This page describes how to add clients to a bare-metal cluster.
Clients are used to run applications that need to access the WEKA filesystems. They do not contribute CPUs or drives to the cluster and only connect to the cluster to use its filesystems.
A default WEKA installation uses the cgroups functionality to limit or isolate resources for WEKA sole usage. For example, using specific CPUs.
WEKA backends and clients that serve protocols can run on a supported OS with cgroups V1 (legacy).
Customers using a supported OS with cgroups V2 (hierarchy) or want to modify the cgroups usage can either set it during the agent installation or edit the service configuration file (see Modify the cgroups usage).
If the OS is configured with cgroups hybrid mode, which operates with cgroups V1 and V2, WEKA uses V1 by default.
cgroups, short for control groups, is a feature in the Linux kernel that allows for fine-grained resource allocation and management of system resources such as CPU, memory, and I/O. It enables administrators to create hierarchies of processes and allocate system resources based on defined rules.
With cgroups, processes can be organized into groups, and each group can be allocated specific amounts of resources such as CPU time, memory, or network bandwidth. This allows for more efficient use of system resources and can help prevent individual processes from monopolizing system resources.
To use the WEKA filesystems from a client, just call the mount command. The mount command automatically installs the software version, and there is no need to join the client to the cluster.
To mount a filesystem in this method, first, install the WEKA agent from one of the backend instances and then mount the filesystem.
# Agent Installation (one time)
curl http://Backend-1:14000/dist/v1/install | sh
# Creating a mount point (one time)
mkdir -p /mnt/weka
# Mounting a filesystem
mount -t wekafs -o net=eth0 backend-1/my_fs /mnt/weka
For the first mount, this installs the WEKA software and automatically configures the client. For more information on mount and configuration options, refer to Mount a filesystem using the stateless clients feature.
Each client must have a unique IP and FQDN.
Customers using a supported OS with cgroups V2 (hierarchy ) or want to modify the cgroups usage can set the cgroups usage during the agent installation or edit the service configuration file.
The cgroups setting includes the following modes:
auto: WEKA tries using cgroups V1 (default). If it fails, the cgroups is set to
force: WEKA always uses cgroups V1. If the OS does not support it, WEKA fails.
none: WEKA never uses cgroups. Even if it runs on OS with cgroups V1. For example, use this option to prevent a conflict with third-party components, such as the slurm workload manager.
In the agent installation command line, specify the required cgroups mode.
curl http://Backend-1:14000/dist/v1/install | WEKA_CGROUPS_MODE=none sh
- 1.Open the service configuration file
/etc/wekaio/service.confand add one of the following:
- 2.Restart the WEKA agent service.
- 3.Verify the cgroups setting by running the
weka local statuscommand.
[root@weka-cluster] #weka local status
Weka v4.2.0 (CLI build 4.2.0)
Cgroups: mode=auto, enabled=true
Containers: 1/1 running (1 weka)
Nodes: 2/2 running (2 READY)
Adding instances that do not contribute resources to the cluster but are used for mounting filesystems is possible. It is recommended to use the previously described method for adding client instances for mounting purposes. However, in some cases, adding them to the cluster permanently is helpful. For example, use these instances as NFS/SMB servers, which are always expected to be up.
Install the WEKA software.
All clients in a WEKA system cluster must use the same software version as the backends or a maximum of one version back. The backend containers must run the same WEKA software version except during upgrades (as managed by the upgrade process).
weka cluster container add
Once the client is in the stem mode (this is the mode defined immediately after running the
install.shcommand), use the following command line on the client to add it to the cluster:
weka -H <backend-hostname> cluster container add <client-hostname>
Parameters in the command line
On completion of this stage, the container-id of the newly added container will be received. Make a note of it for the next steps.
weka cluster container cores
To configure the new container as a client, run the following command:
weka cluster container cores <container-id> <cores> --frontend-dedicated-cores=<frontend-dedicated-cores>
Parameters in the command line
weka cluster container net add
If the new client is to communicate with the WEKA cluster over the kernel UDP stack, running this command is unnecessary.
If a high-performance client is required and the appropriate network NIC is available, use the following command to configure the networking interface used by the client to communicate with the WEKA cluster:
weka cluster container net add <container-id> <device> --ips=<ips> --netmask=<netmask> --gateway=<gateway>
Parameters in command line
When configuring an InfiniBand client, do not pass the
InfiniBand/Ethernet clients can only join a cluster with the same network technology connectivity. It is possible to mix InfiniBand and Ethernet clients in the same cluster as long as the cluster backends are connected to both network technologies.
weka cluster container apply
After successfully configuring the container and its network device, run the following command to finalize the configuration by activating the container:
weka cluster container apply <container-id> [--force]
Parameters in command line