Add nvidia_smi input to monitor nvidia GPUs (#4026)

This commit is contained in:
Jack Zampolin 2018-04-17 13:40:55 -07:00 committed by Daniel Nelson
parent bcf1cf59c1
commit 3046f957d5
4 changed files with 232 additions and 0 deletions

View File

@ -65,6 +65,7 @@ import (
_ "github.com/influxdata/telegraf/plugins/inputs/nsq_consumer" _ "github.com/influxdata/telegraf/plugins/inputs/nsq_consumer"
_ "github.com/influxdata/telegraf/plugins/inputs/nstat" _ "github.com/influxdata/telegraf/plugins/inputs/nstat"
_ "github.com/influxdata/telegraf/plugins/inputs/ntpq" _ "github.com/influxdata/telegraf/plugins/inputs/ntpq"
_ "github.com/influxdata/telegraf/plugins/inputs/nvidia_smi"
_ "github.com/influxdata/telegraf/plugins/inputs/openldap" _ "github.com/influxdata/telegraf/plugins/inputs/openldap"
_ "github.com/influxdata/telegraf/plugins/inputs/opensmtpd" _ "github.com/influxdata/telegraf/plugins/inputs/opensmtpd"
_ "github.com/influxdata/telegraf/plugins/inputs/passenger" _ "github.com/influxdata/telegraf/plugins/inputs/passenger"

View File

@ -0,0 +1,47 @@
# `nvidia-smi` Input Plugin
This plugin uses a query on the [`nvidia-smi`](https://developer.nvidia.com/nvidia-system-management-interface) binary to pull GPU stats including memory and GPU usage, temp and other.
### Configuration
```toml
# Pulls statistics from nvidia GPUs attached to the host
[[inputs.nvidia_smi]]
## Optional: path to nvidia-smi binary, defaults to $PATH via exec.LookPath
# bin_path = /usr/bin/nvidia-smi
## Optional: timeout for GPU polling
# timeout = 5s
```
### Metrics
- measurement: `nvidia_smi`
- tags
- `name` (type of GPU e.g. `GeForce GTX 170 Ti`)
- `compute_mode` (The compute mode of the GPU e.g. `Default`)
- `index` (The port index where the GPU is connected to the motherboard e.g. `1`)
- `pstate` (Overclocking state for the GPU e.g. `P0`)
- `uuid` (A unique identifier for the GPU e.g. `GPU-f9ba66fc-a7f5-94c5-da19-019ef2f9c665`)
- fields
- `fan_speed` (integer, percentage)
- `memory_free` (integer, KB)
- `memory_used` (integer, KB)
- `memory_total` (integer, KB)
- `temperature_gpu` (integer, degrees C)
- `utilization_gpu` (integer, percentage)
- `utilization_memory` (integer, percentage)
### Sample Query
The below query could be used to alert on the average temperature of the your GPUs over the last minute
```
SELECT mean("temperature_gpu") FROM "nvidia_smi" WHERE time > now() - 5m GROUP BY time(1m), "index", "name", "host"
```
### Example Output
```
nvidia_smi,compute_mode=Default,host=8218cf,index=0,name=GeForce\ GTX\ 1070,pstate=P2,uuid=GPU-823bc202-6279-6f2c-d729-868a30f14d96 fan_speed=100i,memory_free=7563i,memory_total=8112i,memory_used=549i,temperature_gpu=53i,utilization_gpu=100i,utilization_memory=90i 1523991122000000000
nvidia_smi,compute_mode=Default,host=8218cf,index=1,name=GeForce\ GTX\ 1080,pstate=P2,uuid=GPU-f9ba66fc-a7f5-94c5-da19-019ef2f9c665 fan_speed=100i,memory_free=7557i,memory_total=8114i,memory_used=557i,temperature_gpu=50i,utilization_gpu=100i,utilization_memory=85i 1523991122000000000
nvidia_smi,compute_mode=Default,host=8218cf,index=2,name=GeForce\ GTX\ 1080,pstate=P2,uuid=GPU-d4cfc28d-0481-8d07-b81a-ddfc63d74adf fan_speed=100i,memory_free=7557i,memory_total=8114i,memory_used=557i,temperature_gpu=58i,utilization_gpu=100i,utilization_memory=86i 1523991122000000000
```

View File

@ -0,0 +1,149 @@
package nvidia_smi
import (
"bufio"
"fmt"
"os"
"os/exec"
"strconv"
"strings"
"time"
"github.com/influxdata/telegraf"
"github.com/influxdata/telegraf/internal"
"github.com/influxdata/telegraf/plugins/inputs"
)
var (
measurement = "nvidia_smi"
metrics = "fan.speed,memory.total,memory.used,memory.free,pstate,temperature.gpu,name,uuid,compute_mode,utilization.gpu,utilization.memory,index"
metricNames = [][]string{
[]string{"fan_speed", "field"},
[]string{"memory_total", "field"},
[]string{"memory_used", "field"},
[]string{"memory_free", "field"},
[]string{"pstate", "tag"},
[]string{"temperature_gpu", "field"},
[]string{"name", "tag"},
[]string{"uuid", "tag"},
[]string{"compute_mode", "tag"},
[]string{"utilization_gpu", "field"},
[]string{"utilization_memory", "field"},
[]string{"index", "tag"},
}
)
// NvidiaSMI holds the methods for this plugin
type NvidiaSMI struct {
BinPath string
Timeout time.Duration
metrics string
}
// Description returns the description of the NvidiaSMI plugin
func (smi *NvidiaSMI) Description() string {
return "Pulls statistics from nvidia GPUs attached to the host"
}
// SampleConfig returns the sample configuration for the NvidiaSMI plugin
func (smi *NvidiaSMI) SampleConfig() string {
return `
## Optional: path to nvidia-smi binary, defaults to $PATH via exec.LookPath
# bin_path = /usr/bin/nvidia-smi
## Optional: timeout for GPU polling
# timeout = 5s
`
}
// Gather implements the telegraf interface
func (smi *NvidiaSMI) Gather(acc telegraf.Accumulator) error {
if _, err := os.Stat(smi.BinPath); os.IsNotExist(err) {
return fmt.Errorf("nvidia-smi binary not at path %s, cannot gather GPU data", smi.BinPath)
}
data, err := smi.pollSMI()
if err != nil {
return err
}
err = gatherNvidiaSMI(data, acc)
if err != nil {
return err
}
return nil
}
func init() {
inputs.Add("nvidia_smi", func() telegraf.Input {
return &NvidiaSMI{
BinPath: "/usr/bin/nvidia-smi",
Timeout: 5 * time.Second,
metrics: metrics,
}
})
}
func (smi *NvidiaSMI) pollSMI() (string, error) {
// Construct and execute metrics query
opts := []string{"--format=noheader,nounits,csv", fmt.Sprintf("--query-gpu=%s", smi.metrics)}
ret, err := internal.CombinedOutputTimeout(exec.Command(smi.BinPath, opts...), smi.Timeout)
if err != nil {
return "", err
}
return string(ret), nil
}
func gatherNvidiaSMI(ret string, acc telegraf.Accumulator) error {
// First split the lines up and handle each one
scanner := bufio.NewScanner(strings.NewReader(ret))
for scanner.Scan() {
tags, fields, err := parseLine(scanner.Text())
if err != nil {
return err
}
acc.AddFields(measurement, fields, tags)
}
if err := scanner.Err(); err != nil {
return fmt.Errorf("Error scanning text %s", ret)
}
return nil
}
func parseLine(line string) (map[string]string, map[string]interface{}, error) {
tags := make(map[string]string, 0)
fields := make(map[string]interface{}, 0)
// Next split up the comma delimited metrics
met := strings.Split(line, ",")
// Make sure there are as many metrics in the line as there were queried.
if len(met) == len(metricNames) {
for i, m := range metricNames {
// First handle the tags
if m[1] == "tag" {
tags[m[0]] = strings.TrimSpace(met[i])
continue
}
// Then parse the integers out of the fields
out, err := strconv.ParseInt(strings.TrimSpace(met[i]), 10, 64)
if err != nil {
return tags, fields, err
}
fields[m[0]] = out
}
// Return the tags and fields
return tags, fields, nil
}
// If the line is empty return an emptyline error
return tags, fields, fmt.Errorf("Different number of metrics returned (%d) than expeced (%d)", len(met), len(metricNames))
}

View File

@ -0,0 +1,35 @@
package nvidia_smi
import (
"testing"
)
func TestParseLineStandard(t *testing.T) {
line := "85, 8114, 553, 7561, P2, 61, GeForce GTX 1070 Ti, GPU-d1911b8a-f5c8-5e66-057c-486561269de8, Default, 100, 93, 1\n"
tags, fields, err := parseLine(line)
if err != nil {
t.Fail()
}
if tags["name"] != "GeForce GTX 1070 Ti" {
t.Fail()
}
if temp, ok := fields["temperature_gpu"].(int); ok && temp == 61 {
t.Fail()
}
}
func TestParseLineEmptyLine(t *testing.T) {
line := "\n"
_, _, err := parseLine(line)
if err == nil {
t.Fail()
}
}
func TestParseLineBad(t *testing.T) {
line := "the quick brown fox jumped over the lazy dog"
_, _, err := parseLine(line)
if err == nil {
t.Fail()
}
}