<s>
High-availability	B-General_Concept
clusters	I-General_Concept
(	O
also	O
known	O
as	O
HA	B-General_Concept
clusters	I-General_Concept
,	O
fail-over	B-General_Concept
clusters	B-Architecture
)	O
are	O
groups	O
of	O
computers	O
that	O
support	O
server	B-Application
applications	I-Application
that	O
can	O
be	O
reliably	O
utilized	O
with	O
a	O
minimum	O
amount	O
of	O
down-time	O
.	O
</s>
<s>
They	O
operate	O
by	O
using	O
high	B-Application
availability	I-Application
software	I-Application
to	O
harness	O
redundant	O
computers	O
in	O
groups	O
or	O
clusters	B-Architecture
that	O
provide	O
continued	O
service	O
when	O
system	O
components	O
fail	O
.	O
</s>
<s>
Without	O
clustering	O
,	O
if	O
a	O
server	B-Application
running	O
a	O
particular	O
application	O
crashes	O
,	O
the	O
application	O
will	O
be	O
unavailable	O
until	O
the	O
crashed	O
server	B-Application
is	O
fixed	O
.	O
</s>
<s>
HA	O
clustering	O
remedies	O
this	O
situation	O
by	O
detecting	O
hardware/software	O
faults	O
,	O
and	O
immediately	O
restarting	O
the	O
application	O
on	O
another	O
system	O
without	O
requiring	O
administrative	O
intervention	O
,	O
a	O
process	O
known	O
as	O
failover	B-General_Concept
.	O
</s>
<s>
For	O
example	O
,	O
appropriate	O
file	O
systems	O
may	O
need	O
to	O
be	O
imported	O
and	O
mounted	O
,	O
network	B-Architecture
hardware	O
may	O
have	O
to	O
be	O
configured	O
,	O
and	O
some	O
supporting	O
applications	B-Application
may	O
need	O
to	O
be	O
running	O
as	O
well	O
.	O
</s>
<s>
HA	B-General_Concept
clusters	I-General_Concept
are	O
often	O
used	O
for	O
critical	O
databases	B-Application
,	O
file	O
sharing	O
on	O
a	O
network	B-Architecture
,	O
business	O
applications	B-Application
,	O
and	O
customer	O
services	O
such	O
as	O
electronic	O
commerce	O
websites	O
.	O
</s>
<s>
HA	B-General_Concept
cluster	I-General_Concept
implementations	O
attempt	O
to	O
build	O
redundancy	O
into	O
a	O
cluster	O
to	O
eliminate	O
single	O
points	O
of	O
failure	O
,	O
including	O
multiple	O
network	B-Architecture
connections	O
and	O
data	O
storage	O
which	O
is	O
redundantly	O
connected	O
via	O
storage	O
area	O
networks	O
.	O
</s>
<s>
HA	B-General_Concept
clusters	I-General_Concept
usually	O
use	O
a	O
heartbeat	B-Operating_System
private	O
network	B-Architecture
connection	O
which	O
is	O
used	O
to	O
monitor	O
the	O
health	O
and	O
status	O
of	O
each	O
node	O
in	O
the	O
cluster	O
.	O
</s>
<s>
One	O
subtle	O
but	O
serious	O
condition	O
all	O
clustering	O
software	O
must	O
be	O
able	O
to	O
handle	O
is	O
split-brain	B-General_Concept
,	O
which	O
occurs	O
when	O
all	O
of	O
the	O
private	O
links	O
go	O
down	O
simultaneously	O
,	O
but	O
the	O
cluster	O
nodes	O
are	O
still	O
running	O
.	O
</s>
<s>
HA	B-General_Concept
clusters	I-General_Concept
often	O
also	O
use	O
quorum	B-General_Concept
witness	O
storage	O
(	O
local	O
or	O
cloud	O
)	O
to	O
avoid	O
this	O
scenario	O
.	O
</s>
<s>
A	O
witness	O
device	O
cannot	O
be	O
shared	O
between	O
two	O
halves	O
of	O
a	O
split	O
cluster	O
,	O
so	O
in	O
the	O
event	O
that	O
all	O
cluster	O
members	O
cannot	O
communicate	O
with	O
each	O
other	O
(	O
e.g.	O
,	O
failed	O
heartbeat	B-Operating_System
)	O
,	O
if	O
a	O
member	O
cannot	O
access	O
the	O
witness	O
,	O
it	O
cannot	O
become	O
active	O
.	O
</s>
<s>
Not	O
every	O
application	O
can	O
run	O
in	O
a	O
high-availability	B-General_Concept
cluster	I-General_Concept
environment	O
,	O
and	O
the	O
necessary	O
design	O
decisions	O
need	O
to	O
be	O
made	O
early	O
in	O
the	O
software	O
design	O
phase	O
.	O
</s>
<s>
In	O
order	O
to	O
run	O
in	O
a	O
high-availability	B-General_Concept
cluster	I-General_Concept
environment	O
,	O
an	O
application	O
must	O
satisfy	O
at	O
least	O
the	O
following	O
technical	O
requirements	O
,	O
the	O
last	O
two	O
of	O
which	O
are	O
critical	O
to	O
its	O
reliable	O
function	O
in	O
a	O
cluster	O
,	O
and	O
are	O
the	O
most	O
difficult	O
to	O
satisfy	O
fully	O
:	O
</s>
<s>
The	O
application	O
must	O
be	O
able	O
to	O
use	O
shared	O
storage	O
(	O
NAS/SAN	O
)	O
.	O
</s>
<s>
A	O
number	O
of	O
these	O
constraints	O
can	O
be	O
minimized	O
through	O
the	O
use	O
of	O
virtual	O
server	B-Application
environments	O
,	O
wherein	O
the	O
hypervisor	O
itself	O
is	O
cluster-aware	O
and	O
provides	O
seamless	O
migration	O
of	O
virtual	O
machines	O
(	O
including	O
running	O
memory	O
state	O
)	O
between	O
physical	O
hosts	O
—	O
see	O
.	O
</s>
<s>
A	O
key	O
difference	O
between	O
this	O
approach	O
and	O
running	O
cluster-aware	O
applications	B-Application
is	O
that	O
the	O
latter	O
can	O
deal	O
with	O
server	B-Application
application	I-Application
crashes	O
and	O
support	O
live	O
"	O
rolling	O
"	O
software	O
upgrades	O
while	O
maintaining	O
client	O
access	O
to	O
the	O
service	O
(	O
e.g.	O
</s>
<s>
Failover	B-General_Concept
clustering	I-General_Concept
falls	O
under	O
planning	O
,	O
creating	O
and	O
configuring	O
also	O
troubleshooting	O
.	O
</s>
<s>
The	O
most	O
common	O
size	O
for	O
an	O
HA	B-General_Concept
cluster	I-General_Concept
is	O
a	O
two-node	O
cluster	O
,	O
since	O
that	O
is	O
the	O
minimum	O
required	O
to	O
provide	O
redundancy	O
,	O
but	O
many	O
clusters	B-Architecture
consist	O
of	O
many	O
more	O
,	O
sometimes	O
dozens	O
of	O
nodes	O
.	O
</s>
<s>
The	O
attached	O
diagram	O
is	O
a	O
good	O
overview	O
of	O
a	O
classic	O
HA	B-General_Concept
cluster	I-General_Concept
,	O
with	O
the	O
caveat	O
that	O
it	O
does	O
not	O
make	O
any	O
mention	O
of	O
quorum/witness	O
functionality	O
(	O
see	O
above	O
)	O
.	O
</s>
<s>
This	O
normally	O
refers	O
to	O
clusters	B-Architecture
that	O
have	O
multiple	O
services	O
running	O
simultaneously	O
;	O
in	O
the	O
single	O
service	O
case	O
,	O
this	O
degenerates	O
to	O
active/passive	O
.	O
</s>
<s>
N+M	O
—	O
In	O
cases	O
where	O
a	O
single	O
cluster	O
is	O
managing	O
many	O
services	O
,	O
having	O
only	O
one	O
dedicated	O
failover	B-General_Concept
node	O
might	O
not	O
offer	O
sufficient	O
redundancy	O
.	O
</s>
<s>
N-to-1	O
—	O
Allows	O
the	O
failover	B-General_Concept
standby	O
node	O
to	O
become	O
the	O
active	O
one	O
temporarily	O
,	O
until	O
the	O
original	O
node	O
can	O
be	O
restored	O
or	O
brought	O
back	O
online	O
,	O
at	O
which	O
point	O
the	O
services	O
or	O
instances	O
must	O
be	O
failed-back	O
to	O
it	O
in	O
order	O
to	O
restore	O
high	O
availability	O
.	O
</s>
<s>
N-to-N	O
—	O
A	O
combination	O
of	O
active/active	O
and	O
N+M	O
clusters	B-Architecture
,	O
N	O
to	O
N	O
clusters	B-Architecture
redistribute	O
the	O
services	O
,	O
instances	O
or	O
connections	O
from	O
the	O
failed	O
node	O
among	O
the	O
remaining	O
active	O
nodes	O
,	O
thus	O
eliminating	O
(	O
as	O
with	O
active/active	O
)	O
the	O
need	O
for	O
a	O
'	O
standby	O
 '	O
node	O
,	O
but	O
introducing	O
a	O
need	O
for	O
extra	O
capacity	O
on	O
all	O
active	O
nodes	O
.	O
</s>
<s>
The	O
terms	O
logical	O
host	O
or	O
cluster	O
logical	O
host	O
is	O
used	O
to	O
describe	O
the	O
network	B-Protocol
address	I-Protocol
that	O
is	O
used	O
to	O
access	O
services	O
provided	O
by	O
the	O
cluster	O
.	O
</s>
<s>
HA	B-General_Concept
clusters	I-General_Concept
usually	O
use	O
all	O
available	O
techniques	O
to	O
make	O
the	O
individual	O
systems	O
and	O
shared	O
infrastructure	O
as	O
reliable	O
as	O
possible	O
.	O
</s>
<s>
Disk	B-General_Concept
mirroring	I-General_Concept
(	O
or	O
Redundant	O
Arrays	O
of	O
Independent	O
Disks	O
—	O
RAID	O
)	O
so	O
that	O
failure	O
of	O
internal	O
disks	O
does	O
not	O
result	O
in	O
system	O
crashes	O
.	O
</s>
<s>
The	O
Distributed	B-Application
Replicated	I-Application
Block	I-Application
Device	I-Application
is	O
one	O
example	O
.	O
</s>
<s>
Redundant	O
network	B-Architecture
connections	O
so	O
that	O
single	O
cable	O
,	O
switch	O
,	O
or	O
network	B-Architecture
interface	O
failures	O
do	O
not	O
result	O
in	O
network	B-Architecture
outages	O
.	O
</s>
<s>
Redundant	O
storage	O
area	O
network	B-Architecture
(	O
SAN	O
)	O
connections	O
so	O
that	O
single	O
cable	O
,	O
switch	O
,	O
or	O
interface	O
failures	O
do	O
not	O
lead	O
to	O
loss	O
of	O
connectivity	O
to	O
the	O
storage	O
(	O
this	O
would	O
violate	O
shared	B-Operating_System
nothing	I-Operating_System
architecture	I-Operating_System
)	O
.	O
</s>
<s>
These	O
features	O
help	O
minimize	O
the	O
chances	O
that	O
the	O
clustering	O
failover	B-General_Concept
between	O
systems	O
will	O
be	O
required	O
.	O
</s>
<s>
In	O
such	O
a	O
failover	B-General_Concept
,	O
the	O
service	O
provided	O
is	O
unavailable	O
for	O
at	O
least	O
a	O
little	O
while	O
,	O
so	O
measures	O
to	O
avoid	O
failover	B-General_Concept
are	O
preferred	O
.	O
</s>
<s>
For	O
instance	O
,	O
the	O
Apache	B-Application
Cassandra	I-Application
API	B-General_Concept
Hector	B-Operating_System
defines	O
three	O
ways	O
to	O
configure	O
a	O
failover	B-General_Concept
:	O
</s>
