CentOS 7.x Rocks cluster 7.0 Build compute server
By default compute nodes are named compute-0-0.local, compute-0-1.local (assuming .local is used as domain name for private network). The compute nodes are setup via Pxeboot. They should be connected only to private network and then booted via network. Ideally compute nodes should be configured to boot via network as primary option and then via hard-disk as secondary.
To setup compute node use:
- Before doing network boot on compute node, on master server, run below command:
- On the ncurses based popup Choose Compute, it would display: 'insert-ethers is waiting for new compute nodes.'
- After this boot the compute node via network.
- When the frontend machine receives the DHCP request from the compute node, It will display, "Discovered a new appliance with MAC"
- insert-ethers has discovered a compute node. The "( )" next to compute-0-0 indicates the node has not yet requested a kickstart file. You will see this type of output for each compute node that is successfully identified by insert-ethers.
- 00:13:72:ba:c8:df Compute-0-0 ()
- Kickstart files are retrieved via HTTPS. If there was an error during the transmission, the error code will be visible instead of (*).
- 00:13:72:ba:c8:df Compute-0-0 (*)
- The compute node has successfully requested a kickstart file from the frontend. If there are no more compute nodes, you may now quit insert-ethers by pressing F8.
- Compute node will intall automatically.
- After successful installation of compute node, it will restart.
- We can list of hosts which are part of rocks cluster using
- rocks list host
If we want to give some other hostname then we can use:
insert-ethers --hostname <desired-name>
insert-ethers --hostname gpu
When we use option such as --hostname, we can setup only one node at a time. Once the node successfully requests pxeboot file, insert-ethers exits automatically.
For information on other options taken by insert-ethers refer http://central-7-0-x86-64.rocksclusters.org/roll-documentation/base/7.0/insert-ethers.html
ssh to compute node
To ssh compute node simply use:
ssh compute-0-0 ssh gpu
There is automatic authorized key setup from master to all compute. So no password is required. No prompt would be there to accept ssh fingerprint / key either. It would directly get connected.
Solving httpd not started issue while using insert-ethers
It is possible for httpd to fail to start with errors such as below in /var/log/httpd/error_log file:
[Thu Jul 07 06:35:56.564016 2022] [auth_digest:error] [pid 8490] (2)No such file or directory: AH01762: Failed to create shared memory segment on file /run/httpd/authdigest_shm.8490 [Thu Jul 07 06:35:56.564030 2022] [auth_digest:error] [pid 8490] (2)No such file or directory: AH01760: failed to initialize shm - all nonce-count checking and one-time noncesdisabled
To solve this use:
mkdfir /run/httpd chown apache:apache /run/httpd systemctl start httpd
Configure ntp client
Configure ntp client on all compute nodes using Configure basic ntp server and client
Optionally for new compute node installations also automate NTP client setup via CentOS 7.x Rocks cluster 7.0 Customize compute node during installation
Configure history retention
It is important to store command line history for more no. of lines along with timestamps on the cluster. To configure same on compute nodes use Storing date / time along with commands in history
Optionally for new compute node installations also automate history configuration via CentOS 7.x Rocks cluster 7.0 Customize compute node during installation