►
From YouTube: Ceph Testing Weekly 2018-09-12
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).
B
B
Be
here
today
later
next
week,
oh
man
I,
hope
he's
okay,
his
family's
alright,
it
seemed
too
critical,
just
very
urgent
sure.
Okay,
I
wasn't
sure.
If
we're
gonna
get
anyone
else
or
not,
I
don't
have
a
ton
for
this.
I've
been
working
on
downstream
testing
stuff,
mostly
last
week.
So
I'll
talk
about
here.
Really:
okay,.
D
C
D
C
D
C
C
D
C
Yeah
I
just
landed
this.
You
merge
this
I,
remember
you
or
Gregor
in
there's
this
ticket
yeah,
but
the
problem
is
that
the
workers
and
by
I
mean
the
this.
This
in
each
file
is
for
two
toll,
Japan's
tag
and
CLI,
and
it's
deploying
everything
within
the
I
mean
all
services
within
the
one
user
to
told
you,
oh
no,
it's
using
Ubuntu
and
deploys
and
this
init
file
is
managing
all
of
them
and
parallels
papito
workers.
Everything
in
one
script.
C
C
E
But
cute
what
Kier
is
working
on?
He
explained
it
to
me
yesterday.
It
things
he
what
he's
working
on
is
an
ansible
recipe
that
would
deploy
a
tautology,
a
full
tooth,
ology
environment
in
OVH,
which
seems
to
me
to
be
a
hugely
useful
thing,
because
we
could
build
on
that
build
integration.
Tests
on
that
yeah.
C
D
There
is
kind
of
like
a
last
mile
issue
where
it
it
it
and
get
you
a
hundred
percent
of
the
way
there,
but
it
could
be
extended
like
if
we
wrote
if
we
had
if
we
had
a
service,
that
that,
for
example,
managed
teeth,
ology
worker
processes,
the
tooth
ology
role
that
we
have
in
SEF,
CI,
Mensa
bloke,
could
be
extended
to
say,
tip
system,
D
unit
file
for
the
service
and
and
tell
the
system's.
You
know
to
enable
it.
So
we
what
I'm
getting
at
is.
D
We
could
use
CFC
amenable
for
this
from
from
what
I
understand
to
be
the
goal
we
wouldn't
have
to,
but
but
yes,
FC,
amenable
isn't
isn't
quite
just
used
for
provisioning
or
machines.
It's
it's
kind
of
a
collection
of
roles
that
that
we
we
keep
all
of
the
I.
Don't
think
this
is
a
lie.
I
think
we
keep
all
of
the
the
ansible
that
that
is
relevant
for
managing
our
lab
in
that
repository
and
things
like
paddles
and
people
hold
each
other
in
psychology
or
part
of
that
I.
E
E
D
E
D
Yeah
we
we
certainly
assume
that
that
D
s
exists
in
in
the
lab
that
we're
deploying
on
said
I'm,
not
sure,
I
kind
of
doubt.
If
there
are
a
lot
of
other
dependencies
that
could
cause
problems
there,
we
we
were
pretty
careful
to
when
we
started
the
AFC
mantel
project,
because
we
didn't
use
ansible
forever.
D
Well,
we
started
using
it.
We
put
a
lot
of
effort
into
keeping
the
configuration
separate
from
the
actual
machinery
of
deploying
things
and
the
the
intent
there
was
that
it
could
be
used
by
by
other
labs
and
when
we
we
do
use
it
in
multiple
labs
right,
because
we
have
the
sepia
one
and
we
have
the
the
Red
Hat
internal
lab
that
we
use
for
a
lot
of
downstream
stuff
and
AFCI
mountable
manages
both
of
those
and
the
difference
between
the
way
those
two
deployments
work
is
they
have.
Each
lab
has
a
different.
E
Right
possible,
but
is
it
feasible,
for
example,
when
we
you
know
we're
working
on
this
right
now?
Our
stated
objective
is
to
is
to
get
this.
You
know
to
be
deploying
pathology,
you
know
in
OVH,
and
let's
say
you
know
we
did
all
of
this
work
and
we,
you
know
open
15p
ours.
You
know
to
step
C
and
hence
able
to
make
it
work.
You
know
it
is,
isn't
there
going
to
be
some
trepidation?
You
know
on
your
part
to
merging
pr's,
because
you're
testing
am
ansible
is
so
important
into
your
infrastructure.
C
C
C
D
Yeah
I
mean
certainly
tools.
You
know,
schools
that
don't
make
sense
to
be
written
in
ansible
can
can
be
written
in
Python.
Absolutely
those
tools
can
be
run
from
ansible
if
necessary
or
manually
or
some
other
way,
but
so
real
quick
to
answer
Nathan's
question
about
how
worried
we
would
be
about
merging
FC
metz,
we'll
pull
requests,
I.
D
C
D
E
D
No
I
I
feel
similarly
I
think
that
it,
it
might
be
better
if,
if
and
and
same
same
for
Pattinson
copito
in
you
know,
I
think
that
yeah,
when
we
started
doing
this,
there
didn't
seem
to
be
a
way
to
make
that
work
smoothly,
because
we
we
wanted.
We
wanted
to
have
a
repo
that
we
want
to
have
a
place
where
we
could
run
everything
from.
D
We
didn't
necessarily
care
if
you
know
there
if
these
roles,
if
their
primary
residence
was
in
one
repo,
but
there
wasn't
at
least
at
the
time,
a
way
to
do.
What's
the
what's
the
term
I'm
even
looking
for
here,
but.
D
Yeah,
there
wasn't
a
way
to
at
the
time
to
avoid
having
a
monolithic
repo,
but,
but
maybe
other
people
have
had
this
problem,
maybe
there's
a
solution
now.
I
would
certainly
at
least
personally
I'd
be
I'd,
be
really
open
to
it
like.
If
we
found
a
way
that
we
could,
we
could
continue
to
run
the
roles
in
the
way
that
we're
doing
now,
while
also
having
the
roles
actually
live
in
different
repos
in
the
projects
that
they
they
actually
deploy.
Oh
I.
E
D
That
makes
a
little
more
logical
sense
to
me
if
we
can
figure
out
a
way
to
to
do
those
things
and
still
have
a
way
to
run
like
if
there's
a
way
to
almost
like
stimulate
them
into
the
Ceph,
see
a
mental
repository,
but
not
actually
a
sibling.
You
know
describing
something
that
I
don't
even
know
if
it
exists.
Well,
we're
talking
about.
E
Two
different
things
that
I'm
talking
about
about
making
what?
Maybe
you
haven't
even
considered
that
in
in
in
the
interests
of
being
able
to
integration
test
the
system
as
a
whole,
so
it
makes
sense
to
have
the
whole
system
in
a
single
repo
in
the
whole
set
by
the
whole
system.
I
mean
tooth
ology,
thefts
and
ansible
paddles
poopie,
all
of
that
in
a
single
repo,
so
that
we
can
open
a
full
request
and
run
integration
tests
on
it
and
merge
it.
E
E
D
C
D
E
I'm,
just
thinking
of
it
in
terms
of
integration
testing,
from
a
practical
standpoint,
you
you
can,
you
can
have
separate
projects,
but
but
they
don't
necessarily
have
to
be
implemented
in
separate
repos.
They
could
be
implemented
in
separate
directories
right
in
a
single
repo,
but
I
mean
I.
Don't
want
to
have
a
long
debate,
you
know
if
it,
but
but
there
doesn't
definitely
would
have
to
be
some
kind
of
versioning.
You
know
so
that,
and
like
you
said,
I
mean
you
know
some
way
to
define
it.
E
You
know
yeah
like
which
version
of
both
veto
is.
Is
this
version
of
thorgy
designed
to
use,
and
then
it
can
just
install
that
version,
though
to
have
it
make
it
as
a
Python
module?
You
know
that
that's
also
of
a
way
to
do
that,
but
over
time
I've
come
to
appreciate
the
fact
self.
Is
this
huge
monolithic
thing
that
come
to
appreciate
the
stability
of
having
almost
the
entire
system
in
a
single.
D
B
D
E
E
E
D
D
C
That
we
need
to
provide
for
answerable
scripts.
At
least
I
didn't
know
if
you
guys
are
already
providing
this.
But
when
I
was
thinking
about
waiting
in
separate
ansible,
40.2
touji,
I
would
suggest
you
have
a
single
property
file,
not
Spirit
alone,
all
or
all
the
roles,
but
one
property
file
where
in
listed
all
the
versions
that
should
be
deployed
to
the
cluster.
E
Yeah
that
would
work
it,
but
we
have
to.
We
have
to
talk
then
about
versioning
right
in
all
of
these
different
repos.
So
each
each
repo
has
to
have
a
versioning
mechanism
and
and
then
a
way
or
or
if
we,
if
we
use
github.
If
we
use
a
single
master
branch,
then
we
have
to
have
a
tag,
no
way
of
tagging,
but
that's
versioning.
D
Yeah
we
we
can,
it
would
certainly
be
possible
in,
for
example,
the
ansible
rules
that
we
have
right
now
to
abstract
out
the
repositories
and
it
would,
it
would
also
be
possible
to
tell
you
know,
say
the
excuse
me:
the
ansible
get
module
to
check
out
a
certain
branch
or
tag
or
sha-1,
so
whatever
versioning
we
use,
we
can
make
ansible
check
out
that
version
I
mean
that
that
that's
going
to
be
fine
right.
Those
things
may
not
be
parametrized
right
now,
but
that
that
could
certainly
happen.
D
You
know
for
the
test
deployment
and
then
run
like
today.
It
would
be
this
FC
amenable
rules,
you
know
a
different,
you
know
in
the
future.
It
could
be.
You
know
the
rules
exist
in
the
repositories
of
the
projects
or
whatever,
but
but
the
point
is
we
could
maintain.
A
single
inventory
configuration
that
we
that
we
pointed
is
these
roles
that
wherever
they
live
and
it
could
contain
things
like
well
use
this
repository
use.
This
version
set
these
other
parameters.
C
C
C
D
So
the
the
way,
the
pattern
that
we're
using
is
is
we
have
the
Ceph
CI
mantle
repo
and
then
for
each
of
our
labs.
We
use
what
we
call
a
secrets
repository,
which
contains
the
inventory.
We
called
it
secrets
because
we
just
wanted
to
remind
ourselves.
Whenever
we
talked
about
the
repos
that
there
there
are
encrypted
secrets
there,
there
happen
to
be
encrypted
secrets
in
them.
Those
are
not
related
to
tooth
ology
or
paddles
or
papito,
except
for
maybe
there's
like
a
database
password
somewhere,
but
the
the
encryption
of
anything
is
optional
in
these
things.
D
C
D
C
Is
that
this
it
is
currently
not
usable
for
newcomers
because,
let's
say
you're
a
newbie
and
you
want
to
set
a
bit
collagen.
So
what's
your
steps
to
you,
you
set
up
your
own
instant,
its
disaster
because
you
need
so
many
pieces.
It's
it's!
It's
not
just
about
just
hey
check
out
Tala
G
and
then
run
some
script
and
it
will
roll
up
all
the
all
the
set
up
with
everything
yeah.
D
C
C
And
this
is
very
related
to
the
to
the
integration
tests,
because
integration
test
shouldn't
be
so
complicated
for
setting
up
environment
because
it
just
need
to
run
some
basic
command
to
run
everything
for
you
and
I
I
understand
it's!
It's
not
possible
right
now.
It
you
make
it
so
easy,
so
I
mean
we
have
to.
C
C
C
C
C
D
C
E
D
E
E
D
Yes,
totally
I
I
think
it's
I
think
it
would
be
a
good
idea
to
look
at
at
least
using
pip
to
install
these
things,
and
if
we,
if
we
we
do,
need
further
action
to
be
taken,
then
we
can
start
doing
that
in
ansible.
You
know
I'm
not
married
to
the
idea
of
you
of
using
git
clone
to
deploy
things
right,
so
so
yeah,
let's,
let's
look
at
making
these
things
more
usable
via
pip
I,
think
I
totally
agree.
There.
C
C
Know,
but
there
was
another
idea:
I
don't
know
if
Nathan
was
sounded
it
not
about
packaging,
the
two
toge
itself
as
a
distributable
rpm
and
the
same
for
Poole,
Peter
and
pedals.
So
you
don't
need
to
check
out
them
to
the
you
know
to
the
separate
user
directories
you
just
can
install
it
to
the
system.
D
C
It's
actually
can
work
with
people.
Just
the
thing
is
that
we
can
yeah.
You
know,
I,
don't
strong
opinion,
how?
How
should
we
go
with?
You
know,
development,
setup
and
production
if
there
should
be
different
or
if
they
can
be
different
like
like
the
you
have
production
and
you
have
separate
users,
but
for
development.
It's
it's
not
so
important
to
have
different
users
right
and
you
can
everything
install.
We
keep
command
ins
in
virtual
environment
and
running
everything
there.
D
Yes,
certainly
for
testing,
we
could
all
use
we
could
use.
You
know
the
same
user
I
think
that
I
mean
whether
used
pip
or
our
p.m.
you
can
install
things
globally.
You
and-
and
you
can
then
choose
to
run
things
as
separate
users
or
not
if
it
makes
things
a
lot
easier,
maybe
for
testing,
we
run
everything
as
the
same
user.
I
don't
have
strong
ideas.
E
D
E
D
E
D
E
B
D
D
B
B
E
D
C
C
C
E
D
It's
it's
not
it's
not
just
cuz,
because
I
think
that
smells
better
or
anything
I
mean
picture
your
a
brand
new
you.
You
know
your
your
someone
who
they
just
started.
Looking
at
stuff
and
self
testing,
for
example,
and
you
just
started
looking
at
tooth
ology
right
now,
it's
I
think
it's
really
intimidating
I!
Think
it's
I
think
it's
more
intimidating.
If,
when
when
you
clone
the
repo
or
whatever,
now
you
have
every
service,
that's
even
related
to
tooth
ology
and
and
and
it
feels
like
you
have
to
understand
them
as
well,
but
you.
E
Do
have
to
understand
them
Zach,
you
can't
run
tooth
ology
without
paddles
and
poopy
toe
Canyon,
or
that
gets
actually
gets
to
my
my
question
about
interactive
a
couple
lines
down
in
the
minutes.
Is
there
a
way
to
run
to
theology
kind
of
in
a
minimal
way
without
without
the
workers
and
without
sweets
that
can
you
run
a
single
job?
That's
a
something
that
this
idea
of
running
a
single.
C
Probably
initially
you
could
run,
but
right
now
the
worker
I
mean
only
worker
which
will
just
can't
accept
fake
sweets
from
installed
II,
but
whenever
it's
run
it
tries
to
connect
to
pedals,
and
if
pedals
is
not
running,
it
will
break
hold
the
test.
The
it's
not
possibly
right
now
to
run
any
tests
without
pedals.
E
D
E
Then
the
the
dashboard
stuff
in
the
in
the
database
stuff
and
all
of
this
heavy-duty
lab
stuff
can
be
on
top
of
that,
as
where
you
you
want
to
set
up
your
own
lab,
okay.
Well,
that's
the
next
step
after
you've
started
running
to
ecology,
this
minimal
tooth
ology,
but
that
doesn't
exist
yet
is
mine.
My
understanding,
yeah.
B
C
C
B
E
Yeah,
so
we
could
we
could
fit.
We
could
continue
the
integration
testing
I
mean
we
have
an
actionable
conclusion.
I
know
how
to
I've
done
it.
Isaiah
is
Asian
before
so.
That's
something
I
could
conceivably
work
on
so
I'm.
Happy
that
we're
moving
this.
This
conversation
has
moved
us.
You
know
one
centimeter
closer
to
integration,
testing,
awesome.
B
Okay,
so
interactive:
oh
no,.
C
E
B
B
Well,
I
mean
you
can
do
anything
with
it
when
the
when
it's
being
run
by
it's
by
a
worker
process,
right,
yeah,
a
schedule.
Job
can't
be
interactive
right,
yeah,
that's
what
I'm
saying
dude
ology
sweetie
takes
it.
Then
it's
an
an
error
that
it
got
plumbed
through.
Somehow
I
mean
I'm
Nathan.
Then
we
didn't
have
suite
or
scheduling
we
just
liked
brand
tooth
ology
with
you
know:
yeah
mold
commands
I'm
with
you
mo
fragments
on
the
command
line,
and
then
it
went
out
and
did
that
from
like
the
bash
shell.
You
were
running
and.
B
D
D
D
D
So
that's
the
it's
a
job
configuration
to
here
that
you
would
pass
it
yeah
you'd
pass
it.
You
know
the
full
job
configuration
and
then
also
pointed
out
the
targets
which
can
be
embedded
in
the
job
configuration
optionally
right.
You
can
pass
an
arbitrary
amount,
a
number
of
EML
files
that
just
get
merged
together
so
that
target
stanza.
D
We
tend
to
keep
separate
just
you
know
for
convenience,
but
it
could
be
l1
and-
and
you
do
this
from
a
terminal
and
when
it
you
know,
if,
if
you're,
you
could
be
using
the
interactive
tasks
or
the
interactive
on
failure
or
I
forget
what
that
flag
is
called
option.
Two,
you
know
if
it
hits
an
unhandled
exception.
It'll
drop
you
into
the
prompt,
but
then
the
job
runs
until
the
point
at
which
it
wants
to
be
interactive
and
then
there's
your
interpreter.
E
D
D
B
D
B
D
Then
use
interactive
on
I
forget
if
it's
interactive
on
error,
I
think
the
check.
If
you
want
it
to
go
interactive,
always
at
a
certain
point,
then
you
insert
the
interactive
task
into
the
the
tasks
list
at
the
point
where
you
want
to
be
interactive.
So
if
both
then,
whichever
happens
first
will
happen.
So
if
you
get
an
error
it
before
the
interactive
task
is
invoked,
then
you'll
get
the
prompt
and
then,
when
you
exit
out
of
that
prompt,
you
should
also
get
another
prompt
when
the
interactive
task
is
invoked.
C
C
So
the
urine
test
question
regarding
reading
tests,
so
I've
been
trying
to
run
it
by
27
talks,
target
and
I
figured
out
that
the
talks
has
a
bug
and
you
can't
run
it
in
from
virtual
environment
because
it's
branchy,
we
use
some
components
from
the
base
system
instead
of
the
ocean
environment
there.
But
later
I
was
trying
to
do
this
on
on
Stan
and
openSUSE
as
well
too.
C
Finally,
I
got
some
successfully
that
I
was
able
to
run
at
PI
27
on
upstream,
and
that
I
found
that
there
are
a
lot
of
a
lot
of
failures,
but
they
are
skipped,
but
anyway
they
are
trying
to
connect
with
Shimon,
etc.
It's
not
kind
of
unit
test.
It
requires
some
infrastructure.
Why?
Why,
if
I
want
to
run
unit
tests,
it's
it
wants
to
connect
to
some
other
resources.
So
you
probably
need
to
play
this
in
two
separate
Suites.
D
A
C
D
Yeah,
well,
it's
hard
to
know.
It's
I'd
like
to
see
what
you're,
seeing
because
one
one
thing
that
you
might
be
seeing
is
some
tests
do
things.
They
know
won't
work
to
verify
that
they
won't
work
right.
So
there
are
tests
that
generate
error
messages
and
then
pass
and
that's
correct
behavior.
You
might
also
be
seeing
bugs
in
tests,
but
but
but
it's
hard.
D
E
D
C
E
E
D
D
E
So
but
but
last
week
you
were
on
the
station's
active
I
was
we
were
discussing
this
and
I
and
I
told
David
already
that
404
the
OVH
and
we
generate
OVH
images
for
use
with
thorgy
OpenStack,
so
I'm
sure
that
you
know
we
can
provide
the
openSUSE
image
OVH
image
to
him
and
he
could
you
know,
put
it
wherever
it
needs
to
be
put
and
and
then
tried
to
use
it,
and
probably
that
would
get
us
90%
of
the
way.
Oh
yeah.
E
E
D
Make
sure
he
knows
that
that
if
he
needs
some
help,
he
he
could
ask
you
and
you
might
be
able
to
help
ya.
Setting
up
a
fog
image
is
mostly
just
getting
a
machine
installed
and
then
and
then
creating
an
image
based
on
that
installed
machine.
So
I'm
sure
you
can
figure
out
how
to
install
openSUSE
yeah
butBut.