CL-Softwares
/
telegraf
8
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

temporarily remove RAS plugin

This commit is contained in:
Steven Soroka 2020-10-16 14:44:15 -04:00
parent a422c8f93b
commit 933925e8e3
6 changed files with 0 additions and 611 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

1
CHANGELOG.md
View File

@ -73,7 +73,6 @@
- [nsd](/plugins/inputs/nsd/README.md) add nsd input plugin - Contributed by @gearnode
- [opcua](/plugins/inputs/opcua/README.md) Add OPC UA input plugin - Contributed by InfluxData
- [proxmox](/plugins/inputs/proxmox/README.md) Proxmox plugin - Contributed by @effitient
- [ras](/plugins/inputs/ras/README.md) New input plugin for RAS (Reliability, Availability and Serviceability) - Contributed by @p-zak
- [win_eventlog](/plugins/inputs/win_eventlog/README.md) Windows eventlog input plugin - Contributed by @simnv
#### New Output Plugins

1
plugins/inputs/all/all.go
View File

@ -141,7 +141,6 @@ import (
_ "github.com/influxdata/telegraf/plugins/inputs/puppetagent"
_ "github.com/influxdata/telegraf/plugins/inputs/rabbitmq"
_ "github.com/influxdata/telegraf/plugins/inputs/raindrops"
_ "github.com/influxdata/telegraf/plugins/inputs/ras"
_ "github.com/influxdata/telegraf/plugins/inputs/redfish"
_ "github.com/influxdata/telegraf/plugins/inputs/redis"
_ "github.com/influxdata/telegraf/plugins/inputs/rethinkdb"

58
plugins/inputs/ras/README.md
View File

@ -1,58 +0,0 @@
# RAS Daemon Input Plugin
The `RAS` plugin gathers and counts errors provided by [RASDaemon](https://github.com/mchehab/rasdaemon).
### Configuration
```toml
[[inputs.ras]]
## Optional path to RASDaemon sqlite3 database.
## Default: /var/lib/rasdaemon/ras-mc_event.db
# db_path = ""
```
In addition `RASDaemon` runs, by default, with `--enable-sqlite3` flag. In case of problems with SQLite3 database please verify this is still a default option.
### Metrics
- ras
- tags:
- socket_id
- fields:
- memory_read_corrected_errors
- memory_read_uncorrectable_errors
- memory_write_corrected_errors
- memory_write_uncorrectable_errors
- cache_l0_l1_errors
- tlb_instruction_errors
- cache_l2_errors
- upi_errors
- processor_base_errors
- processor_bus_errors
- internal_timer_errors
- smm_handler_code_access_violation_errors
- internal_parity_errors
- frc_errors
- external_mce_errors
- microcode_rom_parity_errors
- unclassified_mce_errors
Please note that `processor_base_errors` is aggregate counter measuring the following MCE events:
- internal_timer_errors
- smm_handler_code_access_violation_errors
- internal_parity_errors
- frc_errors
- external_mce_errors
- microcode_rom_parity_errors
- unclassified_mce_errors
### Permissions
This plugin requires access to SQLite3 database from `RASDaemon`. Please make sure that user has required permissions to this database.
### Example Output
```
ras,host=ubuntu,socket_id=0 external_mce_base_errors=1i,frc_errors=1i,instruction_tlb_errors=5i,internal_parity_errors=1i,internal_timer_errors=1i,l0_and_l1_cache_errors=7i,memory_read_corrected_errors=25i,memory_read_uncorrectable_errors=0i,memory_write_corrected_errors=5i,memory_write_uncorrectable_errors=0i,microcode_rom_parity_errors=1i,processor_base_errors=7i,processor_bus_errors=1i,smm_handler_code_access_violation_errors=1i,unclassified_mce_base_errors=1i 1598867393000000000
ras,host=ubuntu level_2_cache_errors=0i,upi_errors=0i 1598867393000000000
```

294
plugins/inputs/ras/ras.go
View File

@ -1,294 +0,0 @@
// +build !windows
package ras
import (
"database/sql"
"strconv"
"strings"
"time"
_ "github.com/mattn/go-sqlite3"
"github.com/influxdata/telegraf"
"github.com/influxdata/telegraf/plugins/inputs"
)
type Ras struct {
DbPath string
latestTimestamp time.Time
cpuSocketCounters map[int]metricCounters
serverCounters metricCounters
}
type machineCheckError struct {
Id int
Timestamp string
SocketId int
ErrorMsg string
MciStatusMsg string
}
type metricCounters map[string]int64
const (
mceQuery = `
SELECT
id, timestamp, error_msg, mcistatus_msg, socketid
FROM mce_record
WHERE timestamp > ?
`
defaultDbPath = "/var/lib/rasdaemon/ras-mc_event.db"
dateLayout = "2006-01-02 15:04:05 -0700"
memoryReadCorrected = "memory_read_corrected_errors"
memoryReadUncorrected = "memory_read_uncorrectable_errors"
memoryWriteCorrected = "memory_write_corrected_errors"
memoryWriteUncorrected = "memory_write_uncorrectable_errors"
instructionCache = "cache_l0_l1_errors"
instructionTLB = "tlb_instruction_errors"
levelTwoCache = "cache_l2_errors"
upi = "upi_errors"
processorBase = "processor_base_errors"
processorBus = "processor_bus_errors"
internalTimer = "internal_timer_errors"
smmHandlerCode = "smm_handler_code_access_violation_errors"
internalParity = "internal_parity_errors"
frc = "frc_errors"
externalMCEBase = "external_mce_errors"
microcodeROMParity = "microcode_rom_parity_errors"
unclassifiedMCEBase = "unclassified_mce_errors"
)
func (r *Ras) SampleConfig() string {
return `
## Optional path to RASDaemon sqlite3 database.
## Default: /var/lib/rasdaemon/ras-mc_event.db
# db_path = ""
`
}
func (r *Ras) Description() string {
return "RAS plugin exposes counter metrics for Machine Check Errors provided by RASDaemon (sqlite3 output is required)."
}
func (r *Ras) Gather(acc telegraf.Accumulator) error {
db, err := connectToDB(r.DbPath)
if err != nil {
return err
}
defer db.Close()
rows, err := db.Query(mceQuery, r.latestTimestamp)
if err != nil {
return err
}
defer rows.Close()
for rows.Next() {
mcError, err := fetchMachineCheckError(rows)
if err != nil {
return err
}
tsErr := r.updateLatestTimestamp(mcError.Timestamp)
if tsErr != nil {
return err
}
r.updateCounters(mcError)
}
addCpuSocketMetrics(acc, r.cpuSocketCounters)
addServerMetrics(acc, r.serverCounters)
return nil
}
func (r *Ras) updateLatestTimestamp(timestamp string) error {
ts, err := parseDate(timestamp)
if err != nil {
return err
}
if ts.After(r.latestTimestamp) {
r.latestTimestamp = ts
}
return nil
}
func (r *Ras) updateCounters(mcError *machineCheckError) {
if strings.Contains(mcError.ErrorMsg, "No Error") {
return
}
r.initializeCpuMetricDataIfRequired(mcError.SocketId)
r.updateSocketCounters(mcError)
r.updateServerCounters(mcError)
}
func newMetricCounters() *metricCounters {
return &metricCounters{
memoryReadCorrected: 0,
memoryReadUncorrected: 0,
memoryWriteCorrected: 0,
memoryWriteUncorrected: 0,
instructionCache: 0,
instructionTLB: 0,
processorBase: 0,
processorBus: 0,
internalTimer: 0,
smmHandlerCode: 0,
internalParity: 0,
frc: 0,
externalMCEBase: 0,
microcodeROMParity: 0,
unclassifiedMCEBase: 0,
}
}
func (r *Ras) updateServerCounters(mcError *machineCheckError) {
if strings.Contains(mcError.ErrorMsg, "CACHE Level-2") && strings.Contains(mcError.ErrorMsg, "Error") {
r.serverCounters[levelTwoCache] += 1
}
if strings.Contains(mcError.ErrorMsg, "UPI:") {
r.serverCounters[upi] += 1
}
}
func connectToDB(server string) (*sql.DB, error) {
return sql.Open("sqlite3", server)
}
func (r *Ras) initializeCpuMetricDataIfRequired(socketId int) {
if _, ok := r.cpuSocketCounters[socketId]; !ok {
r.cpuSocketCounters[socketId] = *newMetricCounters()
}
}
func (r *Ras) updateSocketCounters(mcError *machineCheckError) {
r.updateMemoryCounters(mcError)
r.updateProcessorBaseCounters(mcError)
if strings.Contains(mcError.ErrorMsg, "Instruction TLB") && strings.Contains(mcError.ErrorMsg, "Error") {
r.cpuSocketCounters[mcError.SocketId][instructionTLB] += 1
}
if strings.Contains(mcError.ErrorMsg, "BUS") && strings.Contains(mcError.ErrorMsg, "Error") {
r.cpuSocketCounters[mcError.SocketId][processorBus] += 1
}
if (strings.Contains(mcError.ErrorMsg, "CACHE Level-0") ||
strings.Contains(mcError.ErrorMsg, "CACHE Level-1")) &&
strings.Contains(mcError.ErrorMsg, "Error") {
r.cpuSocketCounters[mcError.SocketId][instructionCache] += 1
}
}
func (r *Ras) updateProcessorBaseCounters(mcError *machineCheckError) {
if strings.Contains(mcError.ErrorMsg, "Internal Timer error") {
r.cpuSocketCounters[mcError.SocketId][internalTimer] += 1
r.cpuSocketCounters[mcError.SocketId][processorBase] += 1
}
if strings.Contains(mcError.ErrorMsg, "SMM Handler Code Access Violation") {
r.cpuSocketCounters[mcError.SocketId][smmHandlerCode] += 1
r.cpuSocketCounters[mcError.SocketId][processorBase] += 1
}
if strings.Contains(mcError.ErrorMsg, "Internal parity error") {
r.cpuSocketCounters[mcError.SocketId][internalParity] += 1
r.cpuSocketCounters[mcError.SocketId][processorBase] += 1
}
if strings.Contains(mcError.ErrorMsg, "FRC error") {
r.cpuSocketCounters[mcError.SocketId][frc] += 1
r.cpuSocketCounters[mcError.SocketId][processorBase] += 1
}
if strings.Contains(mcError.ErrorMsg, "External error") {
r.cpuSocketCounters[mcError.SocketId][externalMCEBase] += 1
r.cpuSocketCounters[mcError.SocketId][processorBase] += 1
}
if strings.Contains(mcError.ErrorMsg, "Microcode ROM parity error") {
r.cpuSocketCounters[mcError.SocketId][microcodeROMParity] += 1
r.cpuSocketCounters[mcError.SocketId][processorBase] += 1
}
if strings.Contains(mcError.ErrorMsg, "Unclassified") || strings.Contains(mcError.ErrorMsg, "Internal unclassified") {
r.cpuSocketCounters[mcError.SocketId][unclassifiedMCEBase] += 1
r.cpuSocketCounters[mcError.SocketId][processorBase] += 1
}
}
func (r *Ras) updateMemoryCounters(mcError *machineCheckError) {
if strings.Contains(mcError.ErrorMsg, "Memory read error") {
if strings.Contains(mcError.MciStatusMsg, "Corrected_error") {
r.cpuSocketCounters[mcError.SocketId][memoryReadCorrected] += 1
} else {
r.cpuSocketCounters[mcError.SocketId][memoryReadUncorrected] += 1
}
}
if strings.Contains(mcError.ErrorMsg, "Memory write error") {
if strings.Contains(mcError.MciStatusMsg, "Corrected_error") {
r.cpuSocketCounters[mcError.SocketId][memoryWriteCorrected] += 1
} else {
r.cpuSocketCounters[mcError.SocketId][memoryWriteUncorrected] += 1
}
}
}
func addCpuSocketMetrics(acc telegraf.Accumulator, cpuSocketCounters map[int]metricCounters) {
for socketId, data := range cpuSocketCounters {
tags := map[string]string{
"socket_id": strconv.Itoa(socketId),
}
fields := make(map[string]interface{})
for errorName, count := range data {
fields[errorName] = count
}
acc.AddCounter("ras", fields, tags)
}
}
func addServerMetrics(acc telegraf.Accumulator, counters map[string]int64) {
fields := make(map[string]interface{})
for errorName, count := range counters {
fields[errorName] = count
}
acc.AddCounter("ras", fields, map[string]string{})
}
func fetchMachineCheckError(rows *sql.Rows) (*machineCheckError, error) {
mcError := &machineCheckError{}
err := rows.Scan(&mcError.Id, &mcError.Timestamp, &mcError.ErrorMsg, &mcError.MciStatusMsg, &mcError.SocketId)
if err != nil {
return nil, err
}
return mcError, nil
}
func parseDate(date string) (time.Time, error) {
return time.Parse(dateLayout, date)
}
func init() {
inputs.Add("ras", func() telegraf.Input {
defaultTimestamp, _ := parseDate("1970-01-01 00:00:01 -0700")
return &Ras{
DbPath: defaultDbPath,
latestTimestamp: defaultTimestamp,
cpuSocketCounters: map[int]metricCounters{
0: *newMetricCounters(),
},
serverCounters: map[string]int64{
levelTwoCache: 0,
upi: 0,
},
}
})
}

254
plugins/inputs/ras/ras_test.go
View File

@ -1,254 +0,0 @@
// +build !windows
package ras
import (
"fmt"
"testing"
"github.com/influxdata/telegraf/testutil"
"github.com/stretchr/testify/assert"
)
func TestUpdateCounters(t *testing.T) {
ras := newRas()
for _, mce := range testData {
ras.updateCounters(&mce)
}
assert.Equal(t, 1, len(ras.cpuSocketCounters), "Should contain counters only for single socket")
for metric, value := range ras.cpuSocketCounters[0] {
if metric == processorBase {
// processor_base_errors is sum of other seven errors: internal_timer_errors, smm_handler_code_access_violation_errors,
// internal_parity_errors, frc_errors, external_mce_errors, microcode_rom_parity_errors and unclassified_mce_errors
assert.Equal(t, int64(7), value, fmt.Sprintf("%s should have value of 7", processorBase))
} else {
assert.Equal(t, int64(1), value, fmt.Sprintf("%s should have value of 1", metric))
}
}
for metric, value := range ras.serverCounters {
assert.Equal(t, int64(1), value, fmt.Sprintf("%s should have value of 1", metric))
}
}
func TestUpdateLatestTimestamp(t *testing.T) {
ras := newRas()
ts := "2020-08-01 15:13:27 +0200"
testData = append(testData, []machineCheckError{
{
Timestamp: "2019-05-20 08:25:55 +0200",
SocketId: 0,
ErrorMsg: "",
MciStatusMsg: "",
},
{
Timestamp: "2018-02-21 12:27:22 +0200",
SocketId: 0,
ErrorMsg: "",
MciStatusMsg: "",
},
{
Timestamp: ts,
SocketId: 0,
ErrorMsg: "",
MciStatusMsg: "",
},
}...)
for _, mce := range testData {
err := ras.updateLatestTimestamp(mce.Timestamp)
assert.NoError(t, err)
}
assert.Equal(t, ts, ras.latestTimestamp.Format(dateLayout))
}
func TestMultipleSockets(t *testing.T) {
ras := newRas()
cacheL2 := "Instruction CACHE Level-2 Generic Error"
overflow := "Error_overflow Corrected_error"
testData = []machineCheckError{
{
Timestamp: "2019-05-20 08:25:55 +0200",
SocketId: 0,
ErrorMsg: cacheL2,
MciStatusMsg: overflow,
},
{
Timestamp: "2018-02-21 12:27:22 +0200",
SocketId: 1,
ErrorMsg: cacheL2,
MciStatusMsg: overflow,
},
{
Timestamp: "2020-03-21 14:17:28 +0200",
SocketId: 2,
ErrorMsg: cacheL2,
MciStatusMsg: overflow,
},
{
Timestamp: "2020-03-21 17:24:18 +0200",
SocketId: 3,
ErrorMsg: cacheL2,
MciStatusMsg: overflow,
},
}
for _, mce := range testData {
ras.updateCounters(&mce)
}
assert.Equal(t, 4, len(ras.cpuSocketCounters), "Should contain counters for four sockets")
for _, metricData := range ras.cpuSocketCounters {
for metric, value := range metricData {
if metric == levelTwoCache {
assert.Equal(t, int64(1), value, fmt.Sprintf("%s should have value of 1", levelTwoCache))
} else {
assert.Equal(t, int64(0), value, fmt.Sprintf("%s should have value of 0", metric))
}
}
}
}
func TestMissingDatabase(t *testing.T) {
var acc testutil.Accumulator
ras := newRas()
ras.DbPath = "/tmp/test.db"
err := ras.Gather(&acc)
assert.Error(t, err)
}
func TestEmptyDatabase(t *testing.T) {
ras := newRas()
assert.Equal(t, 1, len(ras.cpuSocketCounters), "Should contain default counters for one socket")
assert.Equal(t, 2, len(ras.serverCounters), "Should contain default counters for server")
for metric, value := range ras.cpuSocketCounters[0] {
assert.Equal(t, int64(0), value, fmt.Sprintf("%s should have value of 0", metric))
}
for metric, value := range ras.serverCounters {
assert.Equal(t, int64(0), value, fmt.Sprintf("%s should have value of 0", metric))
}
}
func newRas() *Ras {
defaultTimestamp, _ := parseDate("1970-01-01 00:00:01 -0700")
return &Ras{
DbPath: defaultDbPath,
latestTimestamp: defaultTimestamp,
cpuSocketCounters: map[int]metricCounters{
0: *newMetricCounters(),
},
serverCounters: map[string]int64{
levelTwoCache: 0,
upi: 0,
},
}
}
var testData = []machineCheckError{
{
Timestamp: "2020-05-20 07:34:53 +0200",
SocketId: 0,
ErrorMsg: "MEMORY CONTROLLER RD_CHANNEL0_ERR Transaction: Memory read error",
MciStatusMsg: "Error_overflow Corrected_error",
},
{
Timestamp: "2020-05-20 07:35:11 +0200",
SocketId: 0,
ErrorMsg: "MEMORY CONTROLLER RD_CHANNEL0_ERR Transaction: Memory read error",
MciStatusMsg: "Uncorrected_error",
},
{
Timestamp: "2020-05-20 07:37:50 +0200",
SocketId: 0,
ErrorMsg: "MEMORY CONTROLLER RD_CHANNEL2_ERR Transaction: Memory write error",
MciStatusMsg: "Uncorrected_error",
},
{
Timestamp: "2020-05-20 08:14:51 +0200",
SocketId: 0,
ErrorMsg: "MEMORY CONTROLLER WR_CHANNEL2_ERR Transaction: Memory write error",
MciStatusMsg: "Error_overflow Corrected_error",
},
{
Timestamp: "2020-05-20 08:15:31 +0200",
SocketId: 0,
ErrorMsg: "corrected filtering (some unreported errors in same region) Instruction CACHE Level-0 Read Error",
MciStatusMsg: "Error_overflow Corrected_error",
},
{
Timestamp: "2020-05-20 08:16:32 +0200",
SocketId: 0,
ErrorMsg: "Instruction TLB Level-0 Error",
MciStatusMsg: "Error_overflow Corrected_error",
},
{
Timestamp: "2020-05-20 08:16:56 +0200",
SocketId: 0,
ErrorMsg: "No Error",
MciStatusMsg: "Error_overflow Corrected_error",
},
{
Timestamp: "2020-05-20 08:17:24 +0200",
SocketId: 0,
ErrorMsg: "Unclassified",
MciStatusMsg: "Error_overflow Corrected_error",
},
{
Timestamp: "2020-05-20 08:17:41 +0200",
SocketId: 0,
ErrorMsg: "Microcode ROM parity error",
MciStatusMsg: "Error_overflow Corrected_error",
},
{
Timestamp: "2020-05-20 08:17:48 +0200",
SocketId: 0,
ErrorMsg: "FRC error",
MciStatusMsg: "Error_overflow Corrected_error",
},
{
Timestamp: "2020-05-20 08:18:18 +0200",
SocketId: 0,
ErrorMsg: "Internal parity error",
MciStatusMsg: "Error_overflow Corrected_error",
},
{
Timestamp: "2020-05-20 08:18:34 +0200",
SocketId: 0,
ErrorMsg: "SMM Handler Code Access Violation",
MciStatusMsg: "Error_overflow Corrected_error",
},
{
Timestamp: "2020-05-20 08:18:54 +0200",
SocketId: 0,
ErrorMsg: "Internal Timer error",
MciStatusMsg: "Error_overflow Corrected_error",
},
{
Timestamp: "2020-05-20 08:21:23 +0200",
SocketId: 0,
ErrorMsg: "BUS Level-3 Generic Generic IO Request-did-not-timeout Error",
MciStatusMsg: "Error_overflow Corrected_error",
},
{
Timestamp: "2020-05-20 08:23:23 +0200",
SocketId: 0,
ErrorMsg: "External error",
MciStatusMsg: "Error_overflow Corrected_error",
},
{
Timestamp: "2020-05-20 08:25:31 +0200",
SocketId: 0,
ErrorMsg: "UPI: COR LL Rx detected CRC error - successful LLR without Phy Reinit",
MciStatusMsg: "Error_overflow Corrected_error",
},
{
Timestamp: "2020-05-20 08:25:55 +0200",
SocketId: 0,
ErrorMsg: "Instruction CACHE Level-2 Generic Error",
MciStatusMsg: "Error_overflow Corrected_error",
},
}

3
plugins/inputs/ras/ras_windows.go
View File

@ -1,3 +0,0 @@
// +build windows
package ras