►
From YouTube: Grafana Mimir Community Call 2022-04-28
Description
Second Grafana Mimir Community Call, in which we discussed coming features in Mimir 2.1 and asked the community about their issues with getting started with Mimir.
Join us on the last Thursday of every month. Please see the linked document for meeting notes, agenda and details on how to join: https://docs.google.com/document/d/1E4jJcGicvLTyMEY6cUFFZUg_I8ytrBuW8r5yt1LyMv4/view
A
A
So
I
will
just
list
some
of
them
and
I
know
some
people
who
are
working
on
those
features
are
in
this
course
in
this
call.
So
I
may
point
at
you
and
ask
for
more
details.
So
first
one
is
actually
about
the
ruler
in
the
past
ruler.
A
A
No,
so
this
feature
is
already
ready
and
it
will
be
part
of
mimir
2.1.
The
next
feature
that
oh
jesus,
I'm
sorry,
I'm
not
sure
how
to
pronounce
your
name.
Yes,
I
saw
satisfied
cases
works
on
is
a
support
for
out
of
order
samples,
so
jesus,
if
you
want
to
take
over
and
tell
us
a
little
bit
more
about
it.
Now
is
your
time.
B
Okay,
so
I
wanted
to
use
this
call
to
share
the
design
doc
that
we've
been
working
on
for
the
past
couple
of
months
is
public,
so
anyone
can
can
read
and
suggest
modifications.
B
I'm
super
excited
right
now
because
I
sure
did
not
do
this
on
purpose
but
like
we
are
ingesting
samples
now
in
the
depth
cluster
since
three
minutes
ago.
So
like
I'm
a
bit
yeah
looking
at
the
graphs
while
while
attending
this
goal
as
well,
yeah
and
yeah,
so
we
expect
to
have
support
for
out
of
some
out
of
order
samples
in
like
maybe
a
month
from
now
or
perhaps
a
bit
later
like
we.
B
We
need
to
be
exhaustive
with
tests,
but
just
let
everybody
know
that
we
are
working
on
this
and
yeah.
Any
suggestion
is
very
welcome.
B
So
yeah
the
design
the
design
we've
made
could
induce
samples,
like,
I
think,
and
I
said,
a
hundred
years
old,
but
we
are
actually
setting
an
allowance
parameter
in
mimir
and
also
permitted
to
promote
csdp,
basically,
which
is
called
the
allowance
and
and
like
you
said
it
is
exaggeration,
so
you
can
say
30
minutes
one
hour.
B
The
reason
why
we
do
this
is
because
we
want
to
be
careful
writing
out
the
for
the
samples
because
of
the
design
can
have
performance
like
a
performance
impact
on
on
the
database.
So
that's
why
we
don't
want
to
allow
very
old
samples
to
come
in.
It
could
do
it,
but
we,
I
think
we
think
it's
it's
more
common
to
have.
You
know
like
five
minute
old
samples
than
a
year
old
sample,
so
the
use
case
we
are
trying
to
solve.
B
C
Yeah,
I
was
wondering
if
you
could
share
more
details
about
the
testing
plan.
B
B
We
are,
I
think,
at
the
moment
we
are
sending
like
700
samples
to
the
cluster
with
ingestors
and
and
we
are
seeing
how
it
handles.
We
are
concerned
about
memory,
mapping,
memory
usage
and
the
compaction,
so
that's
things
we
need
to
look
at
very
carefully.
B
I
guess
in
the
following
days
and
and
then
the
other
thing
that
we
wanted
to
explore
was
this
issue
you
mentioned
about
adding
latency
in
the
investors
and
see
how
it
began,
how
it
behaves
and
also
steve.
The
other
day
came
up
with
mentioned
an
issue
about
that.
B
Sometimes
we
get
up
too
many
out
of
bound
samples,
and
that
causes
an
issue
in
mimir,
and
so
we
we
want
to
to
also
test
that,
like
like,
send
more
more
and
more
of
the
further
examples
that
go
far
far
in
the
time
and
see
and
see
how
this
is
handled.
Yeah-
and
I
mean
I
mean
if
anyone
has
any
suggestions
for
the
tests
they
are
welcome.
C
And
you
also
mentioned
that
you're
going
to
introduce
in
you
introduced
in
tsdp
an
option
to
configure
how
far
in
the
past,
we
can
ingest
how
to
order
samples.
I'm
wondering
if
there's
also
an
option
to
completely
disable
the
how
to
order
logic
or
if
it's
the
same
option
or
there's
no.
B
B
Yeah
then,
there's
also
a
an
interesting
detail
here.
Out
of
order
introduces
a
new
type
of
chunk
to
the
prometheus
dsdb.
So
that's
a
breaking
change
that
makes
newer
versions
not
compatible
with
old
versions.
So,
like
a
rollback
from
this
version
would
not
be
possible,
but
we
have
done
a
change
in
the
middle
where
we
we
made
prometheus
discard
unknown
chunks
in
case
it
finds
any
so
like
that's
a
step
in
the
middle
for
this,
and
it
introduces
the
possibility
to
roll
back
in
case.
There
was
any
any
trouble
that
we
didn't.
B
So
the
way,
what
can
you
repeat
the
question?
Sorry.
D
B
B
So
out
of
order
chunks
get
compacted
into
overlapping
blocks.
That's
that's
what
I
know,
and
I
don't
know
then,
when
blocks
get
merged
the
in
order
data
and
now
the
further
blocks
get
merged.
I
don't
know,
what's
happening
with
the
compressor.
B
So
the
way
you
the
compactor,
you
mean
the
comparator
component
in
beaming.
Right
that
reads
the
block
and
and
there's
it
yes,
but
it
actually
uses
the
prometheus
compactor
code
yeah.
So
the
way
this
is
yeah,
so
we
are
leveraging
on
that
compactor.
So,
basically,
when
when,
when
the
tsdb
compacts
the
the
memory
map
chunks,
it
will
generate
a
persistent
block
for
in-order
data
and
then
potentially
overlapping
blocks
for
out-of-order
data,
and
then
we
are
leveraging
on
the
compactor
to
to
see
them
all
and
immerse
them.
A
A
D
Yeah
sure
so
we
already
have
support
for
backfilling
blocks
into
the
beer,
but
it's
a
manual
procedure
where
one
will
upload
the
old
blocks
into
object
storage,
so
the
compactor
will
take
care
we'll
make
use
of
them.
So
what
I'm
doing
is
basically
add
the
support
into
into
mimir
for
sending
blocks
for
backfilling.
D
D
D
The
in
addition
to
the
api
api,
there
is
also
I'm
also
adding
some
support
to
to
the
mirror
tool.
So
you
can,
you
can
perform
backfilling
from
the
command
line
by
by
that
tool
and
the
the
work
is
currently
at
the
prototype
stage,
but
the
the
the
basic
functionality
has
been
implemented.
D
So
I'm
hoping
it
will
be
done
in
not
too
long.
Although
there
are
some
architectural
questions
being
discussed
in
the
in
the
pr
specifically
which
component
in
which
to
put
the
the
backfill
api
at
the
moment,
it's
in
the
in
the
distributor
component,
but
there's
but
there's
discussion
as
to
whether
we
should
move
it
perhaps
directly
to
the
compactor
instance
to
to
avoid
to
avoid
the
distributor
having
to
send
the
the
files
to
to
investors.
D
For
for
us,
because
the
investors
have
access
to
the
object,
storage
and
the
the
distributor
does
not.
D
There
are
plans
to
implement
that,
but
it's
not
part
of
the
prototype
yet
because
it's
it's
more
effective
to
just
implement
the
the
let's
say,
the
the
basic
parts
first
and
then
we'll
get
to
validation.
Later
and
sponsor.
You
know,
nailed
the
architectural
questions,
for
example,
and
and
yes.
A
A
E
So
maybe,
as
a
short
interjection,
only
two
things
first
feel
more
than
welcome
to
write
yourself
into
the
agenda
into
the
attendee
list,
in
particular,
if
you're
external,
I
messed
up
that,
should
have
been
world
writable
from
the
start
marker
just
fixed
sorry.
The
other
thing
is
in
particular
for
the
external
people.
Please
please,
please
feel
super
welcome
to
pipe
up
at
any
point
in
time
like
this
is
not.
This
is
not
grafana
talking
to
you.
This
is
supposed
to
be
a
conversation.
E
E
A
So
support
for
string
string
trunks-
I
was
talking
about
that.
It's
a
little
optimization
in
the
remote
read
api
that
we
hope
to
land
in
time
for
mimir
2.1.
A
We
also,
hopefully,
will
get
multi
architecture
docker
images
that
we
have
been
working
on
for
some
time
and
yeah.
Hopefully
they
will
be
available
for
all
of
us,
mac,
m1
users
and
also
people
running
on
small
arm
chips.
A
C
Okay,
yeah
the
next
release.
2.1
is
schedule
for
the
week
of
16th
of
may.
The
first
release
judo
was
led
by
me
and
now
we
are
looking
for
a
volunteer
among
member
maintainers
to
lead.
The
next
release
is
anyone
in
the
call
open
to
lead
the
release.
C
Yeah
right
yeah,
I'm
me
and
few
other
people
are
starting
network
to
further
simplify
mimir
from
an
operational
perspective,
both
getting
mimir
up
and
running
on
the
day,
zero,
as
well
as
operating
media
over
the
time
so
yeah.
I
just
have
a
very
open
question
to
you:
you
can.
We
can
discuss
it
right
now,
like
in
a
time
box,
fashion,
10
minutes
or
so,
and
then,
if
you
want
to
reach
out
to
offline
on
slack,
to
continue
the
discussion,
I'm
very
happy
to
do
it.
C
But
basically
my
question
is,
in
your
opinion,
given
your
experience
with
mimir:
what's
your
your
personal
top
priority
to
get
vimeo
easier
to
install
and
operate
like
what's
the
issue
you
got
getting
started
with
mimir
or
if
you
migrated
from
another
system
to
mimir
what
was
the
pain
point
yeah?
C
We
are
receiving
many
questions
on
slack
and
some
of
them
are
about
the
configuration.
So
I
think
just
you
know
scrolling
the
slack
community
channel.
We
we
we
can
learn
a
lot
about
it.
I've
already
seen
a
few
times
the
question
about
how
to
configure
the
gcs
and
s3
bucket
authentication,
and
I
think
that's
one
of
the
one
on
one
of
the
information
which
are
lacking
from
our
documentation.
So
probably
there's
a
good
opportunity
to
improve
the
documentation
about
the
bucket
configuration.
C
What
do
you
think
cryo,
because
you
you're
also
in
touch
with
many
community
users
running
with
running
the
hell
chart.
F
F
Like
plans
in
in
place
to
improve
that,
one
of
which
is
that
there
will
be
support
for
getting
secrets
from
environment
into
the
mimir
config,
so
that
you
don't
have
to
write
it
into
your
config
and
also,
which
would
mean
eventually,
that
the
config
doesn't
have
to
be
a
secret
in
kubernetes,
so
that
you
would
get
like
nice
diff.
When
you,
when
you
change
your
config.
F
And
also
you're
thinking
of
possibly
making
like
wiring
the
storage
settings
into
the
values
and
not
direct
and
kind
of
templating
it
into
the
configuration,
because
right
now
I
see
that
one
challenge
is
that
the
storage
has
to
be
configured
at
multiple
places
and
it's
kind
of
a
hassle
to
know
when
to
set
which
one
and
how
many
places
I
have
to
set
it.
So
one
way
around
that
is,
you
know,
making
the
ham
chart
do
that
for
you.
F
Basically,
the
one
thing
that
we
would
like
to
avoid
is
to
have
like
too
much
magic
in
the
ham
chart,
because
we
already
had
feedback
for
the
old
chart
or
the
enterprise
chart
that
sometimes
it's
hard
to
know
what
the
actual
configuration
will
be
in
the
end,
with
things
coming
in
from
different
sources
and
and
cri
flags
affecting
your
configuration
and
stuff
like
that.
G
We've
also
had
a
a
few
questions
from
community
members
around
capacity
planning
and
we've
got
some
things
in
the
docs,
but
it
doesn't
cover
all
components
like
like
memcache,
and
I
I
think
we
could
also
we
could
add
some
more
information
about,
like
expected
instance
sizes
or
something
like
that.
Like
you
know,
if
you're
in
google
cloud
use
these
type,
if
you're
an
aws
use
these
type,
something
like
that.
C
Sorry,
if
I
typed
him
in
the
storage.
C
I
I
think,
our
our
documentation,
even
if
you
know
we
put
some
effort
to
improve
the
documentation
before
the
lunch
about
the
ruler
and
the
alert
manager.
I
think
it's
not
100
clear
how
they
should
be
configured.
F
And
to
add
to
that,
I
think
it's
kind
of
obvious
that
we
need
like
more
documentation
about
how
to
use
features
with
hemp
chart,
because,
judging
by
the
feedback
on
the
slack
channel
and
and
the
hand
chakra
pool
people
are,
are
clearly
interested
in
the
am
chart.
So
you
know
it
needs
better
documentation.
F
F
No,
I
was
just
wondering
because
if
we
talked
about
new
features
for
mimir,
maybe
I
should
have
put
into
the
agenda
like
upcoming
things,
for
the
hand
chart
as
well
and
the
hem
chart
is.
There
are
pull
requests
to
requests
in
the
works
to
support
multi-zone,
so
zone,
aware
replication
of
injester
storage,
get
phase
or
lag
manager.
F
C
Earlier
you
mentioned
that
that
you
would
like
to
to
to
configure
the
storage
in
a
single
place
in
the
helm
chart,
and
I
actually
would
like
to
do
the
same,
but
in
the
mimic
configuration
that's,
something
loki
did
and
it
works
similar
to
the
prometheus
global
config
option
where
you
come
to
basically
the
defaults
sort
of
defaults,
and
I
I
wanted
to
propose
something
similar
like
having
a
global
entry
in
the
member
configuration
where
you
can
set
some
again
sorry
for
the
repetition
global
configuration
and
then,
if
you
want
on
a
per
component
level,
you
can
override
it
and
the
main
reason
why
we
still
may
want
to
override
it
is
because
for
the
block,
storage
and
ruler
and
dialect
manager,
we
can't
use
the
same
bucket.
C
C
The
solution
was
simple,
was
just
storing
all
the
tsb
blocks
under
a
prefix
in
the
storage
like
under
blocks,
for
example,
so
that
it
doesn't
conflict
with
the
ruler
and
the
alert
manager.
Anyway,
I
think
I
think
that's
something
we
can.
We
can
work
on
handy
if
I'm
not
mistaken,
is
working
on
apr
to
allow
to
configure
the
the
packet
prefix
in
mimir.
C
G
C
Sorry
yeah
so
that
will
unlock
the
the
option
to
run
mimir
with
just
one
single
bucket.
That's
why
I'm
particularly
interested
into
into
that
work,
but
yeah.
Do
you
have
any
any
comment
about
introducing
a
global
conflict,
starting
with
the
storage.
C
A
Well,
we
don't
really
have
more
items,
we
have
open
mic.
Cryo
has
already
already
talked
about
the
new
helm
features,
and
I
put
your
item
speaking
about
global
configuration
for
storage
there,
so
we
are
actually
at
the
end
of
agenda.
Now,
I'm
wondering
oh,
you
have
a
question
about
it.
Anything
hold
you
back
from
trying
there
and
I'm
wondering
if
there
is
anyone
who
did
not
try
me
here
but
would
like
to
but
is
for
whatever
reason
stopped
or
maybe
hold
back
held
back.