►
From YouTube: Geo HA upgrade from 12.1.9 to 12.2.8
Description
In this video we're performing an upgrade from GitLab 12.1.9 to 12.2.8 on a Geo HA setup. We're evaluating the new documentation, which is not merged yet.
Issue: https://gitlab.com/gitlab-org/gitlab/issues/32435
MR under review: https://gitlab.com/gitlab-org/omnibus-gitlab/merge_requests/3577
A
A
A
Updated
22
minutes
ago,
according
to
Dan,
was
the
last
time
I
checked
for
our
newer
grades,
but
the
last
committed
from
10
hours
ago.
So
this
seems
fine,
so
we're
good
to
go
so
I
have
the
documentation.
Here,
it's
update.
It's
the
version
from
the
merge
request.
We
were
working
on
at
mount
or
still
not
merged
yet
so
we're
gonna
follow
that
I
have
my
terminal
windows
open
for
each
machine
so
on
the
left
side
or
the
primary
machines
and
on
the
right
side
or
a
secondary
with
one
more
in
the
Postgres
tracking
database.
A
A
A
A
B
All
the
primary
notes
how
much
air
we
I
think
we
had
that
in,
but
then
we
changed
it
at
some
point,
because
people
said
that
there's
a
lot
of
repetition,
okay,
but
I-
think
we
could
make
a
on
all
nodes
in
brackets
on
the
primary
cluster,
similar
to
what
we
have
yeah
like
I'll.
Just
make
a
note:
okay,.
A
A
A
B
A
C
A
That
from
from
succulence
right,
maybe
yeah
I
have
like
a
really
simple
webpage
and
stuff
I,
really
like
I,
really
like
the
style
they're
using
like
no
hassle
just
to
the
point
yes
Haley
really
enjoy
that.
So
next
step
is
to
the
reconfigure
it
skipping
the
post
appointment,
my
creations:
let's
go
I'm
keeping
the
keynotes
admin
page
open
to
see.
If
anything
changes
there.
A
A
A
B
A
A
C
A
C
A
A
Cannot
remember,
maybe
it's
so
we
have
the
application
wall
here
and
that's
that's
according
to
documentation,
but
I
would
expect
that
an
application
wall
would
not
her
yeah.
So
we
have
to
double
check
that,
but
I'm
not
gonna,
restart
port.
Well,
wouldn't
matter
anyway,
because
it's
not
connecting
to
this
PostgreSQL
anyway,
so
I
so.
C
C
A
B
A
Well,
when
we,
when
we
did
to
the
apt,
install
I,
get
installed,
it
did
we
start
unicorn
and
sidekick,
but
I'm
not
sure,
that's
always
the
case
mm-hm,
but
if
that
would
not
be
the
case,
I
think
we
would
better
like
restart
them
on
the
deploy
node
first
and
then
do
the
upgrades
on
the
other
nodes.
One.
B
B
A
A
C
A
A
A
A
A
A
A
A
C
A
B
A
B
A
A
A
B
B
A
A
B
A
Detracted
any
issue
somewhere
because
four
deployments
and
get
lots
of
counter
doing
a
bunch
of
checks
whether
they
need
to
do
help
or
not,
so
it
check-in
and
a
post
quest
and
reconfigure
locks.
If
this
line
is
in
the
logs
and
they're
checking,
also
I
think
they're,
comparing
like
process
ID
from
the
unicorn
they
had
before
they
did.
We
configure
an
accident
and
see
if
it
changed
and
if
it
didn't,
then
they
still
do
the
hop
from
from
the
ansible
playbooks.
B
A
Now
I'm
thinking
about
it,
but
that's
probably
a
bigger
problem.
It
like
the
whole
AJ
stuff.
So
when
we
saw
like
we
totally
did
have
that
upgrade
the
first
application
note.
We
saw
that
the
second
note
wasn't
really
healthy
and
yeah
I
guess
our
health
checks
should
be
smart
to
take
that
up.
But
if
you
then
have
the
load
balancer
not
sending
any
traffic
to
the
application.
B
A
A
B
B
Necessary
is
not
necessary,
but
if
I
think
I
like
the
idea
of
removing
it,
because
it's
already
long
and
I
mean
this
is
not
really
helping.
Anybody
make
a
decision
right,
it's
like
if
it's
not
strictly
required,
but
we
still
want
it
to
be
done
and
have
to
do
it
in
a
way.
And
then,
when
somebody
really
wants
to
know,
we
say
like
well,
it's
not
strictly
required,
but
you
know.
A
A
B
A
A
A
B
C
A
A
C
A
B
A
B
A
A
A
A
It's
it's
weird
because,
like
you'd
like
to
see
here,
giddily
only
giggling
notice,
also
how
to
it
automatically
so
I
would
expect.
Like
kidnap
reconfigure
knows
when
to
help
things.
Yeah
I
mean
that
makes
sense
in
a
way
so
why
these
instructions
are
still
here,
I'm,
not
sure,
but
we've
seen
these
I've
seen
this
one
so
that
cool
so
I
think
we
could
skip
it
for
now
and
move
on
with
the
post-deployment
migrations
and
checks
on
the
deploy
node
on
the
primary
cluster.
D
B
A
B
A
B
Something
in
between
not
healthy,
not
unhealthy,
but
something
yeah.
It's
like
I
think
that
the
user
interfaces
essentially
says
things
are
problematic,
but
they're
not
really
problematic,
because
we
are
in
the
middle
of
an
upgrade
right
and
I.
Think
that's
that's,
maybe
also
potential
improvement
to
say,
like
hey,
you
are
you're
upgrading
at
the
moment.
You
know
like
this
is
not
the.
This
is
not
unnormal
yeah.
A
B
B
A
A
B
A
A
A
B
Kidnap
about
what
well,
if
I,
see
this
now
I'm,
like
the
exceptions
everywhere
that
I
don't
understand
right
mm-hmm,
but
actually
it's
fine.
So
if
you
run
this
on
the
given
B
note,
it
should
either
not
work
and
say
like
hey.
This
is
not
available
on
a
get
to
you.
Node
or
you
know,
actually
produce
a
different
output,
because
this
I
would
say
this.
This
would
result
in
a
support.
Ticket
right
people
would
say
I
rent
this
I
got
all
of
these
exceptions.
A
A
A
That's
still
interesting,
I
haven't
seen
this
one
before,
but
I
just
consider
it
a
hiccup
I'm
not
really
blocking
for
this
moment,
so
the
documentation
makes
sense.
This
point
that
it
does
not
do
this
on
all
notes
which
yes,
one
thing
for
the
deploy
note:
how
are
people
selecting
a
deploy?
Note
I
think
it
can
be
any
node
yeah.
That's
that's
not
entirely
true,
because
you
cannot
do
it
from
the
giddily
note,
because
you
don't
have
a
database
connection,
for
example,
so.
A
B
A
C
B
B
A
Yeah,
we
don't
have
it
at
this
moment,
but
I
know
Stan
note
somewhere
that
when
you
want
to
the
migration
should
also
need
to
run
that
on
a
machine
that
has
direct
access
to
the
Postgres
database.
Of
course,
if
you're
using
a
load
balancer
for
Postgres,
like
you
like,
or
a
red
line
repair
yeah
daddy,
then
that.
A
A
A
A
A
A
B
A
B
B
D
D
B
C
A
Maybe
we
should
do
it
on
the
Postgres
note
itself,
I
think
that's
yeah
the
thing!
That's
the
problem,
so
it's
so!
This
fourth
class
note
is
still
running
611,
96
11,
but
the
library
for
happy
w
was
updated
and
it's
I
think
it
loads
it
during
run
time
and
then
it's
there's
a
version
which
match
match
between
the
Postgres
version
that
is
running
and
a
post
quest
and
the
library
version
that
is
installed.
Yeah.
A
You
mean
there
the
refresh
table
command,
yes,
yeah,
that's
true,
but
I
was
wondering
whether
I
need
to
restore
it
to
Postgres
node
or
the
tracking
database
and
I
think
it
should
be
the
dragon
database,
because
it's
that
one,
that's
loading
in
the
data
from
the
main
database
and
it
made
database
doesn't
really
know
aw.
Okay,
but
let
me
run
it
first
without
doing
the
tracking
database.
If
it
fails
again,
let's
switch
towards
the
tracking
database
and
if
Stan
fails
then
I
don't
know.
D
B
A
Micro
version
or
patch
version-
oh
yes,
okay,.
A
B
I
think
then,
like
I,
think
there
are
two
things
to
follow
up
here
right.
We
we
have
no
documentation
on
this
behavior
and
skipping.
The
Postgres
version
actually
also
like
the
skipping.
The
Postgres
version
update
does
not
actually
skip
upgrading
the
Postgres
version,
which
is
why
we
are
now
in
the
situation
where,
essentially,
we
are
requiring
a
restart
of
the
tracking
database,
and
that
is
not
something
that
should
happen
right.
I
would
argue.