Node hostname has been wrongly changed.

[Eth4nSheng]

HI,all

I found that when I initially pushed the computing node system and logged into the node, the hostname was the correct node number. However, when I restarted the node and logged in again, the hostname was wrongly changed to localhost. This would cause the cluster to fail to function properly. Could you please advise me on which aspects I should start troubleshooting this issue?

I deploy version is 15.3u1.

Ethan
tks

[Eth4nSheng]

Additionally, this fault phenomenon can be consistently reproduced through the push OS. It is the first time after the push system that the hostname is normal. After restarting the node again, the hostname changes back to localhost. My current solution is to manually change the node names to the correct numbers. However, this is not feasible on a large scale.

[aphmschonewille]

Hi there,

I see both comments. Can i draw the conclusion the name change only happens after a push?
A push to a node is powerfull, easy but can be leading to sometimes undesired outcomes as the image is not having node (name) information (on purpose). What can be synced can be set in the exclude_* variables for the osimage in luna, however a reboot is always the safest.
Can you provide info on what distribution your pushing?

-Antoine

[Eth4nSheng]

hi antoine

Yes, I confirm that it occurred after the system pushed the notification.
At the same time, I also conducted a test. After the system configuration was completed, I attempted to re-execute the push os. When the OS was first installed and started, the hostname was the normal expected corresponding node number. After restarting again, it still changed to localhost.

This is the information of my computing node:
NAME="Rocky Linux"
VERSION="9.7 (Blue Onyx)"
ID="rocky"
ID_LIKE="rhel centos fedora"
VERSION_ID="9.7"
PLATFORM_ID="platform:el9"
PRETTY_NAME="Rocky Linux 9.7 (Blue Onyx)"
ANSI_COLOR="0;32"
LOGO="fedora-logo-icon"
CPE_NAME="cpe:/o:rocky:rocky:9::baseos"
HOME_URL="https://rockylinux.org/"
VENDOR_NAME="RESF"
VENDOR_URL="https://resf.org/"
BUG_REPORT_URL="https://bugs.rockylinux.org/"
SUPPORT_END="2032-05-31"
ROCKY_SUPPORT_PRODUCT="Rocky-Linux-9"
ROCKY_SUPPORT_PRODUCT_VERSION="9.7"
REDHAT_SUPPORT_PRODUCT="Rocky Linux"
REDHAT_SUPPORT_PRODUCT_VERSION="9.7"

[Eth4nSheng]

this is the configuration information of my Luna.

00000 14:03:18 [root@master ~]# luna osimage show compute-image
+-------------------------------------------------------------------------------------+
|                               Osimage => compute-image                              |
+------------------+------------------------------------------------------------------+
| name             | compute-image                                                    |
| grab_filesystems | /, /boot                                                         |
| grab_exclude     | /proc/*, /sys/*, /dev/*, /tmp/*, /run/*, /opt/*, /var/log/*, ... |
| initrdfile       | compute-image-1778739096-initramfs-5.14.0-611.54.1.el9_7.x86_64  |
| kernelversion    | 5.14.0-611.54.1.el9_7.x86_64                                     |
| kernelfile       | compute-image-1778739096-vmlinuz-5.14.0-611.54.1.el9_7.x86_64    |
| kernelmodules    | ipmi_devintf, ipmi_si, ipmi_msghandler                           |
| kerneloptions    | _net.ifnames=0 _biosdevname=0                                    |
| path             | /trinity/images/compute-image                                    |
| imagefile        | compute-image-1778739152.tar.bz2                                 |
| distribution     | redhat                                                           |
| osrelease        | 9                                                                |
| comment          | None                                                             |
| tag              | default                                                          |
+------------------+------------------------------------------------------------------+
000000 14:03:33 [root@master ~]# luna cluster
+----------------------------------------------------------------------------------+
|                               Cluster => mycluster                               |
+----------------------+-----------------------------------------------------------+
| name                 | mycluster                                                 |
| controller           | hostname = master                                         |
|                      | status = None                                             |
|                      | vendor = None                                             |
|                      | serverport = 7050                                         |
|                      | ipaddress = 172.1.1.254                                   |
|                      | luna_config = /trinity/local/luna/daemon/config/luna.ini  |
| technical_contacts   | root@localhost                                            |
| provision_method     | torrent                                                   |
| provision_fallback   | http                                                      |
| nameserver_ip        | 172.1.1.254                                               |
| forwardserver_ip     | 223.5.5.5,8.8.8.8                                         |
| domain_search        | cluster,ib,ipmi                                           |
| ntp_server           | 172.1.1.254                                               |
| security             | False                                                     |
| nextnode_discover    | False                                                     |
| createnode_ondemand  | True                                                      |
| createnode_macashost | False                                                     |
| packing_bootpause    | True                                                      |
| user                 | None                                                      |
| debug                | False                                                     |
+----------------------+-----------------------------------------------------------+
/

[aphmschonewille]

In the image, can you remove etc/NetworkManager/system-connections/* and then repack? Give the node a fresh boot and see if that works. If so, this means the exclude rules for the osimage have to be extended.

[Eth4nSheng]

This is an empty directory, and it contains no configuration files.

[Eth4nSheng]

Could I add a script to the image so that it writes the correct initial hostname to the /etc/hostname file on the system's first boot to resolve this issue?

[Eth4nSheng]

Hi Antoine

I think I've figured out the cause of this problem.

Because I used the diskfull script to push the system to my hard drive, it appeared to boot into the system on the first boot. However, after checking the information in /proc/cmdline, I found that it was actually in the ramdisk system environment.

The image below shows that the sda ​​disk has been correctly configured by the diskfull script, but the current system is still booting from the ramdisk using Luna, and the /etc/hostname file does not exist on the sda ​​system. Therefore, when I reboot, the system will initialize the hostname to localhost.

Do you think my analysis is correct, or what parameters I can add when using the diskfull script to write the correct node name to the disk?

tks

Ethan

[aphmschonewille]

Hi Ethan,

The diskfull should still contain the NM files. And then doing ipxe boot would also give you a node with a hostname and network configs. By looking at your images, yes it looks you are in diskfull install mode. However, the two images are contradicting a bit. The second image is after a the deployment? I see tmpfs there. What do you have configured for 'pre' of the group or node (if you needed to change this at all)?
In any case, the netboot toggle on group/node level would tell the node to continue local (disk) boot. Could you try that? If your local disk is setup correctly, then it should boot with the correct config.

-Antoine

[Eth4nSheng]

+--------------------------------------------------------------------------------------+
|                                Group => compute-group                                |
+---------------------+----------------------------------------------------------------+
| name                | compute-group                                                  |
| domain              | cluster                                                        |
| osimage             | compute-image                                                  |
| osimagetag          | default (default)                                              |
| kerneloptions       | _net.ifnames=0 _biosdevname=0 (compute-image)                  |
| interfaces          | interface = BOOTIF                                             |
|                     |   network = cluster                                            |
|                     | interface = BMC                                                |
|                     |   network = ipmi                                               |
| setupbmc            | True                                                           |
| bmcsetupname        | compute-bmcsetup                                               |
| unmanaged_bmc_users | None                                                           |
| netboot             | True                                                           |
| bootmenu            | False (default)                                                |
| roles               | None                                                           |
| scripts             | diskfull                                                       |
+---------------------+----------------------------------------------------------------+
| prescript           | cat << EOF >> /tmp/my-local-disk.sh                            |
|                     | export MY_LOCAL_DISK_NAME=/dev/sda                             |
|                     | export MY_LOCAL_DISK_SECTORS=                                  |
|                     | 4 More lines...                                                |
+---------------------+----------------------------------------------------------------+
| partscript          | mount -o mpol=interleave -t tmpfs tmpfs /sysroot               |
+---------------------+----------------------------------------------------------------+
| postscript          | echo 'tmpfs / tmpfs mpol=interleave 0 0' >> /sysroot/etc/fstab |
+---------------------+----------------------------------------------------------------+
| provision_interface | BOOTIF (default)                                               |
| provision_method    | torrent (cluster)                                              |
| provision_fallback  | http (cluster)                                                 |
| comment             | None                                                           |
+---------------------+----------------------------------------------------------------+

I did not make any "pre" configuration for the group. At the final stage of the diskfull execution, the disk boot settings would be automatically set as the first boot item. Therefore, when the system was restarted again, it did not go through the PXE boot process but directly called the grub boot from the disk. At this time, there was no information related to the node name in the disk.
Currently, my solution is: Add a script to the image. When the correct node name is obtained during the network boot after pushing, store it in /etc/hostname to solve this problem.this can solve the problem I am currently facing.