[EBPF] gpu: enable periodic uprobe attacher scan (#33630)

DataDog · Feb 4, 2025 · 7eafe83 · 7eafe83
1 parent d4488fa
commit 7eafe83
Show file tree

Hide file tree

Showing 3 changed files with 14 additions and 12 deletions.
diff --git a/pkg/gpu/config/config.go b/pkg/gpu/config/config.go
@@ -26,8 +26,8 @@ type Config struct {
 	ebpf.Config
 	// Enabled indicates whether the GPU monitoring probe is enabled.
 	Enabled bool
-	// ScanTerminatedProcessesInterval is the interval at which the probe scans for terminated processes.
-	ScanTerminatedProcessesInterval time.Duration
+	// ScanProcessesInterval is the interval at which the probe scans for new or terminated processes.
+	ScanProcessesInterval time.Duration
 	// InitialProcessSync indicates whether the probe should sync the process list on startup.
 	InitialProcessSync bool
 	// NVMLLibraryPath is the path of the native libnvidia-ml.so library
@@ -40,11 +40,11 @@ type Config struct {
 func New() *Config {
 	spCfg := pkgconfigsetup.SystemProbe()
 	return &Config{
-		Config:                          *ebpf.NewConfig(),
-		ScanTerminatedProcessesInterval: time.Duration(spCfg.GetInt(sysconfig.FullKeyPath(GPUNS, "process_scan_interval_seconds"))) * time.Second,
-		InitialProcessSync:              spCfg.GetBool(sysconfig.FullKeyPath(GPUNS, "initial_process_sync")),
-		NVMLLibraryPath:                 spCfg.GetString(sysconfig.FullKeyPath(GPUNS, "nvml_lib_path")),
-		Enabled:                         spCfg.GetBool(sysconfig.FullKeyPath(GPUNS, "enabled")),
-		ConfigureCgroupPerms:            spCfg.GetBool(sysconfig.FullKeyPath(GPUNS, "configure_cgroup_perms")),
+		Config:                *ebpf.NewConfig(),
+		ScanProcessesInterval: time.Duration(spCfg.GetInt(sysconfig.FullKeyPath(GPUNS, "process_scan_interval_seconds"))) * time.Second,
+		InitialProcessSync:    spCfg.GetBool(sysconfig.FullKeyPath(GPUNS, "initial_process_sync")),
+		NVMLLibraryPath:       spCfg.GetString(sysconfig.FullKeyPath(GPUNS, "nvml_lib_path")),
+		Enabled:               spCfg.GetBool(sysconfig.FullKeyPath(GPUNS, "enabled")),
+		ConfigureCgroupPerms:  spCfg.GetBool(sysconfig.FullKeyPath(GPUNS, "configure_cgroup_perms")),
 	}
 }
diff --git a/pkg/gpu/consumer.go b/pkg/gpu/consumer.go
@@ -104,7 +104,7 @@ func (c *cudaEventConsumer) Start() {
 	c.wg.Add(1)
 	go func() {
 		c.running.Store(true)
-		processSync := time.NewTicker(c.cfg.ScanTerminatedProcessesInterval)
+		processSync := time.NewTicker(c.cfg.ScanProcessesInterval)
 
 		defer func() {
 			cleanupExit()

diff --git a/pkg/gpu/probe.go b/pkg/gpu/probe.go
@@ -325,9 +325,11 @@ func getAttacherConfig(cfg *config.Config) uprobes.AttacherConfig {
 				},
 			},
 		},
-		EbpfConfig:         &cfg.Config,
-		PerformInitialScan: cfg.InitialProcessSync,
-		SharedLibsLibset:   sharedlibraries.LibsetGPU,
+		EbpfConfig:                     &cfg.Config,
+		PerformInitialScan:             cfg.InitialProcessSync,
+		SharedLibsLibset:               sharedlibraries.LibsetGPU,
+		ScanProcessesInterval:          cfg.ScanProcessesInterval,
+		EnablePeriodicScanNewProcesses: true,
 	}
 }