| 1 | // SPDX-License-Identifier: GPL-3.0-or-later |
| 2 | |
| 3 | package nvidia_smi |
| 4 | |
| 5 | import ( |
| 6 | "context" |
| 7 | _ "embed" |
| 8 | "errors" |
| 9 | "runtime" |
| 10 | "time" |
| 11 | |
| 12 | "github.com/netdata/netdata/go/plugins/pkg/confopt" |
| 13 | "github.com/netdata/netdata/go/plugins/plugin/framework/collectorapi" |
| 14 | ) |
| 15 | |
| 16 | //go:embed "config_schema.json" |
| 17 | var configSchema string |
| 18 | |
| 19 | func init() { |
| 20 | collectorapi.Register("nvidia_smi", collectorapi.Creator{ |
| 21 | JobConfigSchema: configSchema, |
| 22 | Defaults: collectorapi.Defaults{ |
| 23 | UpdateEvery: 10, |
| 24 | }, |
| 25 | Create: func() collectorapi.CollectorV1 { return New() }, |
| 26 | Config: func() any { return &Config{} }, |
| 27 | }) |
| 28 | } |
| 29 | |
| 30 | func New() *Collector { |
| 31 | return &Collector{ |
| 32 | Config: Config{ |
| 33 | Timeout: confopt.Duration(time.Second * 10), |
| 34 | // Disable loop mode on Windows due to go.d.plugin's non-graceful exit |
| 35 | // which can leave `nvidia_smi` processes running indefinitely. |
| 36 | LoopMode: !(runtime.GOOS == "windows"), |
| 37 | }, |
| 38 | binName: "nvidia-smi", |
| 39 | charts: &collectorapi.Charts{}, |
| 40 | gpus: make(map[string]bool), |
| 41 | migs: make(map[string]bool), |
| 42 | } |
| 43 | |
| 44 | } |
| 45 | |
| 46 | type Config struct { |
| 47 | UpdateEvery int `yaml:"update_every,omitempty" json:"update_every"` |
| 48 | AutoDetectionRetry int `yaml:"autodetection_retry,omitempty" json:"autodetection_retry"` |
| 49 | Timeout confopt.Duration `yaml:"timeout,omitempty" json:"timeout"` |
| 50 | BinaryPath string `yaml:"binary_path" json:"binary_path"` |
| 51 | LoopMode bool `yaml:"loop_mode,omitempty" json:"loop_mode"` |
| 52 | } |
| 53 | |
| 54 | type Collector struct { |
| 55 | collectorapi.Base |
| 56 | Config `yaml:",inline" json:""` |
| 57 | |
| 58 | charts *collectorapi.Charts |
| 59 | |
| 60 | exec nvidiaSmiBinary |
| 61 | binName string |
| 62 | |
| 63 | gpus map[string]bool |
| 64 | migs map[string]bool |
| 65 | } |
| 66 | |
| 67 | func (c *Collector) Configuration() any { |
| 68 | return c.Config |
| 69 | } |
| 70 | |
| 71 | func (c *Collector) Init(context.Context) error { |
| 72 | if c.exec == nil { |
| 73 | if runtime.GOOS == "windows" && c.LoopMode { |
| 74 | c.LoopMode = false |
| 75 | } |
| 76 | smi, err := c.initNvidiaSmiExec() |
| 77 | if err != nil { |
| 78 | return err |
| 79 | } |
| 80 | c.exec = smi |
| 81 | } |
| 82 | |
| 83 | return nil |
| 84 | } |
| 85 | |
| 86 | func (c *Collector) Check(context.Context) error { |
| 87 | mx, err := c.collect() |
| 88 | if err != nil { |
| 89 | return err |
| 90 | } |
| 91 | if len(mx) == 0 { |
| 92 | return errors.New("no metrics collected") |
| 93 | } |
| 94 | return nil |
| 95 | } |
| 96 | |
| 97 | func (c *Collector) Charts() *collectorapi.Charts { |
| 98 | return c.charts |
| 99 | } |
| 100 | |
| 101 | func (c *Collector) Collect(context.Context) map[string]int64 { |
| 102 | mx, err := c.collect() |
| 103 | if err != nil { |
| 104 | c.Error(err) |
| 105 | } |
| 106 | |
| 107 | if len(mx) == 0 { |
| 108 | return nil |
| 109 | } |
| 110 | return mx |
| 111 | } |
| 112 | |
| 113 | func (c *Collector) Cleanup(context.Context) { |
| 114 | if c.exec != nil { |
| 115 | if err := c.exec.stop(); err != nil { |
| 116 | c.Errorf("cleanup: %v", err) |
| 117 | } |
| 118 | c.exec = nil |
| 119 | } |
| 120 | } |