shine
shine copied to clipboard
cea-hpc
Shine does not handle nodes which have a DNS name different from their system name
In order to determine if some actions are required locally on a node, shine compares the node name specified in the 'nodes' entry of the nid_map keyword with the node name returned by the 'hostname' command on the node. Additionally, shine uses this same 'nodes' entry to make ssh connections on the node.
In some configurations, the I/O and compute nodes may be reachable by a network name that is different from their system name.
Example: The admin node can access node io1 through a management network only. The DNS name of io1 on this network is: io1-mn The 'io1' DNS name is used for the interconnect network. The system name of io1 as returned by 'hostname': io1 (short name) or io1.domainname (fqdn name).
To allow shine to access this node, the nid_map should be:
nid_map: nodes=io1-mn nids=io1@o2ib0
This nid_map allows shine to make an ssh connection on io1, but, locally
on the node, shine will find that it has nothing to do on this node,
because 'io1-mn' is different from 'io1'.
To allow shine to run a local command on the node, the nid_map should be:
nid_map: nodes=io1 nids=io1@o2ib0
However, shine will not be able to make an ssh connection on the node,
as it is only accessible by its DNS name 'io1-mn'.
A workaround exists, consisting in specifying the system name in the nid_map, and setting up an alias based ssh config on the admin node. But it would be a good thing if shine was able to handle such situations by its own, as this workaround may not always be possible to set up, depending on security rules on the cluster.
Reported by: theryf
