►
From YouTube: Scalability weekly - 2020-03-02
A
A
B
Sure
so
it's
just
at
this
point
we've
been
discussing
on
the
epic
for
some
time
mention
and
it's
basically
a
discussion
about
what
we
like.
We
are
moving
towards
using
the
sidekick
twister
by
default,
said
psyche
and
slightly
twisted.
So
one
thing
that
I
noted
when
investigating
is
that
sidekick
twisted
doesn't
read.
These
site
could
kills
file
to
consider
using
the
weights,
so
basically
the
priorities
which
means
that
all
kills
or
will
take
like
one
priority,
meaning
that
we
are
all
flags.
So
I
started
thinking.
B
If
that
wouldn't
be
a
problem
for
people
that
today
uses
psyche
and
have
their
weights
being
considered
for
each
queue,
and
we
start
like
we
switch
to
using
cyclocross
the
reflect
priority
list
for
the
queues.
Like
imagine
that
there's
a
instance
that
uses
one
single
cube
for
like
blood
the
truth
boots
on
these
shoes
like
very
high
and
then.
B
C
B
C
B
Know
that
this
is
the
the
ones
that
we
define
on
the
today
we
define
of
the
worker
attributes
and
we
generate
this
file.
Then
we
read
this
file
afterwards,
when
we
start
this
I'd
keep
their
normal
cyclic
process
so
just
taking
that
site,
because
it
doesn't
consider
these
priorities
and
we
are,
we
are
going
to
start
using
these
by
default.
So
imagine
that
would
be
mine
might
be
a
problem
in
love
you.
Why.
D
A
Yeah
we've
we've
got
the
weights
in
in
a
file
that
isn't
read
by
default
by
psychic,
and
it's
never
been
read
in
psychic
cluster.
As
far
as
I
can
tell
I'm,
also
I'm
just
going
to
check
this,
but
so
I
put
my
gut
feeling
in
the
dock,
which
is
that,
if
you
are
relying
on
the
weights
to
get
the
higher
priority
queues
processed
on
your
self-managed
instance,
your
throughput
might
be
too
low
anyway,
because
that's
just
that
there's
so
much
queued
at
any
one
point
that
it
psychic
needs
to
make
those
decisions.
D
B
So
normally
the
usage
is
just:
let's
go
through
a
few
extra
processes
which
type
a
cluster
and
website
click.
Medical,
enable
brief
everything.
Ok,
so
basically
the
idea
is
like
configuring,
the
process
extra
process
website,
because
about
using
psyche
I
do
so
that
still
the
approach
that
we
have
today,
which
is
different,
my
background
that
we
are
I,
believe
we
are
using
a
cyclic
way.
So
by
default
right
we
have
different
configurations
for
each
node
and
it's
affected.
So.
D
D
C
A
We
so
if
we
spin
up
psychic
cluster,
we
just
start
a
bunch
of
psychic
processes
that
don't
load
this
file.
If
we
spin
up
psychic
through
omnibus
at
least
we
spin
up
a
psychic
process
that
does
load
this
file,
that's
the
difference
and
the
reason
we
can't
load
that
file
in
psychic
cluster
is
because
it
also
defines
queues
that
need
to
be
processed,
which
is
obviously
against
what
we're
trying
to.
A
D
A
No,
but
we
need
to
consider
whether
we
want
to
so
yeah
I
think
the
way
the
weights
work
is
that
if
you,
what
I
understand
is
basically
they
add
something.
To
this
so
say,
you
have
a
cube
with
a
weight
of
2
and
a
cube
with
the
weight
of
1.
The
Q,
with
the
way
of
2
appears
twice
in
the
list.
The
Q
of
the
way
of
1
appears
once
in
the
list,
and
then
we
just
pick
around
the
entry
from
that
list
to
like
get
so.
D
A
A
A
Have
is
169,
which
means
the
chance
of
a
weight.
5Q
being
picked
is
around
about
three
percent
and
the
chance
of
a
weight
1q,
which
is
the
default
being
picked,
is
around
about
half
a
percent.
So
it
is,
is
still
you
know,
I
guess
it's
still
as
a
ratio,
it's
still
a
difference,
but
like
I,
don't
know,
if
it's
it's.
D
D
Little
bit
the
yeah
but
I
think
you'll
find,
and
this
is
just
totally
that's
a
lot
of
the
cues
that
are
on
the
same
priority.
Clusters
have
got
the
same
weights
right
so
because
we
have
the
priority
clusters,
that's
kind
of
kind
of
normalizing
it
or
not,
and-
and
so
it's
probably
not
exact
but
like
generally
speaking,
that's
the
case.
A
C
A
One
thing
as
well,
there's
already
sort
of
been
discussing
in
the
epic
age
like
QA
tests,
so
they
don't
really
tell
us
a
lot,
but
they
do
tell
us
there's
something
drastically
broken
so
like,
if
they're
expecting
like
a
merge
request
to
be
merged
within
like,
however
long
and
this
change
stops
it
from
being
merged
that
quickly,
because
the
merge
was
a
high
weight
queue
and
now
it's
not.
That
would
tell
us
that.
But,
like
you
said,
it's
not
really
a
strong
proof.
The.
A
Just
not
enough
traffic,
the
only
proof
I
can
think
of
that
would
be
useful,
is
to
somehow
have
I
hidden
up
because,
like
we
could
add
the
option
to
psychic
cost
a
to
add
the
weights.
It's
just
none
of
a
faff
that
we
don't
really
need.
We
could
add
that
as
like
a
backwards
compatibility
thing,
so
we
can.
D
A
D
A
A
So,
like
you
know,
the
concurrency
should
always
be
the
same,
because
it's
the
same
two
groups,
but
they
wanted
to
have
different
processes
on
the
same
node
with
different
concurrency.
So
that
seems
like
there
might
be
some
gaps
here
that
we
need
to
address
any
way
of
like
we
as
Gilad
need
to
address
anyway.
The
other
option
for
backwards
compatibility
might
be
slightly
simpler
as
well,
though,
is,
instead
of
so,
we
back
ports.
I
could
cluster
to
call
so
clusters
available
in
core.
A
We
use
slightly
cluster
by
default,
but
we
just
add
a
switch
and
get
that
Derby.
That
says,
don't
use
psyche
cluster.
The
problem,
then,
is
like
we
do
need
to
get
to
a
point
where
we're
using
psychic
cluster.
Even
if
we're
using
queue
weights,
but
it
does
defer
the
work
of
putting
queue
weights
into
psychic
cluster
because
we
don't
think
we'll
need
it
until
later.
So
that
might
be
another
option,
because.
A
D
C
A
C
A
A
So
is
everybody
happy
as
well?
There
goes
ahead
with
that
and
like
house
some
kind
of
fallback
option
to
not
use
psychic
cluster
and
we
give
that
to
support
somehow
we
probably
do
want
to
document
it,
but
carefully
as
like.
This
is
not
going
to
stick
around
for
long
like
you
know,
we've
got
yet
lab
13.0
coming
up
so
I
guess
we
could
say
like
after
13.0
we
can
remove
the.
If
we're
quick
enough,
I
think.
B
A
A
Weights
on
the
command
line,
so
we
can
pass
the
file
we.
Actually
we
don't
need
to
pass
the
file
because
they're
also
in
or
or
we
could
pop,
we
can
pass
the
weights
on
the
command
line.
I,
just
don't
want
to
do
that
if
we
don't
need
to
because
it'll
be
needlessly
like
ideally,
I
would
like
to
just
get
rid
of
the
weights.
A
D
D
A
And
I
don't
see
those
I
think
that's
a
good
point.
I,
don't
see
those
I,
don't
know
the
founder
is
trying
to
talk
or
not
I
don't
see.
Those
is
exclusive.
I
see
that
we
could
remove
the
explicit
weights
that
we've
set
and
then
reconsidering
consider
reintroducing
weights
in
some
form
in
future
based
on
the
attributes,
rather
than
based
on
something
that
people
don't
really
remember
to
set
in
future.
But.
C
A
And
actually,
maintaining
this
escape
hatch.
Isn't
that
big
of
a
deal?
Because
it's
just
the
file
in
the
repo
which
we
could
just
stop
updating
or
like
minimally
updates
and
the
psychic
entry
point
itself,
which
we
need
to
keep,
because
I
could
cost
the
users
it
in?
Because
it's
part
of
the
gem
so
who.
A
A
D
It's
worth
looking
at
the
at
the
dashboards
that
we
should
put
the
product
so
that
people
will
know
if
their
sidekick
keys
are
saturated.
A
cover
see,
we've
got
pretty
good,
metrics
I.
Think
it's
not
calm,
but
I
don't
know.
What's
the
product
has
got
because,
obviously,
if
saturated
then
this
will
be
off,
workers
are
saturated.
This
will
be
a
problem,
but
if
the
workers
on
such
a
reason
it's
not
a
problem
but
having
visibility
into
that's
kind
of
assault.
B
B
D
That's
the
way
that
we
did
with
Pima
rights
and
it
worked.
It
worked
pretty
well
there
because
their
own
on
Pumas
been
running
well.
I,
Brennan
GDK
has
been
running
payment
for
the
last
year,
or
so
it
could
work
pretty
well
there
and
the,
and
it
definitely
brought
up
a
few
issues
which
was
pretty
useful
as
well.
Yeah
he'll.
A
B
B
B
A
A
D
C
D
So,
like
obviously
this,
this
I
don't
know
it's
kind
of
more
like
are.
These
are
things
that
we
want
to
work
on?
You
know,
there's
there's
always
a
lot
of
talk
about
Redis
and
some
of
those
things
make
a
lot
of
sense,
especially
I.
Think
that
s
members
thing
kind
of
every
time
we
look
at
Redis
gives
me
the
creeps.
I
have.
A
A
Someone
about
that
earlier,
like
about
how
we
kind
of
pretty
much
need
D
normalization,
because
we
need
like
the
just
like
just
shoved.
An
array
in
a
red
is
key
for
one.
We
need
to
get
everything
and
then
we
need
the
set
for
when
we
need
to
check
membership
because
I,
you
know
the
other.
Two
options
are
bad,
but
also
having
the
same
thing
stored
in
two
places,
and
two
data
structures
is
bad,
so
yeah.
D
I
mean
this:
is
this
is
like
a
kind
of
complicated
on
serpent-like
I
would
say:
do
we
really
need
to
always
load
20,000
tags
like
there's,
no
page,
that
shows
you
20,000
tags
right,
so
the
places
where
we're
doing
it
is
scan
and
loading
all
20,000
like.
Why
are
we
doing
that,
because
we're
never
gonna
show
all
of
them
and.
A
A
C
A
A
Yeah
I
mean
I,
think
I.
Think
the
reddest
thing
is
definitely
something
we
need
to
address
right
now.
It
looks
like
quite
a
small
project
but
I'm
sure
the
psychic
things
look
like
quite
a
small
project
when
you
started
it
and
then
it
just
sort
of
ballooned
nice
things
tend
to
do.
I,
I'm
happy
with
that
as
the
top
priority,
because
I
I'm
not
too
worried
about
sidekick,
like
obviously
it
does
show
up
a
bit
in
the
capacity
fine
dashboards.
But
basically
it's
you
know
seems.
C
A
A
And
cash
psychic
cash
seems
fine
and
also
it's
a
cash
like
you
know
we
shouldn't
be
spending
too
much
time,
but
psychic
persistent
or
seem
a
bit
of
a
source
of
issues,
especially
with
the
sessions
thing
you
mentioned.
The
one
thing
I
wanted
to
ask
was
like
and
I,
don't
really
know
how
to
frame
this
in
terms
of
project,
because
I
think
the
way
it
split
up
in
terms
of
projects
is
good
right
now,
but
I
would
like
us
to
spend
some
more
time
working
on
internal,
tooling
and
I.
A
D
A
So
I'm
not
talking
about
working
on
new
stuff
like
in
addition
to
like
once
in
the
dark
I'm
talking
about
like
rearranging
on
the
dark,
because
I
want
to
work
that
stuff's
more
fun
but
also
like
you
know,
we've
got
stackprof.
We've
talked
about
our
beast
by
like
having
by
default
and
stuff
like
that,
like
making
those
sorts
of
things
available
to
people
easily
like
could
provide
some
big
wins,
but
mostly
during
incidents.
Yes,
so.
D
I
I'll
just
give
you
a
bit
of
background.
It's
kind
of
difficult
without
meirin
being
here
as
well,
but
I
didn't
give
any
priority.
There
were
just
kind
of
four
themes:
another
characterization
really
came
from
from
Marin
and
and
I
think
he
felt
that
that
was
stuff
that
we
would
mostly
do
in
between
the
the
other
work.
But
if
we
like,
if
you
feel
like
you
know,
we
can,
we
can
kind
of
push
for
a
little
bit
more.
The
out
of
those
things,
particularly
to
me,
like
I,
just
there's
so
much
stuff.
A
A
A
P1
or
p2
things
on
the
current
board.
If
you
run
out
of
p1
or
p2
things,
then
you
can
take
an
observatory
thing
or
take
a
thing
from
the
current
board,
but
some
yeah
I
guess
this
will
need
to
wait
for
Marin
to
get
back,
but
I
would
love
to
spend
more
time
working
on
this
stuff,
I
think
it's
fun.
That
doesn't
necessarily
mean
we
should
spend
more
time
working
on
it
because
I
think
it's
fun.
Yeah.
D
Like
3
a.m.
for
him,
I
mean
you
know,
but
the
like
I
think
there's
a
lot
of
stuff
down
there
like
the
rollout
of
continues
profiling
like
that,
would
be
something
that
I
would
imagine
he
would
earn
and
the
slow
log
stuff
and
then
kind
of
on
the
other
three
priorities.
I,
don't
think
is
other
than
the
decision
state,
raters
I,
don't
think.
There's
like
a
lot
of
things
on
there
for
him
and
I.
Think
that's
one
of
the
things
that
I'm
a
little
bit
worried
about
at
the
moment
is
that
is
not
enough.
A
A
bit
but
I
also
think
that's
part
of
the
point
of
having
him
on
the
team
because,
like
you
know
with
the
four
of
us,
like
we
generating
stuff
that
they've
done
by
like
a
back
engineer
or
whatever,
because
we
like,
we
know
stuff
that
you
can
see.
That
comes
up
that
we
think
should
be
worked
on
and
I.
Think
Craig
will
do
the
same
from
the
sre
side.
So
great.
If
you're
watching
this
I
guess,
oh
yeah.
B
C
C
D
D
A
D
A
A
A
A
Great
I
just
had
a
quick
question:
we've
got
the
scalability
office
hours
tonight,
I,
don't
know
if
anybody
was
planning
on
joining.
I
probably
can
but
like
I
said
before.
It's
not
a
super
convenient
time
for
me,
but
it
would
be
good
if
somebody
did
so
I
can,
if
nobody
else
can,
but
maybe
there
should
be
more
than
one
of
us
anyway.
Yeah.