►
From YouTube: Geo: Discussing potential improvements to Geo updates
Description
Toon Claes (Backend Developer for Geo) and Fabian Zimmer (Product Manager for Geo) discuss potential improvements when updating Geo. More details are available here: https://gitlab.com/groups/gitlab-org/-/epics/1450
A
A
Okay,
there,
it
is
so
I
actually
think
that's
wrong.
That's
the
first
thing
I
wanted
to
confirm
with
you
I
think.
The
steps
that
I
actually
followed
were
the
steps
that
are
required
for
a
high
availability
upgrade,
and
that
just
happened,
because
I
I
essentially
just
try
to
follow
the
like
flow
of
the
documentation
so
like
we
can
go
through
this,
but
I
think
this
is
actually
the
H,
a
installation
and
the
I
also
I
tested
it
again
because
essentially
I
like
when
I
went
through
this
coached
by
this
is
really
odd.
A
You
know,
like
all
of
those
steps,
a
simple
sort
of
primary
secondary
setup
and
then
I
also
performed,
like
I,
basically
tore
down
my
my
instances
and
upgraded
again
and
I
followed
like
essentially
only
the
like
updating
the
act
steps,
and
it
also
worked
right
so
either.
You
know
like
some
of
the
things
in
the
instructions
are
superfluous
or
I
actually
follow
tha
instructions,
and
they
just
also
work
when
you
have
a
simple
installation
right,
which
makes
sense,
because
it's
essentially
isn't
just
more
stopping
and
starting.
A
So
we
can
okay,
so
so
the
first
thing
I
think
that
we
can
do,
and
that
is
maybe
non-technical,
is
I.
Think
we
can
improve
the
documentation
a
little
bit
for
for
how
to
operate.
You
I
think
you
already
suggested
removing,
or
at
least
moving
to
a
separate
page,
all
of
the
like
15
different
steps
of
you
know:
old
versions,
I
think
that
really
Trotters
the
the
thing
that
makes
it
really
messy.
It
makes
it
really
messy
so
I
think
actually
perversions
for
more
than
the
old
yeah.
A
You
know,
I
think
we
shouldn't
meet
the
information,
but
it
should
be.
You
know
if
you
are
on
a
really
old
version
of
kitchenette
here
you
know:
are
some
extra
instructions
not
like
a
long
list
of
things
on
the
same
page?
Okay,
so
but
I
think
that
can
be
done.
I
think
when
I,
when
I
read
through
the
so
the
first
question
I
sort
of
had
is
when
you
like,
go
to
sort
of
Docs,
not
get
let
of
calm,
and
you
just
type
in
geo
update.
A
But
if
you
are
like
a
large
customer
and
you
have
eight
eight
instances
or
an
880,
you
know
it's
you
just
open
a
terminal
windows
and
so
I
think
we
can
say.
Actually
this
first
infraction
is
kind
of
superfluous,
because
in
the
update
get
leveling
number
two
all
of
those
things
I
explained
again
so
I
think
we
can
just
delete
that.
A
If
you
go
to
update
gitlab,
this
is
this
is
the
first
thing
I
think
that
that
confused
me
so
I
just
went
from
there
to
the
omnibus
packages,
but
I
think
my
my
research
so
far
is
limited
to
you
know
the
Omnibus,
not
the
doctor
installations
or
other
things,
so
that's
good,
something
that
each
to
look
at
because
I'm
not
sure
how
different
that
is.
But
if
you
go
to
on
like
the
omnibus
installation,
instructions
or
update
and
get
laughs
installed
with
the
omnibus
get
lab
package
here,
you
have
like
essentially
in
the
documentation.
A
You
have
top-level
points
on
hey,
you
know.
The
the
first
thing
you
want
to
do
is
like
updating
using
the
official
repositories
right
and
that's
just
essentially
like
up
get
update
and
install
well,
that's
super,
straightforward
and
I
think
you
know
for
essentially
a
regular
installation.
That
is
all
you
need
to
do
right.
A
I
mean
then
we
can
just
remove
that
as
well
and
I
think
that's
good
and
then
I
would
also
argue,
and
this
is
this
is
a
question
for
you,
because
I
don't
know
what
the
would
be
correct
or
a
black
D.
Let's
say
the
preferred
way
is:
if
you
know
you
have
a
regular
normal
get
11
sedation
with
Jia
were
enabled,
and
you
are
updating
now.
Is
there
ever
a
scenario
where
you
don't
read
like
you?
Don't
want
to
test.
A
B
A
B
A
A
Runs
for
two.
Secondly,
so
can
we,
for
example,
run
that
automatically
like?
Essentially,
if
you
run
the
update,
update
and
then
the
apt-get
install
and
and
give
it
all
under
this
end
like
this,
it
detects
hey.
You
know
this
is
the
deal
installation
like
he's
on
a
primary
or
on
a
secondary
during
that
process
of
updating
at
the
end,
you
know
when
you're
done,
why
not
just
run
the
gate
lab?
Do
your
check
I.
A
A
A
Well,
because
it
is
just
a
simple
check
and
it
gives
you
a
a
you
know:
I've
succeeded
installing,
and
if
you,
if
you
write
a
little
note
saying
after
you
know,
you
update
gitlab,
you
know
we
will
automatically
check
if
your
geo
installation
is
working
right
and
otherwise
you
will
see
some
kind
of
of
information
in
a
screen
that
says
hey.
This
is
under
t
yeah.
B
A
B
A
So
we've
done
this,
you
know
and
I
think
this
is
enough
for
the
regular
update
anything
other
than
potentially
like
alternating
or
like
check
at
the
end.
But
I
think
that's
that's
fine,
but
you
know
this
is
where
I
got
confused.
So
let's
say
you
have
an
H
a
deployment
right
or
zero
downtime
updates.
A
So
all
of
these
items
here
on
the
right
are
top-level
items
right,
so
you
have
a
zero
downtime
update
section,
a
single
deployment,
multi
node
geo
deployment,
so
I
honestly
I,
don't
even
fully
understand
what
the
difference
is
between
a
single
deployment
here
and
the
like
what
is
described
here
with
the
official
packages,
but
that's
just
the
way
it
goes
beyond
geo.
What's
beyond
geo
right,
so
I
don't
really
know,
but
what
I
can
see
is
like
okay,
if
I
go
to
zero
downtime
updates
right.
A
B
A
And
then,
if
you
go
to
a
multi,
node
AJ
it
actually
and
like
here,
I,
don't
know
right
but
like
this
is
these
are
the
each
page,
a
instructions
and
then
at
the
same
level
you
know
again
after
that
is
the
geo
deployment
instructions,
yeah
I
think
you
know
these
are
the
instructions
for
an
hae
upgrade,
so,
let's
maybe
I
think
Douglas
knows
right
so
I
said
for
an
Ag
or
or
for
zero.
Downtime
I
think
it's
both
but
I'm,
not
sure.
But
it's
there's
definitely
a
notion
here
that
if
you.
A
Really
follow
these
instructions
in
order
right,
single,
like,
depending
on
your
deployment
type
right,
so
multi,
node
geo,
multi,
node,
h,
a
and
not
quite
sure,
because,
like
this
is
this
is
not
my
first
question
like
clearly.
You
can
have
a
multi
node
h,
a
deployment
and
also
use
kyo
right
so
which
ones
are
actually
the
the
right
instructions.
I
think
it's
those
but
I
think
that's
worth
having.
A
But
we
have
a
primary
node
and
secondary
nodes
right,
so
is
it
like
I
would
assume
that
in
a
age,
a
setup
with
a
with
a
geo
deployment,
the
deploy
node
must
be
the
primary
node
right
because
it
says
you
you
have
to
essentially
dedicate
you
need
to
make
like
designate
a
deploy,
node
and
then
I
think
you
have
to
do
things
like
on
the
deploy
node
first
right
and
that's
similar
to
like
the
things
that
you
need
to
do
with
with
geo
right.
But
it's
unclear.
B
A
But
I
think
I
think
that
is
something
that
we
can
work
on,
because
I
think
like
just
from
like
essentially
writing
down
like
saying
like
okay,
you
know
you
can
have
a
single
like
a
geo
deployment.
You
know
where
you
have
a
primary
and
secondary
said:
it's
not
aj,
follow
those
instructions
and
then
you
can
say
like
if
you
have
an
h,
a
deployment
right,
and
you
also
utilize
to
your
notes.
This
is
what
you
need
to
do.
I
think
that's!
It's
mainly
I.
Everything
I
think
is
here.
B
So,
instead
of
like
try,
the
document
and
the
right
one
try
to
automate
it
and
use
that
tool
and
make
that's
a
product
make
that
into
a
product
that
customers
can
use
and
I
think
that's
why
they
also
wanted
to
build
CHEO
into
it.
So
it's
it's
uniform
for
everything.
How
you,
how
you
install
kit,
lab
I,
think.
A
A
A
So
the
next
thing
is
that
like-
and
this
is
like-
where
I'm
hoping
you
can
help
me
a
little
bit
I,
don't
fully
understand
what
these
settings
in
gitlab
Rd
actually
do
right.
So,
for
example,
we
have
this
insured
gitlab
Auto
migrate,
equals
false,
but
is
set
in
get
get
lab
dot
Rd
and
every
instruction
that
I
I,
see
in
in
here
I
think
has
some.
A
B
There
are
two
types
in
there,
but
yeah
you
have
to
line
with
like
get
logged
will,
so
you
have
to
set
a
negative
dot.
Rb
and
step
forward
is
like
the
skip
a
post
deployed
in
migration.
Yes,
so
those
are
both
like
migrations
and
then
like
update
the
data
schema
to
the
newest
version.
They
need
to
be
an.
A
B
Like
the
the
migration,
the
normal
migrations
are
usually
fast
and
like
the
post-deployment
migrations,
they
run
in
the
background
when,
when,
when
the
new
kill
app
is
already
running
but
I'm
not
sure
how
that
applies
to
zero,
downtime
and
I,
just
think,
like
all
all
migrations,
even
if
they're
like,
before
or
after
of
post-deployment
migrations,
that
they
should
not
take
the
database
down
or
lock
up
on
things
like
that.
Okay
and
then
yep
the
post-deployment,
migrations
or
migrations
that
are
not
needed
for
get
lapto.
B
A
A
We
live
with
that.
This
thing
here
is
not
even
still
relevant
because
that
feels,
but
this
is
just
like
my
feeling
right
now
like
creating
an
empty
fire
with
a
specific
name
to
trigger
something
I,
don't
know
you
know
that
feels
feels
not
very
nice.
You
know.
Do
I
have
to
clean
this
up
later
on
right.
Well,
that's
weird,
because
I
recently
saw
some
virtue.
B
B
But
yet
the
the
reconfigure
command,
it's
it's.
It
does
so
many
things
it's
hard
to
like,
say
what
it
does
and
what
it
not
owes.
It's
really
like
updating
the
whole
configuration
and
problem
for.
We
also
do
the
auto
migrates,
while
while
doing
that,
so
it's
like
the
magic
command
that
fixes
all
things
but
yeah.
A
A
B
That
was
my
biggest
worries
also
like
why.
Why
are
into
reverting
that
change
in
dinner
party
I?
Think
the
Skip
Auto
reconfigure
file
would
be
delayed
automatically
after
one
run
yeah.
That
would
make
sense
to
do
it
like
this.
Then
you
just
created
ones,
run
one
command
and
then
afterwards,
everything's
back
to
normal,
but
with
the
change
in
get
la
partie
other
home
lose
baggage,
does
never
edit
the
get
up
RB
itself,
so
you
have
to
turn
it
back
at
some
point.
I
guess.
A
B
A
Question
is
like,
but
maybe
you
know-
and
this
is
also
maybe
something
I
look
up
in
the
installation
of
gitlab
for
geo.
Maybe
the
default
state
here
after
a
geo
installation
is
that
this
is
set
to
false
right,
and
so
this
is
more
of
a
hey.
You
know
just
make
sure
it
is
right.
I
could
I.
Look
that
up.
A
B
A
Actually,
here
it's
just
like
disabled
temporarily
and
we
enable
it
so
there's
something
we
do
I
think
during
to
get
the
Geo
insulation
right.
Yes,
configure
the
primary
server
so
I
think
here
like
at
the
end,
it
is
set
to
two
again
all
right,
so
it's
it's!
It's
flipped
from
like
its
pitfalls
and
then
I
think
the
end
state
is
yeah.
B
A
A
A
A
You
know
like
this
is
I
think
like
I
just
talked
about
it
that
may
be
sort
of
the
like
most
minimal
iteration
we're
not
even
changing.
You
know
how
its
solid
we're
just
minimizing
like
what
people
actually
need
to
type
in,
because
I
think
to
me.
That
was
maybe
at
least
my
my
frustration.
You
know
if
you
have
to
do
this,
you
have
to
do
it
like
in
the
secondary,
which
is
very,
very
similar.
A
You
have
to
do
it
like
as
many
times
as
we
have
secondaries
right
so
essentially,
you're
you're,
like
putting
like
a
user
in
a
copy
and
paste
loop
for
50
times,
side
and
I.
Think
then
things
can
can
easily
like
be
mistaken.
I'm
just
worried,
like
let's
say,
like
you've,
done
it
three
times
right
and
then
all
of
a
sudden,
your
copy
pasting,
number
files
and
not
number
four.
A
B
A
Think
that
you
know
that
was
my
initial
idea.
This
has
been
essentially
what
what
we
could
do
is
have
a
task
that
says
you
know
like
pseudo,
get
Lautrec
get
lab
geo
update,
but
and
then
you
know
or
like
something
like
something
that
we
think
makes
sense
and
that
break
task
and
essentially
just
execute
those
things
I,
don't
know.
How
is
my
great
task?
A
Sorry,
like
I,
had
a
quick
look
at
the
source
code,
but
he
could
even
so
the
instructions
here
from
four
to
eight
for
the
primary
are
like
very,
very
similar
to
four
to
eight
in
the
secondary
but
they're,
not
exactly
the
same
right
and
I
I,
don't
know
if
the
rake
task
would
also
be
able
to
do
sort
of
detect
am
I
on
the
primary
or
am
I
on
a
secondary,
because
then
you
don't
even
need
to
like
have
a
a
different
command.
It
just
says
essentially.
B
A
You
know
that's
and
if
we
like
yeah,
that's
I
I,
like
that,
you
agree
and
I'm
happy
about
that.
B
I'm
not
completely
sure
about
step
seven
but
arrest
seems
pretty
obvious
obvious
to
me.
I,
never
even
like
when
you
talk
about
spoke
about
adding
the
guillotine
check
to
the
installation
update
process,
and
we
also
don't
need
to
do
that.
Yes,
that's
true
man
that
makes
more
sense
to
just
like
put
it
there
and
yeah.
A
No
and
I
think
what
what
then
would
essentially
happen
is
you
would
run
Zulu
capelet
break
it,
love
Geo,
update
right
and
it
does
all
it
stays
and
then,
at
the
end
it
runs
a
get
litleo
check,
but
and
essentially
the
output
after
that
to
use
the
respite
happy
days,
everything
is
working
and
that
could
be
the
same
on
both
the
primary
and
the
secondary.
Even
though,
and
that
makes
me
sad,
I
already
open
an
issue,
I
think
get
lebryk
G
you
expect
is,
is
not
available
on
the
primary.
A
You
know
make
it
let
breaks
that
is
available
on
primary
note.
That
I
opened
that
a
few
days
ago
also
Faramir
to
see
that
it's
you
know
it's
the
kid
lap
namespace,
but
whatever
yeah
that's
another
thing.
It's
like
the
I
think
the
namespace
should
be
changed,
but
it's
like
on
a
on
a
primary
that
says,
like
this
command
is
only
available
on
a
secondary
and
I.
Think.
B
A
Think
actually
like
it
could
also
display
some
different
information,
for
example,
how
many
times
how
many
secondaries
are
active,
because
the
primary
knows
about
that
right.
So
that
could
be
an
interesting
way
of
saying
you
know.
I
am
the
primary
you
know
like
these
are
like
this
is
my
status
and
I
know
of
these,
for
example,
of
these
files,
secondaries,
all
of
which
I
can
meet
you
at
the
moment.
You
know,
that's,
that's
a
that's!
It.
B
A
Mean
I
think
the
the
Geo
tip
that
do
you
check
should
essentially
like
I
think
this
is
also
a
semantic
like
difference
right.
What
is
the
difference
between
checking
at
the
status,
but
it's
like
why?
Why
do
we
have
two
things
and
they
contain
slightly
different
information?
I
understood
the
the
do
check
to
be
essentially
a
a
I
am
healthy
or
not
right,
where
the
status
gives
you
more
information
about.
You
know
the
syncing
and
you
know
like
what
like
projects
are
available
and
things
like
that,
so
they
have
slightly
different
purposes.
A
A
B
B
A
A
B
A
Yeah,
no
here,
that's
exactly
what
I
did
I
just
followed.
The
updating
using
the
official
repository
destructions,
which
is
literally
just
up,
get
update
up
get
installed,
and
that
was
it
was
possible
for
me
to
upgrade
from
11.11
to
a
12-point
or
3
on
both
the
like
right,
like
on
primary
and
secondary
and.
B
B
A
B
B
B
B
A
A
You
know
you
have
like
four
different
ways
of
modifying
your
environment:
you're,
changing
your
configuration
file,
but
it
left
Rd,
but
you
are
creating
in
key
files
to
like
indicate
some
kind
of
change
in
behavior
right
and
then
you
are
setting
environmental
variables
commands
and
you
used
like
rake
tasks
to
actually
like
do
some
other
things,
so
it's
like
they're
there
for
at
least
four
different
sort
of
concepts
here.
You
know
that
that
are
like
happening
all.
At
the
same
time,
and
at
least
to
me,
you
know
like
as
a
slightly
naive
user
I.
A
Don't
fully
appreciate.
You
know
the
like
like
why.
So,
for
example,
you
why-why-why
create
a
file
right
and
why
not
such
an
environment
variable
right
or
why
is
the
Skip
post-deployment
equals
true,
a
variable
like
this
right?
It's
like.
Why
is
the
notice
skip
post-deployment
migrations
file,
the
same
as
the
like?
It
is
like
the
gift.
Let's
give
it
all
to
me:
Connecticut
right,
I,
don't
know.
B
Yeah
I
have
to
be
I'm,
not
sure
how
it's
all
pseudo
noise-
oh,
that's,
not
true,
I
was
thinking
it's
when
we
like
combine
these
commands
together
in
wrong
rake
tasks.
You
have
to
be
careful
that
at
some
version
things
an
omnibus
do
not
change
that.
The
kid
lab
wheels
code
is
not
aware
of
cost
and
then
our
wake
toss
would
break
if
they
do
something
they
aren't
aware.
We
are
automating.
A
A
But
I
mean
I
think
the
the
first
iteration
that
we're
proposing
is,
we
would
shrink
the
setup
so
number
one
setting
the
gate
learn:
rails
Auto
migrate
to
false.
That's
still
something
that
needs
to
happen.
That
I
think.
Unless
we
have
a
better
idea,
we
will
probably
have
to
do
that
and
then
number
two.
This
like
skip
auto,
run
like
creating
this
file
skip
Auto
reconfigure
is
kind
of
in
the
same
ballpark
or
that
it's
not
it
doesn't
seem
very
geo-specific
right
like
we
can't
really
do
anything
about
that
either
easy
I.
A
B
A
It
just
works
wow,
it's
really
cool
so
and
then
like
essentially
so
these
three
things
they,
they
would
say
the
same,
but
then
from
4
to
8.
You
know
the
suggestion
is
to
combine
all
of
those
in
to
a
Hitler
break
task
when
which
would
be,
for
example,
Zulu
gitlab
break
it
let
geo
update,
and
that
would
be
the
same
for
a
primary
and
secondary,
and
that
would
essentially
reduce
the
amount
of
manual
intervention
spiral.
B
B
A
A
If
you
about
it,
it's
it's
really
better
like
interesting
from
sort
of
a
human
perspective,
because
I
mean
like
we
read
so
many
things
and
you
like
I,
expect
like
documents
to
be
organized
hierarchically
right
and
if
they're,
not,
you
know
like
it's
really
confusing.
You
know
it's
like
sequential
I,
guess,
like
sequential,
is,
like
you,
don't
have
that
just
okay.
A
I
think
then
we
can.
We
can
potentially
continue
the
discussion
in
those
issues
also
asynchronously,
and
so
my
goal,
like
I,
think
we
could
if,
for
example,
by
the
end
of
the
week,
we
have
them
like
these
issues
in
a
in
a
state
where
we
both
understand
them.
You
know
because
we
don't
need
to
like
implement
right
now,
but
once
we
start
working
on
12.2
I
think
I
think
then
that
needs
to
be
actionable.
Potentially
by.
Like
another
engineer,
that's
you
yes,
yeah
cool.