►
From YouTube: 2020 04 01 Sharding WG 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
A
So
for
the
agenda
for
today
we're
going
to
discuss
them.
The
topics
listed
in
the
issue
alvaro
had
listed
a
bunch
of
questions.
So
as
for
enterprise
auction,
we
think
it's
an
option.
We're
still
trying
to
figure
out
pricing,
so
I
am
meeting
with
them
at
11
a.m.
Pacific
time
tomorrow
they
set
up
a
meeting.
Pat's
gonna
join
me,
so
we're
going
to
talk
to
them
tomorrow
find
out.
The
pricing
is
yeah.
I
guess
get
more
of
an
overview
of
what
they
have
to
offer.
C
Categories
be
better
that
yeah
I
think
the
main
thing
that
we
probably
meet
what
you
know
what
to
have
is
the
shard
rebalancing,
which
is
only
a
feature
of
the
enterprise
and
the
ability
to
pin
tenants
to
a
shard,
so
I
think
if,
if
Enterprises
is
not
an
option,
I
don't
know
that
it's
gonna
meet
our
requirements.
Kind
of
what
we
had
talked
about
previously.
A
C
A
I'll
grow
a
you,
prime
minister,
first
part
of
it
we're
having
a
meeting
with
Microsoft
effectively
tomorrow
to
talk
about
cytus
enterprise
and
get
some
licensing
questions
from
them.
So
if
you
have
any
questions
that
you'd
like
for
us
to
ask
them
feel
free
to
just
Adam
here.
D
So
what
is
the
policy
here
from
what
I
understand
the
usual
policy?
Is
that
whatever
that
Lanson
get
look
calm
should
end
up
landing
sooner
layer
on
the
product?
Now,
in
this
case,
how
does
it
feel
under
anonymous
your
solution
and
or
relative
to
the
next
topic?
How
is
going
to
attract
the
licensing
if
it's
open
source
product.
D
B
B
D
Okay,
so
then,
because
I
saw
si
seed
comment
on
on
this
issue
regarding
the
licensing
so
yeah,
what
I'm
trying
to
understand
them.
So
if
we
opted
for
Enterprise,
it
wouldn't
be
a
problem
if
we
opted
for
the
open
source
license,
which
is
a
GPL
and
then
some
enterprise
customers
asked
for
this,
they
maybe
I.
B
D
Clear
I've
been
advocating
for
a
long
time
that
solutions
that
applied
forgivable
comm
do
not
need
to
apply
necessary
to
the
product,
so
I'm
totally
on
the
line.
But
basically
the
seed
comment
is
in
one
line.
Summer
is
saying
that
enterprise
customers,
some
enterprise
customers,
may
not
like
the
a
to
be
a
license,
and
then
I'm
wondering
why
this
is
an
issue
or
might
be
an
issue
if
they
need
not
to
you
be
using
this
yeah.
B
B
D
Well,
on
the
database
side,
message
has
been
a
little
bit
different,
which
is
we
try
not
to
do
anything
that
may
not
land
on
on
the
product
side
and
we
are
making
different
decisions
based
in
this
concern,
which
I
don't
necessarily
agree
with,
because
it's
limiting
us.
It's
limiting
us
based
on
this
future
expectation.
I
would
like
to
have
more
freedom
that,
without
because
give
up
definitely
agree,
needs
different
solutions
and
it's
orders
of
magnitude
from
the
largest
enterprise
customer
discussing
on
the
site,
but
it
obviously
means
different
solutions.
D
A
A
Next
one
was
licensing,
I
think
what
I
was
reading
about,
Mia
Farrow
licensing.
It
may
take
the
opens
of
the
site,
it's
community,
offering
off
the
table
if
SIDS
concerned
about
it
and
a
bunch
of
articles
I
read
about
it
said
it's
questionable
whether
or
not
you
need
to
open
up
your
servers
if
you're,
using
something
with
a
fair
licensing,
so
the
community
cited
community
may
be
a
non-starter
for
us.
Sometime
probably
did
look
into
a
little
further
on
this
reports.
D
There's
there's
actually
a
lot
of
confusion
around
around
this,
and
my
my
personal
take
if
I
may
bring
it
to
the
table,
is
that
the
concerns
about
GPL
or
a
couple
of
licensing
I
do
apply
to
libraries.
If
you
are
building
a
product,
can
you
use
the
library
and
this
library
to
do
that?
Your
whole
product,
even
at
some
point,
is,
might
be
acceptable
for
you.
Maybe
the
library
will
integrate
a
new
feature
into
a
proprietary
product
and
you
don't
want
to
open
up
everything
and
that's
why
a
legal
department
licenses
so.
D
Libraries
now
for
a
part
of
that
you
don't
integrate,
as
as
as
as
as
a
part
of
the
same
entity
as
a
component
since
link
to
component
from
the
software
perspective.
This
is
computer
talking
over
the
network
or
anything
like
that.
I
would
say
a
GPL
or
TBL
rs3
as
anything
else,
and
you
can
do
whatever
you
want
with
that.
You
don't
even
need
to
make
your
own
software
to
be
a
lab
for
them
for
that
matter,
and
they
of
course
enjoy
their
freedom.
Self
opens
where's
the
tradition.
D
So
from
that
perspective,
I
would
believe
that,
as
long
as
it's
not
a
library-
and
this
is
not
a
library,
any
qabil
or
ATPL-
s-
should
be
more
than
fine
in
this
case,
actually,
technically
speaking,
a
bill
also
expands
this
morality
over
the
network,
but
the
drivers
you're
using
which
are
new
components,
are
actually
betting
in
your
application.
There
they're
not
very
literally
less
than
license
so.
D
D
D
D
Actually,
on
the
key
topic
acceptable
licenses
page,
this
is
all
very
addressed
and
the
limitation
specifically
applies
to
libraries.
So
it's
clearly
stated
as
libraries.
Yet
it
still
says
that
in
general,
it's
better
to
check
the
legal
or
request
some
self
approval
that
it
is
design
the
same
way
as
I'm
proposing,
which
is
nice
for
libraries
for
non
libraries,
just
components
you're
using
your
strictly
more
than
five
I
mean
I
opinion
might.
A
C
I'm
so
I
yeah,
I
think
the
biggest
feature
that
we
had
talked
about
is
the
shard
rebalancing,
which
I
think
it's
a
feature
that
we
sort
of
all
agreed
that
we
would
probably
want
to
have,
but
we
could
possibly
write
our
own
yeah.
I
mean
it's
yeah
possible
that
we
can
develop
some
other
solution
around
that
we
would
have
to
look
into
what
that
would
be
or
how
that
wouldn't
fit
in
with
situs.
A
C
A
D
D
D
Homegrown
solution
leave
my
opt
for
our
approach
or
even
have
coordinators.
This
call
directly
the
applications
understand
the
starting
algorithms
and
they
talk
directly
to
the
Sharks,
in
which
case
things
aren't
actually
significantly
different,
and
in
this
case
the
cost
of
effort
may
be
different.
A
So
part
of
the
reason
we
started,
this
working
group
was
to
also
help
with
the
scalability
of
the
database
right.
We
talked
initially
about
the
application
charting
approach,
but
we
focused
this
working
group
on
just
the
database
sharding
piece
of
it,
so
we
can
have
that
horizontal
scaling
at
the
database.
We
still
have
some
ability
to
vertically
scale
because
we're
not
at
the
maximum
provision
server
for
the
database,
but
before
we
get
to
that
point,
they
wanted
to
have
an
option
to
be
able
to
scale
out
the
database
horizontally.
A
So
that
was
the
focus
of
the
group.
The
application
charting
is
not
off
the
table,
but
it's
not
the
focus
of
this
right
now,
although
I
mean,
if
we
make
a
justification
for
it,
if
we
get
enough
of
a
stumbling
block
at
something,
we
could
certainly
consider,
but
it's
not
the
primary
focus
at
the
moment.
Yeah.
D
A
D
Shard
and
needs
to
connect,
and
we
avoid
the
middleware
that
are
the
coordinators,
so
this
is
potentially
more
high
performance,
lower
latency.
We
don't
introduce
potential
failures
derived
from
coordinators,
but
of
course
the
application
needs
to
be
more
complex
because
it
needs
to
understand
all
the
sharding
and
and
how
to
perform.
If
there's
queries
that
need
to
address
our
ocean
starts
at
once,
which
is
it
actually
the
toughest
problem?
But
this
approach
is
the
proton
has
been
used
by
the
large
social
networks.
A
D
I
think
there
will
be
competing
first
of
all
because
the
application,
so,
let's
maybe
label,
bit
different
where
I'm
calling
application.
Surely
shortly,
let's
call
it
maybe
coordinator
less,
it
would
not
require
40
maker
right,
so
the
applications
talks
directly
to
the
shark
either
because
there's
application
code
advanced
and
charting
or
because
there's
a
proxy,
but
it's
sitting
next
to
the
application,
could
be
like
a
sidecar
from
to
reach
idea.
This,
let's
go
in
for
an
outboard
in
Ariel,
a
second
of
this
precise
turn.
This
may
change
the
way
you
shark.
D
A
E
D
So,
strictly
speaking,
what
what
you
need
is
the
same
as
I
touched
us,
which
is
those
tables
that
are
not
charted,
are
replicated
and
then
you'll
have
a
copy
of
all
these
tables
in
all
the
different
charts,
and
for
these
we
could
use
exactly
the
same
technology
that
psychoses
using,
which
is
calcium,
phosphorus
ten,
and
it's
this
logical,
replication
right.
So
the
same
technology
mainly
there's
is
a
little
bit
more
integrated
to
the
product
itself
that
you
know
technology
is
there
to
be
able
to
replicate.
Logically,
these
tables
are
called
replicated
tables.
D
The
the
difference
will
come
if
a
given
query
needs
to
not
only
replicated
tables
that
will
be
more
than
fine,
but
if
they
need
to
go
cross
rights,
so
the
given,
where
you
may
involve
more
than
one
shot.
Yes,
in
that
case,
then
you
start
needing
to
replicate
the
functionality
of
the
coordinator
and
maybe
three
applications,
but
in
this
case
is
what
you
would
be
done
is
to
try
to
understand
the
the
shape
of
the
queries
that
you're
doing
and
and
whether
those
queries
are
actually
hitting
a
single
shot
hitting
several
shots.
D
The
typical
the
ferry,
which
never
the
world's
never
perfect,
that
the
theory
is
that
all
relatively
almost
the
applications
are
typically
a
record-based
or
related
record
based,
in
which
case
they
gather
information
from
a
single
shard
and
then
it's
more
like
OLAP
applications
or
data
warehouse
and
where
you
get
data
from
many
shots,
because
you're
doing
a
lot
of
a
lot
of
data
gathering
which
may
happen
at
the
city
separate
stage
and
others
I
know,
there's
stuff
like
right
now
so
might
happen
at
that
stage.
Where
do
you
have
to
make
everything?
D
D
D
Well,
I
mean
I,
based
on
what
cycles
does
so
they
they
used
to
proceed
to
face-to-face,
commit
ridges
two-phase
commit
is
small
particle,
but
it's
safe.
It's
consistent
strongly
consistent
from
a
distribution
systems
perspective.
So
that's,
okay.
The
concern
here
is
that
right
now,
migrations
are
are
becoming
a
bottleneck
for
the
database,
they're
being
slow,
they're.
D
D
A
Thanks
for
that
clarification,
there
are
a
couple
issues
out
there,
where
we
are
trying
to
push
back
on
when
they
are
doing
migrations
or
notifying
engineers
or
when
they're
doing
migrations
that
will
be
large
and
painful
and
slow
right
and
figuring
out.
Ask
them
to
figure
out
alternatives
to
that.
I
will
dig
up
the
issues
for
those.
D
This
is
totally
off
topic
that
databases
are
happier
if
you
try
to
to
do
operations
that
are
more
kind
of
append-only
versus
heavy
updates
or
heavy
deletes
right.
Of
course,
you
can
do
updates.
Of
course
you
can
do
to
do
this,
but
they're
less
some
things
and
deletes
do
the
more
inserts
just
seeming
like
that.
We
do.
The
database
is
happier
I.
Don't
you
know,
you
know
you
know
levels
so.
A
Yeah
we're
we
do
far
too
many
migrations
right
now
and
we're
trying
to
get
less
of
that
and
coordinated
a
little
better
on
that.
So
I
will
link
those
issues
here
and
thanks
for
bringing
that
up,
see
concerns
over
the
long-term
sustainability
of
siteĆs,
no
idea,
I,
guess
those
are
good
questions.
I
can
ask
tomorrow
when
we
have
the
meeting.
D
D
A
A
E
A
Ya
know
very
moving,
you
know
somewhere,
but
yeah.
They
said
there
and
trying
to
put
it
in
there.
Enterprise
Architect
skew
so
oh
Allah
Cheryl
Oh
notes
from
the
meeting
when
its
Sun
tomorrow,
but
we
have
three
minutes
so
one
of
the
things
that
we've
been
asked
as
a
team
to
come
up
with
goals
by
the
end
of
twelve
ten
for
what
we're
doing
with
situs
and
patent
I
talked
a
bit
about
this
yesterday,
there's
still
some
things.
A
We
saw
some
dependencies
right
so
still
waiting
on
a
way
or
trying
to
find
a
way
to
import
data.
So
we
have
realistic
data
set.
That's
got
the
situs
cluster
set
up
with
the
coordinator
and
three
servers.
If
I
recall
correctly,
the
cost
will
figure
out
tomorrow.
Functionality
we've
already
outlined
a
bit
on
what
what
functionality
we
like
from
enterprise.