►
Description
Speaker: Amrith Kumar, Founder and CTO
This presentation is aimed at IT managers, DevOps or Developers in an IT organization, as it presents an in depth exploration of the architecture and internals of Cassandra databases with OpenStack Trove.
The presentation will start with an overview of the Trove architecture, exploring such concepts as ""How does Trove interact with other OpenStack services"", ""What are the various components of Trove"", ""What are guest agents"", ""How are requests to Trove processed"", and ""How does Trove support multiple database types"".
It will next explain how Trove supports the Apache Cassandra NoSQL database and how to deploy and manage a Cassandra database with Trove. It will then describe Trove's framework for implementing clustering and how Cassandra clusters are deployed through Trove.
It helps the participant understand the internals and architecture of Trove and provides the participant with knowledge that would be useful in assessing, deploying and managing a Cassandra database with Trove.
A
B
Thanks
for
all,
thanks
all
for
coming
I'm
going
to
be
talking
about
OpenStack
and
a
project
on
OpenStack
called
trove
which
does
database
this
service.
How
many
of
you
here
are
even
familiar
with
OpenStack
or
trove?
We
can
guess:
okay,
I'm
sure
all
of
you
are
familiar
with
the
first
two,
so
we
won't
bother
about
that.
B
Look,
occupied,
open,
stat
and
trove,
something
about
the
architecture
of
trove,
how
it
works,
what
it
does
and
then
drill
them
a
little
bit
more
detail
about
Cassandra
itself
and
leave
some
time
for
questions
so
the
OpenStack
mission
OpenStack,
it's
a
cloud
operating
system,
it's
open
source.
You
can
download
it
yet
it
use
it
and
it
basically
gives
you
the
ability
to
manage
your
entire
data
center.
B
Using
a
set
of
open
source
tools
and
trove
is
a
project
in
OpenStack
I'm,
one
of
the
contributors
to
the
project
and
for
those
of
you
who
are
familiar
with
the
general
way
in
which
OpenStack
projects
work.
I'm
on
the
core
review
team
for
the
project.
Troubs
mission
is
to
make
it
easy
for
you
to
use
both
relational
and
non-relational
databases
in
the
openstack
cloud.
B
As
a
point
of
reference
for
those
of
you
who
are
familiar
with
amazon
RDS
is
the
relational
database
service
in
amazon,
but
there's
DynamoDB,
which
is
a
no
sequel
database
and
then
there's
redshift,
imagine
something
which
has
all
three
of
them
in
one
place.
That's
true
true
can
orchestrate
relational
and
non-relational
databases
and
we'll
talk
about
some
of
those
as
well
feel
free
by
the
way
to
stop
any
point
along
the
way
and
ask
questions.
I
know
this
is
towards
the
end
of
the
conference,
so
everybody
is
probably
itching
to
go
as
well.
B
B
B
B
For
those
of
you
who
are
trying
to
magnify
this
picture,
I
apologize,
there's
a
URL
sitting
here
in
the
right-hand
corner.
All
of
these
slides
are
available
online
on
our
website.
Tesoro,
calm,
/,
I
think
slide
/
Cassandra
summit.
If,
if
you
want
this,
particular
image
is
also
available
in
the
OpenStack
documentation.
So
this
is
just
a
schematic
of
what
a
typical
OpenStack
deployment
would
look
like
a
brief
history
of
where
trove
came.
Camera
purple
is
again
database
the
service.
B
So
we
assume
that,
in
order
to
use
TRO,
you
have
a
functioning
OpenStack
environment,
meaning
you
have
Keystone
for
identity
management,
identity
and
access
management.
You
have
noble
functioning,
you
have
storage
with
cinder
and
Swift,
you
have
Neutron
and
you
may
have
a
user
access
console,
but
as
you
progress
beyond
just
running
VMS
with
your
application
in
it,
you
find
there's
a
value
to
have
a
manage
database
and
the
folks
at
HB
and
Rackspace
sure
found
that
we
have.
Purely
by
coincidence,
we
have
one
person
in
the
room
here.
B
B
B
B
The
database
is
a
service
part
trove
or
the
platform
which
we
distribute
is
here
on
the
left,
so
like
every
other
OpenStack
service
trove
exposes
an
API.
This
is
a
standard
restful
api
and
you
can
make
requests
to
this
api.
To
do
things
like
provision
me.
A
cassandra
instance
provision
mia
cassandra,
cluster
provision
me
a
my
sequel
instance.
B
All
of
those
things
come
in
here
there's
a
couple
of
other
components
in
trove
which
we'll
talk
about,
but
let's
talk
about
a
typical,
a
request
which
you
may
make
as
a
user,
you
may
say
and
again
to
use
an
Amazon
analogy.
If
you
wanted
a
my
sequel
database
or
a
Cassandra
database,
you
could
go
to
ec
to
spin
up
an
ec2
instance
install
your
software
and
you
got
a
perfectly
good
running
database
instance
or
you
could
go
to
RDS
and
make
a
request
and
say,
give
me
an
RDS,
my
sequel
instance.
B
Making
a
request
to
trow
is
like
making
a
request
to
RDS,
so
you
make
a
request.
A
trove
and
say
give
me
an
RDS
instance
or
in
this
case
give
me
a
my
sequel
instance
or
Cassandra
instance.
It
is
now
troves
responsibility
to
orchestrate
the
rest
of
OpenStack
to
give
you
a
queryable
endpoint,
so
I
could
make
a
request
to
trove
and
say
give
me
a
Cassandra
instance
and
the
immediate
response
you're
going
to
get
is
going
to
be
your
connection
string
is
this,
and
this
is
your
instance
identifier.
B
You
can
then
go
and
pull
that
instance
identifier
to
figure
out
when
that
instance
comes
online.
It
takes
about
a
minute
you'll
get
a
running
instance
at
that
point.
That
connection
string
is
a
standard
cassandra
connection
string
or
a
my
sequel
connection.
String
trove
is
not
in
the
middle
of
that
data
path.
B
You
interact
with
that
database,
exactly
as
you
will,
if
you
provisioned
it
on
a
bare
machine
and
the
the
the
the
architecture
of
throw
is
like
the
architecture
of
every
other
OpenStack
service,
there's
the
API,
which
does
the
RESTful
API,
it's
just
a
whiskey
server.
All
the
heavy
listing
is
heavy
lifting
is
done
by
a
task
manager.
Service
trove
has
a
trove
task
manager
and
when
you
provision
a
database
instance
or
instances
under
the
covers
your
provisioning,
a
nova
instance,
which
is
a
virtual
machine.
B
B
Storage
for
that
is
attached
from
cinder.
So
you
have
some
group
of
stores,
some
set
of
storage,
which
is
managed
for
you
by
cinder.
Trova
is
going
to
make
a
request
to
sender
and
say
the
user
requested,
a
50
gigabyte
volume
or
a
terabyte
volume.
That
volume
is
now
attached
to
your
instance
and
that's
where
your
data
gets
stored.
Okay,
I'm
going
to
pause
here
and
see.
If
anybody
has
questions
at
this
stage
go
ahead.
Oh
okay,
very
simple!
B
So
for
those
of
you
who
couldn't
hear
the
question,
how
do
you
select
what
kind
of
database
you
want?
The
trove
API
requests
for
create
is
the
same
API
call,
no
matter
what
database
you
want,
but
in
the
request
you
specify
the
data
store.
So
the
data
store
could
be
at
this
point,
my
sequel
or
Percona
or
pxe,
which
is
/
contacts
or
DB
cluster
or
postgres
or
or
Cassandra
or
couch,
or
reddit
or
vertica,
or
do
a
sore
Oracle.
You
specify
the
name
of
the
data
store
for
each
data
store.
B
You
could
have
multiple
versions,
so
you
also
specify
the
version,
so
you
could
have
oracle
11g
in
oracle
12c,
so
you
would
specify
I
want
an
Oracle
11g
instance,
so
you
provide
those
two
parameters:
you're
also
required
to
provide
a
flavor
which
is
a
nova,
flavor
and
you're
required
to
provide
a
size
for
most
data
stores,
except
for
Redis,
which
doesn't
require
a
backing
volume.
Those
are
the
four
parameters
you
provide
on
a
per
data
store
basis.
B
B
There
aren't
too
many
of
those
right
now
what
database
were
you
thinking
about?
Sorry,
the
question
was:
how
easy
is
it
to
add
a
database
which
is
not
supported
but
which
database
did
you
have
in
mind
aerospike,
okay,
so
adding
a
new
database
is
not
it's
not
very
hard.
It's
not
trivial.
It's
a
there's,
a
set
of
things
which
you
need
to
do
so
in
trobe.
You
can
define
a
data
store
to
be
in
one
of
three
levels
of
maturity,
experimental
or
in
test
or
in
production.
B
Now
we'll
talk
in
a
bit
and
I'll
come
back
to
your
question
about
what
it
takes
to
add.
A
datastore
trove
can
do
a
lot
of
things
for
a
database,
I,
don't
know
about
the
architecture
of
aerospike
and
what
all
it
can
do.
But,
let's
talk
about,
say
a
database
like
Cassandra
provisioning.
A
single
instance
of
cassandra
is
not
hard
provisioning.
A
cluster
of
cassandra
is
harder
doing
a
backup.
The
single
instance
of
cassandra
is
not
hard
doing.
A
backup
of
a
cluster
is
harder.
B
Trope
supports
backups,
so
you
need
to
implement
code
in
your
guest
agent,
which
understands
how
to
do
some
of
these
more
esoteric
operations
resizing
an
instance.
So
you
launch
this
instance
with
whatever
database
you
want,
and
you
find
that
you
chose
the
wrong
flavor.
You
want
bigger
instance.
There
needs
to
be
code
in
your
guest
agent,
which
understands
what
to
do
in
order
to
resize
the
instance.
So
those
are
all
incremental
features
which
you
can
add
to
your
guest
agent.
Then
you
can
get
it
to
the
level
where
you
say
it's
test
ready.
B
B
If
your
database
requires,
for
example,
a
licensed
binary
to
be
downloaded
and
a
complicated
click-through
mechanism,
then
you
need
to
provide
a
mechanism
where
the
test
infrastructure
can
in
an
automated
way
do
those
things
so
there's
a
set
of
things
which
we
do
and
I'll
try
not
to
make
this
a
commercial
from
my
company.
But
if
you
want
to
add
aerospike
speak
to
me
after
the
call
and
tell
you
how
to
do
it,
there's
some
money
involved
there.
B
Other
questions,
ok,
so
the
crux
of
this
entire
architecture
is
that
on
a
per
database
basis
or
a
per
data
store
basis,
my
sequel
Cassandra.
What
have
you
there
is
some
custom
code
which
is
implemented
in
a
guest
agent
and
the
guest
agent
understands
how
to
perform
that
specific
operation
for
that
database.
Doing
a
backup
on
oracle
is
different
from
doing
a
back
up
on
my
sequel,
trobe
supports
an
api
called
back
up
the
code.
To
do
the
specific
thing
for
each
of
these
databases
is
buried
in
the
guest
agent.
B
Ok,
that's
about
all
I'm
going
to
talk
about
on
architecture.
Unless
anyone
has
other
questions,
specifically,
ok,
so
I
guess,
the
question
is:
what
are
all
the
things
that
TRO
can
do
for
you,
and
this
is
maybe
a
partial
list.
We've
talked
about
many
of
these
things,
provisioning
a
single
instance
not
very
hard.
You
know
give
me
an
instance
of
this
Tampa
you're
doing
some
development.
You
want
to
spin
up
a
single
instance
of
Cassandra.
You
want
to
run
some
queries
against
it
great.
B
You
can
do
that
some
databases,
support
replication
either
for
performance
reasons
or
high
availability.
Reasons
show
supports
that.
Give
me
a
cluster.
Ok,
all
of
you
probably
heard
a
lot
about
Cassandra's
clustering
capabilities.
Many
databases,
support
clustering,
trove,
supports
clustering,
I
will
caveat
and
say
not
all
databases
support
all
of
these
capabilities
at
this
point,
for
example,
clusters
or
support
in
only
a
couple
of
databases,
my
sequel
with
Percona
vertica,
for
vertical
clusters
and
for
clusters.
Cassandra
clusters
are
coming
we'll
talk
about
that
in
a
second.
B
The
ability
to
backup
do
backup
and
restore
for
small
data
sets.
This
is
very
easy
stream,
the
data
out
somewhere
and
if
you
want
to
restore
it
load
the
data
back.
How
do
you
do
it
when
it's
a
larger
data
set
throat
has
automation
for
that
replication
clustering?
We
talked
about
that
a
very
important
use
case
for
true
this
configuration
management.
So
if
you
happen
to
have
hundreds
or
thousands
of
database
instances,
these
database
instances
has
a
config
file
of
some
kind.
B
Now
assume
you
want
to
make
a
configuration
change
to
all
1000
of
these
doing
this
manually
is
very,
very
time-consuming
and
very
error-prone.
So
trove
gives
you
the
ability,
you
say,
create
a
configuration
group
and
attach
it
to
a
whole
suite
of
instances,
and
now
it's
troves
responsibility.
We
will
talk
about
how
it
does
some
of
these
things
and
resizing
it.
You
have
a
small
instance.
You
want
to
go
to
a
large
instance.
You
have
small
storage,
you
want
to
go
to
large
storage,
so
on
and
so
forth.
B
B
Of
course
you
don't
want
to
have
a
situation
where
you
either
lose
data,
or
somebody
can
break
in
and
steal
your
data
and
in
an
organization
you
typically
don't
have
a
situation
where
you
have
just
one
database,
many
organizations,
many
applications.
These
days
have
multiple
databases,
some
use,
a
sequel
database
and
a
graph
database
and
some
key
value
store,
and
in
the
old
days
you
had
an
IT
organization
which
had
all
the
expertise
on
every
database
known
to
man.
Now,
that's
no
longer
the
case.
Very
often,
this
is
fragmented
to
the
organization.
B
If
this
is
the
problem
you
face
in
an
organization,
one
of
the
good
solutions
is
to
automate
everything
you
possibly
can
do
and
that's
basically
what
trove
does
for
you.
It
addresses
the
issue
of
setup
by
completely
automating
that
it
addresses
the
issue
of
configuration
by
completely
automating
the
ability
to
manage
configuration,
not
for
a
single
instance,
but
for
a
whole
group
of
instances
and
for
all
complicated
configurations,
clusters
replicas
and
all
of
that
Shrove
manages
that,
for
you,
entro
supports
a
whole
bunch
of
different
databases.
B
So
when
one
first
works
with
show,
is
there
any
training
or
I
hate
to
do
this,
but
I
we
wrote
a
book
on
true,
so
you're
happy
to
you
can
happily
buy
a
copy
of
the
book.
I
have
a
couple
here.
If
people
are
interested
in
the
book
on
TRO,
we
offer
training
as
well,
but
true
by
itself
is
not
a
if
you're
an
end
user
of
trove.
It
is.
B
Maybe
one
or
two
commands
later
you
can
get
a
guest
image
for
just
about
any
database.
You
start
with
the
database
you're
familiar
with
in
terms
of
complexity,
I
would
say
an
average
engineer
kind
of
person
a
couple
of
days
and
you'll
have
something
which
you
can
launch
your
first
instance
and
the
nice
thing
about
it
is.
You
can
run
all
of
OpenStack
on
a
modestly
sized
laptop.
So
on
just
about
any
operating
system
stick
vm
were
on
it
run,
it
will
bunt
to
vm
or
red
hat
fedora
vm.
You
can
run
a
red.
B
B
It
gives
you
a
common
API
for
a
whole
bunch
of
different
database
operations
across
a
lot
of
databases,
and
it
allows
you
to
do
standard
constructs
replication,
clustering
h,
a
things
like
that
many
management
operations,
backups
configuration
management
and
things
like
that
and
very
important
thing
at
the
end-
is
even
in
an
organization
where
you
do
not
have
a
lot
of
expertise
on
every
last
detail
about
a
database
you're
using
the
automation
built
into
embodies
the
best
practices
for
that
database.
So
the
guys
who
contributed
the
code
for
my
sequel.
B
Originally
one
of
them
is
sitting
right
here
in
the
in
the
room
operated.
Maybe
thousands
of
servers
running
my
sequel
and
the
best
intelligence
they
had
about
how
to
manage.
That
was
built
into
trove,
the
guys
who
contributed
the
code
for
clustering
operated
thousands
of
nodes
of
in
a
production
environment.
Therefore
they
put
the
intelligence
now
how
to
do
that
into
the
code
for
clustering,
so
you're
getting
all
of
that
as
part
of
truth
right.
B
So,
let's
talk
about
a
couple
of
very
simple
operations
which
Shrove
does
for
you
a
single
instance
provisioning,
a
single
instance,
the
most
common
thing,
which
somebody
starts
with.
What
is
your
request?
What
data
store,
what
version,
what
size
do
you
want
and
what
database?
Of
course,
it's
it's
going
to
tell
you
between
these
two
and
the
flavor
of
the
instance.
B
That's
all
you
need
to
know
what
robe
is
going
to
do
is
validate
your
request
provision,
the
storage
provision,
the
networking
do
all
of
those
things
tie
it
together
apply
the
configuration
and
hand
you
back
a
connection
string
so
effectively.
This
entire
operation
is
automated
for
you,
okay,
relatively
straightforward.
This
is
a
request.
Restful
api
web
UI
CLI,
whatever
you
want
endpoint,
is
a
connection
string.
B
Okay,
let's
try
another
one:
I
took
a
back
up
of
a
database
or
no,
let's
say
I
want
to
take
a
backup
of
a
database
indicate,
which
instance
and
say
you
want
to
backup
troubs,
going
to
verify
that
you
can
take
the
backup.
Do
you
have
the
privilege
to
do
that?
It
will
execute
database
specific
code
to
do
the
backup
now,
depending
on
your
database.
So
my
sequel,
you
can
do
a
backup
in
one
or
three
different
ways.
You
can
take
a
backup
as
a
my
sequel
dump.
B
You
can
do
a
backup
as
apricot
extra
DB
backup
or
as
an
incremental
backup.
All
three
are
supported,
so
you
could
specify
I
want
an
incremental
backup,
in
which
case
you
specify
what
the
parent
was
sure
will
take
care
of
all
the
my
sequel
specific
code
to
do
that,
and
it
will
push
the
back
up
onto
Swift.
All
done
you
get
at
the
end
of
it
is
a
completion
status,
yes,
no,
otherwise,
okay!
B
B
Now
that
you
created
a
backup,
what
can
you
do
with
the
backup?
Okay,
I
took
a
backup.
Give
me
an
instance
based
on
that
backup,
right,
so
load
from
that
snapshot
again
same
three
things
and
identify
the
backup
trobe
will
create
the
storage
provision.
The
instance
load,
the
backup
for
you
and
hand
you
the
same
connections
right
all
done
for
you.
B
Similar
I
have
a
single
instance
of
a
database
provision
of
replica
for
me
I'm.
Finding
that
my
master
is
getting
beat
up.
Give
me
a
read
replica.
All
you
need
to
tell
me
is
the
same
three
things
and
you
tell
me
what
the
master
is.
True
is
going
to
launch
the
replica
for
you
and
give
you
the
connection
strength.
Now
you
have
two
databases:
load
balancers,
not
provided
by
a
true
if
you
want
to
have
a
proxy.
That's
up
to
you.
B
Another
simple
use
case
provision
a
cluster,
and
this
is
something
which
is
going
to
be
supported
for
Cassandra.
Very
soon
give
me
your
cluster
configuration
in
the
case
of,
for
example,
when
you
say
give
me
a
cluster,
you
specify
how
many
data
nodes
you
want.
That's
it
you
will
get
that
many
data
nodes.
You
will
get
some
number
of
query
routers.
Some
number
of
config
servers,
all
provision
for
you
and
up
and
ready
for
you
with
a
completion
status
and
you'll
get
a
connection
string
behind
the
scenes.
There's
a
complete
cluster.
B
B
And
once
you
have,
the
database
is
up
and
running.
I
now
want
to
apply
some
change
to
it.
So
I'm
running
the
database
and
I
want
to
change
the
amount
of
memory
I
allocate
to
something
I
want
to
change
some
config
parameter,
and
I
want
to
do
this
not
for
one
database,
but
for
a
hundred
tell
me,
the
configuration
group
create
a
configuration
group
and
the
instance
or
instances
at
this
point.
It's
troves
responsibility
to
verify
your
request
is
correct
and
then
apply
those
changes
and
there
are
some
changes
which
may
require
reboot.
B
B
B
However,
all
of
the
commands,
which
all
of
these
requests
can
be
done,
either
as
a
API
call
or
a
UI,
but
also
from
a
CLI
cron
works
just
fine
and
that's
typically,
the
way
in
which
people
who
have
deployed
trove
do
some
of
these,
but
again
there's
a
RESTful
API
available.
If
you
want
to
build
some
automation,
a
round
trope
to
do
some
of
these
things
for
your
own
deployment,
you
could
build
it
through
that
other
questions.
Okay,.
B
B
B
B
B
My
company
provides
a
distribution
of
trove
which
does
support
additional
capabilities
and
databases.
One
of
them
is
Oracle,
and
also
in
trove,
not
all
databases.
It's
this
is
important
to
understand.
Not
all
databases
support
all
capabilities.
You
may
be
able
to
do
a
lot
more
with
MySQL,
which
has
production
grade
guest
agents.
Then
you
may
be
able
to
do
at
this
point
would
say:
db2
Express,
which
has
an
experimental
guest
agent
at
this
point.
I.
B
B
So
we
talked
a
lot
about
a
bunch
of
databases-
let's
not
specifically
about
Cassandra
trova
supported
Cassandra
from
the
very
beginning,
from
from
the
very
first
release
of
trove,
there
was
support
for
Cassandra
and
you
could-
and
you
can
today,
provision
a
single
instance
of
Cassandra
use
it
for
development
purposes
and
the
Cassandra
which
it
provisions
is
Cassandra
from
community
Apache.
Cassandra
support
for
clustered
configurations
is
something
which
is
coming
soon,
that'll
probably
be
in
January,
and
support
for
datastax
enterprise
is
something
which
we're
adding,
which
will
come
out
a
little
bit
later.
B
That
will
be
in
our
enterprise.
Product
it'll,
probably
not
be
in
the
community
version
of
trove.
For
a
while
licensing
reasons,
and
I've
talked
a
little
bit
about
my
company,
I've
mentioned
it.
So
who
are
we
where
the
company?
Who
excuse
me
we're
now
the
largest
contributor
to
trove?
We
were
not
the
people
who
originated
this
project.
The
project
was
originated
by
people
at
HP
and
Rackspace,
and
we
cannot
claim
credit
for
that.
B
We
came
onto
the
project
well
after
they
had
laid
a
solid
foundation,
but
at
this
part,
were
at
this
point
were
the
largest
contributor
to
the
project.
We
have
the
largest
team
of
people
working
on
it
on
the
open
source
version
of
the
project.
At
least
we
provide
a
enterprise-ready
version
of
trove,
which
you
can
use
its.
You
can
download
it
on
our
website,
wws
or
calm.
It
is
based
on
trove.
B
It
has
several
extensions
to
trove
and
some
of
the
things
which
make
our
product
different
or
that
we
provide
guests
images
for
many
of
these
databases
drove
itself
does
not
provide
guests
images.
If
you
download
tro
from
the
community,
you
get
elements,
you
build
your
own
images.
If
you
want,
of
course,
we
provide
support
and
services.
We
support
Oracle
today,
which
is
not
supported
in
the
community
trove
and
we
release
our
product
every
three
months,
OpenStack
releases
every
six,
so
you
have
some
early
access
to
features.
B
B
So
the
question
yeah.
Well,
everybody
heard
the
question.
Let
me
think
about
that
for
a
second
I
believe
the
Cassandra
cluster
code
is
going
back
to
the
community.
So,
yes,
the
code
to
support
community
Cassandra
will
be
in
the
in
the
community
version
of
OpenStack,
a
community
version
of
OpenStack
drove
the
code
which
supports
datastax
enterprise
will
be
in
our
enterprise
product.
B
D
B
No
matter
what
the
database
troop
does
not
implement
its
own
code
for
backup
we
do.
Whatever
the
database
vendor
says,
you
should
do
for
backup,
I
believe
the
recommended
way
of
doing
a
backup
for
a
Cassandra
instance.
One
is
no
tool.
Snapshot,
I'm,
not
Cassandra,
certified
I.
Believe
a
bunch
of
you
here
are
so
correct
me.
B
One
option
is
to
leverage
op
center
and
do
the
OP
Center
way
of
doing
a
snapshot
of
a
cluster,
and
the
other
is
to
reinvent
that
and
do
those
same
commands
one
note
at
a
time:
I,
don't
know
what
the
final
implementation
is
going
to
be.
It's
going
to
be
one
of
those
two,
it's
still
a
work
in
progress.
B
B
B
B
Every
component
in
OpenStack
reports,
events
to
salamat,
ER
trove
reports,
several
events,
it's
a
matter
of
extending
trove
to
potentially
report
baseline
events,
performance
events
to
it.
We
do
get
some
sense
of
it
because
we
know
that
launching
an
instance
should
take
about
22
minutes,
at
least
that
I
might
on
my
laptop
on
a
production
environment.
I
can
do
it
in
slightly
less.
We
do
time
how
long
it
takes
if
a
certain
instance
takes
six
minutes
to
launch.
That's
a
good
indication
that
there's
something
screwed
up
with
that.
B
The
question
is:
can
you
use
instance
store
rather
than
using
cinder?
So
Cassandra
users
have
asked
about
this
a
lot
because
Cassandra's
more
data,
staxus
recommendation.
As
you
use
local
storage,
you
can
create
a
flavor
which
goes
to
a
hypervisor
where
storage
is
guaranteed
to
be
local,
so
trove
doesn't
get
involved
in
that
where's.
The
storage.
At
this
point
we
give
hints
to
the
underlying
projects
on
where
they
should
go
provision
there
piece
parts.
Therefore,
you
can
achieve
exactly
what
you're
talking
about.
B
That
is
something
which
we
do
when
so,
for
example,
when
you
graduate
a
instance
from
a
development
instance
to
a
production
instance,
you
could
say
on
the
development
instance:
I
want
to
run
on
spinning
rust
in
a
production.
I
want
to
run
on
locally
attached
SSD.
That's
exactly
how
you
would
do
this
by
saying
this
flavor
only
works
with
this
hypervisor.
B
This
hypervisor
is
only
supported
on
this
set
of
instances
or
this
set
of
physical
hardware
and
that
physical
Hardware
has
local
SSDs
as
opposed
to
the
other
one,
which
is
a
development
flavor
which
connects
to
an
ass.
So
it's
not
a
direct
solution,
but
trove
doesn't
directly
get
involved
in
storage,
so
we
have
to
use
the
other
underlying
projects
for
that.