fix: add shell-level timeout to perf stat capability probes in metadata collection

cmd/metrics/metadata.go

@@ -147,7 +147,7 @@ var baseMetadataScripts = []script.ScriptDefinition{
 	{
 		Name: scriptPerfSupportedEvents,
 		ScriptTemplate: `# Parse perf list JSON output to extract Hardware events and cstate/power events
-perf list --json 2>/dev/null | awk '
+timeout 30 perf list --json 2>/dev/null | awk '
 BEGIN {
     in_hardware_event = 0
     event_name = ""
@@ -192,7 +192,7 @@ BEGIN {
 	{
 		Name: scriptPerfAllSupportedEvents,
 		ScriptTemplate: `# Parse perf list JSON output to extract Hardware events and cstate/power events
-perf list --json 2>/dev/null | awk '
+timeout 30 perf list --json 2>/dev/null | awk '
 BEGIN {
     event_name = ""
 }
@@ -226,52 +226,52 @@ BEGIN {
 	},
 	{
 		Name:           scriptPerfStatInstructions,
-		ScriptTemplate: "perf stat -a -e instructions sleep 1",
+		ScriptTemplate: "timeout 30 perf stat -a -e instructions sleep 1",
 		Depends:        []string{"perf"},
 	},
 	{
 		Name:           scriptPerfStatRefCycles,
-		ScriptTemplate: "perf stat -a -e ref-cycles sleep 1",
+		ScriptTemplate: "timeout 30 perf stat -a -e ref-cycles sleep 1",
 		Depends:        []string{"perf"},
 	},
 	{
 		Name:           scriptPerfStatPEBS,
-		ScriptTemplate: "perf stat -a -e INT_MISC.UNKNOWN_BRANCH_CYCLES sleep 1",
+		ScriptTemplate: "timeout 30 perf stat -a -e INT_MISC.UNKNOWN_BRANCH_CYCLES sleep 1",
 		Architectures:  []string{cpus.X86Architecture},
 		Depends:        []string{"perf"},
 	},
 	{
 		Name:           scriptPerfStatOCR,
-		ScriptTemplate: "perf stat -a -e OCR.READS_TO_CORE.LOCAL_DRAM sleep 1",
+		ScriptTemplate: "timeout 30 perf stat -a -e OCR.READS_TO_CORE.LOCAL_DRAM sleep 1",
 		Architectures:  []string{cpus.X86Architecture},
 		Depends:        []string{"perf"},
 	},
 	{
 		Name:           scriptPerfStatTMA,
-		ScriptTemplate: "perf stat -a -e '{topdown.slots, topdown-bad-spec}' sleep 1",
+		ScriptTemplate: "timeout 30 perf stat -a -e '{topdown.slots, topdown-bad-spec}' sleep 1",
 		Architectures:  []string{cpus.X86Architecture},
 		Depends:        []string{"perf"},
 	},
 	{
 		Name:           scriptPerfStatAMDUncoreProbe,
-		ScriptTemplate: `perf stat -a -e "l3/event=0x4,umask=0xff,enallcores=0x1,enallslices=0x1,threadmask=0x3,name='l3_lookup_state.all_coherent_accesses_to_l3'/" sleep 1`,
+		ScriptTemplate: `timeout 30 perf stat -a -e "l3/event=0x4,umask=0xff,enallcores=0x1,enallslices=0x1,threadmask=0x3,name='l3_lookup_state.all_coherent_accesses_to_l3'/" sleep 1`,
 		Architectures:  []string{cpus.X86Architecture},
 		Vendors:        []string{cpus.AMDVendor},
 		Depends:        []string{"perf"},
 	},
 	{
 		Name:           scriptPerfStatFixedInstr,
-		ScriptTemplate: "perf stat -a -e '{{{.InstructionsList}}}' sleep 1",
+		ScriptTemplate: "timeout 30 perf stat -a -e '{{{.InstructionsList}}}' sleep 1",
 		Depends:        []string{"perf"},
 	},
 	{
 		Name:           scriptPerfStatFixedCycles,
-		ScriptTemplate: "perf stat -a -e '{{{.CpuCyclesList}}}' sleep 1",
+		ScriptTemplate: "timeout 30 perf stat -a -e '{{{.CpuCyclesList}}}' sleep 1",
 		Depends:        []string{"perf"},
 	},
 	{
 		Name:           scriptPerfStatFixedRefCycles,
-		ScriptTemplate: "perf stat -a -e '{{{.RefCyclesList}}}' sleep 1",
+		ScriptTemplate: "timeout 30 perf stat -a -e '{{{.RefCyclesList}}}' sleep 1",
 		Depends:        []string{"perf"},
 	},
 	{
@@ -300,15 +300,25 @@ BEGIN {
 	},
 }
 
+// metadataScriptTimeout bounds how long any single metadata script may run.
+// Metadata collection is a bounded probing phase -- every script here reads a
+// sysfs/procfs value, runs a tool, or runs 'perf stat ... sleep 1' -- so none has
+// a legitimate reason to run for long. Without a bound, a probe that wedges (e.g.
+// a perf event that hangs the PMU on some virtualized instance types) stalls
+// collection indefinitely, because the controller waits on it forever.
+const metadataScriptTimeout = 60
+
 // getMetadataScripts returns the list of scripts to run for metadata collection.
-// It copies the base definitions and applies template replacements and privilege settings.
+// It copies the base definitions and applies template replacements, privilege
+// settings, and the metadata script timeout.
 func getMetadataScripts(noRoot bool, noSystemSummary bool, numGPCounters int) ([]script.ScriptDefinition, error) {
 	metadataScripts := make([]script.ScriptDefinition, 0, len(baseMetadataScripts))
 
 	// Copy base scripts and apply settings
 	for _, baseDef := range baseMetadataScripts {
 		scriptDef := baseDef
 		scriptDef.Superuser = !noRoot
+		scriptDef.Timeout = metadataScriptTimeout
 
 		// Apply template replacements for fixed counter scripts
 		switch scriptDef.Name {
@@ -339,6 +349,10 @@ func getMetadataScripts(noRoot bool, noSystemSummary bool, numGPCounters int) ([
 	if !noSystemSummary {
 		for _, scriptName := range app.TableDefinitions[app.SystemSummaryTableName].ScriptNames {
 			scriptDef := script.GetScriptByName(scriptName)
+			// Only tighten the budget; never loosen one a script set for itself.
+			if scriptDef.Timeout == 0 || scriptDef.Timeout > metadataScriptTimeout {
+				scriptDef.Timeout = metadataScriptTimeout
+			}
 			metadataScripts = append(metadataScripts, scriptDef)
 		}
 	}