From a970249cc104f3a2f190ab5b7d94c0af7bd92c07 Mon Sep 17 00:00:00 2001
From: Qiu Jian <qiujian@yunionyun.com>
Date: Sat, 30 Oct 2021 11:46:16 +0800
Subject: [PATCH] fix: delayed probing GPU on host init and add disable_gpu
 option

---
 pkg/hostman/hostinfo/hostinfo.go               | 16 ++++++++++------
 pkg/hostman/isolated_device/isolated_device.go | 13 +++++++++----
 pkg/hostman/options/options.go                 |  4 +++-
 3 files changed, 22 insertions(+), 11 deletions(-)

diff --git a/pkg/hostman/hostinfo/hostinfo.go b/pkg/hostman/hostinfo/hostinfo.go
index 4253132bb0..acfd29f4ed 100644
--- a/pkg/hostman/hostinfo/hostinfo.go
+++ b/pkg/hostman/hostinfo/hostinfo.go
@@ -311,11 +311,7 @@ func (h *SHostInfo) parseConfig() error {
 		h.MasterNic = nil
 	}
 
-	if man, err := isolated_device.NewManager(h); err != nil {
-		return fmt.Errorf("NewIsolatedManager: %v", err)
-	} else {
-		h.IsolatedDeviceMan = man
-	}
+	h.IsolatedDeviceMan = isolated_device.NewManager(h)
 
 	return nil
 }
@@ -1473,7 +1469,15 @@ func (h *SHostInfo) uploadStorageInfo() {
 	go storageman.StartSyncStorageSizeTask(
 		time.Duration(options.HostOptions.SyncStorageInfoDurationSecond) * time.Second,
 	)
-	h.getIsolatedDevices()
+	var err error
+	if !options.HostOptions.DisableGPU {
+		err = h.IsolatedDeviceMan.ProbePCIDevices()
+	}
+	if err != nil {
+		h.onFail(errors.Wrap(err, "Probe PCI device failed"))
+	} else {
+		h.getIsolatedDevices()
+	}
 }
 
 func (h *SHostInfo) onSyncStorageInfoSucc(storage storageman.IStorage, storageInfo jsonutils.JSONObject) {
diff --git a/pkg/hostman/isolated_device/isolated_device.go b/pkg/hostman/isolated_device/isolated_device.go
index e912a496bb..270094fd31 100644
--- a/pkg/hostman/isolated_device/isolated_device.go
+++ b/pkg/hostman/isolated_device/isolated_device.go
@@ -96,17 +96,22 @@ type IsolatedDeviceManager struct {
 	DetachedDevices []*CloudDeviceInfo
 }
 
-func NewManager(host IHost) (*IsolatedDeviceManager, error) {
+func NewManager(host IHost) *IsolatedDeviceManager {
 	man := &IsolatedDeviceManager{
 		host:            host,
 		Devices:         make([]IDevice, 0),
 		DetachedDevices: make([]*CloudDeviceInfo, 0),
 	}
-	err := man.fillPCIDevices()
-	return man, err
+	// Do probe laster - Qiu Jian
+	// err := man.fillPCIDevices()
+	return man
 }
 
-func (man *IsolatedDeviceManager) fillPCIDevices() error {
+func (man *IsolatedDeviceManager) ProbePCIDevices() error {
+	if len(man.Devices) > 0 {
+		// already probed, skip
+		return nil
+	}
 	// only support gpu by now
 	gpus, err := getPassthroughGPUS()
 	if err != nil {
diff --git a/pkg/hostman/options/options.go b/pkg/hostman/options/options.go
index 5a870b4ee6..af4a892b06 100644
--- a/pkg/hostman/options/options.go
+++ b/pkg/hostman/options/options.go
@@ -143,7 +143,9 @@ type SHostOptions struct {
 	DisableProbeKubelet bool   `help:"Disable probe kubelet config" default:"false"`
 	KubeletRunDirectory string `help:"Kubelet config file path" default:"/var/lib/kubelet"`
 
-	DisableKVM bool `help:"force disable KVM" default:"false"`
+	DisableKVM bool `help:"force disable KVM" default:"false" json:"disable_kvm"`
+
+	DisableGPU bool `help:"force disable GPU" default:"false" json:"disable_gpu"`
 }
 
 var (