►
From YouTube: Database Group Meeting - 2022-05-04
Description
This is the regular application database team meeting, where we discuss team matters and review ongoing work.
A
B
Welcome
to
the
tuesday
april
5th
2022
database
team
meeting
a
quick
reminder
at
the
end
of
the
meeting.
We're
going
to
do
a
quick
short
mini
live
retro.
So,
while
we're
talking
just
think
about
something
positive
from
last
week
and
think
about
something
that
could
have
gone
better
from
the
last
week
and
we'll
talk
about
it,
then
cool
so
infradev
issues,
I
don't
think
there
are
any
today,
looks
like
there
are
none
how
exciting.
B
C
D
C
Some
discussion
in
there
so
there's
a
there's,
a
proposal
or
there's
some
proposal,
basically
of
an
idea
of
how
to
implement
that
and
then
I
think
also
andreas
made
a
good
point
too,
which
is
like.
Maybe
we
don't
even
really
need
that
right
now
so
yeah,
it's
it's
in
progress.
I
would
say.
B
Cool,
so
I
think
this
is
done
because
we
now
have
so
last
week
we
merged
all
556
yaml
files
with
feature
categories
in
them.
That's
pretty
much
the
only
thing
they've
gotten
there
right
now
are
feature
categories,
and
I
think
that's
it.
B
So
we
haven't
announced
it
yet
because
we're
still
gonna
add
a
cop
or
a
test
or
something
that
will
check
to
make
sure
people
have
added
their
tables
as
files
before
they
converge
and
then,
after
that,
I'm
gonna
file
issues
for
all
of
the
teams
that
have
tables
that
were
assigned
in
future
categories
to
go
through
and
check
their
their
tables
to
see
if
they're
assigned
correctly
so
yeah.
So
that's
cool
to
have
some
progress
there,
andreas,
how
auto
vacuum
running.
E
We
are
working
on
the
groundwork,
so
we
basically
one
step,
is
to
be
able
to
take
put
the
migration
on
hold
for
a
short
period
of
time
and
we're
just
discussing
how
to
do
that,
and
then
I
think
the
sort
of
the
other
groundwork
is
to.
I
think
we
would
like
to.
In
the
long
run.
E
We
would
like
to
be
able
to
deal
with
a
lot
more
indicators,
so
I
would
like
to
prepare
that
a
little
bit
so
that
we
we
have
a
way
of
you,
know,
reasoning
about
signals
and
and
that
stuff
that's
what
I'm
working
on
right
now.
That
said,
I
don't
have
a
lot
of
time
for
that.
In
any
case,.
C
B
Enough
all
right
cool
thanks
simon.
It
display
results
of
background
migrations.
F
Yep
I
had
a
few
mr's
that
were
kind
of
stacked
up
for
this
that
have
started
merging
out
so
the
last
one
which
is
actually
displaying
them.
I'm
working
on
now,
hopefully.
A
So
about
the
improved
batches
background
migration
log.
So
when
I
was
doing
a
documentation,
so
we
have
a
section
that
it's
called
like
the
debugging
and
I
found
that
we
are
not
logging.
All
of
the
parameters
to
our
old
blocks
about
the
battery
background
migrations,
for
example,
the
class
name,
job
parameters
etc.
And
I
just
created
an
mr,
where
we
are
adding
these
fields
to
the
logs.
B
Cool
and
then
you've
also
got
the
documentation.
A
In
progress,
yeah
documentation,
yeah
I've
created
a
like
a
first
iteration
and
collecting
feedback
now
and
yeah
we'll
see.
B
G
Small
plant,
a
small
note
on
that,
even
though
we
switched
the
martial
is
back
as
a
technical
writer,
because
we
started
the
review
with
amy.
I
think
that
you
can
continue
the
review
with
amy,
because
this
is
a
large
document.
It's
350
lines,
so
there's
no
no
need,
if
she's
already
on
top
of
it
and
yeah.
B
Yeah
amy
said
as
much
before,
so
she
she
is
more
than
happy
to
wrap
up
work
that
we
started
with
her.
So
please
feel
free
to
assign
that
back
to
her
for
another
review
or
the
final
review,
because
she'll
do
some
cleanup
and
stuff
in
there.
A
B
Cool
all
right
so
that
pat
sub
patrick
boiler
plate.
C
Yeah
that
one
there
was
a
couple,
mrs
merged
already-
that
was
kind
of
setting
up
the
groundwork
for
that
change,
and
then
I
have
a
couple
more
that
are
ready
to
go
for
review
this
week.
Sometime
and
it's
you
know,
swing
well,.
B
B
Just
a
reminder
that
dimitri
is
available
as
a
back-end
maintainer
for
batch
background,
migrations
work
so
feel
free
to
leverage
him.
If
you
need
someone
who
will
hopefully
have
context
going
forward
and
also
don't
forget
that
all
of
us
are
qualified,
back-end
reviewers
so
don't
be
afraid
to
assign
first
round
reviews
out
to
one
of
us.
B
All
right,
I
don't
know
that
there's
any
change.
C
B
Simon
is
this:
did
this
merge
the
yeah?
It
did.
B
Close
it
good
job
giannis
and
I
are
horrible
people
and
have
not
yet
worked
on
updating
the
database
team
page,
but
we
should
do
that
because
with
hiring
having
having
up-to-date
information
in
there
is
going
to
be
really
good.
B
So
this
is
closed,
but
there's
a
new
issue
that
I
didn't
add
the
right
labels
to,
but
I'll
do
that
afterwards.
So
we
have
defined
targets
and
we're
meeting
with
I'm,
I'm
gonna
go
to
the
development
team
staff
meeting
today
and
we're
gonna
present
these
to
the
other
directors
and
so
hopefully
for
next
quarter.
People
will
start
adding
a
bunch
more
reviewers
and
that'll
take
some
of
their
reviewer
load
off.
So
that's.
E
Yeah,
that's
great
in
in
one
sentence
or
two:
is
there
some
notion
of
this
local
idea
going
on
in
those
discussions
as
well.
B
Not
in
not
as
relates
exactly
to
this
one,
I
think
the
so
we
haven't
included
local
as
an
option,
because
it's
not
ready
yet.
That
said,
I
planned
on
scheduling
us
to
try
and
add
the
local
flag
in
the
next
milestone,
hoping
that
we
could
try
and
wrap
up
batch
background
migrations,
the
mvp
first
and
then
so
next
milestone.
B
I
it
seems
that
way
at
the
moment
I
yeah
I
I
think
I
think
so.
E
Okay,
that's
great
because
I
feel
like
I.
I
lose
a
lot
of
time
because
of
not
having
local
reviews,
and
I
would
rather
prioritize
on
that.
You
know
rather
not
do
as
many
reviews
and
rather
spend
the
time
on
implementing
things
that
I'm
missing
to
sort
of
help
the
systematic
change.
I.
B
I
totally
agree-
and
I
I
it
is-
it
is
our
intention
to
add
the
local
reviewers
part
of
that.
We
also
just
need
more
reviewers
and
maintainers,
and
so
it
I
think
it
will
work
in
both
ways.
But
yes,
I
I
I'm
there.
E
Okay,
great
and
for
the
implementation
is
that
covered
in
this
issue
that
we
have
for
local
maintainers,
or
is
it
just
a
discussion.
B
I
think
that
I've
updated
that
to
issue
to
be
deliverables
with
okay.
The
idea
of
that
being
the
issue
to
to
okay,
quick
finish
that
so
yeah
yep
cool
all
right.
B
E
E
E
I
will
have
to
take
a
look
whether
that's
or
you
know,
what's
happening
in
the
default
gtk
settings.
Maybe
that's
just
me.
B
No
nice,
that's
a
that's
a
good
catch
all
right!
I'm
gonna!
Stop
sharing
so
for
next
week,
diego
you're
on
triage
rotation,
and
I
think
that's
that's
it
for
that.
Simon!
Congratulations!
B
E
B
E
G
Myra
so
myra
got
approvals
from
all
containers.
B
All
right
hiring
update,
so
we
are
now
hiring
next
quarter,
which
means
that
I'm
spending
a
lot
of
time,
trolling,
linkedin
and
doing
other
things.
If
you're
interested
in
sharing
something
to
your
network.
Let
me
know-
and
I
will
I
can
help
you
write-
that
or
feel
free
to
just
share
the
job
posting
or
the
video
from
our
recruiter
or
one
of
giannis's
videos
from
the
past
or
if
he
records
another
video,
so
hoping
to
find
a
couple
of
really
great
folks
to
come
and
join
us.
G
G
B
All
right,
you
want
to
see
the
next
item.
G
Oh,
it's
just
a
small
reminder
that
marcia
is
back,
and
so
we
now
assign
issues
to
her
and
also
a
reminder
to.
Let's
remember
to.
B
Good
call
out,
thanks
dennis
all
right
ben.
D
I
I
also
wanted
to
ask
about
this
validate
constraint
thing
that
simon
and
patrick
are
well
familiar
with
you.
Do
you
want
to
hold
that
till
after
the
retro.
G
G
D
B
Talk
about
that
first,
before
we
go
on
to
the
retro,
I
think
I
think
this
is.
This
is
a
good
use
of
our
time.
So.
D
Okay,
so
customer
customers
running
gitlab
on
a
single
bare
metal
server.
It's
pretty
big.
It's
got
384
gig
of
ram.
Otherwise
it's
yeah,
an
omnibus
install
of
gitlab
with
postgresql
on
that
server.
D
D
D
And
they
tried
running
the
migrations
in
manually,
just
just
with
the
weight
task
gave
up
on
that
at
some
point,
but
they
had
a
couple
of
emergencies
over
that
weekend
and
I
can't
remember
the
exact
sequence,
but
we
ended
up
advising
them
well,
it
looks
like
it's
the
validate
validate
constraint,
that's
taken
a
long
time
so
we'll
get
to
what
that
does
in
a
second,
so
cancel
running
the
migrations
manually
set
that
migration
up
running
all
the
others
into
your
database.
D
All
the
outstanding
migrations
set,
the
cons
set
the
migration
back
down
again
and
then
and
then
kick
it
all
off
again.
It's
it's
just
going
to
take
some
time.
It
is
what
it
is
that
has
now
been
running
since
the
weekend
before
last,
so
we're
looking
what
nine
days
or
something
the
two
tables
are:
130
million
rows
and
rounding
error
in
terms
of
the
events
table
being
slightly
larger.
D
D
D
And
yeah
we've
checked,
we've
checked
for
locks.
The
process
has
all
the
locks,
it
needs
it's
running
and
indeed
it's
blocking
auto
vacuum
and
also
analyze
from
bonnie,
because
it's
got
the
lock
that
they
need
to
run
from
what
we
can
see
from
the
data
about
the
table.
D
Basically,
half
the
table
is
dead,
two
balls,
it's
it's
like
100
million
dead,
tuples
to
133
million
live
ones
which
is
consistent
with
it.
They
have
basically
haven't
written
the
primary
key
on
every
record
had
some
time
to
auto
vacuum,
some
of
them
and
and
then
not
been
able
to
to
complete.
D
We
also
don't
have
an
analyze,
that's
current,
so
the
last
update
I
put
on
the
ticket
yesterday,
I
I
advised
them.
Okay
can
cancel
so
yeah
can't
cancel
can't
cancel
the
migration
it
should
have
completed
by
now.
It
ran
in
42
seconds
on
their
test
environment.
D
D
Nfs
yeah,
that's
block
storage.
I
got
I
got
them
to
they.
They
showed
the
df
output
for
for
the
postgres
locations,
but
they
do
have
some
nfs
somewhere.
There's
bind
amounts
and
stuff.
It's
it's
a
bit
messy
greg
and
my
team
was
found
that
out
on
an
emergency
call
this
week,
but
yeah
I
got.
D
I
only
really
care
about
the
database
state
data
and
that
sort
of
block
storage
I
mean
it
may
well
not
be
local
block
storage,
it
may
well
be
iscsi
or
or
or
something
I
don't
know
they.
They
clearly
have
some
kind
of
callback
nas.
So
I
don't
know
exactly
what
it's
on
blitz,
block,
storage
and
and
balance
the
probabilities
it's
going
to
be
similar.
I
would
guess,
and
and
certainly
yeah
they've
got,
the
table
sizes
are
in
the
order
of,
I
think
the
largest
one
was
33
gigabytes.
B
330
million
push
events
is
a
lot
of
push
events
for
a
single
bare
metal
server.
Are
they
I
assume
they're,
trying
to
run
this
without
downtime?
Does
that
mean
they're
receiving
a
high
rate
of
pushes
that
could
be
adding
keys
to
this
like
new
values
to
this
table
at
a
rate?
D
Well,
ideally,
ideally
it
would
be
yes,
but
unfortunately
it
can't
insert
into
the
push
events
table
because
the
trigger's
missing
so
the
the
trigger
that
updates
the
new
primary
key
is
missing
because
it
was
pulled
out
by
one
of
the
subsequent
migrations
and
so
every
record
insert
inserts
the
event
id
into
the
old
primary
key
and
tries
to
zero
into
the
into
the
field
for
the
new
primary
key.
That
fails
the
constraint,
because
there's
no
zero
record
in
the
events
table,
so
they
can't
insert
any
records
into
that
table
at
the
moment.
F
D
Yeah
I
mean
I'm
the
last
update
and
I'm
kind
of
you
know
looking
to
run
this
past
year.
So
it's
useful
you're
telling
me
the
things,
I'm
I'm
suggesting
vacuum
and
analyze.
I
I
definitely
wanted
to
get
one
on
on
that.
D
G
Not
geo
just
replicas
for
global,
oh.
D
Yes,
yes,
any
any
sort
of
yes,
any
sort
of.
D
D
D
The
vacuum
is
going
to
play
a
part
in
in
maintaining
the
database
of
the
index,
but
I
also
would
just
want
to
eliminate
the
possibility
that
a
key
index
for
this
constraint,
validation
is
actually
basically
corrupted
or
something.
E
Have
you
had
a
look
at
whether
it's
valid,
so
an
index
has
a
chance
of
existing
but
being
in
an
invalid
state
when
you
create
it
and
then
it's
not
being
used
so
on
first
side
it
might
look
like
it
exists,
but
it's
not
being
used
really.
D
That
would
show
up
show
up
in
a
slash
ds.
E
Plus
yeah
yeah,
it
should
say
invalid
if
it's
actually
that.
C
D
Yes,
that's
my
recollection
as
well.
Certainly
push
event.
Payload
is
fine.
It's
got
the
invalid
constraint,
but
yeah
the
indexes
are
fine
and
events
I'll
track
down
double.
E
Check,
what's
the
difference
to
the
test
system,
then
did
I
hear
that
correctly
that
they
managed
to
run
that
migration
so
validating
this
constraint
within
like
one
minute
yeah,
and
it
takes
multiple
days
on
their
production
system
yeah
and
one
difference
that
we
identified
as
the
traffic
coming
in
so
on
production?
There
is
traffic
on
tests.
There
is
none.
Is
there
any
other
difference?
Do
we
know.
D
No,
not
not
that
I've
been
able
to
tease
out
now
I've
I've
offered
them
a
call
to
just
to
discuss
the
next
step.
So
it
may
well
be
that
in
discussion
we
we
find
some
more
relevant
differences.
F
D
C
D
Yeah
so
I
mean,
I
think
I
think
the
state
of
vacuum
and
analyze
kind
of
sticks
out
at
me.
At
the
moment
I
mean
I
I
I
did
check
you
can't
want
to
explain,
alter
table,
but
I
guess
that
fundamentally
order
table
validate
constraint
is
doing
something
which
will
go
through
the
you
know.
It's
it's
gonna.
It's
gonna
cover
strategy,
different
strategies
for
doing
that,
and
it's
gonna
have
to
pick
one,
but
it
doesn't
look
like
you
can
expose
what
those
are.
F
G
G
D
No,
we
got,
we
got
a
listing
of
locks
and
all
the
processes.
Yesterday,
late
and
and
yum
it's
got,
it's
got
about
20
25
locks,
but
it's
got
them
all.
They're
all
they're,
all
true.
F
One
way
you
could
see
what
it's
doing
is
we
could
run
some
stuff
over
pg
station
user
tables
to
see
like
where
is
it
actually
spending
most
of
its
disc
time
to
see
if
it's
chewing
at
a
table
or
not?
That
will
be
kind
of
complicated
if
they're
also
running
production
on
this
at
the
same
time,
but
maybe
they
could
take
a
couple
snapshots
with
it
running
and
then
turn
off
the
valley
constraint.
Take
a
couple
snapshots.
D
E
Do
we
have
pg
stat
activity
available
like
process
information
on
a
high
level.
D
E
C
D
There's
the
ticket
there
I'll
find
another
third.
D
Yes,
if
they
book
one
I'll
I'll,
let
you
know
it's
gonna,
it's
gonna,
be
it's
gonna
be
late
because
they're
they're
in
the
u.s,
but
you
know
kind
of
this
sort
of
time,
maybe
or
a
bit
later.
I
would
guess
I
mean
it's
not
gonna
be
much
later,
because
I'm
not
available.
A
B
D
No,
not
necessarily
that
that's
definitely
a
possibility
is
that
we
we
fix
this
up
manually.
The
drawback,
obviously,
is
we
have
tested
code
for
doing
it
the
other
way,
because
we
could
we
could
do
it
with
downtime
but
use
the
migration.
E
I
was
thinking
like
as
a
way
to
rule
out
other
stuff
sort
of
interfering
with
with
that
in
terms
of
locks
or
just
put
it
in
maintenance
mode
to
have
the
database
online
and
run
the
migration.
But
that's
a
downtime
for
sure.
D
Yeah
yeah
I
mean
they're
talking
about
they
were
talking
about
using
ip
tables
to
firewall
off
gitlab,
so
that
the
analyzer
would
be
able
to
the
analyzer
vacuum
would
be
able
to
run
so.
Downtime
is
definitely
on
the
cards
and
and
yeah
and-
and
I
mean
the
migration,
if
you're
running
the
migration,
the
weight
tasks,
you
don't
need
sidekick
or
right
puma,
because
it's
self-contained,
so
we
could
just
shut
everything
down.