►
From YouTube: 2020 04 08 Sharding Working Group Sync
Description
No description was provided for this meeting.
If this is YOUR meeting, an easy way to fix this is to add a description to your video, wherever mtngs.io found it (probably YouTube).
A
All
right,
I
threw
a
couple
items
on
the
agenda
to
talk
about
today.
So
first
was
the
situs
break
down
on
costs
if
we're
gonna
go
back
to
them
with
a
smaller
number,
hopefully
of
chords
that
we're
gonna.
Try
to
support
for
production.
I
just
wanted
to
run
through
the
assumptions,
and
maybe
you
can
get
a
breakdown
of
what
the
eight
servers
that
we
have
in
production.
Now,
that's
based
off
the
number
that
Hosea
gave
so
know
that
all
the
assumptions,
the
current
costs-
and
please
correct
me
if
I'm
wrong
on
any
of
these.
B
Deleting
the
licensing
cost
I
should
know
about
per
CPU.
Is
that
based
on
our
CPU
utilization
or
allocated
cores
because
David,
because
the
the
the
actual
utilization
is
way
lower
than
the
allocated,
where
we're
over
we're
oversubscribed
Berber
we're
over
allocated
on
cores
bye,
bye,
like
8,
5,
x,
cog,
to
10x?
So
even
though
we
we
say,
we
have
a
thousand
cores
in
in
production,
we're
actually
only
using
like
10.
B
D
C
Number,
of
course,
that
we
have
right
now
we
should
actually
be
looking
also
from
the
perspective
of
what
the
number,
of
course
and
servers
and
starts
will
be
in
the
insiders
target
cluster
right.
So
absolutely
so,
for
example,
let's
say
broth
knowledge
right,
let's
say
we're:
gonna
have
a
charge
to
say
something
and
let's
say
it's
gonna
charge
is
going
to
be,
as
many
saying.
Definitely
let
number,
of
course,
actually
we'll
spread
the
work
out
monster.
C
C
B
But
if
you
look,
if
there's
load
balancing
among
the
HH
ARDS,
the
the
the
actual,
the
actual
total
usage,
the
individual
shards
will
be
smaller,
so
that
we're
not
we're
not
actually
we're,
not
we're
not
like
fully
oversubscribed
here,
because
we
can
do
like
we,
it's
not
like.
We
just
have
to
X
for
every
H
a
replica
there,
because
they're
they're
not
they're,
not
like
100%
idle,
or
at
least
they
shouldn't
be
in
a
in
architecture.
Yeah.
C
Yeah
I
know
absolutely
I'm
just
saying
that
even
taking
into
account
future
growth
and
some
considerations
about
what
is
the
minimum
number
of
portrait
art
anyway
to
be
able
to
absorb,
despite
assuming
exact
charts,
may
not
be
evenly
distributed
at
some
point.
So
basically,
let's
say
you
know
what
is
the
number
of
words
we're
gonna
have
for
sharp?
Where
is
the
number
of
charts
we're
gonna
have?
What
is
the
number
of
replicas
first
chart?
C
A
C
Yeah,
so
it's
a
little
bit
yeah
they're
under
been.
C
B
C
C
D
C
By
the
way,
this
brings
me
to
another
point,
which
is
how
scientists
is
gonna
hand,
handle
read-only
workloads
as
far
as
I
know,
there's
something
special
there,
so
they
will
not
of
load,
read
workloads
to
the
replicas
of
the
charts
that
we
will
have,
in
other
words
our
replicas,
so
it's
chart
is
gonna,
have
one
or
another
or
two
notes,
488
workers
right.
We
don't
want
to
to
fail
either
one
chart
fails,
but
those
charts
are
gonna,
be
strictly
speaking.
As
far
as
I
know,
hot
standbys,
they're
not
gonna,
be
go.
C
I,
don't
think
so
because
they
actually
don't
support
DHEA,
so
I
hae
on
the
shortest
on
you.
So
that's
what
leads
me
to
guess
that
they
are
not
gonna.
This
extra
work,
because
they're
gonna
be
not
supporting
any
worker.
They're
gonna
be
citing
sitting
idle,
yes
to
be
verified
back
based
on
my
understanding
that
they
said
just
doesn't
have
the
concept
of
Ages
or
replicas
leads
me
to
believe
that,
obviously,
the
planner
is
not
gonna,
be
too
big
a
word
about
something
it
doesn't
know
about.
C
So,
in
summary,
this
means
that
we're
going
to
need
to
dimension
the
charts
course
based
on
the
workload
of
the
master
now
and
then
multiply
this
by
two
or
by
three
four-one
replicas
or
two
replicas
for
88
purposes,
recommended
three
for
h8
purposes
of
different
souls
because
of
the
availability
of
availability
zones.
That
will
multiply
everything
by
3
and
have
2/3
of
the
course
either.
D
C
B
A
A
Something
back
to
al
girls
back
of
the
napkin
minimum
would
be
300
cores
and
if
they
go
with
allocated,
that's
still
420
a
year
120,000
here
and
that's
minimum
and
small
okay
I
have
a
call
set
up
with
them,
find
out
about
licensing
alternatives.
They
they
do
mention
a
site
license:
option
I'm,
trying
to
get
some
more
data
from
that,
and
you
have
an
email
out
to
legal
to
find
out
more
about
their
concerns
with
a
GPL
so
more
to
come
on
that
anything
else.
We
want
to
talk
about
on
cross
breakdown.
D
Do
we
have
any
idea
what
the
expected
like
return
of
investment
is
from
that
coming,
that
that
would
be
an
interesting
number
to
also
get
right.
Yeah
we're
basically
motivating
sharding
with
we
want
to
open
up
the
floodgates
particular
come.
So
surely
that
there's
going
to
be
a
return
of
investment.
A
Alvaro
and
Jose
you
may
not
have
been
on
the
first
call
when
we
talked
about
the
advantages
or
reasonings
for
implementing
sharding
was
something
that
Eric
brought
up.
Is
there
limiting
sales
on
comm
right
now,
because
they're
concerned
that
the
database
is
a
single
point
of
failure
and
that
we
are
going
to
hit
a
scalability
wall
unless
we
do
something
about
it,
and
sharding
has
been
the
chosen
solution
for
that
for
a
while.
A
A
It's
nothing
else
on
the
situs
cost
breakdown,
I
figured
we'd,
move
on
to
data
population
of
situs
cluster
that
Pat's
working
on
so
right
now,
Pat
is
using
a
script
that
was
created
by
the
quality
folks
to
populate
the
situs
cluster
and
it's
very
slow.
How
many?
What?
How
much
do
you
have
in
this
side?
This
cluster
now.
A
A
D
A
D
No,
as
that
script
was
very
basic
like
it
was
basically
identifying
the
the
sensitive
information
and
then
running
updates,
or
you
know,
masking
some
text,
columns
or
whatever,
but
it
basically
just
updates
through
those
tables.
I
would
imagine
that
takes
a
long
time.
If
we
talk
about
some,
you
know
anonymizing
five
terabytes
worth
of
data.
D
F
You
know,
build
how
we're
going
to
like
a
simple
way
to
automate,
adding
the
columns
and
all
the
populating
all
the
data
and
stuff
that
needs
to
be
set
up
to
even
get
it
to
run
in
situs
as
well,
so
that
would
I
mean
I
if
we
can
get
the
staging
data
easily.
That
would
be
good
for
me
and
I
think
to
do
something
in
the
inner
room.
F
F
F
Mean
any
like
yeah
anything
like
that
works
for
me:
I
guess:
I,
don't
particularly
care
on
the
method.
I
guess
I'm
not
really
sure
how
I
would
get
it,
though
yeah.
A
A
Solver
Oh
added
a
note
about
including
number,
of
course
for
staging
dev
environments.
That's
good
call
have
to
ask
Microsoft
about
licensing
that
too,
if
their
license
is
only
for
production
or
for
everything
we're
using.
That
would
greatly
complicate
things,
so
thank
you.
I
will
ask
next
time
we
talk.
Oh
all
right.
Any
other
topics
folks
want
to
cover
today.
D
A
An
an
email
out
to
legal
right
now
and
I
didn't
see
anything
in
my
inbox
this
morning,
so
I
gave
them
a
super
high
level
overview
of
what
we're
thinking
right.
If
we
can
use
it
just
out
of
the
box,
is
that
gonna
be
a
problem
and
then,
if
we
build
on
top
of
it,
what
do
we
need
to
be
concerned
about
so
I
expect
they'll
set
up
a
meeting
to
ask
more
questions
or
they'll
just
send
an
email
back
saying
no,
and
we
won't
have
to
talk
about
it
anymore.