IBM RS/6000 SP, Руководство по определению проблем

Страница: 173 / 316

This soft copy for use by IBM employees only.
node with the next highest IP address (and so on), until the ring is completed by
the node with the lowest IP address
pinging to the Control Workstation.
If there is no response to several retries of the ping packet within the timeout
period, that heartbeat daemon will notify the group leader that connectivity has
been lost over en0, which will then take that node out of the ring.
The
group leader will notify all the nodes of this action so that the ring can be
re-established by missing out the node from which there is no response over
en0. The heartbeat daemon on the Control Workstation will pass the information
about this node up to the host_responds daemon on the Control Workstation,
which in turn passes it up to the SDR daemon to update the
host_responds
object class of the SDR.
The spmon command uses this object to find out the host_responds status of all
the nodes. It may take a few seconds before the loss of connectivity is reflected
by the node changing from green to red on the
spmon GUI due to the timeout
period mentioned before and the time taken for spmon to again read the object
class and refresh the
host_responds GUI.
The heartbeat daemon on the Control Workstation will send out regular ping
packets (known as proclaim packets) to any nodes that have been marked as
missing from the ring in the host_responds object class, to test whether
connectivity has returned, and will reintegrate them back into the ring if that is
the case.
If a node goes red on the spmon GUI, run the following command:
# SDRGetObjects host_responds
Here is an example of the output:
node_number host_responds
1 1
3 1
5 1
6 1
7 1
8 1
9 1
10 1
11 1
12 0
In this example, all the nodes have en0 connectivity except node 12, which has
an entry of 0 (signifying a loss of connectivity).
In such a case, first test the connectivity manually by pinging the en0 interface.
If the
ping is successful, wait a few seconds to see if spmon
will refresh. If it does
not, then check that the heartbeat, SDR, and heartbeat daemons are running on
the Control Workstation, and that the heartbeat daemon is running on the
problematic node.
Examples have already been given for heartbeat and SDR, but following is an
example for the host_responds daemon. It shows how to find out if it is running,
the inittab entries, and how to start it if it has died:
# ps -ef | grep hr
Following is an example of the process name output from this command:
Chapter 5. System Partitioning
153