zhangmeng916 commented on a change in pull request #1307: URL: https://github.com/apache/helix/pull/1307#discussion_r490534836
########## File path: helix-core/src/main/java/org/apache/helix/model/TrieClusterTopology.java ########## @@ -0,0 +1,244 @@ +package org.apache.helix.model; + +/* + * Licensed to the Apache Software Foundation (ASF) under one + * or more contributor license agreements. See the NOTICE file + * distributed with this work for additional information + * regarding copyright ownership. The ASF licenses this file + * to you under the Apache License, Version 2.0 (the + * "License"); you may not use this file except in compliance + * with the License. You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, + * software distributed under the License is distributed on an + * "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY + * KIND, either express or implied. See the License for the + * specific language governing permissions and limitations + * under the License. + */ + +import java.util.ArrayDeque; +import java.util.Arrays; +import java.util.Deque; +import java.util.HashMap; +import java.util.HashSet; +import java.util.LinkedHashMap; +import java.util.List; +import java.util.Map; +import java.util.Set; +import java.util.stream.Collectors; + +import org.apache.helix.HelixException; +import org.slf4j.Logger; +import org.slf4j.LoggerFactory; + + +/** + * This is a class that uses a trie data structure to represent cluster topology. Each node + * except the terminal node represents a certain domain in the topology, and an terminal node + * represents an instance in the cluster. + */ +public class TrieClusterTopology { + private static Logger logger = LoggerFactory.getLogger(TrieClusterTopology.class); + private static final String DELIMITER = "/"; + private static final String CONNECTOR = "_"; + + private final TrieNode _rootNode; + private final String[] _topologyKeys; + + public TrieClusterTopology(final List<String> liveNodes, + final Map<String, InstanceConfig> instanceConfigMap, ClusterConfig clusterConfig) { + if (instanceConfigMap == null || !instanceConfigMap.keySet().containsAll(liveNodes)) { + throw new HelixException(String.format("Config for instances %s is not found!", + instanceConfigMap == null ? liveNodes : liveNodes.removeAll(instanceConfigMap.keySet()))); + } + // A list of all keys in cluster topology, e.g., a cluster topology defined as + // /group/zone/rack/instance will return ["group", "zone", "rack", "instance"]. + _topologyKeys = Arrays.asList(clusterConfig.getTopology().trim().split(DELIMITER)).stream() + .filter(str -> !str.isEmpty()).collect(Collectors.toList()).toArray(new String[0]); + _rootNode = new TrieNode(new HashMap<>(), DELIMITER); + constructTrie(instanceConfigMap); + } + + /** + * Return the topology of a cluster as a map. The key of the map is the first level of + * domain, and the value is a set of string that represents the path to each end node in that + * domain. E.g., assume the topology is defined as /group/zone/rack/instance, the result may be { + * ["group_0": {"zone_0/rack_0/instance_0", "zone_1/rack_1/instance_1"}], ["group_1": {"zone_1 + * /rack_1/instance_1", "zone_1/rack_1/instance_2"}]} + */ + public Map<String, Set<String>> getClusterTopology() { + return getTopologyUnderDomain(new HashMap<>()); + } + + /** + * Return the topology under a certain domain as a map. The key of the returned map is the next + * level domain, and the value is a set of string that represents the path to each end node in + * that domain. + * @param domain A map defining the domain name and its value, e.g. {["group": "1"], ["zone", + * "2"]} + * @return the topology under the given domain, e.g. {["rack_0": {"instance_0", "instance_1"}, + * ["rack_1": {"instance_2", "instance_3"}]} + */ + public Map<String, Set<String>> getTopologyUnderDomain(Map<String, String> domain) { + LinkedHashMap<String, String> orderedDomain = validateAndOrderDomain(domain); + TrieNode startNode = getStartNode(orderedDomain); + Map<String, TrieNode> children = startNode.getChildren(); + Map<String, Set<String>> results = new HashMap<>(); + children.entrySet().forEach(child -> { + String key = child.getKey(); + results.put(key, + truncatePath(getPathUnderNode(child.getValue()), child.getValue().getPath() + DELIMITER)); + }); + return results; + } + + /** + * Validate the domain provided has continuous fields in cluster topology definition. If it + * has, order the domain based on cluster topology definition. E.g. if the cluster topology is + * /group/zone/rack/instance, and domain is provided as {["zone": "1"], ["group", "2"]} will be + * reordered in a LinkedinHashMap as {["group", "2"], ["zone": "1"]} + */ + private LinkedHashMap<String, String> validateAndOrderDomain(Map<String, String> domain) { + LinkedHashMap<String, String> orderedDomain = new LinkedHashMap<>(); + if (domain == null) { + throw new IllegalArgumentException("The domain should not be null"); + } + for (int i = 0; i < domain.size(); i++) { + if (!domain.containsKey(_topologyKeys[i])) { + throw new IllegalArgumentException(String + .format("The input domain is not valid, the key %s is required", _topologyKeys[i])); + } else { + orderedDomain.put(_topologyKeys[i], domain.get(_topologyKeys[i])); + } + } + return orderedDomain; + } + + /** + * Truncate each path in the given set and only retain path starting from current node's + * children to each end node. + * @param toRemovePath The path from root to current node. It should be removed so that users + * can get a better view. + */ + private Set<String> truncatePath(Set<String> paths, String toRemovePath) { + Set<String> results = new HashSet<>(); + paths.forEach(path -> { + String truncatedPath = path.replace(toRemovePath, ""); + results.add(truncatedPath); + }); + return results; + } + + /** + * Return all the paths from a TrieNode as a set. + * @param node the node from where to collect all the nodes' paths. + * @return All the paths under the node. + */ + private Set<String> getPathUnderNode(TrieNode node) { + Set<String> resultMap = new HashSet<>(); + Deque<TrieNode> nodeStack = new ArrayDeque<>(); + nodeStack.push(node); + while (!nodeStack.isEmpty()) { + node = nodeStack.pop(); + if (node.getChildren().isEmpty()) { + resultMap.add(node.getPath()); + } else { + for (TrieNode child : node.getChildren().values()) { + nodeStack.push(child); + } + } + } + return resultMap; + } + + private TrieNode getStartNode(LinkedHashMap<String, String> domain) { + TrieNode curNode = _rootNode; + TrieNode nextNode; + for (Map.Entry<String, String> entry : domain.entrySet()) { + nextNode = curNode.getChildren().get(entry.getKey() + CONNECTOR + entry.getValue()); + if (nextNode == null) { + throw new IllegalArgumentException(String + .format("The input domain %s does not have the value %s", entry.getKey(), + entry.getValue())); + } + curNode = nextNode; + } + return curNode; + } + + private void removeInvalidInstanceConfig(Map<String, InstanceConfig> instanceConfigMap) { + Set<String> toRemoveConfig = new HashSet<>(); + for (String instanceName : instanceConfigMap.keySet()) { + Map<String, String> domainAsMap = instanceConfigMap.get(instanceName).getDomainAsMap(); + if (domainAsMap.isEmpty()) { + logger.info(String.format("Domain for instance %s is not set", instanceName)); + toRemoveConfig.add(instanceName); + } else { + for (String key : _topologyKeys) { + String value = domainAsMap.get(key); + if (value == null || value.length() == 0) { + logger.info(String.format("Domain %s for instance %s is not set", + domainAsMap.get(key), instanceName)); + toRemoveConfig.add(instanceName); + break; + } + } + } + } + toRemoveConfig.forEach(entry -> instanceConfigMap.remove(entry)); + } + + /** + * Constructs a trie based on the provided instance config map. It loops through all instance + * configs and constructs the trie in a top down manner. + */ + private void constructTrie(Map<String, InstanceConfig> instanceConfigMap) { + removeInvalidInstanceConfig(instanceConfigMap); + Map<String, Map<String, String>> instanceDomainsMap = new HashMap<>(); + instanceConfigMap.entrySet().forEach( + entry -> instanceDomainsMap.put(entry.getKey(), entry.getValue().getDomainAsMap())); + + for (Map.Entry<String, Map<String, String>> entry : instanceDomainsMap.entrySet()) { + TrieNode curNode = _rootNode; + String path = ""; + for (int i = 0; i < _topologyKeys.length; i++) { + String key = _topologyKeys[i] + CONNECTOR + entry.getValue().get(_topologyKeys[i]); Review comment: There're some offline discussion, and the conclusion is that trie is a good fit for cluster topology, as also mentioned here: https://github.com/apache/helix/issues/447 and should apply to both rebalance and user facing APIs, meaning that we only need one class for both uses. In this PR, we will create a generic trie structure with validation and construction logic. Topology APIs will use this structure and do whatever traverse needed. Laster, `topology` logic will also use this trie structure and implement the functions it need. The new trie structure is not fundamentally different from the tree in `topology` now, but with a clearer organization/validation logic and present in a more generic way. `topology` should be able to use it, if something mismatch, we can always come back to adjust the structure. ---------------------------------------------------------------- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: [email protected] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
