RowStatus (from RFC1903)
RowStatus ::= TEXTUAL-CONVENTION
STATUS current
DESCRIPTION
"The RowStatus textual convention is used to manage the
creation and deletion of conceptual rows, and is used as the
value of the SYNTAX clause for the status column of a
conceptual row (as described in Section 7.7.1 in RFC1902.)
The status column has six defined values:
- `active', which indicates that the conceptual row is
available for use by the managed device;
- `notInService', which indicates that the conceptual
row exists in the agent, but is unavailable for use by
the managed device (see NOTE below);
- `notReady', which indicates that the conceptual row
exists in the agent, but is missing information
necessary in order to be available for use by the
managed device;
- `createAndGo', which is supplied by a management
station wishing to create a new instance of a
conceptual row and to have its status automatically set
to active, making it available for use by the managed
device;
- `createAndWait', which is supplied by a management
station wishing to create a new instance of a
conceptual row (but not make it available for use by
the managed device); and,
- `destroy', which is supplied by a management station
wishing to delete all of the instances associated with
an existing conceptual row.
Whereas five of the six values (all except `notReady') may
be specified in a management protocol set operation, only
three values will be returned in response to a management
protocol retrieval operation: `notReady', `notInService' or
`active'. That is, when queried, an existing conceptual row
has only three states: it is either available for use by
the managed device (the status column has value `active');
it is not available for use by the managed device, though
the agent has sufficient information to make it so (the
status column has value `notInService'); or, it is not
available for use by the managed device, and an attempt to
make it so would fail because the agent has insufficient
information (the state column has value `notReady').
NOTE WELL
This textual convention may be used for a MIB table,
irrespective of whether the values of that table's
conceptual rows are able to be modified while it is
active, or whether its conceptual rows must be taken
out of service in order to be modified. That is, it is
the responsibility of the DESCRIPTION clause of the
status column to specify whether the status column must
not be `active' in order for the value of some other
column of the same conceptual row to be modified. If
such a specification is made, affected columns may be
changed by an SNMP set PDU if the RowStatus would not
be equal to `active' either immediately before or after
processing the PDU. In other words, if the PDU also
contained a varbind that would change the RowStatus
value, the column in question may be changed if the
RowStatus was not equal to `active' as the PDU was
received, or if the varbind sets the status to a value
other than 'active'.
Also note that whenever any elements of a row exist, the
RowStatus column must also exist.
To summarize the effect of having a conceptual row with a
status column having a SYNTAX clause value of RowStatus,
consider the following state diagram:
STATE
+--------------+-----------+-------------+-------------
| A | B | C | D
| |status col.|status column|
|status column | is | is |status column
ACTION |does not exist| notReady | notInService| is active
--------------+--------------+-----------+-------------+-------------
set status |noError ->D|inconsist- |inconsistent-|inconsistent-
column to | or | entValue| Value| Value
createAndGo |inconsistent- | | |
| Value| | |
--------------+--------------+-----------+-------------+-------------
set status |noError see 1|inconsist- |inconsistent-|inconsistent-
column to | or | entValue| Value| Value
createAndWait |wrongValue | | |
--------------+--------------+-----------+-------------+-------------
set status |inconsistent- |inconsist- |noError |noError
column to | Value| entValue| |
active | | | |
| | or | |
| | | |
| |see 2 ->D| ->D| ->D
--------------+--------------+-----------+-------------+-------------
set status |inconsistent- |inconsist- |noError |noError ->C
column to | Value| entValue| |
notInService | | | |
| | or | | or
| | | |
| |see 3 ->C| ->C|wrongValue
--------------+--------------+-----------+-------------+-------------
set status |noError |noError |noError |noError
column to | | | |
destroy | ->A| ->A| ->A| ->A
--------------+--------------+-----------+-------------+-------------
set any other |see 4 |noError |noError |see 5
column to some| | | |
value | | see 1| ->C| ->D
--------------+--------------+-----------+-------------+-------------
(1) goto B or C, depending on information available to the
agent.
(2) if other variable bindings included in the same PDU,
provide values for all columns which are missing but
required, then return noError and goto D.
(3) if other variable bindings included in the same PDU,
provide values for all columns which are missing but
required, then return noError and goto C.
(4) at the discretion of the agent, the return value may be
either:
inconsistentName: because the agent does not choose to
create such an instance when the corresponding
RowStatus instance does not exist, or
inconsistentValue: if the supplied value is
inconsistent with the state of some other MIB object's
value, or
noError: because the agent chooses to create the
instance.
If noError is returned, then the instance of the status
column must also be created, and the new state is B or C,
depending on the information available to the agent. If
inconsistentName or inconsistentValue is returned, the row
remains in state A.
(5) depending on the MIB definition for the column/table,
either noError or inconsistentValue may be returned.
NOTE: Other processing of the set request may result in a
response other than noError being returned, e.g.,
wrongValue, noCreation, etc.
Conceptual Row Creation
There are four potential interactions when creating a
conceptual row: selecting an instance-identifier which is
not in use; creating the conceptual row; initializing any
objects for which the agent does not supply a default; and,
making the conceptual row available for use by the managed
device.
Interaction 1: Selecting an Instance-Identifier
The algorithm used to select an instance-identifier varies
for each conceptual row. In some cases, the instance-
identifier is semantically significant, e.g., the
destination address of a route, and a management station
selects the instance-identifier according to the semantics.
In other cases, the instance-identifier is used solely to
distinguish conceptual rows, and a management station
without specific knowledge of the conceptual row might
examine the instances present in order to determine an
unused instance-identifier. (This approach may be used, but
it is often highly sub-optimal; however, it is also a
questionable practice for a naive management station to
attempt conceptual row creation.)
Alternately, the MIB module which defines the conceptual row
might provide one or more objects which provide assistance
in determining an unused instance-identifier. For example,
if the conceptual row is indexed by an integer-value, then
an object having an integer-valued SYNTAX clause might be
defined for such a purpose, allowing a management station to
issue a management protocol retrieval operation. In order
to avoid unnecessary collisions between competing management
stations, `adjacent' retrievals of this object should be
different.
Finally, the management station could select a pseudo-random
number to use as the index. In the event that this index
was already in use and an inconsistentValue was returned in
response to the management protocol set operation, the
management station should simply select a new pseudo-random
number and retry the operation.
A MIB designer should choose between the two latter
algorithms based on the size of the table (and therefore the
efficiency of each algorithm). For tables in which a large
number of entries are expected, it is recommended that a MIB
object be defined that returns an acceptable index for
creation. For tables with small numbers of entries, it is
recommended that the latter pseudo-random index mechanism be
used.
Interaction 2: Creating the Conceptual Row
Once an unused instance-identifier has been selected, the
management station determines if it wishes to create and
activate the conceptual row in one transaction or in a
negotiated set of interactions.
Interaction 2a: Creating and Activating the Conceptual Row
The management station must first determine the column
requirements, i.e., it must determine those columns for
which it must or must not provide values. Depending on the
complexity of the table and the management station's
knowledge of the agent's capabilities, this determination
can be made locally by the management station. Alternately,
the management station issues a management protocol get
operation to examine all columns in the conceptual row that
it wishes to create. In response, for each column, there
are three possible outcomes:
- a value is returned, indicating that some other
management station has already created this conceptual
row. We return to interaction 1.
- the exception `noSuchInstance' is returned,
indicating that the agent implements the object-type
associated with this column, and that this column in at
least one conceptual row would be accessible in the MIB
view used by the retrieval were it to exist. For those
columns to which the agent provides read-create access,
the `noSuchInstance' exception tells the management
station that it should supply a value for this column
when the conceptual row is to be created.
- the exception `noSuchObject' is returned, indicating
that the agent does not implement the object-type
associated with this column or that there is no
conceptual row for which this column would be
accessible in the MIB view used by the retrieval. As
such, the management station cannot issue any
management protocol set operations to create an
instance of this column.
Once the column requirements have been determined, a
management protocol set operation is accordingly issued.
This operation also sets the new instance of the status
column to `createAndGo'.
When the agent processes the set operation, it verifies that
it has sufficient information to make the conceptual row
available for use by the managed device. The information
available to the agent is provided by two sources: the
management protocol set operation which creates the
conceptual row, and, implementation-specific defaults
supplied by the agent (note that an agent must provide
implementation-specific defaults for at least those objects
which it implements as read-only). If there is sufficient
information available, then the conceptual row is created, a
`noError' response is returned, the status column is set to
`active', and no further interactions are necessary (i.e.,
interactions 3 and 4 are skipped). If there is insufficient
information, then the conceptual row is not created, and the
set operation fails with an error of `inconsistentValue'.
On this error, the management station can issue a management
protocol retrieval operation to determine if this was
because it failed to specify a value for a required column,
or, because the selected instance of the status column
already existed. In the latter case, we return to
interaction 1. In the former case, the management station
can re-issue the set operation with the additional
information, or begin interaction 2 again using
`createAndWait' in order to negotiate creation of the
conceptual row.
NOTE WELL
Regardless of the method used to determine the column
requirements, it is possible that the management
station might deem a column necessary when, in fact,
the agent will not allow that particular columnar
instance to be created or written. In this case, the
management protocol set operation will fail with an
error such as `noCreation' or `notWritable'. In this
case, the management station decides whether it needs
to be able to set a value for that particular columnar
instance. If not, the management station re-issues the
management protocol set operation, but without setting
a value for that particular columnar instance;
otherwise, the management station aborts the row
creation algorithm.
Interaction 2b: Negotiating the Creation of the Conceptual
Row
The management station issues a management protocol set
operation which sets the desired instance of the status
column to `createAndWait'. If the agent is unwilling to
process a request of this sort, the set operation fails with
an error of `wrongValue'. (As a consequence, such an agent
must be prepared to accept a single management protocol set
operation, i.e., interaction 2a above, containing all of the
columns indicated by its column requirements.) Otherwise,
the conceptual row is created, a `noError' response is
returned, and the status column is immediately set to either
`notInService' or `notReady', depending on whether it has
sufficient information to make the conceptual row available
for use by the managed device. If there is sufficient
information available, then the status column is set to
`notInService'; otherwise, if there is insufficient
information, then the status column is set to `notReady'.
Regardless, we proceed to interaction 3.
Interaction 3: Initializing non-defaulted Objects
The management station must now determine the column
requirements. It issues a management protocol get operation
to examine all columns in the created conceptual row. In
the response, for each column, there are three possible
outcomes:
- a value is returned, indicating that the agent
implements the object-type associated with this column
and had sufficient information to provide a value. For
those columns to which the agent provides read-create
access (and for which the agent allows their values to
be changed after their creation), a value return tells
the management station that it may issue additional
management protocol set operations, if it desires, in
order to change the value associated with this column.
- the exception `noSuchInstance' is returned,
indicating that the agent implements the object-type
associated with this column, and that this column in at
least one conceptual row would be accessible in the MIB
view used by the retrieval were it to exist. However,
the agent does not have sufficient information to
provide a value, and until a value is provided, the
conceptual row may not be made available for use by the
managed device. For those columns to which the agent
provides read-create access, the `noSuchInstance'
exception tells the management station that it must
issue additional management protocol set operations, in
order to provide a value associated with this column.
- the exception `noSuchObject' is returned, indicating
that the agent does not implement the object-type
associated with this column or that there is no
conceptual row for which this column would be
accessible in the MIB view used by the retrieval. As
such, the management station cannot issue any
management protocol set operations to create an
instance of this column.
If the value associated with the status column is
`notReady', then the management station must first deal with
all `noSuchInstance' columns, if any. Having done so, the
value of the status column becomes `notInService', and we
proceed to interaction 4.
Interaction 4: Making the Conceptual Row Available
Once the management station is satisfied with the values
associated with the columns of the conceptual row, it issues
a management protocol set operation to set the status column
to `active'. If the agent has sufficient information to
make the conceptual row available for use by the managed
device, the management protocol set operation succeeds (a
`noError' response is returned). Otherwise, the management
protocol set operation fails with an error of
`inconsistentValue'.
NOTE WELL
A conceptual row having a status column with value
`notInService' or `notReady' is unavailable to the
managed device. As such, it is possible for the
managed device to create its own instances during the
time between the management protocol set operation
which sets the status column to `createAndWait' and the
management protocol set operation which sets the status
column to `active'. In this case, when the management
protocol set operation is issued to set the status
column to `active', the values held in the agent
supersede those used by the managed device.
If the management station is prevented from setting the
status column to `active' (e.g., due to management station
or network failure) the conceptual row will be left in the
`notInService' or `notReady' state, consuming resources
indefinitely. The agent must detect conceptual rows that
have been in either state for an abnormally long period of
time and remove them. It is the responsibility of the
DESCRIPTION clause of the status column to indicate what an
abnormally long period of time would be. This period of
time should be long enough to allow for human response time
(including `think time') between the creation of the
conceptual row and the setting of the status to `active'.
In the absence of such information in the DESCRIPTION
clause, it is suggested that this period be approximately 5
minutes in length. This removal action applies not only to
newly-created rows, but also to previously active rows which
are set to, and left in, the notInService state for a
prolonged period exceeding that which is considered normal
for such a conceptual row.
Conceptual Row Suspension
When a conceptual row is `active', the management station
may issue a management protocol set operation which sets the
instance of the status column to `notInService'. If the
agent is unwilling to do so, the set operation fails with an
error of `wrongValue'. Otherwise, the conceptual row is
taken out of service, and a `noError' response is returned.
It is the responsibility of the DESCRIPTION clause of the
status column to indicate under what circumstances the
status column should be taken out of service (e.g., in order
for the value of some other column of the same conceptual
row to be modified).
Conceptual Row Deletion
For deletion of conceptual rows, a management protocol set
operation is issued which sets the instance of the status
column to `destroy'. This request may be made regardless of
the current value of the status column (e.g., it is possible
to delete conceptual rows which are either `notReady',
`notInService' or `active'.) If the operation succeeds,
then all instances associated with the conceptual row are
immediately removed."
SYNTAX INTEGER {
-- the following two values are states:
-- these values may be read or written
active(1),
notInService(2),
-- the following value is a state:
-- this value may be read, but not written
notReady(3),
-- the following three values are
-- actions: these values may be written,
-- but are never read
createAndGo(4),
createAndWait(5),
destroy(6)
}