feat: More optimal IterateHierarchyV2 and iterateChildrenV2 [#600] (#601

) * chore: More optimal IterateHierarchyV2 and iterateChildrenV2 [#600] Closes #600 The existing (effectively v1) implementations are suboptimal since they don't construct a graph before the iteration. They search for children by looking at all namespace resources and checking `isParentOf`, which can give `O(tree_size * namespace_resources_count)` time complexity. The v2 algorithms construct the graph and have `O(namespace_resources_count)` time complexity. See more details in the linked issues. Signed-off-by: Andrii Korotkov <[email protected]> * improvements to graph building Signed-off-by: Michael Crenshaw <[email protected]> * use old name Signed-off-by: Michael Crenshaw <[email protected]> * chore: More optimal IterateHierarchyV2 and iterateChildrenV2 [#600] Closes #600 The existing (effectively v1) implementations are suboptimal since they don't construct a graph before the iteration. They search for children by looking at all namespace resources and checking `isParentOf`, which can give `O(tree_size * namespace_resources_count)` time complexity. The v2 algorithms construct the graph and have `O(namespace_resources_count)` time complexity. See more details in the linked issues. Signed-off-by: Andrii Korotkov <[email protected]> * finish merge Signed-off-by: Michael Crenshaw <[email protected]> * chore: More optimal IterateHierarchyV2 and iterateChildrenV2 [#600] Closes #600 The existing (effectively v1) implementations are suboptimal since they don't construct a graph before the iteration. They search for children by looking at all namespace resources and checking `isParentOf`, which can give `O(tree_size * namespace_resources_count)` time complexity. The v2 algorithms construct the graph and have `O(namespace_resources_count)` time complexity. See more details in the linked issues. Signed-off-by: Andrii Korotkov <[email protected]> * discard unneeded copies of child resources as we go Signed-off-by: Michael Crenshaw <[email protected]> * remove unnecessary comment Signed-off-by: Michael Crenshaw <[email protected]> * make childrenByUID sparse Signed-off-by: Michael Crenshaw <[email protected]> * eliminate duplicate map Signed-off-by: Michael Crenshaw <[email protected]> * fix comment Signed-off-by: Michael Crenshaw <[email protected]> * add useful comment back Signed-off-by: Michael Crenshaw <[email protected]> * use nsNodes instead of dupe map Signed-off-by: Michael Crenshaw <[email protected]> * remove unused struct Signed-off-by: Michael Crenshaw <[email protected]> * skip invalid APIVersion Signed-off-by: Michael Crenshaw <[email protected]> --------- Signed-off-by: Andrii Korotkov <[email protected]> Signed-off-by: Michael Crenshaw <[email protected]> Co-authored-by: Michael Crenshaw <[email protected]>
argoproj · Jul 18, 2024 · 6b2984e · 6b2984e
1 parent 7d150d0
commit 6b2984e
Show file tree

Hide file tree

Showing 5 changed files with 405 additions and 34 deletions.
diff --git a/pkg/cache/cluster.go b/pkg/cache/cluster.go
@@ -122,6 +122,9 @@ type ClusterCache interface {
  // IterateHierarchy iterates resource tree starting from the specified top level resource and executes callback for each resource in the tree.
  // The action callback returns true if iteration should continue and false otherwise.
  IterateHierarchy(key kube.ResourceKey, action func(resource *Resource, namespaceResources map[kube.ResourceKey]*Resource) bool)
+ // IterateHierarchyV2 iterates resource tree starting from the specified top level resources and executes callback for each resource in the tree.
+ // The action callback returns true if iteration should continue and false otherwise.
+ IterateHierarchyV2(keys []kube.ResourceKey, action func(resource *Resource, namespaceResources map[kube.ResourceKey]*Resource) bool)
  // IsNamespaced answers if specified group/kind is a namespaced resource API or not
  IsNamespaced(gk schema.GroupKind) (bool, error)
  // GetManagedLiveObjs helps finding matching live K8S resources for a given resources list.
@@ -1023,6 +1026,107 @@ func (c *clusterCache) IterateHierarchy(key kube.ResourceKey, action func(resour
  }
 }
 
+// IterateHierarchy iterates resource tree starting from the specified top level resources and executes callback for each resource in the tree
+func (c *clusterCache) IterateHierarchyV2(keys []kube.ResourceKey, action func(resource *Resource, namespaceResources map[kube.ResourceKey]*Resource) bool) {
+ c.lock.RLock()
+ defer c.lock.RUnlock()
+ keysPerNamespace := make(map[string][]kube.ResourceKey)
+ for _, key := range keys {
+ _, ok := c.resources[key]
+ if !ok {
+ continue
+ }
+ keysPerNamespace[key.Namespace] = append(keysPerNamespace[key.Namespace], key)
+ }
+ for namespace, namespaceKeys := range keysPerNamespace {
+ nsNodes := c.nsIndex[namespace]
+ graph := buildGraph(nsNodes)
+ visited := make(map[kube.ResourceKey]int)
+ for _, key := range namespaceKeys {
+ visited[key] = 0
+ }
+ for _, key := range namespaceKeys {
+ // The check for existence of key is done above.
+ res := c.resources[key]
+ if visited[key] == 2 || !action(res, nsNodes) {
+ continue
+ }
+ visited[key] = 1
+ if _, ok := graph[key]; ok {
+ for _, child := range graph[key] {
+ if visited[child.ResourceKey()] == 0 && action(child, nsNodes) {
+ child.iterateChildrenV2(graph, nsNodes, visited, func(err error, child *Resource, namespaceResources map[kube.ResourceKey]*Resource) bool {
+ if err != nil {
+ c.log.V(2).Info(err.Error())
+ return false
+ }
+ return action(child, namespaceResources)
+ })
+ }
+ }
+ }
+ visited[key] = 2
+ }
+ }
+}
+
+func buildGraph(nsNodes map[kube.ResourceKey]*Resource) map[kube.ResourceKey]map[types.UID]*Resource {
+ // Prepare to construct a graph
+ nodesByUID := make(map[types.UID][]*Resource, len(nsNodes))
+ for _, node := range nsNodes {
+ nodesByUID[node.Ref.UID] = append(nodesByUID[node.Ref.UID], node)
+ }
+
+ // In graph, they key is the parent and the value is a list of children.
+ graph := make(map[kube.ResourceKey]map[types.UID]*Resource)
+
+ // Loop through all nodes, calling each one "childNode," because we're only bothering with it if it has a parent.
+ for _, childNode := range nsNodes {
+ for i, ownerRef := range childNode.OwnerRefs {
+ // First, backfill UID of inferred owner child references.
+ if ownerRef.UID == "" {
+ group, err := schema.ParseGroupVersion(ownerRef.APIVersion)
+ if err != nil {
+ // APIVersion is invalid, so we couldn't find the parent.
+ continue
+ }
+ graphKeyNode, ok := nsNodes[kube.ResourceKey{Group: group.Group, Kind: ownerRef.Kind, Namespace: childNode.Ref.Namespace, Name: ownerRef.Name}]
+ if ok {
+ ownerRef.UID = graphKeyNode.Ref.UID
+ childNode.OwnerRefs[i] = ownerRef
+ } else {
+ // No resource found with the given graph key, so move on.
+ continue
+ }
+ }
+
+ // Now that we have the UID of the parent, update the graph.
+ uidNodes, ok := nodesByUID[ownerRef.UID]
+ if ok {
+ for _, uidNode := range uidNodes {
+ // Update the graph for this owner to include the child.
+ if _, ok := graph[uidNode.ResourceKey()]; !ok {
+ graph[uidNode.ResourceKey()] = make(map[types.UID]*Resource)
+ }
+ r, ok := graph[uidNode.ResourceKey()][childNode.Ref.UID]
+ if !ok {
+ graph[uidNode.ResourceKey()][childNode.Ref.UID] = childNode
+ } else if r != nil {
+ // The object might have multiple children with the same UID (e.g. replicaset from apps and extensions group).
+ // It is ok to pick any object, but we need to make sure we pick the same child after every refresh.
+ key1 := r.ResourceKey()
+ key2 := childNode.ResourceKey()
+ if strings.Compare(key1.String(), key2.String()) > 0 {
+ graph[uidNode.ResourceKey()][childNode.Ref.UID] = childNode
+ }
+ }
+ }
+ }
+ }
+ }
+ return graph
+}
+
 // IsNamespaced answers if specified group/kind is a namespaced resource API or not
 func (c *clusterCache) IsNamespaced(gk schema.GroupKind) (bool, error) {
  if isNamespaced, ok := c.namespacedResources[gk]; ok {