►
From YouTube: AMA with SREs (Public Livestream)
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).
A
A
A
A
B
Hi,
thank
you
so
yeah.
So
questions
regards
to
how
to
deal
with
control
the
arrows
on
sentry
so
basically
to
give
a
little
bit
of
context,
I
test
a
future
flag
and
good
luck
on
that
I
enabled
this
morning
and
this
feature
we
raise
the
control
the
order
to
century
and
especially
now
that
we're
watching
how
the
feature
flag,
enabled
behaves
I
think
the
zero
is
super
useful,
considering
we
can
get
not
only
the
exact
project
and
controller
method
that
relates
to
it,
but
also
the
user
that
got
affected
by
it.
B
Since
this
is
a
controlled
error
which
expect
to
happen
from
time
to
time,
so
it
should
keep
happening
and
I
think
it's
okay
to
happen.
What
is
your
preferred
take
on
how
to
deal
with
this
on
sentry?
So
here's
a
couple
of
suggestions:
should
we
ignore
the
zero
on
sentry?
Should
we
ignore
until
the
next
end
users
get
affected,
so
we
get
a
little
kiss?
B
C
C
So
I'm
Ben
I'm
on
the
observability
team
or
observe
observability
working
group.
We
don't
actually
have
any
kind
of
active,
alerting
or
or
SLO
kind
of
behavior
with
sentry
it's
kind
of
a
an
available
debugging
resource.
But
we
don't.
We
don't
act
directly
on
anything
that
we
get
in
sentry,
so
there's
no.
Nobody
will
get
paged
if
you
send
stuff
to
century
at
this
time.
So
it's
it's
totally
fine
to
just
we
it's
kind
of
ignored
by
default.
C
If,
if
we're
getting
errors,
well,
then
we'll
go
and
look
and
see
if
there's
anything
new
in
century
and
so
I
think
it's
fine
to
just
to
you
know:
do
it
the
way
you've
done
it?
That's
actually
great
I
love
having
controlled
errors
and
controlled
messages
to
catch
these.
If
it
is
I,
guess
I
don't
know
enough
about
what
the
feature
flag
and
the
error
is
as
to
whether
this
is
critical
enough,
that
we
should
have
a
metric
for
it
and
have
some
kind
of
SLO
for
this
error.
C
B
A
E
D
A
C
So
this
is
this
is
a
loaded
question,
because
I
kind
of
know
the
answer
already,
but
I
figured
I'd,
give
somebody
a
chance
to
talk
about
it
in
the
previous
AMA
we
talked
about
how
is
Puma
going
for
people
who
don't
know.
We've
been
moving.
Our
rail
stack
from
a
multi-process
to
a
multi-process
multi-threaded
serving
mode
and
I'm,
hoping
somebody
that
worked
on
that
can
can
talk
about
the
current
state.
C
C
You
I
guess
I
guess:
I
can
go
ahead
and
answer
my
own
question.
It's
done.
We
we
finished
rolling
out
Puma
and
now
we
have
a
significant
amount
of
improved
rails
capacity
without
as
much
additional
cost,
and
actually
it
should
reduce
our
cost.
In
terms
of
how
much
how
many
web
pages
we
can
serve
for
how
many
CPUs
in
memory
we
spend
on
it
so
awesome
job
all
around
to
the
s-series
and
developers
that
collaborated
to
get
that
done.
A
H
G
My
network
connection
is
a
little
flaky
I'm,
not
sure.
If
you
couldn't
hear
me.
G
Is
fine,
okay,
fantastic
yeah,
so
so
sure,
so
that
means
that,
for
example,
if
so,
only
one
so
within
within
the
Pumbaa
process,
as
I
understand
it,
only
one
thread
is
actively
running
on
CPU
at
a
time.
But
when
the
thread
enters
or
in
terms
of
white
state
like
for
a
remote
service
call,
another
thread
can
open
up
can
switch
on
CPU,
but
any
any
logical
resource
is
held
by
the
thread
that
just
went
off.
Cpu
are
retained
by
that
thread.
G
G
Concurrently
and
that's
then
another
and
another
thread
can
do
the
same
if
we
don't
have.
If
we
don't
have
a
maximum
number
of
connections
that
that,
if
the
high-water
mark
ever
exceeds
the
provision
capacity
in
the
connection
pools,
then
one
thread
could
end
up
waiting
well,
even
in
a
wait
state,
even
though
it
would
otherwise
be
running.
Both
do
that
does
that
make
sense.
This
is
so
much
easier
with
a
white
board.
A
G
As
far
as
I
know,
we
know
there
was
extensive
discussions
about
this
prior
to
rolling
Puma
out
and
I
was
in
and
out
of
that
discussion.
So
I,
don't
know
where
we
settled
on
the
sizing
for
the
for
the
connection
pools
but
I'm,
just
just
as
a
specific
concrete
example,
but
really
I
just
wanted
to
call
it
out,
because
because
they
were
there
because
the
the
the
nature
of
resource
contention
got
a
little
bit
more
complex,
and/or
interesting
as
a
result
of
trying
to
leverage
the
greater
efficiencies
of
having
multiple
threads
within
the
process.
G
H
C
We've
just
put
it
in
a
queue
to
wait,
and
you
can
see
that
while
we
were
running
on
unicorn,
we
would
have
quite
high
spikes
of
hundreds
of
requests
during
an
during
the
during
the
day,
just
waiting
waiting,
waiting,
waiting
and
now
with
Puma.
It's
almost
nothing,
and
so
basically,
we've
we
can
now
as
soon
as
a
request
comes
in
through
the
web
or
the
API
with
it.
Ruby
can
start
working
on
that
without
having
to
contend
on
the
resource.
C
F
Hey
everyone
question
related
to
the
recent
announcement
from
Google
in
terms
of
their
plans
to
of
how
they
plan
to
price
their
kubernetes
cluster
is
consumption
and
we
have
plans
on
migrating.
Most
of
it
lapse
components
to
kubernetes
clusters
right,
so
just
wanted
to
get
a
sense
of.
Did
you
all
see
that
come
in?
What
is
it
a
complete
surprise
and
how
y'all
plan
to
you
know,
try
to
keep
costs
low
and
still
complete
the
migration
I
think.
I
We're
probably
hi
I'm
on
the
observability
working
group
as
well.
I'm
Craig
I
think
we're
probably
going
to
spend
far
more
on
the
node
pools
than
the
charge
for
the
control
plane.
I
think
that
comes
it's
ten
cents
an
hour,
so
seventy
two
dollars
per
month
per
cluster
and
we're
only
going
to
have
one
cluster
per
environment.
Maybe
two.
If
we
have
to
split
something
like
logging
out,
so
it'll
probably
be
a
rounding
error.
You
know
GCP
bill.
I
I
J
Anyone's
interested,
this
is
Brandon
aghhhhhh
answer
that
question
I
lead
up
the
Alliance's,
so
I
get
to
do
much
of
this
and
was
with
the
team
at
Google
before
so.
This
is
this
was
a
backtrack
on
Google,
which
is
a
poor
choice,
because
you
should
never
put
something
free
and
then
charge
for
it.
Generally.
J
That's
a
bad
idea,
I
think
what
a
lot
of
people
didn't
realize
is
there,
if
you
generally
what
we're
doing
isn't
what
everyone
does,
which
is
large
kubernetes
clusters
in
a
few
of
them,
other
people
have
treated
Cooper
neighs
clusters
like
their
VMs
and
so
that
overhead,
rapidly
kind
of
gets
different.
Cost-Wise
Azure
pulled
this
off
in
Google
response,
so,
as
your
container
service
was
like,
hey
we're
going
to
give
this
for
free,
Google
responded,
Amazon
never
has
Amazon
is
always
charged
something
for
the
control
plane,
so
for
eks.
J
So
this
continues
to
be
with
eks
and
with
GK
they
they're
now
more
or
less
in
line
with
what
is
standard
with
the
with
those
as
you're
being
different.
You
can
make
some
guesses
on
that.
I
have
assumptions
as
to
why,
as
yours
has
remained
free,
but
all
sorts
of
interesting
questions
we
can
chat
about,
but
contact
twice,
I
hope
that
helps
there.
You
go.
I
C
Yeah
I
think
the
the
biggest
impact
and/or
concern
from
our
end
was.
We
have
quite
a
lot
of
kubernetes
demo
clusters
and
test
clusters
that
we
were
spinning
up
and
just
leading
because
they
were
free
now
we're
gonna
have
to
track
and
shut
them
down
and
and
tell
people
to
create
life
cycles
for
their
demos
and
tests.
D
A
E
A
G
I
am
not
aware
of
new
alerts.
There
was
I,
I
didn't
feel
it
was
necessarily
appropriate
to
speculate
in
in
in
this
forum,
but
there
was
there
was
one
regression
events
that
I
thought
might
possibly
have
been
related
to
to
resource
contention
within
between
green
threads
in
Puma,
but
I
didn't
I,
wasn't
present
to
collect
to
collect
data,
and
we
didn't
have
the
right
tool
thing
installed
anyway
at
the
time
to
collect
evidence
to
refute
that
hypothesis,
so
I'm.
So
at
this
point,
I'm
not
aware
of
any
regressions
ya.
A
Know
and
from
a
rule
of
process,
to
answer
your
question
a
little
more
gentleness
and
we
did
roll
this
out
gradually
across
the
fleet
and
I.
Think
Ben
was
trying
to
ask
when
he
was
asking
for
job
you're
asking
that
question
Jarvis
one
of
the
people
involved
in
the
rollout
so
I
think
they
may
have
discovered
some
things
in
that
role.
I
was
not
deeply
involved
on
that.
G
That
was
my
impression
as
well
like
particularly
the
the
concurrency
level
I
I
think
was
was
a
topic
of
great
great
debate
where,
where
there's
arguably
some
benefit
even
to
having
Puma
with
with
one
thread
but
but
being
able
to
scale
it
to
two
or
more
threads
per
process,
was
anyway
a
topic
of
discussion.
I
think
I
think
different
people
had
strong
differing
views
on
that
and
I
did
not
have
a
strong
opinion,
so
I
can't
really
voice
any
more
thoughts
on.
E
J
Just
a
tactical
question:
so
I
don't
bug
people,
and
this
is
actually
mostly
because
I
feel
like
I,
like
you
Dave
far
too
often
just
a
good
place.
If
we're
looking
for
pirates
in
timing
on
Oh,
what
are
we
moving
to
kubernetes
and
stuff
I'd
love
to
just
read
it
and
not
have
to
ask
questions
and
me,
but
if
you
guys
have
a
place,
you
guys
would
say
I'd
love
to
just
go.
Do
my
homework
currently
I
could.
A
Give
some
places
at
the
moment
but
I
feel
like
it's
actually
likely.
You
know,
yeah
Ben's
type
in
ethics,
I
think
there
are
certain
ethics,
where
we're
trying
to
call
us
information
on
projects
that
we're
working
on
there
are
longer
lived
than
a
particular
month
or
quarter,
and
the
other
thing
I
would
say,
though,
is-
and
it
was
brought
up
in
their
group
conversation
actually
yesterday
by
Jerry,
with
the
new
VP
coming
in
I'm
kind
of
waiting
to
see
how
they
want
to
organize
some
of
that
stuff.
A
So
I'll
answer
with
the
epochs
and
things
that
we
do
have
around
that
and
then
that
structure
also
across
you
know
what
is
the
delivery
team
working
on
with
their
things
related
to
Marin?
So
some
of
those
questions
are
you
know
it's
gonna
have
to
tactically
filter
from
director
level
down
to
manager
level
like
what
particular
project
are
we
looking
at
a
lot.
A
D
A
J
Going
again,
which
I
feel
bad,
but
if
no
no
have
you
all
looked
so
when
I
say
kubernetes
we're
like
a
JDK
I,
don't
as
I'm
curious
on
the
team,
have
you
all
looked
at
anthos
and
all
the
other
pieces
that
they
offer
above
it?
If
so,
what
are
your
thoughts,
if
not
totally
fair
but
I'm,
just
curious
if
anyone's
poked
around
the
managed,
SEO
and
all
the
other
parts
that
they
like
around
anthos
and
your
thoughts,
the
answer
could
be:
don't
have
any
it's
fine,
I'm,
just
curious.
H
I'd
never
heard
of
it,
I
wasn't
saying
I.
If
we've
looked
at
it,
it's
been
at
an
individual
level
and
I
haven't
heard
of
anything
amongst.
That's
me,
where
we've
had
any
kind
of
concerted
effort
to
consider
something
like
that,
strategically
we're
primarily
focus
on
just
moving
like
moving
initial
batches
and
services
over
the
communities.
First,
yeah
accepting
minor
in
efficiencies
along
the
way,
and
then
those
sorts
of
things
would
be
okay,.
D
A
Then,
to
add
to
what
little,
why
do
you
know
of
anthos
I
think
we
would
actually
love
to
have
some
other
input
from
the
company
and
the
business
in
terms
of
what
we
want
to
do
for
business
continuity,
and
things
like
that
anthos
could
have
a
hand
in
managing
coronaries
cluster.
So
if
we
decided
to
go
with
a
multi
cloud
strategy
or
something
like
that,
so
I
think
again
as
we
emerge
our
strategy
and
what
we
want
to
do
with
those
things.
That's
where
I
Anthon
have
a
picture,
but
yeah
I
agree
with.
J
And
then,
probably
at
the
thousand
or
sorry
10,000
foot
level
to
my
levels
right
sounds
like.
Is
that
something
it
may
be
a
month
or
two
out
would
be
worth
digging
into
and
just
spending
like
a
brainstorming
session
with
we
have
help.
Would
you
guys
like
to
do
that?
Yeah
I,
think
that
would
be
good,
be.