Problems with host DNS resolution when configuring services

[verdurin]

Following the README procedure, I found that stackhpc.nfs failed at the point of trying to mount NFS because it couldn't resolve the hostname of the Slurm control node:

TASK [stackhpc.nfs : mount the filesystem] ***************************************************************************************************************************************************************************************************
fatal: [slurmtest03-login-0]: FAILED! => {
    "changed": false
}

MSG:

Error mounting /home: mount.nfs: Failed to resolve server slurmtest03-control: Name or service not known

fatal: [slurmtest03-compute-1]: FAILED! => {
    "changed": false
}

MSG:

Error mounting /home: mount.nfs: Failed to resolve server slurmtest03-control: Name or service not known

fatal: [slurmtest03-compute-0]: FAILED! => {
    "changed": false
}

MSG:

Error mounting /home: mount.nfs: Failed to resolve server slurmtest03-control: Name or service not known

I see there has been previous discussion about whether to use IPs or hostnames.

Can work around this for now with custom SSH config, but I think this should be documented more explicitly.

[verdurin]

A similar error occurred with MySQL, and I applied the same fix i.e. adding an IP address in the environment vars.

[sjpb]

The workaround for a lack of working internal dns is to enable the etc_hosts role.

In environments/$your_env/inventory/groups, at line 54 (probably) you will have this:

[etc_hosts]
# Hosts to manage /etc/hosts e.g. if no internal DNS. See ansible/roles/etc_hosts/README.md

you need to enable this by adding hosts to this group e.g.

[etc_hosts]
# Hosts to manage /etc/hosts e.g. if no internal DNS. See ansible/roles/etc_hosts/README.md
cluster

edit: we should probably turn this on by default TBH

[verdurin]

Ah, I thought there should be provision for this, and I just hadn't found it. Will try that now, thanks.

[sjpb]

You can see all the available groups in https://github.com/stackhpc/ansible-slurm-appliance/blob/main/environments/common/inventory/groups. Usually there is a role of the same name with documentation.

[verdurin]

I think it would be friendlier to newbies to have this on by default, yes.