►
From YouTube: 2020 04 06 Database Sharding Working Group
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).
B
B
A
Fix
any
questions
on
the
post-crisis
11,
all
right,
so
couple
items
came
up
last
week,
while
we
were
working
through
sharding
and
looking
at
situs.
So
it's
one
of
validate
assumptions.
So
everybody
that's
in
this
working
groups
on
the
missing
page,
sorry,
so
we're
working
on
sharding
implementation
for
dot-com.
Only.
This
is
not
something,
at
least
in
the
initial
iteration
that
we
would
shift
to
self-managed.
A
C
I
think
that's
reasonable,
as
I
mentioned
last
week.
I'm
still
also,
this
is
different
story.
I.
Don't
phosphorous
actually
distributed
this
product
right,
so
I
started
kind
of
like
a
dependency.
You
see,
you
need
a
phosphorous
and
then,
if,
if
that
would
be
distributed
as
a
product
like
that,
like
phosphorus
or
dependency,
like
give
an
example,
then
you
could
do
anything
you
want
absolutely
transparently.
So
thank
you
for
some
discussion.
A
That's
my
god
and
we've
dropped
support
for
my
sequel,
so
it's
basically
Leon
support
Postgres
at
this
point
in
time,
okay
and
then
I
just
wanted
to
do
a
quick
clarification
on
situs.
There
were
some
concerns
that
you
know
if
situs
ultimately
doesn't
work
out.
Have
we
wasted
a
lot
of
time
and
effort
on
it,
and
the
answer
is
yeah
I
mean
if
situs
ultimately
doesn't
work
out,
then
some
time
is
lost,
but
ultimately
we're
still
following
the
path
that
we
laid
out.
A
A
That
being
said,
there
are
a
couple
things
that
we
are
concerned
about
with
situs.
First,
is
the
enterprise
license
making
Christopher
typing
about
comparing
to
our
overall
spin?
So
right
now
the
enterprise
license.
It
looks
like
it
would
be
about
1.05
million
per
year.
I,
don't
know
how
this
compares
to
our
overall
stem
Chris,
the
first
question
down
there.
Does
anybody
else
know
what
our
overall
spend
is
for
infrastructure
for
supporting
comm
yeah
a
lot
I
can
get
you
the
cost.
On
give
me
a
second
I'm,
not
sure
the
cost
of
the
database
itself
is.
A
E
F
F
F
So
if
we
spend
some
money
on
site
us
to
make
it
scale
more
horizontally,
we
can
reduce
them
how
the
vertical
size
of
the
individual
database
nodes
so
as
a
what
is
this
as
a
percent
of
total
database
spend
because
this
is
this-
is
within
the
common
scope
of
just
database
service
and
has
nothing
to
do
with
the
rest
of
like
it.
It
doesn't
change
what
we
spend
on
monitoring.
It
doesn't
change
what
we
spent
on
API.
It
changes
what
we
spend
on
database,
yeah,
I.
A
Think
I
see
both
points
right.
It
does
and
we're
still
doing
some
research
on
what
the
overall
return
on
investment.
If
we
went
forward
with
cytus,
because
yeah
I
think
you're
right,
then
it's
about
30k
a
month
on
the
infrastructure
deals
for
the
database
and
cytus
did
tell
us
that
they
have
seen
that
you
can
reduce
by
2
to
3
X.
If
you
implement
siteís,
you
can
reduce
the
number
of
cores
that
you're
using
you
know.
So
if
we
cut
it
in
half,
there's
it's
not
gonna,
be
a
million
point.
A
Five
per
year,
it'll
still
be
I,
can't
remember:
Pat
did
a
quick
calculation
still
seven
or
eight
hundred
per
year,
just
on
database
costs
additional
cost,
but
I
also
see
Christopher's
point.
You
know
this
is
a
1%
increase
on
overall
hosting
of
the
site.
Then
it
might
be
worth
it
at
that
point
in
time,
but
if
this
is
a
20%
increase
on
just
the
database
portion
of
the
overall
com
host,
that
might
be
harder
to
justify
so.
F
F
So
we
that
we
have
that's.
We
actually
have
way
more
than
that.
We
have
about
1200
cores,
yeah
and
I
offer
like
I
plan
it
out,
in
other
conversations,
we're
over
scaled
by
at
least
two
X
comfortably,
so
we're
only.
We
only
actually
utilize
about
250
cores
at
most,
like
we're,
probably
only
in
the
200
like
200
core
range
for
what
we
actually
need
for
licensing
so.
C
Million
others,
ok,
I'm
still
I
do
it.
We
need
to
take
into
account
that
Silas
or
any
sharding
architecture
for
that
matter,
it's
going
to
require
more
words
than
we
have
right
now,
because
each
start
needs
to
have
its
own
height
availability.
So
at
least
there's
gonna
be
two
instances
for
a
start
and
I
would
not
recommend
actually
less
than
three
so
I
would
recommend
three
instances
per
chart,
plus
the
coordinators,
which
are
separate.
D
C
C
F
I,
don't
think
it's
I,
don't
think.
That's
gonna
actually
change
the
number
of
cores
that
are
actually
utilized,
so
we'll
probably
utilize
few
more
cores,
but
it's
not
gonna,
be
a
doubling
it'll,
probably
be
you
know,
adding
situs
I'm,
gonna
rough
guess,
it'll
be
like
a
10
or
20
percent
overhead,
but,
like
I
said
we
have,
we
are.
We
are
wait.
F
So
it's
one
of
those
things
where
the
bigger
your
data
set,
the
more
exponential
your
your
your
core
time
goes
and
I've
seen
this
I've
seen
this
quite
obviously,
as
the
load
goes
from,
say:
10,000
requests
per
second
to
11,000,
the
CPU
time
could
double
and
so
I'm
not
quite
worried
about
this.
So
much.
F
A
Gonna
interrupt
so
yes,
there's
a
long.
A
detailed
work.
I
literally
just
threw
together
a
doc
down
here,
I'm
all
up
items,
so
we
can
further
break
down
and
consider
the
conversation.
Excuse
me
all
of
these
are
valid
points
and
there's
a
lot
more
work
to
do
to
break
down
the
cost.
The
pricing
literally
came
in
on
Thursday
the
number
the
760
cores
number
came
from
Hosea,
so
we
have
12
servers,
96
cores
of
peas,
we're
actually
not
going
to
be
using
four
of
them.
A
So
that's
where
that
number
came
from
and
there's
the
additional
wording
on
the
situs
website,
where
it
says
they
can
consider
a
site
license,
no
that's
from
when
they
were
siteís
and
not
necessarily
a
subsidiary
of
Microsoft.
So
that
might
not
be
an
option.
That's
not
something
came
up
during
the
conversation,
but
it's
I
will
ask
the
representatives
and
Tio's
like
hey.
Is
there
a
site
license
because
the
pricing
that
we're
looking
at
right
now
is?
A
E
Spend
down
and
increasing
it
by
ten
percent,
which
is
a
little
bit
high,
a
little
bit
high
of
an
estimate,
but
twelve
million
twelve,
roughly
twelve
million
one
twelfth
of
that
ten
percent.
You
know
that!
That's
that's!
That's
a
that's
a
pretty
daunting!
So
my
request
is,
is
you
know
in
the
next
week?
If,
if
we
can
get
some
events,
details
of
kind
of
expectations
around
this
and
come
up
with
them
like
here's,
the
model
of
what
we
currently
spend?
E
Here's
the
model
of
what
we
would
expect
to
spend
based
on
some
analysis
and
it
did
seems
like
it-
should
be
pretty
straightforward
kind
of
not
quite
back
in
the
napkin
but
back
of
the
Google
Doc
kind
of
calculation
because,
like
like
you,
said
great
glue,
a
million
dollars
just
isn't
I
I,
don't
know
if
I
can
realistically
put
a
proposal
forward.
That
says
we're
going
to
increase,
spend
five
million
dollars.
I,
wouldn't.
A
I,
just
don't
think
I
mean
from
what
we've
seen
so
far
of
the
features
that
they
offer
and
what
we've
been
able
to
test
with
I.
Just
don't
think
the
values
there
at
that
price,
okay,
I,
think
I
think.
Ultimately,
we
just
cost
us
a
lot
more
in
the
long
run.
So
we'll
continue
to
break
this
down
and
I
will
have
the
follow-up
email
conversation
with
Microsoft
and
we
will
try
and
cut
it
as
much
to
the
bone
as
we
can
getting
it
down
to
a
realistic
number.
You
know
with
over-provisioning
with
anything
else.
A
A
Kind
of
jumped
around
admits,
situs
and
again,
like
I,
said,
if
cytus
turns
out
to
be
a
no-go
for
us,
it's
there's
some
good
information
on
features
that
we
would
like
to
have
on
sharding
that
you
know.
Situs
Enterprise
offers
some
nice
kind
of
tenant
rebalancing
and
a
couple
other
features.
If
we
can't
have
those,
if
we
can't
use
enterprise,
it's
something
we
would
have
to
build
ourselves,
and
it's
just
the
timeline
and
invested
team
on
that
one.
A
The
other
concern
is
with
the
community
license
and
jumping
down
here,
so
community
might
still
be
something
we
can
use.
Situs
is
an
extension
and
if
we
use
it,
as
is
we're,
probably
okay
with
the
a
GPL
license,
because
we're
not
actually
linking
into
any
of
the
code,
it's
just
something
that
we're
installing
and
running
on
PostScript
sequel.
That
being
said,
we
still
need
to
make
sure
that
legal
is
okay
with
it.
A
The
concern
that
I
have
is
that
if
we
use
community
license,
and
then
we
build
on
top
of
that
to
get
the
other
features
that
we
want
do
we
then
need
to
open
it
up.
We're
legal
would
be
concerned
because,
right
now
a
EPO
is
on
the
list
of
licenses
that
we
don't
like
or
support.
Then
it's
in
the
handbook
so
again
more
research
to
be
done
there
on
licensing.
A
So
other
going
straight
down,
talk
about
that
so
and
then
what
else
has
been
done
so
Pat
measured
the
results
on
implementing
cascading
deletes
without
foreign
keys.
It
seems,
like
the
differences,
were
pretty
minimal
and
it's
acceptable.
There's
some
implementation
details
that
we're
talking
about
on
the
best
path
forward.
On
that
one.
A
That's
it.
We
won't
straight
down
the
agenda
on
the
next
steps,
talk
about
the
more
detailed
breakdown
on
site,
scoffs
and
trade-offs,
and
then
make
you
wanna
verbalize.
Your
question
for
next
follow-up.
Sure.
D
Yeah,
it
sounds
like
we're
on
track
and
the
earlier.
The
dates
are
in
for
next
steps
to
better
does
it
heads
up
that
this
affects
Melia,
our
team
member
Russia
has
extended
the
no
work
holiday
to
May.
Nearly
has
graciously
been
contributing
all
this
time,
but
we
also
need
to
comply
by
the
local
laws
and
regulation,
so
the
earlier
the
dates
the
better.
So
we
can
plan
around
it.
That's
an
ongoing
development
in
that
part
of
the
region,
so.
H
A
The
other
topics
anybody
wants
to
cover
today-
oh
you
know
what
blocker
so
Pat
has
the
situs
cluster
set
up.
The
he's
been
chatting
with
Nalia
about
their
ability
to
build
a
big
test,
set
the
concern
that
he
has
right
now
is
it's
limited
in
the
number
of
projects.
I
think
it's
only
one
project
right
with
a
bunch
of
data
in
it
doesn't
really
it's
probably
not
the
best
test
set,
so
we're
looking
into
alternatives
to
get
something.
That's
more
representative
of
production
data,
and
we
just
talked
about
that
like
20
minutes
ago.
A
So
I
don't
have
a
good
follow
up
for
that
one,
but
we
will
follow
up
this
week
to
see.
If
we
can.
You
know
there
was
a
script
that
exists
out
there
to
anonymize
production
data.
So
maybe
there's
a
way.
We
can
revive
that
one
and
have
a
representative
test
that
from
a
production
like
environment,
so
we're
looking
in
the
alternatives
are
and
what
we
can
use
for
our
situs
cluster
testing.
They
have
a
bigger
test
set
and
something
that's
more
representative
of
production
so
that
we
can
see
what
the
performance
improvements
are.