Server Compatibility Test
The ska-pst chart includes a self-test functionality for validating server configuration compatibility. This is enabled through the chart values field selftest.enabled to true.
| Test | Description |
|---|---|
| NVME detection | Confirms the existence of expected NVME mount. |
| NVME write test | Tests the write speed of detected NVME mount. |
| NVIDIA GPU and driver detection | Confirms the existence of NVIDIA GPU device(s) and installed driver. |
| Mellanox NIC detection | Confirms the existence of Mellanox NIC devices. |
| Shared memory mount | Confirms the existence of shared memory through the path /dev/shm |
| shmmax configured | Confirms shmmax configuration present in /proc/sys/kernel/shmmax |
| shmmin configured | Confirms shmmin configuration present in /proc/sys/kernel/shmmin |
| Memlock | Confirms if the allowed lockable memory is set to at least the minimum advised requirement |
| CUDA H2D and D2H | Confirms if CUDA host to device data transfer and its reverse meets the advised minimum requirement |
Executing the self-test through Kubernetes in the ska-pst chart
Executing the self-test as a k8s Job ensures that the advised server configurations are inherited by the ska-pst-core containers through Kubernetes plugins and configurations.
Post installation of the ska-pst chart, the ska-pst-selftest pod can be interrogated to determine the selftest results.
Executing the self-test through docker directly on a baremetal server
The ska-pst self-test script can also be executed through a container engine such as docker on a baremetal server. A motivation to execute the self-test script in this layer is to confirm if the containerisation engine (an abstraction layer beneath Kubernetes) configuration is configured as expected.
