HPE CMU 8.0: Failed password for root on node with passwordless login

[brs]

Hi, I'm setting up a server with CMU to manage a small cluster and I'm having an issue which I can't seem to solve.

The CMU server (CentOS 7.2) is configured with passwordless login, which works fine, to the node(s) on the cluster (also CentOS 7.2). However, as soon as CMU is up and running it starts trying to login on the node(s), and I get messages like this on the node (/var/log/secure):

Oct 24 15:33:50 noatun1 sshd[21682]: Failed password for root from 10.230.98.252 port 55444 ssh2
Oct 24 15:33:50 noatun1 sshd[21682]: Connection closed by 10.230.98.252 [preauth]

As well as an annoying 1000+ failed attempts to login message whenever I log in to the node.

In MainMonitoringDaemon_cmuserver.log I get the following.

[25-Oct-2016_12:08:39] [CMUSecReelector   ]  [Warning] Most likely password- less rsh/rcp or ssh/scp does not work to host ->10.230.98.254<-
[25-Oct-2016_12:08:47] [CMUSecReelector   ]  [Warning] Could not perform checksum check MonitSrHelperGetSumInFile

Does anybody have any clues on what could be the issue. If I (through the CMU GUI) select ssh login on the node, the terminal opens and log in without any issues (or error messages).

Sincerely

Brian Højen-Sørensen

[Dave-SBS]

You don't say if the compute nodes can ssh without password back to the head node.  That is the first thing that comes to mind.

[Abhishekc · Accepted solution]

Hello,

For monitoring to work properly, please 

• Make sure password less ssh works between management node & all the compute nodes and among all compute nodes using IP/hostname. Also, compute nodes should be able to do password less ssh to itself.

• Check if management node is not taking too long to connect to compute nodes in that problematic network entity via ssh due to improper gateway/DNS settings. It should be less than 4 secs.

BTW, Is this a customer cluster ? If yes, please let us know the customer details and raise an issue with HPE local support center.

[brs]

Thanks,

I have first gotten around to it now, but that was indeed the issue. The nodes were not set up with passwordless login using ip's - only hostname, which I hadn't thought of (since it worked with hostname).

On top of that I had a problem with a mix of hostbased and keybased passwordless logins.

Cheers,

Brian