/*
 * 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.
 */
package org.apache.solr.common.cloud;

import java.lang.invoke.MethodHandles;
import java.util.ArrayList;
import java.util.Arrays;
import java.util.Collection;
import java.util.Collections;
import java.util.EnumSet;
import java.util.HashMap;
import java.util.HashSet;
import java.util.LinkedHashMap;
import java.util.List;
import java.util.Map;
import java.util.Map.Entry;
import java.util.Objects;
import java.util.Set;
import java.util.SortedSet;
import java.util.TreeSet;
import java.util.concurrent.ConcurrentHashMap;
import java.util.concurrent.CountDownLatch;
import java.util.concurrent.ExecutorService;
import java.util.concurrent.Future;
import java.util.concurrent.RejectedExecutionException;
import java.util.concurrent.TimeUnit;
import java.util.concurrent.TimeoutException;
import java.util.concurrent.atomic.AtomicBoolean;
import java.util.concurrent.atomic.AtomicReference;
import java.util.function.Predicate;
import java.util.function.UnaryOperator;
import java.util.stream.Collectors;

import org.apache.solr.client.solrj.cloud.autoscaling.AutoScalingConfig;
import org.apache.solr.common.AlreadyClosedException;
import org.apache.solr.common.Callable;
import org.apache.solr.common.SolrCloseable;
import org.apache.solr.common.SolrException;
import org.apache.solr.common.SolrException.ErrorCode;
import org.apache.solr.common.params.AutoScalingParams;
import org.apache.solr.common.params.CollectionAdminParams;
import org.apache.solr.common.params.CoreAdminParams;
import org.apache.solr.common.util.ExecutorUtil;
import org.apache.solr.common.util.ObjectReleaseTracker;
import org.apache.solr.common.util.Pair;
import org.apache.solr.common.util.SolrNamedThreadFactory;
import org.apache.solr.common.util.Utils;
import org.apache.zookeeper.KeeperException;
import org.apache.zookeeper.KeeperException.NoNodeException;
import org.apache.zookeeper.WatchedEvent;
import org.apache.zookeeper.Watcher;
import org.apache.zookeeper.Watcher.Event.EventType;
import org.apache.zookeeper.data.Stat;
import org.slf4j.Logger;
import org.slf4j.LoggerFactory;

import static java.util.Collections.EMPTY_MAP;
import static java.util.Collections.emptyMap;
import static java.util.Collections.emptySet;
import static java.util.Collections.emptySortedSet;
import static org.apache.solr.common.util.Utils.fromJSON;

public class ZkStateReader implements SolrCloseable {
  public static final int STATE_UPDATE_DELAY = Integer.getInteger("solr.OverseerStateUpdateDelay", 2000);  // delay between cloud state updates
  private static final Logger log = LoggerFactory.getLogger(MethodHandles.lookup().lookupClass());

  public static final String BASE_URL_PROP = "base_url";
  public static final String NODE_NAME_PROP = "node_name";
  public static final String CORE_NODE_NAME_PROP = "core_node_name";
  public static final String ROLES_PROP = "roles";
  public static final String STATE_PROP = "state";
  // if this flag equals to false and the replica does not exist in cluster state, set state op become no op (default is true)
  public static final String FORCE_SET_STATE_PROP = "force_set_state";
  /**
   * SolrCore name.
   */
  public static final String CORE_NAME_PROP = "core";
  public static final String COLLECTION_PROP = "collection";
  public static final String ELECTION_NODE_PROP = "election_node";
  public static final String SHARD_ID_PROP = "shard";
  public static final String REPLICA_PROP = "replica";
  public static final String SHARD_RANGE_PROP = "shard_range";
  public static final String SHARD_STATE_PROP = "shard_state";
  public static final String SHARD_PARENT_PROP = "shard_parent";
  public static final String NUM_SHARDS_PROP = "numShards";
  public static final String LEADER_PROP = "leader";
  public static final String SHARED_STORAGE_PROP = "shared_storage";
  public static final String PROPERTY_PROP = "property";
  public static final String PROPERTY_PROP_PREFIX = "property.";
  public static final String PROPERTY_VALUE_PROP = "property.value";
  public static final String MAX_AT_ONCE_PROP = "maxAtOnce";
  public static final String MAX_WAIT_SECONDS_PROP = "maxWaitSeconds";
  public static final String STATE_TIMESTAMP_PROP = "stateTimestamp";
  public static final String COLLECTIONS_ZKNODE = "/collections";
  public static final String LIVE_NODES_ZKNODE = "/live_nodes";
  public static final String ALIASES = "/aliases.json";
  public static final String CLUSTER_STATE = "/clusterstate.json";
  public static final String CLUSTER_PROPS = "/clusterprops.json";
  public static final String COLLECTION_PROPS_ZKNODE = "collectionprops.json";
  public static final String REJOIN_AT_HEAD_PROP = "rejoinAtHead";
  public static final String SOLR_SECURITY_CONF_PATH = "/security.json";
  public static final String SOLR_AUTOSCALING_CONF_PATH = "/autoscaling.json";
  public static final String SOLR_AUTOSCALING_EVENTS_PATH = "/autoscaling/events";
  public static final String SOLR_AUTOSCALING_TRIGGER_STATE_PATH = "/autoscaling/triggerState";
  public static final String SOLR_AUTOSCALING_NODE_ADDED_PATH = "/autoscaling/nodeAdded";
  public static final String SOLR_AUTOSCALING_NODE_LOST_PATH = "/autoscaling/nodeLost";
  public static final String SOLR_PKGS_PATH = "/packages.json";

  public static final String DEFAULT_SHARD_PREFERENCES = "defaultShardPreferences";
  public static final String REPLICATION_FACTOR = "replicationFactor";
  public static final String MAX_SHARDS_PER_NODE = "maxShardsPerNode";
  public static final String AUTO_ADD_REPLICAS = "autoAddReplicas";
  public static final String MAX_CORES_PER_NODE = "maxCoresPerNode";
  public static final String PULL_REPLICAS = "pullReplicas";
  public static final String NRT_REPLICAS = "nrtReplicas";
  public static final String TLOG_REPLICAS = "tlogReplicas";
  public static final String READ_ONLY = "readOnly";

  public static final String ROLES = "/roles.json";

  public static final String CONFIGS_ZKNODE = "/configs";
  public final static String CONFIGNAME_PROP = "configName";

  public static final String LEGACY_CLOUD = "legacyCloud";
  public static final String SAMPLE_PERCENTAGE = "samplePercentage";

  /**
   * @deprecated use {@link org.apache.solr.common.params.CollectionAdminParams#DEFAULTS} instead.
   */
  @Deprecated
  public static final String COLLECTION_DEF = "collectionDefaults";

  public static final String URL_SCHEME = "urlScheme";

  private static final String SOLR_ENVIRONMENT = "environment";

  public static final String REPLICA_TYPE = "type";

  /**
   * A view of the current state of all collections; combines all the different state sources into a single view.
   */
  protected volatile ClusterState clusterState;

  private static final int GET_LEADER_RETRY_INTERVAL_MS = 50;
  private static final int GET_LEADER_RETRY_DEFAULT_TIMEOUT = Integer.parseInt(System.getProperty("zkReaderGetLeaderRetryTimeoutMs", "4000"));
  ;

  public static final String LEADER_ELECT_ZKNODE = "leader_elect";

  public static final String SHARD_LEADERS_ZKNODE = "leaders";
  public static final String ELECTION_NODE = "election";

  /**
   * Collections tracked in the legacy (shared) state format, reflects the contents of clusterstate.json.
   */
  private Map<String, ClusterState.CollectionRef> legacyCollectionStates = emptyMap();

  /**
   * Last seen ZK version of clusterstate.json.
   */
  private int legacyClusterStateVersion = 0;

  /**
   * Collections with format2 state.json, "interesting" and actively watched.
   */
  private final ConcurrentHashMap<String, DocCollection> watchedCollectionStates = new ConcurrentHashMap<>();

  /**
   * Collections with format2 state.json, not "interesting" and not actively watched.
   */
  private final ConcurrentHashMap<String, LazyCollectionRef> lazyCollectionStates = new ConcurrentHashMap<>();

  /**
   * Collection properties being actively watched
   */
  private final ConcurrentHashMap<String, VersionedCollectionProps> watchedCollectionProps = new ConcurrentHashMap<>();

  /**
   * Collection properties being actively watched
   */
  private final ConcurrentHashMap<String, PropsWatcher> collectionPropsWatchers = new ConcurrentHashMap<>();

  private volatile SortedSet<String> liveNodes = emptySortedSet();

  private volatile Map<String, Object> clusterProperties = Collections.emptyMap();

  private final ZkConfigManager configManager;

  private ConfigData securityData;

  private final Runnable securityNodeListener;

  private ConcurrentHashMap<String, CollectionWatch<DocCollectionWatcher>> collectionWatches = new ConcurrentHashMap<>();

  // named this observers so there's less confusion between CollectionPropsWatcher map and the PropsWatcher map.
  private ConcurrentHashMap<String, CollectionWatch<CollectionPropsWatcher>> collectionPropsObservers = new ConcurrentHashMap<>();

  private Set<CloudCollectionsListener> cloudCollectionsListeners = ConcurrentHashMap.newKeySet();

  private final ExecutorService notifications = ExecutorUtil.newMDCAwareCachedThreadPool("watches");

  private Set<LiveNodesListener> liveNodesListeners = ConcurrentHashMap.newKeySet();

  private Set<ClusterPropertiesListener> clusterPropertiesListeners = ConcurrentHashMap.newKeySet();

  /**
   * Used to submit notifications to Collection Properties watchers in order
   **/
  private final ExecutorService collectionPropsNotifications = ExecutorUtil.newMDCAwareSingleThreadExecutor(new SolrNamedThreadFactory("collectionPropsNotifications"));

  private static final long LAZY_CACHE_TIME = TimeUnit.NANOSECONDS.convert(STATE_UPDATE_DELAY, TimeUnit.MILLISECONDS);

  private Future<?> collectionPropsCacheCleaner; // only kept to identify if the cleaner has already been started.

  /**
   * Get current {@link AutoScalingConfig}.
   *
   * @return current configuration from <code>autoscaling.json</code>. NOTE:
   * this data is retrieved from ZK on each call.
   */
  public AutoScalingConfig getAutoScalingConfig() throws KeeperException, InterruptedException {
    return getAutoScalingConfig(null);
  }

  /**
   * Get current {@link AutoScalingConfig}.
   *
   * @param watcher optional {@link Watcher} to set on a znode to watch for config changes.
   * @return current configuration from <code>autoscaling.json</code>. NOTE:
   * this data is retrieved from ZK on each call.
   */
  public AutoScalingConfig getAutoScalingConfig(Watcher watcher) throws KeeperException, InterruptedException {
    Stat stat = new Stat();

    Map<String, Object> map = new HashMap<>();
    try {
      byte[] bytes = zkClient.getData(SOLR_AUTOSCALING_CONF_PATH, watcher, stat, true);
      if (bytes != null && bytes.length > 0) {
        map = (Map<String, Object>) fromJSON(bytes);
      }
    } catch (KeeperException.NoNodeException e) {
      // ignore
    }
    map.put(AutoScalingParams.ZK_VERSION, stat.getVersion());
    return new AutoScalingConfig(map);
  }

  private static class CollectionWatch<T> {

    int coreRefCount = 0;
    Set<T> stateWatchers = ConcurrentHashMap.newKeySet();

    public boolean canBeRemoved() {
      return coreRefCount + stateWatchers.size() == 0;
    }

  }

  public static final Set<String> KNOWN_CLUSTER_PROPS = Set.of(
      LEGACY_CLOUD,
      URL_SCHEME,
      AUTO_ADD_REPLICAS,
      CoreAdminParams.BACKUP_LOCATION,
      DEFAULT_SHARD_PREFERENCES,
      MAX_CORES_PER_NODE,
      SAMPLE_PERCENTAGE,
      SOLR_ENVIRONMENT,
      CollectionAdminParams.DEFAULTS);

  /**
   * Returns config set name for collection.
   * TODO move to DocCollection (state.json).
   *
   * @param collection to return config set name for
   */
  public String readConfigName(String collection) throws KeeperException {

    String configName = null;

    String path = COLLECTIONS_ZKNODE + "/" + collection;
    log.debug("Loading collection config from: [{}]", path);

    try {
      byte[] data = zkClient.getData(path, null, null, true);
      if (data == null) {
        log.warn("No config data found at path {}.", path);
        throw new KeeperException.NoNodeException("No config data found at path: " + path);
      }

      ZkNodeProps props = ZkNodeProps.load(data);
      configName = props.getStr(CONFIGNAME_PROP);

      if (configName == null) {
        log.warn("No config data found at path{}. ", path);
        throw new KeeperException.NoNodeException("No config data found at path: " + path);
      }
    } catch (InterruptedException e) {
      SolrZkClient.checkInterrupted(e);
      log.warn("Thread interrupted when loading config name for collection {}", collection);
      throw new SolrException(ErrorCode.SERVER_ERROR, "Thread interrupted when loading config name for collection " + collection, e);
    }

    return configName;
  }


  private final SolrZkClient zkClient;

  private final boolean closeClient;

  private volatile boolean closed = false;

  private Set<CountDownLatch> waitLatches = ConcurrentHashMap.newKeySet();

  public ZkStateReader(SolrZkClient zkClient) {
    this(zkClient, null);
  }

  public ZkStateReader(SolrZkClient zkClient, Runnable securityNodeListener) {
    this.zkClient = zkClient;
    this.configManager = new ZkConfigManager(zkClient);
    this.closeClient = false;
    this.securityNodeListener = securityNodeListener;
    assert ObjectReleaseTracker.track(this);
  }


  public ZkStateReader(String zkServerAddress, int zkClientTimeout, int zkClientConnectTimeout) {
    this.zkClient = new SolrZkClient(zkServerAddress, zkClientTimeout, zkClientConnectTimeout,
        // on reconnect, reload cloud info
        new OnReconnect() {
          @Override
          public void command() {
            try {
              ZkStateReader.this.createClusterStateWatchersAndUpdate();
            } catch (KeeperException e) {
              log.error("A ZK error has occurred", e);
              throw new ZooKeeperException(SolrException.ErrorCode.SERVER_ERROR, "A ZK error has occurred", e);
            } catch (InterruptedException e) {
              // Restore the interrupted status
              Thread.currentThread().interrupt();
              log.error("Interrupted", e);
              throw new ZooKeeperException(SolrException.ErrorCode.SERVER_ERROR, "Interrupted", e);
            }
          }
        });
    this.configManager = new ZkConfigManager(zkClient);
    this.closeClient = true;
    this.securityNodeListener = null;

    assert ObjectReleaseTracker.track(this);
  }

  public ZkConfigManager getConfigManager() {
    return configManager;
  }

  /**
   * Forcibly refresh cluster state from ZK. Do this only to avoid race conditions because it's expensive.
   * <p>
   * It is cheaper to call {@link #forceUpdateCollection(String)} on a single collection if you must.
   *
   * @lucene.internal
   */
  public void forciblyRefreshAllClusterStateSlow() throws KeeperException, InterruptedException {
    synchronized (getUpdateLock()) {
      if (clusterState == null) {
        // Never initialized, just run normal initialization.
        createClusterStateWatchersAndUpdate();
        return;
      }
      // No need to set watchers because we should already have watchers registered for everything.
      refreshCollectionList(null);
      refreshLiveNodes(null);
      refreshLegacyClusterState(null);
      // Need a copy so we don't delete from what we're iterating over.
      Collection<String> safeCopy = new ArrayList<>(watchedCollectionStates.keySet());
      Set<String> updatedCollections = new HashSet<>();
      for (String coll : safeCopy) {
        DocCollection newState = fetchCollectionState(coll, null);
        if (updateWatchedCollection(coll, newState)) {
          updatedCollections.add(coll);
        }
      }
      constructState(updatedCollections);
    }
  }

  /**
   * Forcibly refresh a collection's internal state from ZK. Try to avoid having to resort to this when
   * a better design is possible.
   */
  //TODO shouldn't we call ZooKeeper.sync() at the right places to prevent reading a stale value?  We do so for aliases.
  public void forceUpdateCollection(String collection) throws KeeperException, InterruptedException {

    synchronized (getUpdateLock()) {
      if (clusterState == null) {
        log.warn("ClusterState watchers have not been initialized");
        return;
      }

      ClusterState.CollectionRef ref = clusterState.getCollectionRef(collection);
      if (ref == null || legacyCollectionStates.containsKey(collection)) {
        // We either don't know anything about this collection (maybe it's new?) or it's legacy.
        // First update the legacy cluster state.
        log.debug("Checking legacy cluster state for collection {}", collection);
        refreshLegacyClusterState(null);
        if (!legacyCollectionStates.containsKey(collection)) {
          // No dice, see if a new collection just got created.
          LazyCollectionRef tryLazyCollection = new LazyCollectionRef(collection);
          if (tryLazyCollection.get() != null) {
            // What do you know, it exists!
            log.debug("Adding lazily-loaded reference for collection {}", collection);
            lazyCollectionStates.putIfAbsent(collection, tryLazyCollection);
            constructState(Collections.singleton(collection));
          }
        }
      } else if (ref.isLazilyLoaded()) {
        log.debug("Refreshing lazily-loaded state for collection {}", collection);
        if (ref.get() != null) {
          return;
        }
        // Edge case: if there's no external collection, try refreshing legacy cluster state in case it's there.
        refreshLegacyClusterState(null);
      } else if (watchedCollectionStates.containsKey(collection)) {
        // Exists as a watched collection, force a refresh.
        log.debug("Forcing refresh of watched collection state for {}", collection);
        DocCollection newState = fetchCollectionState(collection, null);
        if (updateWatchedCollection(collection, newState)) {
          constructState(Collections.singleton(collection));
        }
      } else {
        log.error("Collection {} is not lazy or watched!", collection);
      }
    }

  }

  /**
   * Refresh the set of live nodes.
   */
  public void updateLiveNodes() throws KeeperException, InterruptedException {
    refreshLiveNodes(null);
  }

  public Integer compareStateVersions(String coll, int version) {
    DocCollection collection = clusterState.getCollectionOrNull(coll);
    if (collection == null) return null;
    if (collection.getZNodeVersion() < version) {
      if (log.isDebugEnabled()) {
        log.debug("Server older than client {}<{}", collection.getZNodeVersion(), version);
      }
      DocCollection nu = getCollectionLive(this, coll);
      if (nu == null) return -1;
      if (nu.getZNodeVersion() > collection.getZNodeVersion()) {
        if (updateWatchedCollection(coll, nu)) {
          synchronized (getUpdateLock()) {
            constructState(Collections.singleton(coll));
          }
        }
        collection = nu;
      }
    }

    if (collection.getZNodeVersion() == version) {
      return null;
    }

    if (log.isDebugEnabled()) {
      log.debug("Wrong version from client [{}]!=[{}]", version, collection.getZNodeVersion());
    }

    return collection.getZNodeVersion();
  }

  public synchronized void createClusterStateWatchersAndUpdate() throws KeeperException,
      InterruptedException {
    // We need to fetch the current cluster state and the set of live nodes

    log.debug("Updating cluster state from ZooKeeper... ");

    // Sanity check ZK structure.
    if (!zkClient.exists(CLUSTER_STATE, true)) {
      throw new SolrException(ErrorCode.SERVICE_UNAVAILABLE,
          "Cannot connect to cluster at " + zkClient.getZkServerAddress() + ": cluster not found/not ready");
    }

    // on reconnect of SolrZkClient force refresh and re-add watches.
    loadClusterProperties();
    refreshLiveNodes(new LiveNodeWatcher());
    refreshLegacyClusterState(new LegacyClusterStateWatcher());
    refreshStateFormat2Collections();
    refreshCollectionList(new CollectionsChildWatcher());
    refreshAliases(aliasesManager);

    if (securityNodeListener != null) {
      addSecurityNodeWatcher(pair -> {
        ConfigData cd = new ConfigData();
        cd.data = pair.first() == null || pair.first().length == 0 ? EMPTY_MAP : Utils.getDeepCopy((Map) fromJSON(pair.first()), 4, false);
        cd.version = pair.second() == null ? -1 : pair.second().getVersion();
        securityData = cd;
        securityNodeListener.run();
      });
      securityData = getSecurityProps(true);
    }

    collectionPropsObservers.forEach((k, v) -> {
      collectionPropsWatchers.computeIfAbsent(k, PropsWatcher::new).refreshAndWatch(true);
    });
  }

  private void addSecurityNodeWatcher(final Callable<Pair<byte[], Stat>> callback)
      throws KeeperException, InterruptedException {
    zkClient.exists(SOLR_SECURITY_CONF_PATH,
        new Watcher() {

          @Override
          public void process(WatchedEvent event) {
            // session events are not change events, and do not remove the watcher
            if (EventType.None.equals(event.getType())) {
              return;
            }
            try {
              synchronized (ZkStateReader.this.getUpdateLock()) {
                log.debug("Updating [{}] ... ", SOLR_SECURITY_CONF_PATH);

                // remake watch
                final Watcher thisWatch = this;
                final Stat stat = new Stat();
                final byte[] data = getZkClient().getData(SOLR_SECURITY_CONF_PATH, thisWatch, stat, true);
                try {
                  callback.call(new Pair<>(data, stat));
                } catch (Exception e) {
                  log.error("Error running collections node listener", e);
                }
              }
            } catch (KeeperException.ConnectionLossException | KeeperException.SessionExpiredException e) {
              log.warn("ZooKeeper watch triggered, but Solr cannot talk to ZK: [{}]", e.getMessage());
            } catch (KeeperException e) {
              log.error("A ZK error has occurred", e);
              throw new ZooKeeperException(ErrorCode.SERVER_ERROR, "", e);
            } catch (InterruptedException e) {
              // Restore the interrupted status
              Thread.currentThread().interrupt();
              log.warn("Interrupted", e);
            }
          }

        }, true);
  }

  /**
   * Construct the total state view from all sources.
   * Must hold {@link #getUpdateLock()} before calling this.
   *
   * @param changedCollections collections that have changed since the last call,
   *                           and that should fire notifications
   */
  private void constructState(Set<String> changedCollections) {

    Set<String> liveNodes = this.liveNodes; // volatile read

    // Legacy clusterstate is authoritative, for backwards compatibility.
    // To move a collection's state to format2, first create the new state2 format node, then remove legacy entry.
    Map<String, ClusterState.CollectionRef> result = new LinkedHashMap<>(legacyCollectionStates);

    // Add state format2 collections, but don't override legacy collection states.
    for (Map.Entry<String, DocCollection> entry : watchedCollectionStates.entrySet()) {
      result.putIfAbsent(entry.getKey(), new ClusterState.CollectionRef(entry.getValue()));
    }

    // Finally, add any lazy collections that aren't already accounted for.
    for (Map.Entry<String, LazyCollectionRef> entry : lazyCollectionStates.entrySet()) {
      result.putIfAbsent(entry.getKey(), entry.getValue());
    }

    this.clusterState = new ClusterState(liveNodes, result, legacyClusterStateVersion);

    if (log.isDebugEnabled()) {
      log.debug("clusterStateSet: legacy [{}] interesting [{}] watched [{}] lazy [{}] total [{}]",
          legacyCollectionStates.keySet().size(),
          collectionWatches.keySet().size(),
          watchedCollectionStates.keySet().size(),
          lazyCollectionStates.keySet().size(),
          clusterState.getCollectionStates().size());
    }

    if (log.isTraceEnabled()) {
      log.trace("clusterStateSet: legacy [{}] interesting [{}] watched [{}] lazy [{}] total [{}]",
          legacyCollectionStates.keySet(),
          collectionWatches.keySet(),
          watchedCollectionStates.keySet(),
          lazyCollectionStates.keySet(),
          clusterState.getCollectionStates());
    }

    notifyCloudCollectionsListeners();

    for (String collection : changedCollections) {
      notifyStateWatchers(collection, clusterState.getCollectionOrNull(collection));
    }

  }

  /**
   * Refresh legacy (shared) clusterstate.json
   */
  private void refreshLegacyClusterState(Watcher watcher) throws KeeperException, InterruptedException {
    try {
      final Stat stat = new Stat();
      final byte[] data = zkClient.getData(CLUSTER_STATE, watcher, stat, true);
      final ClusterState loadedData = ClusterState.load(stat.getVersion(), data, emptySet(), CLUSTER_STATE);
      synchronized (getUpdateLock()) {
        if (this.legacyClusterStateVersion >= stat.getVersion()) {
          // Nothing to do, someone else updated same or newer.
          return;
        }
        Set<String> updatedCollections = new HashSet<>();
        for (String coll : this.collectionWatches.keySet()) {
          ClusterState.CollectionRef ref = this.legacyCollectionStates.get(coll);
          // legacy collections are always in-memory
          DocCollection oldState = ref == null ? null : ref.get();
          ClusterState.CollectionRef newRef = loadedData.getCollectionStates().get(coll);
          DocCollection newState = newRef == null ? null : newRef.get();
          if (newState == null) {
            // check that we haven't just migrated
            newState = watchedCollectionStates.get(coll);
          }
          if (!Objects.equals(oldState, newState)) {
            updatedCollections.add(coll);
          }
        }
        this.legacyCollectionStates = loadedData.getCollectionStates();
        this.legacyClusterStateVersion = stat.getVersion();
        constructState(updatedCollections);
      }
    } catch (KeeperException.NoNodeException e) {
      // Ignore missing legacy clusterstate.json.
      synchronized (getUpdateLock()) {
        this.legacyCollectionStates = emptyMap();
        this.legacyClusterStateVersion = 0;
        constructState(Collections.emptySet());
      }
    }
  }

  /**
   * Refresh state format2 collections.
   */
  private void refreshStateFormat2Collections() {
    for (String coll : collectionWatches.keySet()) {
      new StateWatcher(coll).refreshAndWatch();
    }
  }

  // We don't get a Stat or track versions on getChildren() calls, so force linearization.
  private final Object refreshCollectionListLock = new Object();

  /**
   * Search for any lazy-loadable state format2 collections.
   * <p>
   * A stateFormat=1 collection which is not interesting to us can also
   * be put into the {@link #lazyCollectionStates} map here. But that is okay
   * because {@link #constructState(Set)} will give priority to collections in the
   * shared collection state over this map.
   * In fact this is a clever way to avoid doing a ZK exists check on
   * the /collections/collection_name/state.json znode
   * Such an exists check is done in {@link ClusterState#hasCollection(String)} and
   * {@link ClusterState#getCollectionsMap()} methods
   * have a safeguard against exposing wrong collection names to the users
   */
  private void refreshCollectionList(Watcher watcher) throws KeeperException, InterruptedException {
    synchronized (refreshCollectionListLock) {
      List<String> children = null;
      try {
        children = zkClient.getChildren(COLLECTIONS_ZKNODE, watcher, true);
      } catch (KeeperException.NoNodeException e) {
        log.warn("Error fetching collection names: [{}]", e.getMessage());
        // fall through
      }
      if (children == null || children.isEmpty()) {
        lazyCollectionStates.clear();
        return;
      }

      // Don't lock getUpdateLock() here, we don't need it and it would cause deadlock.
      // Don't mess with watchedCollections, they should self-manage.

      // First, drop any children that disappeared.
      this.lazyCollectionStates.keySet().retainAll(children);
      for (String coll : children) {
        // We will create an eager collection for any interesting collections, so don't add to lazy.
        if (!collectionWatches.containsKey(coll)) {
          // Double check contains just to avoid allocating an object.
          LazyCollectionRef existing = lazyCollectionStates.get(coll);
          if (existing == null) {
            lazyCollectionStates.putIfAbsent(coll, new LazyCollectionRef(coll));
          }
        }
      }
    }
  }

  // We don't get a Stat or track versions on getChildren() calls, so force linearization.
  private final Object refreshCollectionsSetLock = new Object();
  // Ensures that only the latest getChildren fetch gets applied.
  private final AtomicReference<Set<String>> lastFetchedCollectionSet = new AtomicReference<>();

  /**
   * Register a CloudCollectionsListener to be called when the set of collections within a cloud changes.
   */
  public void registerCloudCollectionsListener(CloudCollectionsListener cloudCollectionsListener) {
    cloudCollectionsListeners.add(cloudCollectionsListener);
    notifyNewCloudCollectionsListener(cloudCollectionsListener);
  }

  /**
   * Remove a registered CloudCollectionsListener.
   */
  public void removeCloudCollectionsListener(CloudCollectionsListener cloudCollectionsListener) {
    cloudCollectionsListeners.remove(cloudCollectionsListener);
  }

  private void notifyNewCloudCollectionsListener(CloudCollectionsListener listener) {
    listener.onChange(Collections.emptySet(), lastFetchedCollectionSet.get());
  }

  private void notifyCloudCollectionsListeners() {
    notifyCloudCollectionsListeners(false);
  }

  private void notifyCloudCollectionsListeners(boolean notifyIfSame) {
    synchronized (refreshCollectionsSetLock) {
      final Set<String> newCollections = getCurrentCollections();
      final Set<String> oldCollections = lastFetchedCollectionSet.getAndSet(newCollections);
      if (!newCollections.equals(oldCollections) || notifyIfSame) {
        cloudCollectionsListeners.forEach(listener -> listener.onChange(oldCollections, newCollections));
      }
    }
  }

  private Set<String> getCurrentCollections() {
    Set<String> collections = new HashSet<>();
    collections.addAll(legacyCollectionStates.keySet());
    collections.addAll(watchedCollectionStates.keySet());
    collections.addAll(lazyCollectionStates.keySet());
    return collections;
  }

  private class LazyCollectionRef extends ClusterState.CollectionRef {
    private final String collName;
    private long lastUpdateTime;
    private DocCollection cachedDocCollection;

    public LazyCollectionRef(String collName) {
      super(null);
      this.collName = collName;
      this.lastUpdateTime = -1;
    }

    @Override
    public synchronized DocCollection get(boolean allowCached) {
      gets.incrementAndGet();
      if (!allowCached || lastUpdateTime < 0 || System.nanoTime() - lastUpdateTime > LAZY_CACHE_TIME) {
        boolean shouldFetch = true;
        if (cachedDocCollection != null) {
          Stat exists = null;
          try {
            exists = zkClient.exists(getCollectionPath(collName), null, true);
          } catch (Exception e) {
          }
          if (exists != null && exists.getVersion() == cachedDocCollection.getZNodeVersion()) {
            shouldFetch = false;
          }
        }
        if (shouldFetch) {
          cachedDocCollection = getCollectionLive(ZkStateReader.this, collName);
          lastUpdateTime = System.nanoTime();
        }
      }
      return cachedDocCollection;
    }

    @Override
    public boolean isLazilyLoaded() {
      return true;
    }

    @Override
    public String toString() {
      return "LazyCollectionRef(" + collName + ")";
    }
  }

  // We don't get a Stat or track versions on getChildren() calls, so force linearization.
  private final Object refreshLiveNodesLock = new Object();
  // Ensures that only the latest getChildren fetch gets applied.
  private final AtomicReference<SortedSet<String>> lastFetchedLiveNodes = new AtomicReference<>();

  /**
   * Refresh live_nodes.
   */
  private void refreshLiveNodes(Watcher watcher) throws KeeperException, InterruptedException {
    synchronized (refreshLiveNodesLock) {
      SortedSet<String> newLiveNodes;
      try {
        List<String> nodeList = zkClient.getChildren(LIVE_NODES_ZKNODE, watcher, true);
        newLiveNodes = new TreeSet<>(nodeList);
      } catch (KeeperException.NoNodeException e) {
        newLiveNodes = emptySortedSet();
      }
      lastFetchedLiveNodes.set(newLiveNodes);
    }

    // Can't lock getUpdateLock() until we release the other, it would cause deadlock.
    SortedSet<String> oldLiveNodes, newLiveNodes;
    synchronized (getUpdateLock()) {
      newLiveNodes = lastFetchedLiveNodes.getAndSet(null);
      if (newLiveNodes == null) {
        // Someone else won the race to apply the last update, just exit.
        return;
      }

      oldLiveNodes = this.liveNodes;
      this.liveNodes = newLiveNodes;
      if (clusterState != null) {
        clusterState.setLiveNodes(newLiveNodes);
      }
    }
    if (oldLiveNodes.size() != newLiveNodes.size()) {
      if (log.isInfoEnabled()) {
        log.info("Updated live nodes from ZooKeeper... ({}) -> ({})", oldLiveNodes.size(), newLiveNodes.size());
      }
    }
    if (log.isDebugEnabled()) {
      log.debug("Updated live nodes from ZooKeeper... {} -> {}", oldLiveNodes, newLiveNodes);
    }
    if (!oldLiveNodes.equals(newLiveNodes)) { // fire listeners
      liveNodesListeners.forEach(listener -> {
        if (listener.onChange(new TreeSet<>(oldLiveNodes), new TreeSet<>(newLiveNodes))) {
          removeLiveNodesListener(listener);
        }
      });
    }
  }

  public void registerClusterPropertiesListener(ClusterPropertiesListener listener) {
    // fire it once with current properties
    if (listener.onChange(getClusterProperties())) {
      removeClusterPropertiesListener(listener);
    } else {
      clusterPropertiesListeners.add(listener);
    }
  }

  public void removeClusterPropertiesListener(ClusterPropertiesListener listener) {
    clusterPropertiesListeners.remove(listener);
  }

  public void registerLiveNodesListener(LiveNodesListener listener) {
    // fire it once with current live nodes
    if (listener.onChange(new TreeSet<>(getClusterState().getLiveNodes()), new TreeSet<>(getClusterState().getLiveNodes()))) {
      removeLiveNodesListener(listener);
    }

    liveNodesListeners.add(listener);
  }

  public void removeLiveNodesListener(LiveNodesListener listener) {
    liveNodesListeners.remove(listener);
  }

  /**
   * @return information about the cluster from ZooKeeper
   */
  public ClusterState getClusterState() {
    return clusterState;
  }

  public Object getUpdateLock() {
    return this;
  }

  public void close() {
    this.closed = true;

    notifications.shutdownNow();

    waitLatches.parallelStream().forEach(c -> {
      c.countDown();
    });

    ExecutorUtil.shutdownAndAwaitTermination(notifications);
    ExecutorUtil.shutdownAndAwaitTermination(collectionPropsNotifications);
    if (closeClient) {
      zkClient.close();
    }
    assert ObjectReleaseTracker.release(this);
  }

  @Override
  public boolean isClosed() {
    return closed;
  }

  public String getLeaderUrl(String collection, String shard, int timeout) throws InterruptedException {
    ZkCoreNodeProps props = new ZkCoreNodeProps(getLeaderRetry(collection, shard, timeout));
    return props.getCoreUrl();
  }

  public Replica getLeader(Set<String> liveNodes, DocCollection docCollection, String shard) {
    Replica replica = docCollection != null ? docCollection.getLeader(shard) : null;
    if (replica != null && liveNodes.contains(replica.getNodeName())) {
      return replica;
    }
    return null;
  }

  public Replica getLeader(String collection, String shard) {
    if (clusterState != null) {
      DocCollection docCollection = clusterState.getCollectionOrNull(collection);
      Replica replica = docCollection != null ? docCollection.getLeader(shard) : null;
      if (replica != null && getClusterState().liveNodesContain(replica.getNodeName())) {
        return replica;
      }
    }
    return null;
  }

  public boolean isNodeLive(String node) {
    return liveNodes.contains(node);

  }

  /**
   * Get shard leader properties, with retry if none exist.
   */
  public Replica getLeaderRetry(String collection, String shard) throws InterruptedException {
    return getLeaderRetry(collection, shard, GET_LEADER_RETRY_DEFAULT_TIMEOUT);
  }

  /**
   * Get shard leader properties, with retry if none exist.
   */
  public Replica getLeaderRetry(String collection, String shard, int timeout) throws InterruptedException {

    AtomicReference<Replica> leader = new AtomicReference<>();
    try {
      waitForState(collection, timeout, TimeUnit.MILLISECONDS, (n, c) -> {
        if (c == null)
          return false;
        Replica l = getLeader(n, c, shard);
        if (l != null) {
          leader.set(l);
          return true;
        }
        return false;
      });
    } catch (TimeoutException e) {
      throw new SolrException(ErrorCode.SERVICE_UNAVAILABLE, "No registered leader was found after waiting for "
          + timeout + "ms " + ", collection: " + collection + " slice: " + shard + " saw state=" + clusterState.getCollectionOrNull(collection)
          + " with live_nodes=" + clusterState.getLiveNodes());
    }
    return leader.get();
  }

  /**
   * Get path where shard leader properties live in zookeeper.
   */
  public static String getShardLeadersPath(String collection, String shardId) {
    return COLLECTIONS_ZKNODE + "/" + collection + "/"
        + SHARD_LEADERS_ZKNODE + (shardId != null ? ("/" + shardId)
        : "") + "/leader";
  }

  /**
   * Get path where shard leader elections ephemeral nodes are.
   */
  public static String getShardLeadersElectPath(String collection, String shardId) {
    return COLLECTIONS_ZKNODE + "/" + collection + "/"
        + LEADER_ELECT_ZKNODE + (shardId != null ? ("/" + shardId + "/" + ELECTION_NODE)
        : "");
  }


  public List<ZkCoreNodeProps> getReplicaProps(String collection, String shardId, String thisCoreNodeName) {
    return getReplicaProps(collection, shardId, thisCoreNodeName, null);
  }

  public List<ZkCoreNodeProps> getReplicaProps(String collection, String shardId, String thisCoreNodeName,
                                               Replica.State mustMatchStateFilter) {
    return getReplicaProps(collection, shardId, thisCoreNodeName, mustMatchStateFilter, null);
  }

  public List<ZkCoreNodeProps> getReplicaProps(String collection, String shardId, String thisCoreNodeName,
                                               Replica.State mustMatchStateFilter, Replica.State mustNotMatchStateFilter) {
    //TODO: We don't need all these getReplicaProps method overloading. Also, it's odd that the default is to return replicas of type TLOG and NRT only
    return getReplicaProps(collection, shardId, thisCoreNodeName, mustMatchStateFilter, null, EnumSet.of(Replica.Type.TLOG, Replica.Type.NRT));
  }

  public List<ZkCoreNodeProps> getReplicaProps(String collection, String shardId, String thisCoreNodeName,
                                               Replica.State mustMatchStateFilter, Replica.State mustNotMatchStateFilter, final EnumSet<Replica.Type> acceptReplicaType) {
    assert thisCoreNodeName != null;
    ClusterState clusterState = this.clusterState;
    if (clusterState == null) {
      return null;
    }
    final DocCollection docCollection = clusterState.getCollectionOrNull(collection);
    if (docCollection == null || docCollection.getSlicesMap() == null) {
      throw new ZooKeeperException(ErrorCode.BAD_REQUEST,
          "Could not find collection in zk: " + collection);
    }

    Map<String, Slice> slices = docCollection.getSlicesMap();
    Slice replicas = slices.get(shardId);
    if (replicas == null) {
      throw new ZooKeeperException(ErrorCode.BAD_REQUEST, "Could not find shardId in zk: " + shardId);
    }

    Map<String, Replica> shardMap = replicas.getReplicasMap();
    List<ZkCoreNodeProps> nodes = new ArrayList<>(shardMap.size());
    for (Entry<String, Replica> entry : shardMap.entrySet().stream().filter((e) -> acceptReplicaType.contains(e.getValue().getType())).collect(Collectors.toList())) {
      ZkCoreNodeProps nodeProps = new ZkCoreNodeProps(entry.getValue());

      String coreNodeName = entry.getValue().getName();

      if (clusterState.liveNodesContain(nodeProps.getNodeName()) && !coreNodeName.equals(thisCoreNodeName)) {
        if (mustMatchStateFilter == null || mustMatchStateFilter == Replica.State.getState(nodeProps.getState())) {
          if (mustNotMatchStateFilter == null || mustNotMatchStateFilter != Replica.State.getState(nodeProps.getState())) {
            nodes.add(nodeProps);
          }
        }
      }
    }
    if (nodes.size() == 0) {
      // no replicas
      return null;
    }

    return nodes;
  }

  public SolrZkClient getZkClient() {
    return zkClient;
  }

  /**
   * Get a cluster property
   * <p>
   * N.B. Cluster properties are updated via ZK watchers, and so may not necessarily
   * be completely up-to-date.  If you need to get the latest version, then use a
   * {@link ClusterProperties} instance.
   *
   * @param key          the property to read
   * @param defaultValue a default value to use if no such property exists
   * @param <T>          the type of the property
   * @return the cluster property, or a default if the property is not set
   */
  @SuppressWarnings("unchecked")
  public <T> T getClusterProperty(String key, T defaultValue) {
    T value = (T) Utils.getObjectByPath(clusterProperties, false, key);
    if (value == null)
      return defaultValue;
    return value;
  }

  /**
   * Same as the above but allows a full json path as a list of parts
   *
   * @param keyPath      path to the property example ["collectionDefauls", "numShards"]
   * @param defaultValue a default value to use if no such property exists
   * @return the cluster property, or a default if the property is not set
   */
  public <T> T getClusterProperty(List<String> keyPath, T defaultValue) {
    T value = (T) Utils.getObjectByPath(clusterProperties, false, keyPath);
    if (value == null)
      return defaultValue;
    return value;
  }

  /**
   * Get all cluster properties for this cluster
   * <p>
   * N.B. Cluster properties are updated via ZK watchers, and so may not necessarily
   * be completely up-to-date.  If you need to get the latest version, then use a
   * {@link ClusterProperties} instance.
   *
   * @return a Map of cluster properties
   */
  public Map<String, Object> getClusterProperties() {
    return Collections.unmodifiableMap(clusterProperties);
  }

  private final Watcher clusterPropertiesWatcher = event -> {
    // session events are not change events, and do not remove the watcher
    if (Watcher.Event.EventType.None.equals(event.getType())) {
      return;
    }
    loadClusterProperties();
  };

  @SuppressWarnings("unchecked")
  private void loadClusterProperties() {
    try {
      while (true) {
        try {
          byte[] data = zkClient.getData(ZkStateReader.CLUSTER_PROPS, clusterPropertiesWatcher, new Stat(), true);
          this.clusterProperties = ClusterProperties.convertCollectionDefaultsToNestedFormat((Map<String, Object>) Utils.fromJSON(data));
          log.debug("Loaded cluster properties: {}", this.clusterProperties);

          for (ClusterPropertiesListener listener : clusterPropertiesListeners) {
            listener.onChange(getClusterProperties());
          }
          return;
        } catch (KeeperException.NoNodeException e) {
          this.clusterProperties = Collections.emptyMap();
          log.debug("Loaded empty cluster properties");
          // set an exists watch, and if the node has been created since the last call,
          // read the data again
          if (zkClient.exists(ZkStateReader.CLUSTER_PROPS, clusterPropertiesWatcher, true) == null)
            return;
        }
      }
    } catch (KeeperException | InterruptedException e) {
      log.error("Error reading cluster properties from zookeeper", SolrZkClient.checkInterrupted(e));
    }
  }

  /**
   * Get collection properties for a given collection. If the collection is watched, simply return it from the cache,
   * otherwise fetch it directly from zookeeper. This is a convenience for {@code getCollectionProperties(collection,0)}
   *
   * @param collection the collection for which properties are desired
   * @return a map representing the key/value properties for the collection.
   */
  public Map<String, String> getCollectionProperties(final String collection) {
    return getCollectionProperties(collection, 0);
  }

  /**
   * Get and cache collection properties for a given collection. If the collection is watched, or still cached
   * simply return it from the cache, otherwise fetch it directly from zookeeper and retain the value for at
   * least cacheForMillis milliseconds. Cached properties are watched in zookeeper and updated automatically.
   * This version of {@code getCollectionProperties} should be used when properties need to be consulted
   * frequently in the absence of an active {@link CollectionPropsWatcher}.
   *
   * @param collection     The collection for which properties are desired
   * @param cacheForMillis The minimum number of milliseconds to maintain a cache for the specified collection's
   *                       properties. Setting a {@code CollectionPropsWatcher} will override this value and retain
   *                       the cache for the life of the watcher. A lack of changes in zookeeper may allow the
   *                       caching to remain for a greater duration up to the cycle time of {@link CacheCleaner}.
   *                       Passing zero for this value will explicitly remove the cached copy if and only if it is
   *                       due to expire and no watch exists. Any positive value will extend the expiration time
   *                       if required.
   * @return a map representing the key/value properties for the collection.
   */
  public Map<String, String> getCollectionProperties(final String collection, long cacheForMillis) {
    synchronized (watchedCollectionProps) { // making decisions based on the result of a get...
      Watcher watcher = null;
      if (cacheForMillis > 0) {
        watcher = collectionPropsWatchers.compute(collection,
            (c, w) -> w == null ? new PropsWatcher(c, cacheForMillis) : w.renew(cacheForMillis));
      }
      VersionedCollectionProps vprops = watchedCollectionProps.get(collection);
      boolean haveUnexpiredProps = vprops != null && vprops.cacheUntilNs > System.nanoTime();
      long untilNs = System.nanoTime() + TimeUnit.NANOSECONDS.convert(cacheForMillis, TimeUnit.MILLISECONDS);
      Map<String, String> properties;
      if (haveUnexpiredProps) {
        properties = vprops.props;
        vprops.cacheUntilNs = Math.max(vprops.cacheUntilNs, untilNs);
      } else {
        try {
          VersionedCollectionProps vcp = fetchCollectionProperties(collection, watcher);
          properties = vcp.props;
          if (cacheForMillis > 0) {
            vcp.cacheUntilNs = untilNs;
            watchedCollectionProps.put(collection, vcp);
          } else {
            // we're synchronized on watchedCollectionProps and we can only get here if we have found an expired
            // vprops above, so it is safe to remove the cached value and let the GC free up some mem a bit sooner.
            if (!collectionPropsObservers.containsKey(collection)) {
              watchedCollectionProps.remove(collection);
            }
          }
        } catch (Exception e) {
          throw new SolrException(ErrorCode.SERVER_ERROR, "Error reading collection properties", SolrZkClient.checkInterrupted(e));
        }
      }
      return properties;
    }
  }

  private class VersionedCollectionProps {
    int zkVersion;
    Map<String, String> props;
    long cacheUntilNs = 0;

    VersionedCollectionProps(int zkVersion, Map<String, String> props) {
      this.zkVersion = zkVersion;
      this.props = props;
    }
  }

  static String getCollectionPropsPath(final String collection) {
    return COLLECTIONS_ZKNODE + '/' + collection + '/' + COLLECTION_PROPS_ZKNODE;
  }

  @SuppressWarnings("unchecked")
  private VersionedCollectionProps fetchCollectionProperties(String collection, Watcher watcher) throws KeeperException, InterruptedException {
    final String znodePath = getCollectionPropsPath(collection);
    // lazy init cache cleaner once we know someone is using collection properties.
    if (collectionPropsCacheCl…