infiniband

Documentation

Overview

Package infiniband provides utilities to query infiniband status.

Index

• Variables
• func CountInfinibandClass() int
• func CountInfinibandClassBySubDir(dir string) int
• func CountInfinibandPCIBuses(ctx context.Context) (int, error)
• func SupportsInfinibandPortRate(gpuProductName string) int
• func SupportsInfinibandProduct(gpuProductName string) bool
• func ValidateIbstatOutput(s string) error
• type IBStatCard
• type IBStatCards
  • func ParseIBStat(input string) (IBStatCards, error)
  • func (cards IBStatCards) CheckPortsAndRate(atLeastPorts int, atLeastRate int) error
  • func (cards IBStatCards) Match(expectedPhysicalState string, expectedState string, atLeastRate int) []string
• type IBStatPort
• type IbstatOutput
  • func GetIbstatOutput(ctx context.Context, ibstatCommands []string) (*IbstatOutput, error)

Variables

var (
	ErrIbstatOutputBrokenStateDown        = errors.New("ibstat output unexpected; found State: Down (check the physical switch)")
	ErrIbstatOutputBrokenPhysicalDisabled = errors.New("ibstat output unexpected; found Physical state: Disabled (check the physical switch)")
)

Functions

func CountInfinibandClass

func CountInfinibandClass() int

Counts the directories in "/sys/class/infiniband". Returns 0 if the directory does not exist.

func CountInfinibandClassBySubDir

func CountInfinibandClassBySubDir(dir string) int

Count the sub-directories under the specified directory.

func CountInfinibandPCIBuses

func CountInfinibandPCIBuses(ctx context.Context) (int, error)

lspci | grep -i infiniband 1a:00.0 Infiniband controller: Mellanox Technologies MT2910 Family [ConnectX-7] 3c:00.0 Infiniband controller: Mellanox Technologies MT2910 Family [ConnectX-7] 1a:00.0 Ethernet controller: Mellanox Technologies MT2910 Family [ConnectX-7] 1b:00.0 Ethernet controller: Mellanox Technologies MT2892 Family [ConnectX-6 Dx]

func SupportsInfinibandPortRate

func SupportsInfinibandPortRate(gpuProductName string) int

Returns the default non-zero per-port rate in GB/sec if the product supports infiniband.

func SupportsInfinibandProduct

func SupportsInfinibandProduct(gpuProductName string) bool

Returns true if the product supports infiniband. e.g., "NVIDIA A100" "NVIDIA H100"

func ValidateIbstatOutput

func ValidateIbstatOutput(s string) error

Types

type IBStatCard

type IBStatCard struct {
	Name            string     `json:"CA name"`
	Type            string     `json:"CA type"`
	NumPorts        string     `json:"Number of ports"`
	FirmwareVersion string     `json:"Firmware version"`
	HardwareVersion string     `json:"Hardware version"`
	NodeGUID        string     `json:"Node GUID"`
	SystemImageGUID string     `json:"System image GUID"`
	Port1           IBStatPort `json:"Port 1"`
}

type IBStatCards

type IBStatCards []IBStatCard

func ParseIBStat

func ParseIBStat(input string) (IBStatCards, error)

ParseIBStat parses ibstat output and returns YAML representation

func (IBStatCards) CheckPortsAndRate

func (cards IBStatCards) CheckPortsAndRate(atLeastPorts int, atLeastRate int) error

CheckPortsAndRate checks if the number of active IB ports matches expectations

func (IBStatCards) Match

func (cards IBStatCards) Match(expectedPhysicalState string, expectedState string, atLeastRate int) []string

Match returns the IB port names whose physical state, state, and "Port 1"."Rate" match the expected values. The specified rate is the threshold for "Port 1"."Rate", where it evaluates with ">=" operator (e.g., count all the cards whose rate is >= 400).

If the `expectedPhysicalState` is empty, it matches all states. If the `expectedState` is empty, it matches all states.

type IBStatPort

type IBStatPort struct {
	State         string `json:"State"`
	PhysicalState string `json:"Physical state"`
	Rate          int    `json:"Rate"`
	BaseLid       int    `json:"Base lid"`
	LinkLayer     string `json:"Link layer"`
}

type IbstatOutput

type IbstatOutput struct {
	Parsed IBStatCards `json:"parsed,omitempty"`
	Raw    string      `json:"raw"`
	Errors []string    `json:"errors,omitempty"`
}

func GetIbstatOutput

func GetIbstatOutput(ctx context.Context, ibstatCommands []string) (*IbstatOutput, error)