feat: Support running ext cmds as different users

* Now ipmi can be run as root directly in the exporter * Users can still use wrapper approach if they want Signed-off-by: Mahendra Paipuri <[email protected]>
mahendrapaipuri · Dec 14, 2023 · a41800d · a41800d
1 parent 667231a
commit a41800d
Show file tree

Hide file tree

Showing 3 changed files with 37 additions and 7 deletions.
diff --git a/internal/helpers/helpers.go b/internal/helpers/helpers.go
@@ -4,6 +4,7 @@ import (
 	"fmt"
 	"os/exec"
 	"strings"
+	"syscall"
 
 	"github.com/go-kit/log"
 	"github.com/go-kit/log/level"
@@ -24,7 +25,20 @@ func Execute(cmd string, args []string, logger log.Logger) ([]byte, error) {
 	level.Debug(logger).Log("msg", "Executing", "command", cmd, "args", fmt.Sprintf("%+v", args))
 	out, err := exec.Command(cmd, args...).CombinedOutput()
 	if err != nil {
-		err = fmt.Errorf("error running %s: %s", cmd, err)
+		level.Error(logger).Log("msg", "Error executing command", "command", cmd, "args", fmt.Sprintf("%+v", args), "err", err)
+	}
+	return out, err
+}
+
+// Execute command as a given UID and GID and return stdout/stderr
+func ExecuteAs(cmd string, args []string, uid int, gid int, logger log.Logger) ([]byte, error) {
+	level.Debug(logger).Log("msg", "Executing as user", "command", cmd, "args", fmt.Sprintf("%+v", args), "uid", uid, "gid", gid)
+	execCmd := exec.Command(cmd, args...)
+	execCmd.SysProcAttr = &syscall.SysProcAttr{}
+	execCmd.SysProcAttr.Credential = &syscall.Credential{Uid: uint32(uid), Gid: uint32(gid)}
+	out, err := execCmd.CombinedOutput()
+	if err != nil {
+		level.Error(logger).Log("msg", "Error executing command as user", "command", cmd, "args", fmt.Sprintf("%+v", args), "uid", uid, "gid", gid, "err", err)
 	}
 	return out, err
 }
diff --git a/pkg/collector/ipmi.go b/pkg/collector/ipmi.go
@@ -28,10 +28,14 @@ type impiCollector struct {
 }
 
 var (
-	ipmiDcmiWrapperExec = kingpin.Flag(
-		"collector.ipmi.dcmi.wrapper.path",
-		"Path to IPMI DCMI executable wrapper.",
+	ipmiDcmiExec = kingpin.Flag(
+		"collector.ipmi.dcmi.exec.path",
+		"Path to IPMI DCMI executable.",
 	).Default("ipmi-dcmi-wrapper").String()
+	ipmiDcmiExecAsRoot = kingpin.Flag(
+		"collector.ipmi.dcmi.exec.run.as.root",
+		"Execute IPMI DCMI command as root. This requires batchjob_exporter to run as root or to have appropriate capabilities (cap_setuid).",
+	).Default("false").Bool()
 	ipmiDCMIPowerMeasurementRegex = regexp.MustCompile(
 		`^Power Measurement\s*:\s*(?P<value>Active|Not\sAvailable).*`,
 	)
@@ -78,16 +82,27 @@ func getValue(ipmiOutput []byte, regex *regexp.Regexp) (string, error) {
 
 // Update implements Collector and exposes IPMI DCMI power related metrics.
 func (c *impiCollector) Update(ch chan<- prometheus.Metric) error {
-	args := []string{""}
-	stdOut, err := helpers.Execute(*ipmiDcmiWrapperExec, args, c.logger)
+	args := []string{"--get-system-power-statistics"}
+	var stdOut []byte
+	var err error
+
+	// Execute ipmi-dcmi command
+	if *ipmiDcmiExecAsRoot {
+		stdOut, err = helpers.ExecuteAs(*ipmiDcmiExec, args, 0, 0, c.logger)
+	} else {
+		stdOut, err = helpers.Execute(*ipmiDcmiExec, args, c.logger)
+	}
 	if err != nil {
 		return err
 	}
+
+	// Parse power consumption from output
 	currentPowerConsumption, err := c.getCurrentPowerConsumption(stdOut)
 	if err != nil {
 		level.Error(c.logger).Log("msg", "Failed to collect IPMI DCMI data", "error", err)
 		return err
 	}
+
 	// Returned value negative == Power Measurement is not avail
 	if currentPowerConsumption > -1 {
 		ch <- prometheus.MustNewConstMetric(c.wattsMetricDesc, prometheus.CounterValue, float64(currentPowerConsumption))
@@ -102,6 +117,7 @@ func (c *impiCollector) getCurrentPowerConsumption(ipmiOutput []byte) (float64,
 	if err != nil {
 		return -1, err
 	}
+
 	// When Power Measurement in 'Active' state - we can get watts
 	if value == "Active" {
 		value, err := getValue(ipmiOutput, ipmiDCMICurrentPowerRegex)

diff --git a/scripts/e2e-test.sh b/scripts/e2e-test.sh
@@ -118,7 +118,7 @@ then
     --path.sysfs="pkg/collector/fixtures/sys" \
     --path.cgroupfs="pkg/collector/fixtures/sys/fs/cgroup" \
     --collector.slurm.job.stat.path="pkg/collector/fixtures/slurmjobstat" \
-    --collector.ipmi.dcmi.wrapper.path="pkg/collector/fixtures/ipmi-dcmi-wrapper.sh" \
+    --collector.ipmi.dcmi.exec.path="pkg/collector/fixtures/ipmi-dcmi-wrapper.sh" \
     --collector.nvidia_gpu \
     --collector.nvidia.gpu.stat.path="pkg/collector/fixtures/gpustat" \
     --web.listen-address "127.0.0.1:${port}" \