►
Description
The seventh installment of our SIG group that’s dedicated to reviewing the challenges that come with building, deploying, maintaining, supporting, and using CMSs on Kubernetes happens IRL from a Birds of a Feather (BoF) session at DrupalCon Amsterdam.
Catch up with the group on GitHub: http://bit.ly/338dXC5
A
B
So
yeah
I
guess
we
can
go
ahead
and
start
for
those
of
you
that
aren't
familiar.
We've
been
doing
about
every
two
weeks,
not
completely
consistent
with
that,
but
have
been
trying
to
get
together
through
the
drupal.org
slack
channels
to
talk
about
issues
related
to
Drupal
and
communities.
We've
had
a
lot
of
conversations
so
far.
You
know.
A
lot
of
the
initial
ones
were
focused
around
file:
storage,
optic,
storage,
being
able
to
work
with
files
and
Drupal
what
that
means.
B
We
had
you
know
some
conversations
about
build
packs
which
had
different
lightning
talks,
but
really
it's
just
getting
off
the
ground.
So
they
said
it's
an
open
forum
for
anybody
that
would
like
to
participate,
not
really
sure
how
to
moderate
this
I
don't
have
an
agenda.
So
one
of
the
things
that
we
like
to
do
is
maybe
you
know
it
give
people
an
opportunity
to
introduce
themselves
if
you're
interested,
maybe
tell
us
a
little
bit
about
how
you
use
kubernetes.
B
Some
of
the
challenges
that
you
face
with
kubernetes
there's
a
lot
of
people
with
common
interest,
our
common
knowledge
and
different
levels
of
experience
with
hosting
to
the
communities.
All
the
way
up
through
production
environments,
so
does
anybody
want
to
jump
in
and
maybe
say
hello
to
the
group,
maybe
maybe
throw
out
a
couple
of
questions
some
of
the
new
I.
C
B
D
B
A
A
A
B
Way,
one
thing
that
we
try
to
do
that
might
need
a
little
bit
more
help
is
there's
a
github
page,
that's
linked
from
the
drupal.org
kubernetes
slack.
It
has
a
list
of
resources,
there's
not
a
lot
here,
but
that
the
way
that
there
are
some
early
patterns
that
are
starting
to
be
linked
and
as
we
open
source
more
and
more
things
specifically
with
build
backs
and
some
of
the
other
stuff
there
that
my
company
focuses
on.
That's
where
we'll
add
those
links
as
well,
so
it
might
be
a
good
place.
B
E
C
C
I
created
Buster
film
day
gives
a
measurer
subscription,
with
a
buster
name
space
on
the
inside
apostolate
I
configured
it
until
first,
amid
the
Canadians
many
says
and
I
tournament
to
help
deployed
him
to
give
up,
but
I've
been
busy.
I
mean
working
on
that
for
the
last
three
weeks
and
I
learned
about
this
before
that
I
didn't
believe
all
about
this.
C
C
B
We
we've
found
that
it
is.
You
know
there
is
that
learning
curve
and
it
does
provide
a
lot
of
functionality,
but
there
also
becomes
a
point
which
specifically
looked
on
to
where
you
can
get
really
difficult
to
debug
issues
or
when
something
breaks.
It's
hard
to
figure
out
sometimes
I
know,
there's
a
lot
of
hope
and
excitement
around
comma
3,
but
generally
I
know
in
our
systems.
We've
kind
of
straight
away
from
started.
Moving
more
towards
that
operator,
fuller
matter
and
use
different
templating
engines
and
items
like
that.
B
B
B
B
C
C
I
will
add,
I
think
we
need
things
that
had
most
distal
active
service
levels,
the
previous
traditions
that
caused
a
lot
of
problems
with
management's
that
we
slice
ISM.
Let's
stress
so
yeah
three
three
I'll
see
what
I
think
is
released
overnight.
So
so
it's
pretty
close
to
actually
stable
release
from
three
which
is
to
us
has
been
a
security
patterns
motion
in
Florence,
but
there's
still
some
but
can
be
difficult
to
manage,
is
easy
to
set
up
and
run
from
that
upgrade
process.
Cackles.
C
Yeah
I'll
get
to
actually
try
help
three,
but
okay,
there's
no
challenges.
Mike
Mike
room
is
typically
the
operator
model
managing
things
by
OCR.
These
can
in
that
way,
yeah,
but
that's
another
alternative,
but
you
have
to
have
an
operator
that's
running
in
your
cluster.
That
often
has
a
specification
rather
custom
resource
definition
that
you
can
actually
just
manage
your
deployments
through
this
through
kubernetes
native
approach
to.
C
So
we
need
you
to
bounce
we've
actually
been
having
something
or
a
love-hate
relationship
with
film,
but
actually
a
lot
of
success
as
well.
One
of
the
things
that
so
yes
templates
in
Yahoo,
like
llamó,
is
one
of
you,
the
worst
languages
or
markup
that
you
have
and
then
making
that
remembering
it's
very
messy,
but
at
the
same
time,
there's
this
helpful
game
called
unionist
and
supreme,
which
we
have
unit
tests
for
every
configuration
option
that
we
have
to
say,
I
want
this
option
and
then
with
these
values.
C
This
is
what
I'm
expecting
these
are
values.
This
is
what
I'm
expecting
it's
been
like,
making
everything
so
much
easier
to
refactor
to
make
sure
like
just
catch
some
indentation
issues
or
things
like
that,
so
we
and
it's
not
actually
that
much
working.
So
with
the
proper
measure,
you
can
actually
get
something
else.
What
use
a
lot
of
it
and
I
guess
one
of
the
things
that's
pretty
nice
is
the
way
that
you
can
separate
the
structure
of
your
files
from
the
kind
of
configuration
options
that
you
give
to
your
users.
C
C
We've
had
some
configuration
for
that
and
later
on,
he
relies
and
that's
our
hidden
important
first
SSL
Certificates
stuff
that
we
wanted
to
do
and
we
switch
out
ambassador
for
traffic
and
didn't
like
we
had
nothing
to
change
further
project
configuration
yeah.
So
that's
a
part
where
it's
actually
pretty
nice
to
have
this
kind
of
abstraction.
That's
like
completely
unrelated
to
good
the
resources.
I
can
Janica.
F
Monitoring
yeah
well
defer
I,
have
to
my
mandate
is
to
do
a
lot,
but
very
little
because
I'm
kind
of
like
a
one-man
band
inside
of
a
organization,
that's
doing
a
lot
of
other
things.
So
we're
we're
just
four
on
gke
and
you
get
stackdriver
basically
free
with
it,
which
is
not.
You
know
it
is
what
it
is,
but
the
nice
part
is
that
you
I
think
by
and
large
many
people
who
are
running
through
hole
don't
actually
monitor
their
application.
Much
at
all,
that's
you
know.
F
F
So
I
think
like
baby
steps
just
to
like
turn
off
DB
log
turn
on
like
monologue,
and
it's
not
entirely
obvious
but
like
if
you,
if
you
don't
JSON
onto
a
single
line
in
your
container
law,
stackdriver
will
parse
it
so,
like
you
know,
it's
very
easy
to
go
from
like
dumping,
something
to
the
container
console
and
then
having
having
stackdriver
consume
that
that
it's
structured.
So,
like
that's
a
that's
a
baby
step,
but
we've
run
into
like
we
run,
we
sort
of
go
anti
patter
and
run
cron
inside
of
our
containers.
F
It's
it's
just
easier
to
ship.
One
container,
it's
a
little
anti-pattern
to
where
you're
supposed
to,
but
we've
had
to
do
a
little
bit
of
extra
sort
of
happiness
to
get
those
logs
to
surface
to
the
container
or
because
they're
running
under
prod,
but
it's
totally
doable
and
it
provides
provides
my
client
like
way
more
visibility
than
they
ever
had
before.
Either
worker
do
more
in
aggregating
them.
C
C
D
B
E
E
E
F
F
C
C
We
actually
we
we
try
to
use
that
and
also
had
issues
and
first
we
wanted
to
be
able
to
migrate
existing
slides
use
new
setup
without
having
to
modify
the
application
itself
so
fly
system.
For
example.
We
didn't
want
to
do
it
because
right,
okay,
well
in
also
also
for
people
7,
it
works
differently
and
so
on.
C
So
what
we,
but
we
wanted
to
use,
object,
storage,
and
so
we
there's
a
project
called
part,
one
which
makes
it
possible
to
mount
multiple
cloud
storage
things
like
anything
from
s3
to
Dropbox
or
really
weird
things
like
that,
and
we
wrote
a
small
CSI
driver.
Thank
it
just
a
wrapper
around
it
and
it
makes
it
possible
to.
C
No,
no,
it's
not
thank
you.
So
it's
a
CSI
driver
and
actually
quite
minimal
performance.
It's
I
mean
it's
nothing.
I
think
it
doesn't
work
for
great
for
very
large
files,
but
we've
seen
it
actually
writing
and
reading
it's
just
fine.
It's
similar
to
you.
You,
like
the
hoser
to
the
NFS
solution
from
google
cloud
and
and
anyway,
both
that
is
going
to
be
cached
by
varnish
and
by
the
CDN
and
so
on.
So
we
decided,
but
actually
it's
quite
overview
for
now
and
yet
so
it's
it's
we're
the
only
ones
using
it
so
far.
C
B
F
F
Yeah
I
mean
I,
think
I,
don't
know
any.
Maybe
you
know
we're
dude
we've
got
probably
maybe
around
100
clients
kind
of
in
total
and
like
it's,
not
a
commodity
product,
so
I
or
not,
I'm,
not
worried
about
like
isolating
people
as
much
as
maybe
other
girls
right
like
as
I
really
yeah,
but,
like
you
know,
our
pods
are
relatively
long
living
compared
to
a
lot
of
others
like
it's
still.
One
posted-
and
you
know
it's
in
like
we're.
F
Not
most
of
our
clients
are
not
scaled
because
their
applications
aren't
each
other
they're,
not
paying
us
for
that.
So,
like
I,
you
know
sometimes
I
think
you
can
over
engineer
some
of
that
right,
like
you
could
totally
over
engineer
the
resource
allocation.
So
like
I
are
taking
the
approach
to
boy.
We
still
monitor
node
level
metrics
and
if
I
you
know,
and
then
I
aggregate
by
namespace
so
like
if
I
see
a
noisy
neighbor,
it's
pretty
easy
to
figure
out
who
the
boys
neighbors.
C
Lived
on,
that's
actually,
because
you
right,
so
you
know
one
of
the
things
that
we
do
with
our
H
a
clusters
we
always
usually
across
three
availability
zones
and
basically
your
social
search
percentage
would
be
read
long.
That
was
quite
the
honor
code
using
so,
and
so
that's
where
we
sort
of
them
to
play
that.
So
there
really
is
my
stage
we
do
have
one
of
those
too.
C
You
notice
that
there
was
also
something
for
us,
really
big
learning
like
making
sure
that
you
understand
what
those
limits
mean
for
CPU
and
from
memory,
because
they
completely
differently
and
and
also
like
making
sure
that
you
treat
things
so
like
the
BGP
application,
certain
requests,
its
users,
a
bunch
of
memory
and
release
memory,
and
this
is
fine
and
then
there's
a
data
base
which
will
keep
that
memory.
And
if
you
set
the
limit
by
order
comic
loop,
it
will
eventually
reach
that
limit.
C
It's
no
matter,
and
so
the
the
kind
of
strategy
that
we
have
for
anything
that
will
keep
that
memory
is
to
set
to
the
request
and
the
limits
to
do
this
value.
Yes,
and
this
way
you
know
what
it
has,
there's
no
kind
of
different
things
like
different
different
database,
limited
and
requesters
are
trying
to
follow
them
to
get
to
get
something
and,
at
the
same
time,
per
CPU
being
mostly
having
much
more
like
set
the
limits
very
well,
their
requests
very
low
and
the
limits
very
high.
C
C
C
B
C
F
Think
I
think
that
conversation
recently
that's
been
forcing
a
lot
of.
Is
this
good
business
conversations
with
clients
because
it
makes
them
it
makes
your
clients
get
very
real
about
what
their
expectations
are.
I
have
a
I
have
a
conversation
most
recently
with
the
client,
but
like
they're,
trying
to
score
work
right
for
a
product
that
we
manage
for
them
and
they
want
all
kinds
of
high
availability
and
they
want
promises
on.
F
Want
all
sorts
of
RPO
and
RT
of
guarantee
tense,
and
it's
like
it
makes
it
real.
You
can
engineer
that
in
kubernetes
pretty
easily,
you
know
it's
just
they
have
to
be
willing
to
pay
for
it.
You
know
I
think
this
is
all
pulled
the
you
know.
The
pool
is
no
longer
for
anybody's
eyes
about
like
what's
behind
the
veil,
and
it's.
F
F
Know
that,
but
we've
been
able
to
like
more
like
the
you
have
to
have
a
conversation
about
what
the
ceiling
is,
but
also
like
the
floor
has
come
up
quite
a
bit
for
us,
like
we
all
of
our
clients,
now,
like
our
own
regional,
persisted
this
so
like,
even
if
they're
only
running
in
one
zone.
At
the
time
like
if
that
node
goes
down
in
the
middle
of
the
night,
it
just
gets
rescheduled
to
a
different
zone
and
like
they're
getting
that
for
free
three
three
eyes.
I
think.
C
When
he
predicted
cluster
there's
also
a
few
things,
that's
all
had
already
come
in
the
classroom
to
DNS
metrics
and
these
kind
of
things
and
they,
but
each
of
them
have
a
different
way
of
auto
scaling
and
setting
resource
dynamically,
and
you
found
a
way
to
manage,
hold
them
in
a
consistent
way,
because
it's
like
the
resource
that
I
said
it's
kind
of
scales
with
the
clustered,
but
we've
seen
that
it
wasn't
always
sufficient,
and
sometimes
especially
with
these
kind
of
life
support
services.
You
just
want
to
make
sure
that
they
won't
be
good.
C
C
C
C
A
Want
to
hear
from
you
more
Drupal
or
in
deployment
on
like
what
is
your
experience
or
your
best
practice
communities,
whereas
any?
How
do
the
deployment
of
website
of
Google
website
plus
doing
the
day,
one
and
day
two
operation
like
upgrading
the
very
own
backups?
First,
our
calcium,
you
have
any
best
practice
recommendation
tool,
I.
F
Mean
I
would
just
like
blue
touch
right,
like
take
as
much
of
that
out
of
band
as
possible
is
because
you're
putting
in
just
because
you're
putting
Drupal
and
if
your
fornetti
is
like
doesn't
mean
that
you
have
to
change
like
your
backup
machines.
Very
like
do
what
works
for
you
like.
If
you're
using
planet
aquatic
you
all-bran
art,
then
you
don't
need
point
timer,
but
really
let
it
take
care
of
doing
a
daily
backup
right
or
what
we're
doing
for
other
backups
is
like.
We
have
other
services
that
do
backups.
E
C
Back
up
the
the
database
once
like
their
mind,
might
be
something
like
you,
don't
even
care
about,
and,
and
we've
seen
that
actually
we
know
it
that
different
like
snapshots
and
things
like
that.
In
the
end
we
just
created
con
job
it
does
it
a
good
old
bicycle
done.
We
have
control
of
gradients,
we
know
exactly
what
it's
doing
mix
it
up
and
put
it
into
objects,
orange
and
same
thing
for
pause,
simple:
it
works.
You
can
just
like
open
or
open
up
the
object,
search,
browser
and
look
at.
E
C
A
How
do
you
keep
this
thing?
Company
just
arresting
first
in
attendance,
backups
a
fire
starts
right.
How
do
you
keep
seeing
with
the
database
to
be
on
the
same
level
that
the
way
it's
coming
in
a
file
already
added
by
the
backup
of
the
files,
because
it
was
attributed
six
hours
before
or
do
you
just
give
the
execution
running
at
the
same
time,
yeah?
So
in
the.
C
C
C
A
B
I,
unfortunately,
where
we're
at
I
highly
encourage
everybody
to
hop
into
the
Drupal
or
kubernetes
slack.
If
you
have
questions
that
you
come
up
with
later
or
if
you
want
a
good
group
of
people
to
talk
to
I
mean
this.
This
is
it.
These
are
the
people
that
are
working
with
kubernetes
and
Drupal,
at
least
in
Europe,
and
there's
a
lot
of
really
smart
people
in
the
room,
so
encourage
you
to
take
advantage
of
that.
It's
been
nice
meeting
everybody
and
thank
you
very
much
by.