summaryrefslogtreecommitdiffstats
path: root/src/go/collectors/go.d.plugin/modules/nvidia_smi/nvidia_smi.go
diff options
context:
space:
mode:
authorDaniel Baumann <daniel.baumann@progress-linux.org>2024-05-05 11:19:16 +0000
committerDaniel Baumann <daniel.baumann@progress-linux.org>2024-07-24 09:53:24 +0000
commitb5f8ee61a7f7e9bd291dd26b0585d03eb686c941 (patch)
treed4d31289c39fc00da064a825df13a0b98ce95b10 /src/go/collectors/go.d.plugin/modules/nvidia_smi/nvidia_smi.go
parentAdding upstream version 1.44.3. (diff)
downloadnetdata-upstream.tar.xz
netdata-upstream.zip
Adding upstream version 1.46.3.upstream
Signed-off-by: Daniel Baumann <daniel.baumann@progress-linux.org>
Diffstat (limited to '')
-rw-r--r--src/go/collectors/go.d.plugin/modules/nvidia_smi/nvidia_smi.go117
1 files changed, 117 insertions, 0 deletions
diff --git a/src/go/collectors/go.d.plugin/modules/nvidia_smi/nvidia_smi.go b/src/go/collectors/go.d.plugin/modules/nvidia_smi/nvidia_smi.go
new file mode 100644
index 000000000..0002d4bf2
--- /dev/null
+++ b/src/go/collectors/go.d.plugin/modules/nvidia_smi/nvidia_smi.go
@@ -0,0 +1,117 @@
+// SPDX-License-Identifier: GPL-3.0-or-later
+
+package nvidia_smi
+
+import (
+ _ "embed"
+ "errors"
+ "time"
+
+ "github.com/netdata/netdata/go/go.d.plugin/agent/module"
+ "github.com/netdata/netdata/go/go.d.plugin/pkg/web"
+)
+
+//go:embed "config_schema.json"
+var configSchema string
+
+func init() {
+ module.Register("nvidia_smi", module.Creator{
+ JobConfigSchema: configSchema,
+ Defaults: module.Defaults{
+ Disabled: true,
+ UpdateEvery: 10,
+ },
+ Create: func() module.Module { return New() },
+ Config: func() any { return &Config{} },
+ })
+}
+
+func New() *NvidiaSMI {
+ return &NvidiaSMI{
+ Config: Config{
+ Timeout: web.Duration(time.Second * 10),
+ UseCSVFormat: false,
+ },
+ binName: "nvidia-smi",
+ charts: &module.Charts{},
+ gpus: make(map[string]bool),
+ migs: make(map[string]bool),
+ }
+
+}
+
+type Config struct {
+ UpdateEvery int `yaml:"update_every,omitempty" json:"update_every"`
+ Timeout web.Duration `yaml:"timeout,omitempty" json:"timeout"`
+ BinaryPath string `yaml:"binary_path" json:"binary_path"`
+ UseCSVFormat bool `yaml:"use_csv_format" json:"use_csv_format"`
+}
+
+type (
+ NvidiaSMI struct {
+ module.Base
+ Config `yaml:",inline" json:""`
+
+ charts *module.Charts
+
+ exec nvidiaSMI
+ binName string
+
+ gpuQueryProperties []string
+
+ gpus map[string]bool
+ migs map[string]bool
+ }
+ nvidiaSMI interface {
+ queryGPUInfoXML() ([]byte, error)
+ queryGPUInfoCSV(properties []string) ([]byte, error)
+ queryHelpQueryGPU() ([]byte, error)
+ }
+)
+
+func (nv *NvidiaSMI) Configuration() any {
+ return nv.Config
+}
+
+func (nv *NvidiaSMI) Init() error {
+ if nv.exec == nil {
+ smi, err := nv.initNvidiaSMIExec()
+ if err != nil {
+ nv.Error(err)
+ return err
+ }
+ nv.exec = smi
+ }
+
+ return nil
+}
+
+func (nv *NvidiaSMI) Check() error {
+ mx, err := nv.collect()
+ if err != nil {
+ nv.Error(err)
+ return err
+ }
+ if len(mx) == 0 {
+ return errors.New("no metrics collected")
+ }
+ return nil
+}
+
+func (nv *NvidiaSMI) Charts() *module.Charts {
+ return nv.charts
+}
+
+func (nv *NvidiaSMI) Collect() map[string]int64 {
+ mx, err := nv.collect()
+ if err != nil {
+ nv.Error(err)
+ }
+
+ if len(mx) == 0 {
+ return nil
+ }
+ return mx
+}
+
+func (nv *NvidiaSMI) Cleanup() {}