| This table contains one entry per interface. It defines
objects which allow a network manager to instruct an agent
to test an interface for various faults. Tests for an
interface are defined in the media-specific MIB for that
interface. After invoking a test, the object ifTestResult
can be read to determine the outcome. If an agent can not
perform the test, ifTestResult is set to so indicate. The
object ifTestCode can be used to provide further test-
specific or interface-specific (or even enterprise-specific)
information concerning the outcome of the test. Only one
test can be in progress on each interface at any one time.
If one test is in progress when another test is invoked, the
second test is rejected. Some agents may reject a test when
a prior test is active on another interface.
Before starting a test, a manager-station must first obtain
'ownership' of the entry in the ifTestTable for the
interface to be tested. This is accomplished with the
ifTestId and ifTestStatus objects as follows:
try_again:
get (ifTestId, ifTestStatus)
while (ifTestStatus != notInUse)
/*
* Loop while a test is running or some other
* manager is configuring a test.
*/
short delay
get (ifTestId, ifTestStatus)
}
/*
* Is not being used right now -- let's compete
* to see who gets it.
*/
lock_value = ifTestId
if ( set(ifTestId = lock_value, ifTestStatus = inUse,
ifTestOwner = 'my-IP-address') == FAILURE)
/*
* Another manager got the ifTestEntry -- go
* try again
*/
goto try_again;
/*
* I have the lock
*/
set up any test parameters.
/*
* This starts the test
*/
set(ifTestType = test_to_run);
wait for test completion by polling ifTestResult
when test completes, agent sets ifTestResult
agent also sets ifTestStatus = 'notInUse'
retrieve any additional test results, and ifTestId
if (ifTestId == lock_value+1) results are valid
A manager station first retrieves the value of the
appropriate ifTestId and ifTestStatus objects, periodically
repeating the retrieval if necessary, until the value of
ifTestStatus is 'notInUse'. The manager station then tries
to set the same ifTestId object to the value it just
retrieved, the same ifTestStatus object to 'inUse', and the
corresponding ifTestOwner object to a value indicating
itself. If the set operation succeeds then the manager has
obtained ownership of the ifTestEntry, and the value of the
ifTestId object is incremented by the agent (per the
semantics of TestAndIncr). Failure of the set operation
indicates that some other manager has obtained ownership of
the ifTestEntry.
Once ownership is obtained, any test parameters can be
setup, and then the test is initiated by setting ifTestType.
On completion of the test, the agent sets ifTestStatus to
'notInUse'. Once this occurs, the manager can retrieve the
results. In the (rare) event that the invocation of tests
by two network managers were to overlap, then there would be
a possibility that the first test's results might be
overwritten by the second test's results prior to the first
results being read. This unlikely circumstance can be
detected by a network manager retrieving ifTestId at the
same time as retrieving the test results, and ensuring that
the results are for the desired request.
If ifTestType is not set within an abnormally long period of
time after ownership is obtained, the agent should time-out
the manager, and reset the value of the ifTestStatus object
back to 'notInUse'. It is suggested that this time-out
period be 5 minutes.
In general, a management station must not retransmit a
request to invoke a test for which it does not receive a
response; instead, it properly inspects an agent's MIB to
determine if the invocation was successful. Only if the
invocation was unsuccessful, is the invocation request
retransmitted.
Some tests may require the interface to be taken off-line in
order to execute them, or may even require the agent to
reboot after completion of the test. In these
circumstances, communication with the management station
invoking the test may be lost until after completion of the
test. An agent is not required to support such tests.
However, if such tests are supported, then the agent should
make every effort to transmit a response to the request
which invoked the test prior to losing communication. When
the agent is restored to normal service, the results of the
test are properly made available in the appropriate objects.
Note that this requires that the ifIndex value assigned to
an interface must be unchanged even if the test causes a
reboot. An agent must reject any test for which it cannot,
perhaps due to resource constraints, make available at least
the minimum amount of information after that test
completes. |
