<s>
Fencing	B-Operating_System
is	O
the	O
process	O
of	O
isolating	O
a	O
node	B-Protocol
of	O
a	O
computer	B-Architecture
cluster	I-Architecture
or	O
protecting	O
shared	O
resources	O
when	O
a	O
node	B-Protocol
appears	O
to	O
be	O
malfunctioning	O
.	O
</s>
<s>
The	O
failed	O
node	B-Protocol
may	O
have	O
control	O
over	O
shared	O
resources	O
that	O
need	O
to	O
be	O
reclaimed	O
and	O
if	O
the	O
node	B-Protocol
is	O
acting	O
erratically	O
,	O
the	O
rest	O
of	O
the	O
system	O
needs	O
to	O
be	O
protected	O
.	O
</s>
<s>
Fencing	B-Operating_System
may	O
thus	O
either	O
disable	O
the	O
node	B-Protocol
,	O
or	O
disallow	O
shared	O
storage	O
access	O
,	O
thus	O
ensuring	O
data	O
integrity	O
.	O
</s>
<s>
A	O
node	B-Protocol
fence	O
(	O
or	O
I/O	B-General_Concept
fence	O
)	O
is	O
a	O
virtual	O
"	O
fence	O
"	O
that	O
separates	O
nodes	O
which	O
must	O
not	O
have	O
access	O
to	O
a	O
shared	O
resource	O
from	O
that	O
resource	O
.	O
</s>
<s>
It	O
may	O
separate	O
an	O
active	O
node	B-Protocol
from	O
its	O
backup	O
.	O
</s>
<s>
Mechanisms	O
such	O
as	O
STONITH	B-Operating_System
are	O
designed	O
to	O
prevent	O
this	O
condition	O
.	O
</s>
<s>
Isolating	O
a	O
node	B-Protocol
means	O
ensuring	O
that	O
I/O	B-General_Concept
can	O
no	O
longer	O
be	O
done	O
from	O
it	O
.	O
</s>
<s>
Fencing	B-Operating_System
is	O
typically	O
done	O
automatically	O
,	O
by	O
cluster	O
infrastructure	O
such	O
as	O
shared	O
disk	O
file	O
systems	O
,	O
in	O
order	O
to	O
protect	O
processes	B-Operating_System
from	O
other	O
active	O
nodes	O
modifying	O
the	O
resources	O
during	O
node	B-Protocol
failures	O
.	O
</s>
<s>
Mechanisms	O
to	O
support	O
fencing	B-Operating_System
,	O
such	O
as	O
the	O
reserve/release	O
mechanism	O
of	O
SCSI	O
,	O
have	O
existed	O
since	O
at	O
least	O
1985	O
.	O
</s>
<s>
Fencing	B-Operating_System
is	O
required	O
because	O
it	O
is	O
impossible	O
to	O
distinguish	O
between	O
a	O
real	O
failure	O
and	O
a	O
temporary	O
hang	B-General_Concept
.	O
</s>
<s>
If	O
the	O
malfunctioning	O
node	B-Protocol
is	O
really	O
down	O
,	O
then	O
it	O
cannot	O
do	O
any	O
damage	O
,	O
so	O
theoretically	O
no	O
action	O
would	O
be	O
required	O
(	O
it	O
could	O
simply	O
be	O
brought	O
back	O
into	O
the	O
cluster	O
with	O
the	O
usual	O
join	O
process	O
)	O
.	O
</s>
<s>
However	O
,	O
because	O
there	O
is	O
a	O
possibility	O
that	O
a	O
malfunctioning	O
node	B-Protocol
could	O
itself	O
consider	O
the	O
rest	O
of	O
the	O
cluster	O
to	O
be	O
the	O
one	O
that	O
is	O
malfunctioning	O
,	O
a	O
split	O
brain	O
condition	O
could	O
ensue	O
,	O
and	O
cause	O
data	O
corruption	O
.	O
</s>
<s>
There	O
are	O
two	O
classes	O
of	O
fencing	B-Operating_System
methods	O
,	O
one	O
which	O
disables	O
a	O
node	B-Protocol
itself	O
,	O
the	O
other	O
disallows	O
access	O
to	O
resources	O
such	O
as	O
shared	O
disks	O
.	O
</s>
<s>
In	O
some	O
cases	O
,	O
it	O
is	O
assumed	O
that	O
if	O
a	O
node	B-Protocol
does	O
not	O
respond	O
after	O
a	O
given	O
time-threshold	O
it	O
may	O
be	O
assumed	O
as	O
non-operational	O
,	O
although	O
there	O
are	O
counterexamples	O
,	O
e.g.	O
</s>
<s>
The	O
STONITH	B-Operating_System
method	O
stands	O
for	O
"	O
Shoot	O
The	O
Other	O
Node	B-Protocol
In	O
The	O
Head	O
"	O
,	O
meaning	O
that	O
the	O
suspected	O
node	B-Protocol
is	O
disabled	O
or	O
powered	O
off	O
.	O
</s>
<s>
For	O
instance	O
,	O
power	O
fencing	B-Operating_System
uses	O
a	O
power	O
controller	O
to	O
turn	O
off	O
an	O
inoperable	O
node	B-Protocol
.	O
</s>
<s>
The	O
node	B-Protocol
may	O
then	O
restart	O
itself	O
and	O
join	O
the	O
cluster	O
later	O
.	O
</s>
<s>
However	O
,	O
there	O
are	O
approaches	O
in	O
which	O
an	O
operator	O
is	O
informed	O
of	O
the	O
need	O
for	O
a	O
manual	O
restart	O
for	O
the	O
node	B-Protocol
.	O
</s>
<s>
The	O
resources	O
fencing	B-Operating_System
approach	O
disallows	O
access	O
to	O
resources	O
without	O
powering	O
off	O
the	O
node	B-Protocol
.	O
</s>
<s>
Persistent	B-Operating_System
reservation	I-Operating_System
fencing	B-Operating_System
uses	O
the	O
SCSI3	O
persistent	B-Operating_System
reservations	I-Operating_System
to	O
block	O
access	O
to	O
shared	O
storage	O
.	O
</s>
<s>
When	O
the	O
cluster	O
has	O
only	O
two	O
nodes	O
,	O
the	O
reserve/release	O
method	O
may	O
be	O
used	O
as	O
a	O
two	O
node	B-Protocol
STONITH	B-Operating_System
whereby	O
upon	O
detecting	O
that	O
node	B-Protocol
B	O
has	O
'	O
failed	O
 '	O
,	O
node	B-Protocol
A	O
will	O
issue	O
the	O
reserve	O
and	O
obtain	O
all	O
resources	O
(	O
e.g.	O
</s>
<s>
Node	B-Protocol
B	O
will	O
be	O
disabled	O
if	O
it	O
tries	O
to	O
do	O
I/O	B-General_Concept
(	O
in	O
case	O
it	O
was	O
temporarily	O
hung	O
)	O
.	O
</s>
<s>
On	O
node	B-Protocol
B	O
the	O
I/O	B-General_Concept
failure	O
triggers	O
some	O
code	O
to	O
kill	O
the	O
node	B-Protocol
.	O
</s>
<s>
Persistent	B-Operating_System
reservation	I-Operating_System
is	O
essentially	O
a	O
match	O
on	O
a	O
key	O
,	O
so	O
the	O
node	B-Protocol
which	O
has	O
the	O
right	O
key	O
can	O
do	O
I/O	B-General_Concept
,	O
otherwise	O
its	O
I/O	B-General_Concept
fails	O
.	O
</s>
<s>
However	O
,	O
it	O
may	O
not	O
always	O
be	O
possible	O
to	O
change	O
the	O
key	O
on	O
the	O
failed	O
node	B-Protocol
.	O
</s>
<s>
STONITH	B-Operating_System
is	O
an	O
easier	O
and	O
simpler	O
method	O
to	O
implement	O
on	O
multiple	O
clusters	O
,	O
while	O
the	O
various	O
approaches	O
to	O
resources	O
fencing	B-Operating_System
require	O
specific	O
implementation	O
approaches	O
for	O
each	O
cluster	O
implementation	O
.	O
</s>
