or
whatever
having
all
these
different
tests
available
is
is
a
great
thing
and
the
T-Rex
has
made
there
there.
It's
not
the
same
as
it
was
a
year
ago,
like
a
year
ago.
You
couldn't
do
these
tests
with
T-Rex
anyways.
It
wasn't
fast
enough
to
I
think
it
was
limited
to
one
core
or
something
it's
now
all
DP
DK
enabled
multi-core
like
it's.
G
It's
actually
can
generate
the
traffic
patterns
that
we
want
at
the
rate
we
want.
The
only
thing
is,
of
course,
we
have
to
point
serious
Technologies
towards
it,
so
I
can
get
to
those
rates,
but
the
software
itself
is
not
limited
anymore,
and
so
we've
successfully
been
able
to
deploy
in
our
Network
and
test.
You
know
first
product
using
the
using
the
T-Rex
reach
the
numbers
now
that
it's
dpdk
enabled
so
we're
we're
pretty
impressed
with
that.