►
From YouTube: SIG - Performance and scale 2022-02-24
Description
Meeting Notes:
https://docs.google.com/document/d/1d_b2o05FfBG37VwlC2Z1ZArnT9-_AEJoQTe7iKaQZ6I/edit#heading=h.tybh
A
Okay,
welcome
to
six
scales,
so
it's
february
24th
and
yourself
as
attendee,
please,
okay
agenda
for
today.
The
only
thing
I
have
here
is
review
some
prs,
and
so
I
I
don't
know
if
we
use
the
whole
time,
but
let's
see
where
it
takes
us.
So
first
one
this
one's
been
open
for
a
while
marcelo.
I
know
you
reviewed
this.
A
I
don't
know
where
how
people
feel
about
this
like
in
terms
of
where
we
are,
if
it's
something
that
we
do
want
to
merge
or
something
we
want
to
wait
and
just
continue
to
make
improvements
on
as
we
progress
through
and
build
some
more
slos
and
further
define
some
of
these
fields,
I
don't
know,
doesn't
have
an
opinion.
A
B
A
Okay,
all
right,
I
can
that's
david
again.
Take
a
look.
Okay,
that's
fine!
All
right!
Okay,
that's
that!
So
the
second
item
marcelo
I
had
I've
done
some
work
on
actually
the
load
generator.
I
want
to
show
you
so
this
is.
This
is
the
pr
that
I
put
together.
This
was
I
put
this
together
kind
of
the
end
of
last
week.
A
I
I
kind
of
wanted
to
just
see
how
this
looks
when
I
just
ended
up
writing
a
bunch
of
things,
but
all
I
did
was
I
kind
of
refactored
the
code
I
wanted
to.
A
I
wanted
to
to
change
a
few
things
kind
of
re-center
the
code
around
a
few
ideas
where,
instead
of
having
like
some
some
apis
or
abstractions
for
just
like
a
few
different
types
of
a
few
different
calls
like,
if
you,
I
think,
I
think,
maybe
the
way
it
classifies
like
we,
I
think
the
way
it
was
is
we
basically
had
we
had
a
lot
of
configuration.
A
We
had
a
lot
of
flexibility.
I
kind
of
wanted
to
focus
that
on
defining
like
different
workload
types,
so
we
like
talked
about
study,
states
and
burst
testing,
and
so
I
wanted
to
center
the
code
more
towards
those
two
types.
So
what
I
mainly
the
biggest
highlight
is
what
I
did
is
I
I
created
I
defined
an
api
and
basically
created
an
interface
for
what
will
be
our
job
types
and
it's
pretty
simple.
It's
it's
just
this
like
we
have
a
load,
generator
define
it.
A
As
you
know
these
functions,
and
this
is
all
it
does
and
then
we'll
we
can
fill
in
all
the
separate
nuances
for
the
types
of
jobs
and
we
just
call
them
based
on
whatever
user,
but
that's
kind
of
what
I
was
thinking.
A
That
would
maybe
make
this
a
little
more
flexible,
and
then
I
added
a
steady
state
job
to
it.
Based
on
these
these
interfaces,
I
don't
know
what
do
you
think
marcelo?
I
don't
know
if
you
had
a
chance
to
take
a
look
at
this.
A
Okay,
yeah,
I
think
kind
of
the
where,
like
I
would
like
to
go
with
this,
is,
is
that
if
we
can,
I
like
I,
I
found
when
I
went
through
this
there's
a
lot
of
similarities
that
I
that
we
can
actually
do
between
births
and
study.
State
like,
for
example,
they
create
in
a
in
study
state,
is
like
the
exact
same
in
the
creating
workload
right.
It's
basically
a
burst
to
create.
A
A
A
lot
is
like
over
down
in
like
the
steady
state
part-
and
this
is
where
I
saw
your
configuration
like
you-
had
all
this
configuration
for
doing
intervals
and
stuff,
and
that's
where
I
think
I
want
to
that's
where
I
want
to
expand
to
do
to
add
that
functionality
back
wherever
it
is
here.
Basically
just
expand.
Oh
yeah
down
here
like
expand
this.
A
These
interfaces
to
to
do
that
kind
of
interval.
Work
because
there's
so
many
types
when
I
was
going
through
this
there's
so
many
types
of
studies
they
test
right,
like
all
I
did
is
I
do
a
really
quick
replacement
of
a
set
number
of
vms
and
but
there's
so
many
different
ways
you
can
solve
this,
like
you
can
do
a
slow
build
up.
A
You
can
do
a
like
a
you,
can
have
a
regulated,
delete,
there's
so
many
things
you
can
do,
but
I
think
they
could
all
be
sort
of
iterations
of
this
steady
state
job
like
we
just
have
one
job.
We
kind
of
define
different
nuances
for
it.
A
Okay,
I
showed
a
picture
of
it
like
so
here's
the
new
config
marcelo,
it's
instead
of
so
mainly
it's
just
I've
turned
count
and
timeout
timeout
for
steady
state
always
runs
to
the
timeout
limits.
It's
just
the
time
that
you
know
the
tesla
run
so,
and
this
just
means
that
we're
going
to
have
five
vms.
I
only
do
five
in
my
my
kubernetes
and
docker
for
some
reason,
and
then
I
do
a
set
of
turn
of
two
and
then
yeah
these
vms.
A
A
A
All
right:
well,
when
you
have
a
chance,
have
a
review
and
see
how
much
you
think
this
is,
I
think
we
can.
I
think
we
can
go
a
few
different
places
with
it,
but
yeah.
Let
me
know
what
you
think.
B
Yeah,
okay,
I
will
I
will
do
that,
but
it
looks
interesting
thanks.
A
A
Yeah,
we
now
have
the
thresholds
there
and
then
I
did
and
then
I
think
I
I
think
I
told
you
this
already
the-
and
I
also
mentioned
just
at
the
at
the
the
summit,
which
is
that
we
can
run
the
the
job.
Now
we
can
run
the
the
pre-submit
job
and
I
fared
if
I
showed
this
in
our
last
call.
But
all
you
need
to
do
now
is
run
this
on
any
of
your
pr's
and
it
will
it'll
run
our
periodic
chest
against
your
code.
B
B
A
Sometimes
yeah
yeah,
I
mean,
I
think
I
think
I
definitely
agree
with
you
like.
I
think
I
think
one
of
the
things
that
it's
going
to
be
really
impactful
for
everybody
is
not
just
this
job
but
also
like
like
when
we
have
this
tool
like
we
have
our
tool
that
we
can
hand
to
people
that
we
know
we
talked
about
like
being
able
to
do
study,
state
and
the
audit
tool.
I
think
combining
those
two
things
was
just
going
to
become
very
powerful
when
we
kind
of
reached
that
point.
A
I
think
this
is
a
good
like
what
we
have
now
is
good,
and
this
is
another
stepping
stone
into
that
direction.
Now,
with
the
steady
state
support,
I
think
this
will
give
us.
Yes
a
lot
of
interesting
things.
I
think
because
this,
like
I
mean
we
could
like
this
dashboard,
is
very
powerful
because
it
shows
when
we
combine
this
plus
the
audit
tool,
like
with
the
thresholds
yeah
and
like
a
larger,
not
like
a
5
vm,
but
like
100
vm
count,
it
would
be
super
cool
to
see.
A
Oh,
that
reminds
me
did
you.
You
have
you're
using
you're.
You
did
you
merge
your
code
that
that
actually
uses
the
that
uses
this
right.
Like
you,
I
saw
I
thought
I
saw
you
have
you
have
you
have
the
your
jobs
right
that
yeah
the
auto
two?
A
Yes,
okay,
so
I
changed,
and
here
are
the
examples
so
that
they
fit
this
format,
but
I
don't
know
if
I
need
to
do
any
other
changes
so
that
it
just
works
but
like,
for
example,
what
I
did
was
I
so
you
were
you.
I
saw
you're
using
the
examples
in
here
like
this
one.
I
changed
these
so
that
they
fit
the
the
schema
change.
B
Yeah,
I'm
using
just
specific
names,
so
I
don't
remember
now
it
was
converts
instead
convert
then
a
range
something
like
that
before
isn't
it.
So
I
don't
know
if
you
can
keep
the
old,
the
old
files
or
if
it's
not
compatible
anymore,.
A
It's
not
yeah,
I
I
it's
not
compatible
when
I,
so
where
would
I
go
to
change
it
like?
I
think
I've
got,
I
would
go
to
your
pr
like
you
can
still
do
the
same
things.
It's
just.
What
I
did
is
I've
broken
the
test
down
to
smaller
chunks.
So,
instead
of
doing
like
a
one
one,
two
three
four
five
test,
we
do
a
one
and
then
we
can,
after
that
run
a
two
or
after
that
one
of
three.
B
B
B
Load
generator-
just
oh
sorry,
the
load
generator.
So
if,
if
it
has
like
some
in
new
parameter
to
run
it
or
it's
just
the
file,
if
it's
just
the
file,
you
don't
need
to
change.
You
just
need
to
change
the
file.
A
Yeah,
I
just
it's
just
the
format
that
changed
of
the
the
input
file
the
workload
file
changed
the
format
change
so
yeah.
I
think.
Maybe
you
just
need
to
you-
have
a
script
right
like
there's
a
script
somewhere,
you
run
in
keyword.
I
think
right
in,
like
certain
tools.
B
It's
ultimate
yeah
ultim
yeah,
it's
hack,
and
it's
like
it's
like
a
bare
scale.
B
I'm
changing
some
of
this
because
you
can
see
the
perfscale
workload
environment
variable.
A
B
The
line
29
yeah-
I
changed
that
when
I
run
the
jobs.
Okay,
so
that's
why
I
change.
I
run
different
jobs,
so
I
have
this
warm
up,
but
you
change
off
the
names
so
yeah,
it's
fine,
so
we
just
need
to
update
the
name
of
the
files.
I
think
then
the
rest
supposedly
should
work
and
then
this
one.
A
You
so
what
does
this
go
to
per
sql?
This
goes
right
to
okay,
yeah.
I
write
the
workload
I
use
the
same.
Okay,
so
then
yeah
this.
This
is
the
variable.
I'll
also
have
to
change
that.
You
update
to
I'll
what
I
can.
So
I'm
assuming
you,
you
put
you
you
reference
other
files
that
are
merged
into
the
repo,
or
do
you
like,
create
files
locally
and
reference
them.
B
A
Yeah,
that's
actually
where
I
want
to
go
with
this
as
I
this
is.
This
is
exactly
what
I
want
to
do,
so
I
want
to
take
this
and
actually
go
internally
and
have
do
some
testing
on
our
larger
zones
get
rid
of
some
together.
Some
data
and
report
back.
A
Yeah,
so
you
said:
you've
got
you.
Do
the
I'll
probably
do
the
same
test?
You
did
you
do
like
the
100
200,
whatever
it
is
up
to
500
yeah
and
I
can
get
some
results,
okay
and
then
I'll
probably
do
the
same
with
steadysave.
That's
the
one!
I'm
really
really
interested
in
seeing
not
just
the
verse
with
the
study
today.
One
is
I'm
really
curious
to
see
how
like
how
it
handles.
B
Yeah,
I
have
also
implementation
for
steady
state
for
cooper.
I
didn't
submit
a
pr
there
because
I'm
waiting
some
other
years
to
you
know
get
merged,
but
it's
it's
a
little
bit
different
implementation.
So
it
might
be
also
be
interesting.
You
know
sure
yeah.
We
can
look.
A
At
it,
yeah
sounds
good,
okay,
cool
all
right
I'll,
take
a
look
at
that
and
make
sure
that's
compatible.
Okay,
cool,
okay!
All
right,
I
think
that's!
That
is
all.
Do
you
have
any?
Do
you
guys
have
anything
else
that
you
want
to
cover
for
today?
Do
you
want
to
review
kind
of?
I
think
that
the
summit
that
summit
went
pretty
well.
I
thought
we
had
some
pretty
good
representation
and
it
seemed
like
there
was
some
good
interest
in.
B
B
B
A
A
At
250
and
then
five
minutes
to
p95.
A
Okay,
okay,
really
cool!
This
is
that's
interesting.
B
A
B
C
A
B
Went
back
to
the
scale,
one,
the
one
that
this
looks
like?
Oh,
yes,
it's
it's!
It's
run.
Three
kind
of
you
know
three,
three,
three
tests,
it's
great
like
100
and
two
hundreds
or
two
hundred
and
then
three
hundred
and
then
four
hundred
and
the
total.
B
B
A
B
A
B
Yeah,
so
when
you
create
a
job,
just
you
know
keep
in
mind
that
you
cannot
run
jobs
in
parallel.
You
know
in
the
cluster
and
in
the,
for
example,
if
it's
include
the
steady
state
you
need
to
ch,
you
know
check
the
pro
you
know
in
the
chrome
table
and
find
that
you
know
some
interval
like
three
hours
interval
between
tests,
something
like
that.
Then
you
can
run
tests.
B
We
cannot
have
too
much
test
because
it
needs
to
fit
in
24
hours,
but
we
can
definitely
run
more
tests
per
day,
something
like
that
or
or
even
change.
Maybe
we
don't
need
to
run
daily
tasks.
It
can
be
every
two
days.
Something
like
that.
You
know
I
don't
know,
but
it
depends
on
the
test
that
we
want
to
run
yeah.
A
Okay,
well
I'll
make
the
changes
in
here
for
whatever
in
here
for
the
for
the
so
that
still
works,
so
it's
compatible,
and
then
I
mean
I
think,
I'll
first,
what
I'll
do
is
I'm
going
to
run
this
internally,
just
the
steady
state,
because
I
I
still
think
there's
some
improvements
to
make
around
this.
I
think
there's
like
a
lot
of
nuances
to
this.
So,
let's
see
where
how
far
I
can
go
with
it,
based
on
the
testing
before
we
can
have
a
job
for
it.
A
A
All
right
cool
yeah
that
was
pretty
cool.
Oh,
how?
How
often
does
this
run
marcelo
this
dedicated
daily.