Merge pull request #6617 from ionos-cloud/update-ionos-sdk

ionoscloud: Update ionos-cloud sdk-go and add metrics

Author: k8s-ci-robot

cluster-autoscaler/cloudprovider/ionoscloud/README.md

@@ -25,12 +25,18 @@ Store the token in a secret:
 kubectl create secret generic cloud-config --from-literal=token=MY_TOKEN
 ```
 
-Edit [`example-values.yaml`](./examples-values.yaml) and deploy using helm:
+Edit [`example-values.yaml`](./example-values.yaml) and deploy using helm:
 
 ```console
 helm install ionoscloud-cluster-autoscaler autoscaler/cluster-autoscaler -f example-values.yaml
 ```
 
+### Configuration
+
+The `IONOS_ADDITIONAL_HEADERS` environment variable can be used to configure the client to send additional headers on
+each Ionos Cloud API request, such as `X-Contract-Number`. This can be useful for users with multiple contracts.
+The format is a semicolon-separated list of key:value pairs, e.g. `IONOS_ADDITIONAL_HEADERS="X-Contract-Number:1234657890"`.
+
 ## Development
 
 The unit tests use mocks generated by [mockery](https://github.com/vektra/mockery/v2). To update them run:
@@ -40,6 +46,11 @@ mockery --inpackage --testonly --case snake --boilerplate-file ../../../hack/boi
 mockery --inpackage --testonly --case snake --boilerplate-file ../../../hack/boilerplate/boilerplate.generatego.txt --name IonosCloudManager
 ```
 
+### Debugging
+
+The global logging verbosity controlled by the `--v` flag is passed on to the Ionos Cloud SDK client logger.
+Verbosity 5 maps to `Debug` and 7 to `Trace`. **Do not enable this in production, as it will print full request and response data.**
+
 ### Build an image
 
 Build and push a docker image in the `cluster-autoscaler` directory:

cluster-autoscaler/cloudprovider/ionoscloud/cache.go

@@ -17,29 +17,20 @@ limitations under the License.
 package ionoscloud
 
 import (
+	"maps"
 	"sync"
-	"time"
 
 	"k8s.io/autoscaler/cluster-autoscaler/cloudprovider"
 	"k8s.io/klog/v2"
 )
 
-const nodeGroupCacheEntryTimeout = 2 * time.Minute
-
-type nodeGroupCacheEntry struct {
-	data cloudprovider.NodeGroup
-	ts   time.Time
-}
-
-var timeNow = time.Now
-
 // IonosCache caches resources to reduce API calls.
 // Cached state includes autoscaling limits, sizes and target sizes, a mapping of instances to node
 // groups, and a simple lock mechanism to prevent invalid node group writes.
 type IonosCache struct {
 	mutex sync.Mutex
 
-	nodeGroups           map[string]nodeGroupCacheEntry
+	nodeGroups           map[string]cloudprovider.NodeGroup
 	nodesToNodeGroups    map[string]string
 	nodeGroupSizes       map[string]int
 	nodeGroupTargetSizes map[string]int
@@ -49,11 +40,11 @@ type IonosCache struct {
 // NewIonosCache initializes a new IonosCache.
 func NewIonosCache() *IonosCache {
 	return &IonosCache{
-		nodeGroups:           map[string]nodeGroupCacheEntry{},
-		nodesToNodeGroups:    map[string]string{},
-		nodeGroupSizes:       map[string]int{},
-		nodeGroupTargetSizes: map[string]int{},
-		nodeGroupLockTable:   map[string]bool{},
+		nodeGroups:           make(map[string]cloudprovider.NodeGroup),
+		nodesToNodeGroups:    make(map[string]string),
+		nodeGroupSizes:       make(map[string]int),
+		nodeGroupTargetSizes: make(map[string]int),
+		nodeGroupLockTable:   make(map[string]bool),
 	}
 }
 
@@ -62,15 +53,7 @@ func (cache *IonosCache) AddNodeGroup(newPool cloudprovider.NodeGroup) {
 	cache.mutex.Lock()
 	defer cache.mutex.Unlock()
 
-	cache.nodeGroups[newPool.Id()] = nodeGroupCacheEntry{data: newPool}
-}
-
-func (cache *IonosCache) removeNodesForNodeGroupNoLock(id string) {
-	for nodeId, nodeGroupId := range cache.nodesToNodeGroups {
-		if nodeGroupId == id {
-			delete(cache.nodesToNodeGroups, nodeId)
-		}
-	}
+	cache.nodeGroups[newPool.Id()] = newPool
 }
 
 // RemoveInstanceFromCache deletes an instance and its respective mapping to the node group from
@@ -79,38 +62,39 @@ func (cache *IonosCache) RemoveInstanceFromCache(id string) {
 	cache.mutex.Lock()
 	defer cache.mutex.Unlock()
 
-	klog.V(5).Infof("Removed instance %s from cache", id)
-	nodeGroupId := cache.nodesToNodeGroups[id]
-	delete(cache.nodesToNodeGroups, id)
-	cache.updateNodeGroupTimestampNoLock(nodeGroupId)
+	if _, ok := cache.nodesToNodeGroups[id]; ok {
+		delete(cache.nodesToNodeGroups, id)
+		klog.V(5).Infof("Removed instance %s from cache", id)
+	}
 }
 
 // SetInstancesCacheForNodeGroup overwrites cached instances and mappings for a node group.
 func (cache *IonosCache) SetInstancesCacheForNodeGroup(id string, instances []cloudprovider.Instance) {
 	cache.mutex.Lock()
 	defer cache.mutex.Unlock()
 
-	cache.removeNodesForNodeGroupNoLock(id)
+	maps.DeleteFunc(cache.nodesToNodeGroups, func(_, nodeGroupID string) bool {
+		return nodeGroupID == id
+	})
 	cache.setInstancesCacheForNodeGroupNoLock(id, instances)
 }
 
 func (cache *IonosCache) setInstancesCacheForNodeGroupNoLock(id string, instances []cloudprovider.Instance) {
 	for _, instance := range instances {
-		nodeId := convertToNodeId(instance.Id)
-		cache.nodesToNodeGroups[nodeId] = id
+		nodeID := convertToNodeID(instance.Id)
+		cache.nodesToNodeGroups[nodeID] = id
 	}
-	cache.updateNodeGroupTimestampNoLock(id)
 }
 
-// GetNodeGroupIds returns an unsorted list of cached node group ids.
-func (cache *IonosCache) GetNodeGroupIds() []string {
+// GetNodeGroupIDs returns an unsorted list of cached node group ids.
+func (cache *IonosCache) GetNodeGroupIDs() []string {
 	cache.mutex.Lock()
 	defer cache.mutex.Unlock()
 
-	return cache.getNodeGroupIds()
+	return cache.getNodeGroupIDs()
 }
 
-func (cache *IonosCache) getNodeGroupIds() []string {
+func (cache *IonosCache) getNodeGroupIDs() []string {
 	ids := make([]string, 0, len(cache.nodeGroups))
 	for id := range cache.nodeGroups {
 		ids = append(ids, id)
@@ -125,26 +109,26 @@ func (cache *IonosCache) GetNodeGroups() []cloudprovider.NodeGroup {
 
 	nodeGroups := make([]cloudprovider.NodeGroup, 0, len(cache.nodeGroups))
 	for id := range cache.nodeGroups {
-		nodeGroups = append(nodeGroups, cache.nodeGroups[id].data)
+		nodeGroups = append(nodeGroups, cache.nodeGroups[id])
 	}
 	return nodeGroups
 }
 
 // GetNodeGroupForNode returns the node group for the given node.
 // Returns nil if either the mapping or the node group is not cached.
-func (cache *IonosCache) GetNodeGroupForNode(nodeId string) cloudprovider.NodeGroup {
+func (cache *IonosCache) GetNodeGroupForNode(nodeID string) cloudprovider.NodeGroup {
 	cache.mutex.Lock()
 	defer cache.mutex.Unlock()
 
-	id, found := cache.nodesToNodeGroups[nodeId]
+	nodeGroupID, found := cache.nodesToNodeGroups[nodeID]
 	if !found {
 		return nil
 	}
-	entry, found := cache.nodeGroups[id]
+	nodeGroup, found := cache.nodeGroups[nodeGroupID]
 	if !found {
 		return nil
 	}
-	return entry.data
+	return nodeGroup
 }
 
 // TryLockNodeGroup tries to write a node group lock entry.
@@ -219,19 +203,3 @@ func (cache *IonosCache) InvalidateNodeGroupTargetSize(id string) {
 
 	delete(cache.nodeGroupTargetSizes, id)
 }
-
-// NodeGroupNeedsRefresh returns true when the instances for the given node group have not been
-// updated for more than 2 minutes.
-func (cache *IonosCache) NodeGroupNeedsRefresh(id string) bool {
-	cache.mutex.Lock()
-	defer cache.mutex.Unlock()
-
-	return timeNow().Sub(cache.nodeGroups[id].ts) > nodeGroupCacheEntryTimeout
-}
-
-func (cache *IonosCache) updateNodeGroupTimestampNoLock(id string) {
-	if entry, ok := cache.nodeGroups[id]; ok {
-		entry.ts = timeNow()
-		cache.nodeGroups[id] = entry
-	}
-}

cluster-autoscaler/cloudprovider/ionoscloud/cache_test.go

@@ -18,16 +18,11 @@ package ionoscloud
 
 import (
 	"testing"
-	"time"
 
 	"github.com/stretchr/testify/require"
 	"k8s.io/autoscaler/cluster-autoscaler/cloudprovider"
 )
 
-func newCacheEntry(data cloudprovider.NodeGroup, ts time.Time) nodeGroupCacheEntry {
-	return nodeGroupCacheEntry{data: data, ts: ts}
-}
-
 func TestCache_AddNodeGroup(t *testing.T) {
 	cache := NewIonosCache()
 	require.Empty(t, cache.GetNodeGroups())
@@ -36,17 +31,14 @@ func TestCache_AddNodeGroup(t *testing.T) {
 }
 
 func TestCache_RemoveInstanceFromCache(t *testing.T) {
-	firstTime := timeNow().Add(-2*time.Minute - 1*time.Second)
 	cache := NewIonosCache()
-	cache.nodeGroups["2"] = newCacheEntry(&nodePool{id: "2"}, firstTime)
+	cache.nodeGroups["2"] = &nodePool{id: "2"}
 	cache.nodesToNodeGroups["b2"] = "2"
 
 	require.NotNil(t, cache.GetNodeGroupForNode("b2"))
-	require.True(t, cache.NodeGroupNeedsRefresh("2"))
 
 	cache.RemoveInstanceFromCache("b2")
 	require.Nil(t, cache.GetNodeGroupForNode("b2"))
-	require.False(t, cache.NodeGroupNeedsRefresh("2"))
 }
 
 func TestCache_SetInstancesCacheForNodeGroup(t *testing.T) {
@@ -65,11 +57,11 @@ func TestCache_SetInstancesCacheForNodeGroup(t *testing.T) {
 
 func TestCache_GetNodeGroupIDs(t *testing.T) {
 	cache := NewIonosCache()
-	require.Empty(t, cache.GetNodeGroupIds())
+	require.Empty(t, cache.GetNodeGroupIDs())
 	cache.AddNodeGroup(&nodePool{id: "1"})
-	require.Equal(t, []string{"1"}, cache.GetNodeGroupIds())
+	require.Equal(t, []string{"1"}, cache.GetNodeGroupIDs())
 	cache.AddNodeGroup(&nodePool{id: "2"})
-	require.ElementsMatch(t, []string{"1", "2"}, cache.GetNodeGroupIds())
+	require.ElementsMatch(t, []string{"1", "2"}, cache.GetNodeGroupIDs())
 }
 
 func TestCache_GetNodeGroups(t *testing.T) {
@@ -139,22 +131,3 @@ func TestCache_GetSetNodeGroupTargetSize(t *testing.T) {
 	require.False(t, found)
 	require.Zero(t, size)
 }
-
-func TestCache_NodeGroupNeedsRefresh(t *testing.T) {
-	fixedTime := time.Now().Round(time.Second)
-	timeNow = func() time.Time { return fixedTime }
-	defer func() { timeNow = time.Now }()
-
-	cache := NewIonosCache()
-	require.True(t, cache.NodeGroupNeedsRefresh("test"))
-
-	cache.AddNodeGroup(&nodePool{id: "test"})
-	require.True(t, cache.NodeGroupNeedsRefresh("test"))
-	cache.SetInstancesCacheForNodeGroup("test", nil)
-	require.False(t, cache.NodeGroupNeedsRefresh("test"))
-
-	timeNow = func() time.Time { return fixedTime.Add(nodeGroupCacheEntryTimeout) }
-	require.False(t, cache.NodeGroupNeedsRefresh("test"))
-	timeNow = func() time.Time { return fixedTime.Add(nodeGroupCacheEntryTimeout + 1*time.Second) }
-	require.True(t, cache.NodeGroupNeedsRefresh("test"))
-}