Some fixes after PR review

Signed-off-by: Aleksandr Zimin <[email protected]>
deckhouse · Mar 14, 2024 · 038a0dd · 038a0dd
1 parent 03e8209
commit 038a0dd
Show file tree

Hide file tree

Showing 13 changed files with 162 additions and 186 deletions.
diff --git a/.github/workflows/build_dev.yml b/.github/workflows/build_dev.yml
@@ -7,6 +7,8 @@ env:
   MODULES_MODULE_SOURCE: ${{ vars.DEV_MODULE_SOURCE }}
   MODULES_REGISTRY_LOGIN: ${{ vars.DEV_MODULES_REGISTRY_LOGIN }}
   MODULES_REGISTRY_PASSWORD: ${{ secrets.DEV_MODULES_REGISTRY_PASSWORD }}
+  # for ex https://user:[email protected]/group
+  SOURCE_REPO: "${{ secrets.SOURCE_REPO }}"
 
 on:
   pull_request:

diff --git a/.github/workflows/build_prod.yml b/.github/workflows/build_prod.yml
@@ -8,6 +8,8 @@ env:
   MODULES_REGISTRY_LOGIN: ${{ vars.PROD_MODULES_REGISTRY_LOGIN }}
   MODULES_REGISTRY_PASSWORD: ${{ secrets.PROD_MODULES_REGISTRY_PASSWORD }}
   MODULES_MODULE_TAG: ${{ github.ref_name }}
+  # for ex https://user:[email protected]/group
+  SOURCE_REPO: "${{ secrets.SOURCE_REPO }}"
 
 on:
   push:

diff --git a/.werf/images.yaml b/.werf/images.yaml
@@ -8,6 +8,8 @@
 image: images/{{ $ctx.ImageName }}
 context: images/{{ $ctx.ImageName }}
 dockerfile: Dockerfile
+args:
+  SOURCE_REPO: {{ env "SOURCE_REPO" | default "https://github.com" }}
 
   {{- /* For werf.inc.yaml render content by providing the ImageName param. */ -}}
   {{- else }}

diff --git a/images/agent/Dockerfile b/images/agent/Dockerfile
@@ -4,6 +4,9 @@ ARG BASE_IMAGE=registry.deckhouse.io/base_images/scratch@sha256:b054705fcc9f2205
 
 #################################
 FROM $UBUNTU_UTILS_BUILDER as util-linux-builder
+ARG SOURCE_REPO
+ARG UTIL_LINUX_VERSION=2.39.3
+
 RUN apt-get update && apt-get install -y \
     build-essential \
     git \
@@ -17,12 +20,13 @@ RUN apt-get update && apt-get install -y \
     flex \
     && rm -rf /var/lib/apt/lists/*
 
-RUN git clone https://github.com/util-linux/util-linux.git /util-linux
-
 WORKDIR /util-linux
-RUN git checkout v2.39.3
-RUN ./autogen.sh
-RUN ./configure LDFLAGS="-static" --enable-static-programs -disable-shared
+
+RUN git clone ${SOURCE_REPO}/util-linux/util-linux.git . && \
+  git checkout v${UTIL_LINUX_VERSION} && \
+  ./autogen.sh && \
+  ./configure LDFLAGS="-static" --enable-static-programs -disable-shared
+
 RUN make LDFLAGS="--static" nsenter
 
 #################################
@@ -42,7 +46,7 @@ RUN GOOS=linux GOARCH=amd64 go build -o sds-node-configurator-agent
 #################################
 FROM --platform=linux/amd64 $BASE_IMAGE
 
-COPY --from=util-linux-builder /util-linux/nsenter /usr/local/bin/flant/nsenter.static
+COPY --from=util-linux-builder /util-linux/nsenter /opt/deckhouse/sds/nsenter.static
 COPY --from=agent-builder /go/src/cmd/sds-node-configurator-agent /go/src/cmd/sds-node-configurator-agent
 
 CMD ["/go/src/cmd/sds-node-configurator-agent"]
diff --git a/images/agent/config/config.go b/images/agent/config/config.go
@@ -21,6 +21,7 @@ import (
 	"fmt"
 	"os"
 	"os/exec"
+	"sds-node-configurator/internal"
 	"sds-node-configurator/pkg/logger"
 	"time"
 )
@@ -80,7 +81,7 @@ func getMachineId() (string, error) {
 		args := []string{"-m", "-u", "-i", "-n", "-p", "-t", "1", "cat", "/etc/machine-id"}
 
 		var stdout bytes.Buffer
-		cmd := exec.Command("/usr/local/bin/flant/nsenter.static", args...)
+		cmd := exec.Command(internal.NSENTERCmd, args...)
 		cmd.Stdout = &stdout
 		err := cmd.Run()
 		if err != nil {

diff --git a/images/agent/internal/const.go b/images/agent/internal/const.go
@@ -29,6 +29,9 @@ const (
 	ResizeDelta                  = "32Mi"
 	KubernetesApiRequestLimit    = 5
 	KubernetesApiRequestTimeout  = 1
+	NSENTERCmd                   = "/opt/deckhouse/sds/nsenter.static"
+	LSBLKCmd                     = "/opt/deckhouse/sds/lsblk.static"
+	LVMCmd                       = "/opt/deckhouse/sds/lvm.static"
 )
 
 var (

diff --git a/images/agent/pkg/controller/lvm_volume_group_watcher.go b/images/agent/pkg/controller/lvm_volume_group_watcher.go
@@ -80,7 +80,7 @@ func RunLVMVolumeGroupWatcherController(
 	}
 
 	createFunc := func(ctx context.Context, e event.CreateEvent, q workqueue.RateLimitingInterface) {
-		log.Info(fmt.Sprintf("[RunLVMVolumeGroupController] event create LVMVolumeGroup, name: %s", e.Object.GetName()))
+		log.Info(fmt.Sprintf("[RunLVMVolumeGroupController] Get event CREATE for resource LVMVolumeGroup, name: %s", e.Object.GetName()))
 
 		request := reconcile.Request{NamespacedName: types.NamespacedName{Namespace: e.Object.GetNamespace(), Name: e.Object.GetName()}}
 		shouldRequeue, err := ReconcileLVMVG(ctx, metrics, e.Object.GetName(), e.Object.GetNamespace(), cfg.NodeName, log, cl)
@@ -92,7 +92,7 @@ func RunLVMVolumeGroupWatcherController(
 	}
 
 	updateFunc := func(ctx context.Context, e event.UpdateEvent, q workqueue.RateLimitingInterface) {
-		log.Info(fmt.Sprintf("[RunLVMVolumeGroupWatcherController] update LVMVolumeGroupn, name: %s", e.ObjectNew.GetName()))
+		log.Info(fmt.Sprintf("[RunLVMVolumeGroupWatcherController] Get event UPDATE for resource LVMVolumeGroup, name: %s", e.ObjectNew.GetName()))
 
 		newLVG, ok := e.ObjectNew.(*v1alpha1.LvmVolumeGroup)
 		if !ok {
@@ -200,39 +200,32 @@ func ReconcileLVMVG(
 		log.Error(err, fmt.Sprintf("[ReconcileLVMVG] error getLVMVolumeGroup, objectname: %s", objectName))
 		return true, err
 	}
-	validation, status, err := ValidateLVMGroup(ctx, cl, metrics, lvg, objectNameSpace, nodeName)
-
 	if lvg == nil {
 		err = errors.New("nil pointer detected")
 		log.Error(err, "[ReconcileLVMVG] requested LVMVG group in nil")
 		return true, err
 	}
 
-	if status.Health == NonOperational {
-		health := status.Health
-		var message string
-		if err != nil {
-			message = err.Error()
-		}
-
-		log.Error(err, fmt.Sprintf("[ReconcileLVMVG] ValidateLVMGroup, resource name: %s, message: %s", lvg.Name, message))
-		err = updateLVMVolumeGroupHealthStatus(ctx, cl, metrics, lvg.Name, lvg.Namespace, message, health)
-		if err != nil {
-			log.Error(err, fmt.Sprintf("[ReconcileLVMVG] error update LVMVolumeGroup %s", lvg.Name))
-			return true, err
-		}
-	}
+	isOwnedByNode, status, err := CheckLVMVGNodeOwnership(ctx, cl, metrics, lvg, objectNameSpace, nodeName)
 
 	if err != nil {
-		log.Error(err, fmt.Sprintf("[ReconcileLVMVG] validationLVMGroup failed, resource name: %s", lvg.Name))
-		return false, err
+		log.Error(err, fmt.Sprintf("[ReconcileLVMVG] error CheckLVMVGNodeOwnership, resource name: %s", lvg.Name))
+		if status.Health == NonOperational {
+			health := status.Health
+			message := status.Message
+			log.Error(err, fmt.Sprintf("[ReconcileLVMVG] ValidateLVMGroup, resource name: %s, health: %s, phase: %s, message: %s", lvg.Name, health, status.Phase, message))
+			err = updateLVMVolumeGroupHealthStatus(ctx, cl, metrics, lvg.Name, lvg.Namespace, message, health)
+			if err != nil {
+				log.Error(err, fmt.Sprintf("[ReconcileLVMVG] error update LVMVolumeGroup %s", lvg.Name))
+				return true, err
+			}
+		}
+		return true, err
 	}
 
-	if validation == false {
-		err = errors.New("resource validation failed")
-		log.Error(err, fmt.Sprintf("[ReconcileLVMVG] validation failed for resource, name: %s", lvg.Name))
-		log.Error(err, fmt.Sprintf("[ReconcileLVMVG] status.Message = %s", status.Message))
-		return false, err
+	if !isOwnedByNode {
+		log.Debug(fmt.Sprintf("[ReconcileLVMVG] resource is not owned by node, name: %s, skip it", lvg.Name))
+		return false, nil
 	}
 
 	log.Info("[ReconcileLVMVG] validation passed")
@@ -272,14 +265,14 @@ func ReconcileLVMVG(
 	}
 	log.Info(fmt.Sprintf(`[ReconcileLVMVG] event was created for resource, name: %s`, lvg.Name))
 
-	existVG, err := ExistVG(lvg.Spec.ActualVGNameOnTheNode, log, metrics)
+	isVgExist, vg, err := GetVGFromNode(lvg.Spec.ActualVGNameOnTheNode, log, metrics)
 	if err != nil {
 		log.Error(err, fmt.Sprintf("[ReconcileLVMVG] error ExistVG, name: %s", lvg.Spec.ActualVGNameOnTheNode))
 		return true, err
 	}
-	if existVG {
-		log.Debug("[ReconcileLVMVG] tries to update ")
-		updated, err := UpdateLVMVolumeGroupTagsName(log, metrics, lvg)
+	if isVgExist {
+		log.Debug("[ReconcileLVMVG] start UpdateLVMVolumeGroupTagsName for r " + lvg.Name)
+		updated, err := UpdateLVMVolumeGroupTagsName(log, metrics, vg, lvg)
 		if err != nil {
 			log.Error(err, fmt.Sprintf("[ReconcileLVMVG] unable to update VG tags on VG, name: %s", lvg.Spec.ActualVGNameOnTheNode))
 			return true, err
@@ -292,16 +285,16 @@ func ReconcileLVMVG(
 		}
 
 		log.Info("[ReconcileLVMVG] validation and choosing the type of operation")
-		extendPVs, shrinkPVs, err := ValidateTypeLVMGroup(ctx, cl, metrics, lvg, log)
+		extendPVs, shrinkPVs, err := ValidateOperationTypeLVMGroup(ctx, cl, metrics, lvg, log)
 		if err != nil {
-			log.Error(err, fmt.Sprintf("[ReconcileLVMVG] error ValidateTypeLVMGroup, name: %s", lvg.Name))
+			log.Error(err, fmt.Sprintf("[ReconcileLVMVG] error ValidateOperationTypeLVMGroup, name: %s", lvg.Name))
 			return true, err
 		}
 
-		if err == nil && extendPVs == nil && shrinkPVs == nil {
-			log.Warning("[ReconcileLVMVG] ValidateTypeLVMGroup FAIL")
-			//todo retry and send message
-		}
+		// if err == nil && extendPVs == nil && shrinkPVs == nil {
+		// 	log.Warning(fmt.Sprintf("[ReconcileLVMVG] ValidateOperationTypeLVMGroup FAIL for resource %s", lvg.Name))
+		// 	//todo retry and send message
+		// }
 
 		log.Debug("----- extendPVs list -----")
 		for _, pvExt := range extendPVs {
@@ -449,6 +442,7 @@ func ReconcileLVMVG(
 			log.Error(err, fmt.Sprintf("[ReconcileLVMVG] error CreateEventLVMVolumeGroup, resource name: %s", lvg.Name))
 		}
 
+		log.Debug("[ReconcileLVMVG] Start CreateVGComplex function for resource " + lvg.Name)
 		err := CreateVGComplex(ctx, cl, metrics, lvg, log)
 		if err != nil {
 			log.Error(err, fmt.Sprintf("[ReconcileLVMVG] unable to CreateVGComplex for resource, name: %s", lvg.Name))