Strange hyperfabric problem

[Bruno Bossier_1]

When I run clic_shutdown, it stops the hyperfabric management daemon, but also unplumbs the hyperfabric interfaces (clic0 and clic1 in my case). When I run clic_start it only starts the hyperfabric management daemon, but does not plumb and configure the clic interfaces. When one of the 2 nodes in my cluster was rebooted or when I ran clic_shutdown on one of the 2 nodes, I have to run :

1. clic_shutdown on both nodes
2. clic_start on both nodes
3. /sbin/init.d/net start on both nodes

in this specific order.

This is very anoying. Any ideas ?

Bruno

[Steven E. Protter]

You might want to look ata the code of clic_shutdown and see if its doing something to that causes the network to go down.

what does ifconfig lan0 or ifconfig lan1 show while its down.

Are these inerfaces acting as a NIC or as SCSI devices to access storage?

SEP+

[Shine_5]

Hello

This is because a new field has been added to the Hyperfabric configuration
file /etc/rc.config.d/clic_global_conf to enable HMP. This field may be missing in
the Hyperfabric software .

Solution : -
If the error coours, reconfigure HyperFabric, using clic_init -c. This will
rebuild a correct configuration file.



Shine

[Bruno Bossier_1]

If you are refering to the field BCOMPAT_ENABLED, then these are both set to N.

Bruno

[Bruno Bossier_1]

- clic_shutdown indeed executes an ifconfig unplumb on all hyperfabric interfaces

- clic_start calls /sbin/init.d/net start to plumb the hyperfabric interfaces

This seems to be OK. It's just that when you ran clic_shutdown on one node, you also need to run clic_shutdown on the other node and then run clic_start on the 2 nodes.

Bruno