com.clickhouse.client.ClickHouseNodes Maven / Gradle / Ivy
Go to download
Show more of this group Show more artifacts with this name
Show all versions of clickhouse-client Show documentation
Show all versions of clickhouse-client Show documentation
Unified Java client for ClickHouse
package com.clickhouse.client;
import java.io.IOException;
import java.util.ArrayList;
import java.util.Collection;
import java.util.Collections;
import java.util.LinkedHashSet;
import java.util.LinkedList;
import java.util.List;
import java.util.Map;
import java.util.Optional;
import java.util.Set;
import java.util.SortedMap;
import java.util.TreeMap;
import java.util.WeakHashMap;
import java.util.Map.Entry;
import java.util.concurrent.ScheduledFuture;
import java.util.concurrent.atomic.AtomicBoolean;
import java.util.concurrent.atomic.AtomicInteger;
import java.util.concurrent.atomic.AtomicReference;
import java.util.concurrent.locks.ReentrantReadWriteLock;
import com.clickhouse.client.ClickHouseNode.Status;
import com.clickhouse.client.config.ClickHouseClientOption;
import com.clickhouse.client.config.ClickHouseDefaults;
import com.clickhouse.data.ClickHouseChecker;
import com.clickhouse.data.ClickHouseFormat;
import com.clickhouse.data.ClickHouseRecord;
import com.clickhouse.data.ClickHouseValues;
import com.clickhouse.logging.Logger;
import com.clickhouse.logging.LoggerFactory;
/**
* A generic node manager for managing one or more nodes which may or may not
* belong to same cluster. It maintains two lists - one for healthy nodes and
* the other for faulty ones. Behind the scene, there's background thread(s) for
* discovering nodes and health check. Besides,
* {@link ClickHouseLoadBalancingPolicy} is used to pickup available node and
* moving node between lists according to its status.
*/
public class ClickHouseNodes implements ClickHouseNodeManager {
private static final Logger log = LoggerFactory.getLogger(ClickHouseNodes.class);
private static final long serialVersionUID = 4931904980127690349L;
private static final Map cache = Collections.synchronizedMap(new WeakHashMap<>());
private static final char[] separators = new char[] { '/', '?', '#' };
/**
* Creates list of managed {@link ClickHouseNode} for load balancing and
* fail-over.
*
* @param endpoints non-empty URIs separated by comma
* @param defaultOptions default options
* @return non-null list of nodes
*/
static ClickHouseNodes create(String endpoints, Map defaultOptions) {
int index = endpoints.indexOf(ClickHouseNode.SCHEME_DELIMITER);
String defaultProtocol = ((ClickHouseProtocol) ClickHouseDefaults.PROTOCOL
.getEffectiveDefaultValue()).name();
if (index > 0) {
defaultProtocol = endpoints.substring(0, index);
if (ClickHouseProtocol.fromUriScheme(defaultProtocol) == ClickHouseProtocol.ANY) {
defaultProtocol = ClickHouseProtocol.ANY.name();
index = 0;
} else {
index += 3;
}
} else {
index = 0;
}
String defaultParams = "";
Set list = new LinkedHashSet<>();
char stopChar = ',';
for (int i = index, len = endpoints.length(); i < len; i++) {
char ch = endpoints.charAt(i);
if (ch == ',' || Character.isWhitespace(ch)) {
index++;
continue;
} else if (ch == '/' || ch == '?' || ch == '#') {
defaultParams = endpoints.substring(i);
break;
}
switch (ch) {
case '(':
stopChar = ')';
index++;
break;
case '{':
stopChar = '}';
index++;
break;
default:
break;
}
int endIndex = i;
for (int j = i + 1; j < len; j++) {
ch = endpoints.charAt(j);
if (ch == stopChar || Character.isWhitespace(ch)) {
endIndex = j;
break;
}
}
if (endIndex > i) {
list.add(endpoints.substring(index, endIndex).trim());
i = endIndex;
index = endIndex + 1;
stopChar = ',';
} else {
String last = endpoints.substring(index);
int sepIndex = last.indexOf(ClickHouseNode.SCHEME_DELIMITER);
int startIndex = sepIndex < 0 ? 0 : sepIndex + 3;
for (char spec : separators) {
sepIndex = last.indexOf(spec, startIndex);
if (sepIndex > 0) {
break;
}
}
if (sepIndex > 0) {
defaultParams = last.substring(sepIndex);
list.add(last.substring(0, sepIndex).trim());
} else {
list.add(last.trim());
}
break;
}
}
list.remove("");
if (list.isEmpty()) {
throw new IllegalArgumentException(
"No valid URI found, please try 'http://localhost:8123' or simply 'localhost:8123' if you don't know protocol");
}
if (list.size() == 1 && defaultParams.isEmpty()) {
endpoints = new StringBuilder().append(defaultProtocol).append(ClickHouseNode.SCHEME_DELIMITER)
.append(list.iterator().next()).toString();
return new ClickHouseNodes(Collections.singletonList(ClickHouseNode.of(endpoints, defaultOptions)));
}
ClickHouseNode defaultNode = ClickHouseNode.of(defaultProtocol + "://localhost" + defaultParams,
defaultOptions);
List nodes = new LinkedList<>();
for (String uri : list) {
nodes.add(ClickHouseNode.of(uri, defaultNode));
}
return new ClickHouseNodes(nodes, defaultNode);
}
/**
* Picks node from {@code source} and put into {@code target}.
*
* @param source list of nodes to pick
* @param selector node selector for filtering out nodes
* @param groupSize maximum number of nodes allowed
* @return non-null selected nodes
*/
static List pickNodes(Collection source, ClickHouseNodeSelector selector,
int groupSize) {
boolean hasSelector = selector != null && selector != ClickHouseNodeSelector.EMPTY;
final List list;
if (groupSize < 1) {
if (hasSelector) {
list = new LinkedList<>();
for (ClickHouseNode node : source) {
if (selector.match(node)) {
list.add(node);
}
}
} else {
list = new ArrayList<>(source);
}
} else {
list = new ArrayList<>(groupSize);
int count = 0;
for (ClickHouseNode node : source) {
if (!hasSelector || selector.match(node)) {
list.add(node);
count++;
}
if (count >= groupSize) {
break;
}
}
}
return list;
}
/**
* Picks node from {@code source} and put into {@code target}.
*
* @param source list of nodes to pick
* @param selector node selector for filtering out nodes
* @param target container for selected nodes
* @param groupSize maximum number of nodes allowed
* @param currentTime current timestamp for filtering
*/
static void pickNodes(Collection source, ClickHouseNodeSelector selector,
Set target, int groupSize, long currentTime) {
boolean hasSelector = selector != null && selector != ClickHouseNodeSelector.EMPTY;
int count = target.size();
for (ClickHouseNode node : source) {
if (!hasSelector || selector.match(node)) {
int interval = node.config.getNodeCheckInterval();
if (interval < 1 || (currentTime - node.lastUpdateTime.get()) >= interval) {
target.add(node);
count++;
}
}
if (groupSize > 0 && count >= groupSize) {
break;
}
}
}
/**
* Build unique key according to the given base URI and options for caching.
*
* @param uri non-null URI
* @param options options
* @return non-empty unique key for caching
*/
public static String buildCacheKey(String uri, Map options) {
if (uri == null) {
throw new IllegalArgumentException("Non-null URI required");
} else if ((uri = uri.trim()).isEmpty()) {
throw new IllegalArgumentException("Non-blank URI required");
}
if (options == null || options.isEmpty()) {
return uri;
}
SortedMap sorted;
if (options instanceof SortedMap) {
sorted = (SortedMap) options;
} else {
sorted = new TreeMap<>();
for (Entry entry : options.entrySet()) {
if (entry.getKey() != null) {
sorted.put(entry.getKey(), entry.getValue());
}
}
}
StringBuilder builder = new StringBuilder(uri).append('|');
for (Entry entry : sorted.entrySet()) {
if (entry.getKey() != null) {
builder.append(entry.getKey()).append('=').append(entry.getValue()).append(',');
}
}
return builder.toString();
}
/**
* Gets or creates list of managed {@link ClickHouseNode} for load balancing
* and fail-over.
*
* @param endpoints non-empty URIs separated by comma
* @return non-null list of nodes
*/
public static ClickHouseNodes of(String endpoints) {
return of(endpoints, Collections.emptyMap());
}
/**
* Gets or creates list of managed {@link ClickHouseNode} for load balancing
* and fail-over.
*
* @param endpoints non-empty URIs separated by comma
* @param options default options
* @return non-null list of nodes
*/
public static ClickHouseNodes of(String endpoints, Map options) {
return cache.computeIfAbsent(buildCacheKey(ClickHouseChecker.nonEmpty(endpoints, "Endpoints"), options),
k -> create(endpoints, options));
}
/**
* Gets or creates list of managed {@link ClickHouseNode} for load balancing
* and fail-over. Since the list will be cached in a {@link WeakHashMap}, as
* long as you hold strong reference to the {@code cacheKey}, same combination
* of {@code endpoints} and {@code options} will be always mapped to the exact
* same list.
*
* @param cacheKey non-empty cache key
* @param endpoints non-empty URIs separated by comma
* @param options default options
* @return non-null list of nodes
*/
public static ClickHouseNodes of(String cacheKey, String endpoints, Map options) {
// TODO discover endpoints from a URL or custom service, for examples:
// discover://(smb://fs1/ch-list.txt),(smb://fs1/ch-dc.json)
// discover:com.mycompany.integration.clickhouse.Endpoints
if (ClickHouseChecker.isNullOrEmpty(cacheKey) || ClickHouseChecker.isNullOrEmpty(endpoints)) {
throw new IllegalArgumentException("Non-empty cache key and endpoints are required");
}
return cache.computeIfAbsent(cacheKey, k -> create(endpoints, options));
}
/**
* Flag for exclusive health check.
*/
protected final AtomicBoolean checking;
/**
* Index for retrieving next node.
*/
protected final AtomicInteger index;
/**
* Lock for read and write {@code nodes} and {@code faultyNodes}.
*/
protected final ReentrantReadWriteLock lock;
/**
* Maximum number of nodes can be used for operation at a time.
*/
protected final int groupSize;
/**
* List of healthy nodes.
*/
protected final LinkedList nodes;
/**
* List of faulty nodes.
*/
protected final LinkedList faultyNodes;
/**
* Reference holding future of scheduled discovery.
*/
protected final AtomicReference> discoveryFuture;
/**
* Reference holding future of scheduled health check.
*/
protected final AtomicReference> healthCheckFuture;
/**
* Load balancing policy.
*/
protected final ClickHouseLoadBalancingPolicy policy;
/**
* Load balancing tags for filtering out nodes.
*/
protected final ClickHouseNodeSelector selector;
/**
* Flag indicating whether it's single node or not.
*/
protected final boolean singleNode;
/**
* Template node.
*/
protected final ClickHouseNode template;
/**
* Constructor for testing purpose.
*
* @param nodes non-empty list of nodes
*/
ClickHouseNodes(Collection nodes) {
this(nodes, nodes.iterator().next());
}
/**
* Default constructor.
*
* @param nodes non-empty list of nodes
* @param template non-null template node
*/
protected ClickHouseNodes(Collection nodes, ClickHouseNode template) {
this.checking = new AtomicBoolean(false);
this.index = new AtomicInteger(0);
this.lock = new ReentrantReadWriteLock();
this.nodes = new LinkedList<>(); // usually just healthy nodes
this.faultyNodes = new LinkedList<>();
this.discoveryFuture = new AtomicReference<>(null);
this.healthCheckFuture = new AtomicReference<>(null);
this.template = template;
this.groupSize = template.config.getIntOption(ClickHouseClientOption.NODE_GROUP_SIZE);
Set tags = new LinkedHashSet<>();
ClickHouseNode.parseTags(template.config.getStrOption(ClickHouseClientOption.LOAD_BALANCING_TAGS), tags);
this.policy = ClickHouseLoadBalancingPolicy
.of(template.config.getStrOption(ClickHouseClientOption.LOAD_BALANCING_POLICY));
this.selector = tags.isEmpty() ? ClickHouseNodeSelector.EMPTY : ClickHouseNodeSelector.of(null, tags);
boolean autoDiscovery = false;
for (ClickHouseNode n : nodes) {
autoDiscovery = autoDiscovery || n.config.isAutoDiscovery();
n.setManager(this);
}
if (autoDiscovery) {
this.singleNode = false;
this.discoveryFuture.getAndUpdate(current -> policy.schedule(current, ClickHouseNodes.this::discover,
template.config.getIntOption(ClickHouseClientOption.NODE_DISCOVERY_INTERVAL)));
} else {
this.singleNode = nodes.size() == 1;
}
this.healthCheckFuture.getAndUpdate(current -> policy.schedule(current, ClickHouseNodes.this::check,
template.config.getIntOption(ClickHouseClientOption.HEALTH_CHECK_INTERVAL)));
}
protected void queryClusterNodes(Collection seeds, Collection allNodes,
Collection newHealthyNodes, Collection newFaultyNodes,
Collection useless) {
for (ClickHouseNode node : seeds) {
ClickHouseNode server = null;
try {
server = node.probe();
} catch (Exception e) {
// ignore
}
if (server == null) {
newFaultyNodes.add(node);
continue;
} else if (!server.equals(node)) {
allNodes.add(server);
useless.add(node);
}
try (ClickHouseClient client = ClickHouseClient.builder().agent(false)
.nodeSelector(ClickHouseNodeSelector.of(server.getProtocol())).build()) {
ClickHouseRequest request = client.read(server)
.format(ClickHouseFormat.RowBinaryWithNamesAndTypes);
String clusterName = server.getCluster();
Set clusters = new LinkedHashSet<>();
if (!ClickHouseChecker.isNullOrEmpty(clusterName)) {
clusters.add(clusterName);
}
boolean isAddress = true;
boolean proceed = false;
try (ClickHouseResponse response = request
.query(clusters.isEmpty() ? policy.getQueryForAllLocalNodes()
: policy.getQueryForClusterLocalNodes())
.params(ClickHouseValues.convertToQuotedString(clusterName))
.executeAndWait()) {
for (ClickHouseRecord r : response.records()) {
int idx = 0;
String cluster = r.getValue(idx++).asString();
clusters.add(cluster);
String addr = r.getValue(idx++).asString();
String name = r.getValue(idx++).asString();
if (!(isAddress = server.getHost().equals(addr)) && !server.getHost().equals(name)) {
log.warn(
"Auto discovery may not work as no host_name and host_address in system.clusters matched with %s",
server);
isAddress = true; // fall back to host_address
}
ClickHouseNode discovered = ClickHouseNode.builder(server).cluster(cluster)
.replica(r.getValue(idx++).asInteger())
.shard(r.getValue(idx++).asInteger(), r.getValue(idx++).asInteger()).build();
if (!discovered.equals(server)) {
allNodes.remove(server);
allNodes.add(discovered);
newHealthyNodes.add(discovered);
useless.add(server);
}
proceed = true;
}
} catch (Exception e) {
log.warn("Failed to query system.clusters using %s, due to: %s", server, e.getMessage());
if (e.getCause() instanceof IOException) {
useless.add(server);
continue;
}
}
if (!proceed) {
continue;
}
String query = policy.getQueryForNonLocalNodes();
int limit = template.config.getIntOption(ClickHouseClientOption.NODE_DISCOVERY_LIMIT);
if (limit > 0) {
query = query + " limit " + limit;
}
try (ClickHouseResponse response = request.query(query)
.params(isAddress ? "host_address" : "host_name",
ClickHouseValues.convertToSqlExpression(clusters))
.executeAndWait()) {
for (ClickHouseRecord r : response.records()) {
int idx = 0;
ClickHouseNode n = ClickHouseNode.builder(server).cluster(r.getValue(idx++).asString())
.addOption(ClickHouseClientOption.AUTO_DISCOVERY.getKey(), "false")
.host(r.getValue(idx++).asString()).replica(r.getValue(idx++).asInteger())
.shard(r.getValue(idx++).asInteger(), r.getValue(idx++).asInteger()).build();
allNodes.add(n);
// we have seed nodes so here we can be more defensive
newFaultyNodes.add(n);
}
} catch (Exception e) {
log.warn("Failed to query system.clusters using %s", server, e);
newHealthyNodes.remove(server);
newFaultyNodes.add(server);
}
}
}
}
/**
* Gets next node available.
*
* @return non-null node
*/
protected ClickHouseNode get() {
return apply(selector);
}
/**
* Checks whether it's single node or not.
*
* @return true if it's single node; false otherwise
*/
public boolean isSingleNode() {
return singleNode;
}
@Override
public ClickHouseNode apply(ClickHouseNodeSelector t) {
lock.readLock().lock();
try {
return policy.get(this, t);
} finally {
lock.readLock().unlock();
}
}
@Override
public ClickHouseNode suggestNode(ClickHouseNode server, Throwable failure) {
lock.readLock().lock();
try {
return policy.suggest(this, server, failure);
} finally {
lock.readLock().unlock();
}
}
@Override
public void update(ClickHouseNode node, Status status) {
lock.writeLock().lock();
try {
if (node.config.getNodeCheckInterval() > 0) {
node.lastUpdateTime.set(System.currentTimeMillis());
}
policy.update(this, node, status);
} finally {
lock.writeLock().unlock();
}
}
/**
* Checks (faulty) node status.
*/
public void check() {
// exclusive access
if (!checking.compareAndSet(false, true)) {
return;
}
Set list = new LinkedHashSet<>();
long currentTime = System.currentTimeMillis();
boolean checkAll = template.config.getBoolOption(ClickHouseClientOption.CHECK_ALL_NODES);
int numberOfFaultyNodes = -1;
lock.readLock().lock();
// TODO:
// 1) minimize the list;
// 2) detect flaky node and check it again later in a less frequent way
try {
pickNodes(faultyNodes, selector, list, groupSize, currentTime);
numberOfFaultyNodes = list.size();
if (checkAll) {
pickNodes(nodes, selector, list, groupSize, currentTime);
}
} finally {
checking.set(false);
lock.readLock().unlock();
}
boolean hasFaultyNode = false;
int count = 0;
try {
for (ClickHouseNode node : list) {
ClickHouseNode n = node.probe();
// probe is faster than ping but it cannot tell if the server works or not
boolean isAlive = false;
try (ClickHouseClient client = ClickHouseClient.builder().agent(false).config(n.config)
.nodeSelector(ClickHouseNodeSelector.of(n.getProtocol())).build()) {
isAlive = client.ping(n, n.config.getConnectionTimeout());
} catch (Exception e) {
// ignore
}
if (!n.equals(node)) {
update(n, Status.MANAGED);
update(node, Status.STANDALONE);
}
boolean wasFaultyBefore = numberOfFaultyNodes == -1 || count < numberOfFaultyNodes;
if (isAlive) {
if (wasFaultyBefore) {
update(n, Status.HEALTHY);
}
} else {
hasFaultyNode = true;
if (!wasFaultyBefore) {
update(n, Status.FAULTY);
}
}
count++;
}
} catch (Exception e) {
log.warn("Unexpected error occurred when checking node status", e);
} finally {
if (checkAll || hasFaultyNode) {
scheduleHealthCheck();
}
}
}
/**
* Discovers nodes in the same cluster by querying against
* {@code system.clusters} table.
*/
public void discover() {
// FIXME too aggressive for a large cluster, NODE_GROUP_SIZE should be
// considered
Set allNodes = new LinkedHashSet<>();
Set seeds = new LinkedHashSet<>();
lock.readLock().lock();
try {
// discover nodes freely only when auto discovery is enabled
for (ClickHouseNode node : nodes) {
if (node.config.isAutoDiscovery()) {
seeds.add(node);
} else {
allNodes.add(node);
}
}
// seeds without protocol
for (ClickHouseNode node : faultyNodes) {
if (node.config.isAutoDiscovery()) {
seeds.add(node);
} else {
allNodes.add(node);
}
}
} finally {
lock.readLock().unlock();
}
if (seeds.isEmpty()) {
return;
}
Set newHealthyNodes = new LinkedHashSet<>();
Set newFaultyNodes = new LinkedHashSet<>();
Set useless = new LinkedHashSet<>();
queryClusterNodes(seeds, allNodes, newHealthyNodes, newFaultyNodes, useless);
lock.readLock().lock();
try {
// check if there's any new node or decommission as needed
for (ClickHouseNode n : nodes) {
if (!allNodes.remove(n)) {
useless.add(n);
}
newHealthyNodes.remove(n); // just in case
}
for (ClickHouseNode n : faultyNodes) {
if (!allNodes.remove(n)) {
useless.add(n);
}
newFaultyNodes.remove(n); // just in case
}
} finally {
lock.readLock().unlock();
}
boolean noUselessNode = useless.isEmpty();
if (allNodes.isEmpty() && noUselessNode) {
return;
}
for (ClickHouseNode n : allNodes) { // all new nodes
update(n, Status.MANAGED);
}
for (ClickHouseNode n : newHealthyNodes) {
update(n, Status.HEALTHY);
}
for (ClickHouseNode n : newFaultyNodes) {
update(n, Status.FAULTY);
}
for (ClickHouseNode n : useless) {
update(n, Status.STANDALONE);
}
if (!noUselessNode) {
scheduleHealthCheck();
}
}
public ClickHouseNode getTemplate() {
return template;
}
@Override
public final List getNodes() {
return getNodes(selector, groupSize);
}
@Override
public List getNodes(ClickHouseNodeSelector selector, int groupSize) {
lock.readLock().lock();
try {
return pickNodes(nodes, selector, groupSize);
} finally {
lock.readLock().unlock();
}
}
@Override
public final List getFaultyNodes() {
return getFaultyNodes(selector, groupSize);
}
@Override
public List getFaultyNodes(ClickHouseNodeSelector selector, int groupSize) {
lock.readLock().lock();
try {
return pickNodes(faultyNodes, selector, groupSize);
} finally {
lock.readLock().unlock();
}
}
@Override
public ClickHouseLoadBalancingPolicy getPolicy() {
return policy;
}
@Override
public ClickHouseNodeSelector getNodeSelector() {
return selector;
}
@Override
public Optional> scheduleDiscovery() {
return Optional.ofNullable(discoveryFuture.getAndUpdate(current -> {
return policy.schedule(current, ClickHouseNodes.this::discover, 0L);
}));
}
@Override
public Optional> scheduleHealthCheck() {
return Optional.ofNullable(healthCheckFuture.getAndUpdate(current -> {
return policy.schedule(current, ClickHouseNodes.this::check, 0L);
}));
}
@Override
public void shutdown() {
for (ScheduledFuture future : new ScheduledFuture[] {
discoveryFuture.get(), healthCheckFuture.get()
}) {
if (future != null && !future.isDone() && !future.isCancelled()) {
future.cancel(true);
}
}
}
@Override
public int hashCode() {
final int prime = 31;
int result = prime + checking.hashCode();
result = prime * result + index.hashCode();
result = prime * result + policy.hashCode();
result = prime * result + selector.hashCode();
result = prime * result + nodes.hashCode();
result = prime * result + faultyNodes.hashCode();
return result;
}
@Override
public boolean equals(Object obj) {
if (this == obj) {
return true;
}
if (obj == null || getClass() != obj.getClass()) {
return false;
}
ClickHouseNodes other = (ClickHouseNodes) obj;
// FIXME ignore order when comparing node lists
return policy.equals(other.policy) && selector.equals(other.selector) && nodes.equals(other.nodes)
&& faultyNodes.equals(other.faultyNodes);
}
@Override
public String toString() {
return new StringBuilder("ClickHouseNodes [checking=").append(checking.get()).append(", index=")
.append(index.get()).append(", lock=r").append(lock.getReadHoldCount()).append('w')
.append(lock.getWriteHoldCount()).append(", nodes=").append(nodes.size()).append(", faulty=")
.append(faultyNodes.size()).append(", policy=").append(policy.getClass().getSimpleName())
.append(", tags=").append(selector.getPreferredTags()).append("]@").append(hashCode()).toString();
}
}