Add the ipmi plugin

This commit is contained in:
张光权 2016-03-17 23:45:29 +08:00 committed by Michele Fadda
parent 798e8d0f9f
commit f95ec898a9
6 changed files with 482 additions and 0 deletions

View File

@ -16,6 +16,7 @@ import (
_ "github.com/influxdata/telegraf/plugins/inputs/haproxy"
_ "github.com/influxdata/telegraf/plugins/inputs/httpjson"
_ "github.com/influxdata/telegraf/plugins/inputs/influxdb"
_ "github.com/influxdata/telegraf/plugins/inputs/ipmi"
_ "github.com/influxdata/telegraf/plugins/inputs/jolokia"
_ "github.com/influxdata/telegraf/plugins/inputs/kafka_consumer"
_ "github.com/influxdata/telegraf/plugins/inputs/leofs"

View File

@ -0,0 +1,50 @@
# Telegraf ipmi plugin
Get bare metal metrics using the command line utility `ipmitool`
see ipmitool(https://sourceforge.net/projects/ipmitool/files/ipmitool/)
The plugin will use the following command to collect remote host sensor stats:
ipmitool -I lan -H 192.168.1.1 -U USERID -P PASSW0RD sdr
## Measurements
- ipmi_sensor:
* Tags: `server`,`host`
* Fields:
- status
- value
## Configuration
```toml
[[inputs.ipmi]]
## specify servers via a url matching:
## [username[:password]@][protocol[(address)]]
## e.g.
## root:passwd@lan(127.0.0.1)
##
servers = ["USERID:PASSW0RD@lan(10.20.2.203)"]
```
## Output
> ipmi_sensor,host=10.20.2.203,inst=Ambient\ Temp status=1i,value=20 1458488465012559455
> ipmi_sensor,host=10.20.2.203,inst=Altitude status=1i,value=80 1458488465012688613
> ipmi_sensor,host=10.20.2.203,inst=Avg\ Power status=1i,value=220 1458488465012776511
> ipmi_sensor,host=10.20.2.203,inst=Planar\ 3.3V status=1i,value=3.28 1458488465012861875
> ipmi_sensor,host=10.20.2.203,inst=Planar\ 5V status=1i,value=4.9 1458488465012944188
> ipmi_sensor,host=10.20.2.203,inst=Planar\ 12V status=1i,value=12.04 1458488465013008485
> ipmi_sensor,host=10.20.2.203,inst=Planar\ VBAT status=1i,value=3.04 1458488465013072508
> ipmi_sensor,host=10.20.2.203,inst=Fan\ 1A\ Tach status=1i,value=2610 1458488465013137932
> ipmi_sensor,host=10.20.2.203,inst=Fan\ 1B\ Tach status=1i,value=1775 1458488465013279896
> ipmi_sensor,host=10.20.2.203,inst=Fan\ 2A\ Tach status=1i,value=1972 1458488465013358177
> ipmi_sensor,host=10.20.2.203,inst=Fan\ 2B\ Tach status=1i,value=1275 1458488465013434023
> ipmi_sensor,host=10.20.2.203,inst=Fan\ 3A\ Tach status=1i,value=2929 1458488465013514567
> ipmi_sensor,host=10.20.2.203,inst=Fan\ 3B\ Tach status=1i,value=2125 1458488465013582616
> ipmi_sensor,host=10.20.2.203,inst=Fan\ 1 status=1i,value=0 1458488465013643746
> ipmi_sensor,host=10.20.2.203,inst=Fan\ 2 status=1i,value=0 1458488465013714887
> ipmi_sensor,host=10.20.2.203,inst=Fan\ 3 status=1i,value=0 1458488465013861854

View File

@ -0,0 +1,39 @@
// command
package ipmi
import (
"bytes"
"fmt"
"os/exec"
"strings"
)
type CommandRunner struct{}
func (t CommandRunner) cmd(conn *Connection, args ...string) *exec.Cmd {
path := conn.Path
opts := append(conn.options(), args...)
if path == "" {
path = "ipmitool"
}
return exec.Command(path, opts...)
}
func (t CommandRunner) Run(conn *Connection, args ...string) (string, error) {
cmd := t.cmd(conn, args...)
var stdout bytes.Buffer
var stderr bytes.Buffer
cmd.Stdout = &stdout
cmd.Stderr = &stderr
err := cmd.Run()
if err != nil {
return "", fmt.Errorf("run %s %s: %s (%s)",
cmd.Path, strings.Join(cmd.Args, " "), stderr.String(), err)
}
return stdout.String(), err
}

View File

@ -0,0 +1,90 @@
// connection
package ipmi
import (
"fmt"
"net"
"strconv"
"strings"
)
// Connection properties for a Client
type Connection struct {
Hostname string
Username string
Password string
Path string
Port int
Interface string
}
func NewConnection(server string) *Connection {
conn := &Connection{}
inx1 := strings.Index(server, "@")
inx2 := strings.Index(server, "(")
inx3 := strings.Index(server, ")")
connstr := server
if inx1 > 0 {
security := server[0:inx1]
connstr = server[inx1+1 : len(server)]
up := strings.Split(security, ":")
conn.Username = up[0]
conn.Password = up[1]
}
if inx2 > 0 {
inx2 = strings.Index(connstr, "(")
inx3 = strings.Index(connstr, ")")
conn.Interface = connstr[0:inx2]
conn.Hostname = connstr[inx2+1 : inx3]
}
return conn
}
func (t *Connection) options() []string {
intf := t.Interface
if intf == "" {
intf = "lan"
}
options := []string{
"-H", t.Hostname,
"-U", t.Username,
"-P", t.Password,
"-I", intf,
}
if t.Port != 0 {
options = append(options, "-p", strconv.Itoa(t.Port))
}
return options
}
// RemoteIP returns the remote (bmc) IP address of the Connection
func (c *Connection) RemoteIP() string {
if net.ParseIP(c.Hostname) == nil {
addrs, err := net.LookupHost(c.Hostname)
if err != nil && len(addrs) > 0 {
return addrs[0]
}
}
return c.Hostname
}
// LocalIP returns the local (client) IP address of the Connection
func (c *Connection) LocalIP() string {
conn, err := net.Dial("udp", fmt.Sprintf("%s:%d", c.Hostname, c.Port))
if err != nil {
// don't bother returning an error, since this value will never
// make it to the bmc if we can't connect to it.
return c.Hostname
}
_ = conn.Close()
host, _, _ := net.SplitHostPort(conn.LocalAddr().String())
return host
}

113
plugins/inputs/ipmi/ipmi.go Normal file
View File

@ -0,0 +1,113 @@
// ipmi
package ipmi
import (
"strconv"
"strings"
"time"
"github.com/influxdata/telegraf"
"github.com/influxdata/telegraf/plugins/inputs"
)
type Ipmi struct {
Servers []string
runner Runner
}
var sampleConfig = `
## specify servers via a url matching:
## [username[:password]@][protocol[(address)]]
## e.g.
## root:passwd@lan(127.0.0.1)
##
servers = ["USERID:PASSW0RD@lan(192.168.1.1)"]
`
func NewIpmi() *Ipmi {
return &Ipmi{
runner: CommandRunner{},
}
}
func (m *Ipmi) SampleConfig() string {
return sampleConfig
}
func (m *Ipmi) Description() string {
return "Read metrics from one or many bare metal servers"
}
func (m *Ipmi) Gather(acc telegraf.Accumulator) error {
if m.runner == nil {
m.runner = CommandRunner{}
}
for _, serv := range m.Servers {
err := m.gatherServer(serv, acc)
if err != nil {
return err
}
}
return nil
}
func (m *Ipmi) gatherServer(serv string, acc telegraf.Accumulator) error {
conn := NewConnection(serv)
res, err := m.runner.Run(conn, "sdr")
if err != nil {
return err
}
lines := strings.Split(res, "\n")
for i := 0; i < len(lines); i++ {
vals := strings.Split(lines[i], "|")
if len(vals) == 3 {
tags := map[string]string{"server": conn.Hostname, "name": trim(vals[0])}
fields := make(map[string]interface{})
if strings.EqualFold("ok", trim(vals[2])) {
fields["status"] = 1
} else {
fields["status"] = 0
}
val1 := trim(vals[1])
if strings.Index(val1, " ") > 0 {
val := strings.Split(val1, " ")[0]
fields["value"] = Atofloat(val)
} else {
fields["value"] = 0.0
}
acc.AddFields("ipmi_sensor", fields, tags, time.Now())
}
}
return nil
}
type Runner interface {
Run(conn *Connection, args ...string) (string, error)
}
func Atofloat(val string) float64 {
f, err := strconv.ParseFloat(val, 64)
if err != nil {
return float64(0)
} else {
return float64(f)
}
}
func trim(s string) string {
return strings.TrimSpace(s)
}
func init() {
inputs.Add("ipmi", func() telegraf.Input {
return &Ipmi{}
})
}

View File

@ -0,0 +1,189 @@
// ipmi_test
package ipmi
import (
"testing"
"github.com/influxdata/telegraf/testutil"
"github.com/stretchr/testify/assert"
"github.com/stretchr/testify/require"
)
const serv = "USERID:PASSW0RD@lan(192.168.1.1)"
const cmdReturn = `
Ambient Temp | 20 degrees C | ok
Altitude | 80 feet | ok
Avg Power | 210 Watts | ok
Planar 3.3V | 3.29 Volts | ok
Planar 5V | 4.90 Volts | ok
Planar 12V | 12.04 Volts | ok
Planar VBAT | 3.05 Volts | ok
Fan 1A Tach | 2610 RPM | ok
Fan 1B Tach | 1775 RPM | ok
Fan 2A Tach | 2001 RPM | ok
Fan 2B Tach | 1275 RPM | ok
Fan 3A Tach | 2929 RPM | ok
Fan 3B Tach | 2125 RPM | ok
Fan 1 | 0x00 | ok
Fan 2 | 0x00 | ok
Fan 3 | 0x00 | ok
Front Panel | 0x00 | ok
Video USB | 0x00 | ok
DASD Backplane 1 | 0x00 | ok
SAS Riser | 0x00 | ok
PCI Riser 1 | 0x00 | ok
PCI Riser 2 | 0x00 | ok
CPU 1 | 0x00 | ok
CPU 2 | 0x00 | ok
All CPUs | 0x00 | ok
One of The CPUs | 0x00 | ok
IOH Temp Status | 0x00 | ok
CPU 1 OverTemp | 0x00 | ok
CPU 2 OverTemp | 0x00 | ok
CPU Fault Reboot | 0x00 | ok
Aux Log | 0x00 | ok
NMI State | 0x00 | ok
ABR Status | 0x00 | ok
Firmware Error | 0x00 | ok
PCIs | 0x00 | ok
CPUs | 0x00 | ok
DIMMs | 0x00 | ok
Sys Board Fault | 0x00 | ok
Power Supply 1 | 0x00 | ok
Power Supply 2 | 0x00 | ok
PS 1 Fan Fault | 0x00 | ok
PS 2 Fan Fault | 0x00 | ok
VT Fault | 0x00 | ok
Pwr Rail A Fault | 0x00 | ok
Pwr Rail B Fault | 0x00 | ok
Pwr Rail C Fault | 0x00 | ok
Pwr Rail D Fault | 0x00 | ok
Pwr Rail E Fault | 0x00 | ok
PS 1 Therm Fault | 0x00 | ok
PS 2 Therm Fault | 0x00 | ok
PS1 12V OV Fault | 0x00 | ok
PS2 12V OV Fault | 0x00 | ok
PS1 12V UV Fault | 0x00 | ok
PS2 12V UV Fault | 0x00 | ok
PS1 12V OC Fault | 0x00 | ok
PS2 12V OC Fault | 0x00 | ok
PS 1 VCO Fault | 0x00 | ok
PS 2 VCO Fault | 0x00 | ok
Power Unit | 0x00 | ok
Cooling Zone 1 | 0x00 | ok
Cooling Zone 2 | 0x00 | ok
Cooling Zone 3 | 0x00 | ok
Drive 0 | 0x00 | ok
Drive 1 | 0x00 | ok
Drive 2 | 0x00 | ok
Drive 3 | 0x00 | ok
Drive 4 | 0x00 | ok
Drive 5 | 0x00 | ok
Drive 6 | 0x00 | ok
Drive 7 | 0x00 | ok
Drive 8 | 0x00 | ok
Drive 9 | 0x00 | ok
Drive 10 | 0x00 | ok
Drive 11 | 0x00 | ok
Drive 12 | 0x00 | ok
Drive 13 | 0x00 | ok
Drive 14 | 0x00 | ok
Drive 15 | 0x00 | ok
All DIMMS | 0x00 | ok
One of the DIMMs | 0x00 | ok
DIMM 1 | 0x00 | ok
DIMM 2 | 0x00 | ok
DIMM 3 | 0x00 | ok
DIMM 4 | 0x00 | ok
DIMM 5 | 0x00 | ok
DIMM 6 | 0x00 | ok
DIMM 7 | 0x00 | ok
DIMM 8 | 0x00 | ok
DIMM 9 | 0x00 | ok
DIMM 10 | 0x00 | ok
DIMM 11 | 0x00 | ok
DIMM 12 | 0x00 | ok
DIMM 13 | 0x00 | ok
DIMM 14 | 0x00 | ok
DIMM 15 | 0x00 | ok
DIMM 16 | 0x00 | ok
DIMM 17 | 0x00 | ok
DIMM 18 | 0x00 | ok
DIMM 1 Temp | 0x00 | ok
DIMM 2 Temp | 0x00 | ok
DIMM 3 Temp | 0x00 | ok
DIMM 4 Temp | 0x00 | ok
DIMM 5 Temp | 0x00 | ok
DIMM 6 Temp | 0x00 | ok
DIMM 7 Temp | 0x00 | ok
DIMM 8 Temp | 0x00 | ok
DIMM 9 Temp | 0x00 | ok
DIMM 10 Temp | 0x00 | ok
DIMM 11 Temp | 0x00 | ok
DIMM 12 Temp | 0x00 | ok
DIMM 13 Temp | 0x00 | ok
DIMM 14 Temp | 0x00 | ok
DIMM 15 Temp | 0x00 | ok
DIMM 16 Temp | 0x00 | ok
DIMM 17 Temp | 0x00 | ok
DIMM 18 Temp | 0x00 | ok
PCI 1 | 0x00 | ok
PCI 2 | 0x00 | ok
PCI 3 | 0x00 | ok
PCI 4 | 0x00 | ok
All PCI Error | 0x00 | ok
One of PCI Error | 0x00 | ok
IPMI Watchdog | 0x00 | ok
Host Power | 0x00 | ok
DASD Backplane 2 | 0x00 | ok
DASD Backplane 3 | Not Readable | ns
DASD Backplane 4 | Not Readable | ns
Backup Memory | 0x00 | ok
Progress | 0x00 | ok
Planar Fault | 0x00 | ok
SEL Fullness | 0x00 | ok
PCI 5 | 0x00 | ok
OS RealTime Mod | 0x00 | ok
`
type runnerMock struct {
out string
err error
}
func newRunnerMock(out string, err error) Runner {
return &runnerMock{
out: out,
err: err,
}
}
func (r runnerMock) Run(conn *Connection, args ...string) (out string, err error) {
if r.err != nil {
return out, r.err
}
return r.out, nil
}
func TestIpmi(t *testing.T) {
i := &Ipmi{
Servers: []string{"USERID:PASSW0RD@lan(192.168.1.1)"},
runner: newRunnerMock(cmdReturn, nil),
}
var acc testutil.Accumulator
err := i.Gather(&acc)
require.NoError(t, err)
assert.Equal(t, acc.NFields(), 266, "non-numeric measurements should be ignored")
}
func TestIpmiConnection(t *testing.T) {
conn := NewConnection(serv)
assert.Equal(t, "USERID", conn.Username)
assert.Equal(t, "lan", conn.Interface)
}