Add procfs fallback to netdev collector (#2509)
Some systems have broken netlink messages due to patched kernels. Since these messages can not be parsed, add a flag to fall back to parsing from `/proc/net/dev`. Fixes: https://github.com/prometheus/node_exporter/issues/2502 Signed-off-by: Ben Kochie <superq@gmail.com> Signed-off-by: Ben Kochie <superq@gmail.com>pull/2517/head
parent
440a132c38
commit
ba8c043079
|
@ -17,28 +17,42 @@
|
||||||
package collector
|
package collector
|
||||||
|
|
||||||
import (
|
import (
|
||||||
|
"fmt"
|
||||||
|
|
||||||
"github.com/go-kit/log"
|
"github.com/go-kit/log"
|
||||||
"github.com/go-kit/log/level"
|
"github.com/go-kit/log/level"
|
||||||
|
|
||||||
"github.com/jsimonetti/rtnetlink"
|
"github.com/jsimonetti/rtnetlink"
|
||||||
|
"github.com/prometheus/procfs"
|
||||||
|
"gopkg.in/alecthomas/kingpin.v2"
|
||||||
|
)
|
||||||
|
|
||||||
|
var (
|
||||||
|
netDevNetlink = kingpin.Flag("collector.netdev.netlink", "Use netlink to gather stats instead of /proc/net/dev.").Default("true").Bool()
|
||||||
)
|
)
|
||||||
|
|
||||||
func getNetDevStats(filter *deviceFilter, logger log.Logger) (netDevStats, error) {
|
func getNetDevStats(filter *deviceFilter, logger log.Logger) (netDevStats, error) {
|
||||||
|
if *netDevNetlink {
|
||||||
|
return netlinkStats(filter, logger)
|
||||||
|
}
|
||||||
|
return procNetDevStats(filter, logger)
|
||||||
|
}
|
||||||
|
|
||||||
|
func netlinkStats(filter *deviceFilter, logger log.Logger) (netDevStats, error) {
|
||||||
conn, err := rtnetlink.Dial(nil)
|
conn, err := rtnetlink.Dial(nil)
|
||||||
if err != nil {
|
if err != nil {
|
||||||
return nil, err
|
return nil, err
|
||||||
}
|
}
|
||||||
defer conn.Close()
|
|
||||||
|
|
||||||
|
defer conn.Close()
|
||||||
links, err := conn.Link.List()
|
links, err := conn.Link.List()
|
||||||
if err != nil {
|
if err != nil {
|
||||||
return nil, err
|
return nil, err
|
||||||
}
|
}
|
||||||
|
|
||||||
return netlinkStats(links, filter, logger), nil
|
return parseNetlinkStats(links, filter, logger), nil
|
||||||
}
|
}
|
||||||
|
|
||||||
func netlinkStats(links []rtnetlink.LinkMessage, filter *deviceFilter, logger log.Logger) netDevStats {
|
func parseNetlinkStats(links []rtnetlink.LinkMessage, filter *deviceFilter, logger log.Logger) netDevStats {
|
||||||
metrics := netDevStats{}
|
metrics := netDevStats{}
|
||||||
|
|
||||||
for _, msg := range links {
|
for _, msg := range links {
|
||||||
|
@ -87,3 +101,47 @@ func netlinkStats(links []rtnetlink.LinkMessage, filter *deviceFilter, logger lo
|
||||||
|
|
||||||
return metrics
|
return metrics
|
||||||
}
|
}
|
||||||
|
|
||||||
|
func procNetDevStats(filter *deviceFilter, logger log.Logger) (netDevStats, error) {
|
||||||
|
metrics := netDevStats{}
|
||||||
|
|
||||||
|
fs, err := procfs.NewFS(*procPath)
|
||||||
|
if err != nil {
|
||||||
|
return metrics, fmt.Errorf("failed to open procfs: %w", err)
|
||||||
|
}
|
||||||
|
|
||||||
|
netDev, err := fs.NetDev()
|
||||||
|
if err != nil {
|
||||||
|
return metrics, fmt.Errorf("failed to parse /proc/net/dev: %w", err)
|
||||||
|
}
|
||||||
|
|
||||||
|
for _, stats := range netDev {
|
||||||
|
name := stats.Name
|
||||||
|
|
||||||
|
if filter.ignored(name) {
|
||||||
|
level.Debug(logger).Log("msg", "Ignoring device", "device", name)
|
||||||
|
continue
|
||||||
|
}
|
||||||
|
|
||||||
|
metrics[name] = map[string]uint64{
|
||||||
|
"receive_bytes": stats.RxBytes,
|
||||||
|
"receive_packets": stats.RxPackets,
|
||||||
|
"receive_errors": stats.RxErrors,
|
||||||
|
"receive_dropped": stats.RxDropped,
|
||||||
|
"receive_fifo": stats.RxFIFO,
|
||||||
|
"receive_frame": stats.RxFrame,
|
||||||
|
"receive_compressed": stats.RxCompressed,
|
||||||
|
"receive_multicast": stats.RxMulticast,
|
||||||
|
"transmit_bytes": stats.TxBytes,
|
||||||
|
"transmit_packets": stats.TxPackets,
|
||||||
|
"transmit_errors": stats.TxErrors,
|
||||||
|
"transmit_dropped": stats.TxDropped,
|
||||||
|
"transmit_fifo": stats.TxFIFO,
|
||||||
|
"transmit_colls": stats.TxCollisions,
|
||||||
|
"transmit_carrier": stats.TxCarrier,
|
||||||
|
"transmit_compressed": stats.TxCompressed,
|
||||||
|
}
|
||||||
|
}
|
||||||
|
|
||||||
|
return metrics, nil
|
||||||
|
}
|
||||||
|
|
|
@ -163,7 +163,7 @@ var links = []rtnetlink.LinkMessage{
|
||||||
func TestNetDevStatsIgnore(t *testing.T) {
|
func TestNetDevStatsIgnore(t *testing.T) {
|
||||||
filter := newDeviceFilter("^veth", "")
|
filter := newDeviceFilter("^veth", "")
|
||||||
|
|
||||||
netStats := netlinkStats(links, &filter, log.NewNopLogger())
|
netStats := parseNetlinkStats(links, &filter, log.NewNopLogger())
|
||||||
|
|
||||||
if want, got := uint64(10437182923), netStats["wlan0"]["receive_bytes"]; want != got {
|
if want, got := uint64(10437182923), netStats["wlan0"]["receive_bytes"]; want != got {
|
||||||
t.Errorf("want netstat wlan0 bytes %v, got %v", want, got)
|
t.Errorf("want netstat wlan0 bytes %v, got %v", want, got)
|
||||||
|
@ -196,7 +196,7 @@ func TestNetDevStatsIgnore(t *testing.T) {
|
||||||
|
|
||||||
func TestNetDevStatsAccept(t *testing.T) {
|
func TestNetDevStatsAccept(t *testing.T) {
|
||||||
filter := newDeviceFilter("", "^💩0$")
|
filter := newDeviceFilter("", "^💩0$")
|
||||||
netStats := netlinkStats(links, &filter, log.NewNopLogger())
|
netStats := parseNetlinkStats(links, &filter, log.NewNopLogger())
|
||||||
|
|
||||||
if want, got := 1, len(netStats); want != got {
|
if want, got := 1, len(netStats); want != got {
|
||||||
t.Errorf("want count of devices to be %d, got %d", want, got)
|
t.Errorf("want count of devices to be %d, got %d", want, got)
|
||||||
|
@ -227,7 +227,7 @@ func TestNetDevLegacyMetricNames(t *testing.T) {
|
||||||
}
|
}
|
||||||
|
|
||||||
filter := newDeviceFilter("", "")
|
filter := newDeviceFilter("", "")
|
||||||
netStats := netlinkStats(links, &filter, log.NewNopLogger())
|
netStats := parseNetlinkStats(links, &filter, log.NewNopLogger())
|
||||||
|
|
||||||
for dev, devStats := range netStats {
|
for dev, devStats := range netStats {
|
||||||
legacy(devStats)
|
legacy(devStats)
|
||||||
|
@ -260,7 +260,7 @@ func TestNetDevLegacyMetricValues(t *testing.T) {
|
||||||
}
|
}
|
||||||
|
|
||||||
filter := newDeviceFilter("", "^enp0s0f0$")
|
filter := newDeviceFilter("", "^enp0s0f0$")
|
||||||
netStats := netlinkStats(links, &filter, log.NewNopLogger())
|
netStats := parseNetlinkStats(links, &filter, log.NewNopLogger())
|
||||||
metrics, ok := netStats["enp0s0f0"]
|
metrics, ok := netStats["enp0s0f0"]
|
||||||
if !ok {
|
if !ok {
|
||||||
t.Error("expected stats for interface enp0s0f0")
|
t.Error("expected stats for interface enp0s0f0")
|
||||||
|
@ -282,7 +282,7 @@ func TestNetDevLegacyMetricValues(t *testing.T) {
|
||||||
|
|
||||||
func TestNetDevMetricValues(t *testing.T) {
|
func TestNetDevMetricValues(t *testing.T) {
|
||||||
filter := newDeviceFilter("", "")
|
filter := newDeviceFilter("", "")
|
||||||
netStats := netlinkStats(links, &filter, log.NewNopLogger())
|
netStats := parseNetlinkStats(links, &filter, log.NewNopLogger())
|
||||||
|
|
||||||
for _, msg := range links {
|
for _, msg := range links {
|
||||||
device := msg.Attributes.Name
|
device := msg.Attributes.Name
|
||||||
|
|
Loading…
Reference in New Issue