All Downloads are FREE. Search and download functionalities are using the official Maven repository.

io.activej.fs.cluster.ClusterActiveFs Maven / Gradle / Ivy

Go to download

Provides tools for building efficient, scalable local, remote or clustered file servers. It utilizes ActiveJ CSP for fast and reliable file transfer.

There is a newer version: 6.0-beta2
Show newest version
/*
 * Copyright (C) 2020 ActiveJ LLC.
 *
 * Licensed 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 io.activej.fs.cluster;

import io.activej.async.function.AsyncBiFunction;
import io.activej.async.function.AsyncFunction;
import io.activej.async.process.AsyncCloseable;
import io.activej.async.service.EventloopService;
import io.activej.bytebuf.ByteBuf;
import io.activej.common.collection.Try;
import io.activej.common.function.FunctionEx;
import io.activej.common.function.SupplierEx;
import io.activej.common.initializer.WithInitializer;
import io.activej.common.ref.RefBoolean;
import io.activej.csp.ChannelConsumer;
import io.activej.csp.ChannelSupplier;
import io.activej.csp.dsl.ChannelConsumerTransformer;
import io.activej.eventloop.Eventloop;
import io.activej.eventloop.jmx.EventloopJmxBeanWithStats;
import io.activej.fs.ActiveFs;
import io.activej.fs.FileMetadata;
import io.activej.fs.exception.FsIOException;
import io.activej.jmx.api.attribute.JmxAttribute;
import io.activej.jmx.api.attribute.JmxOperation;
import io.activej.promise.Promise;
import io.activej.promise.Promises;
import io.activej.promise.jmx.PromiseStats;
import org.jetbrains.annotations.NotNull;
import org.jetbrains.annotations.Nullable;
import org.slf4j.Logger;
import org.slf4j.LoggerFactory;

import java.time.Duration;
import java.util.*;
import java.util.function.Consumer;
import java.util.stream.Stream;

import static io.activej.common.Checks.checkArgument;
import static io.activej.common.Utils.transformIterator;
import static io.activej.csp.dsl.ChannelConsumerTransformer.identity;
import static io.activej.fs.util.RemoteFsUtils.ofFixedSize;
import static io.activej.promise.Promises.first;
import static java.util.Collections.emptyMap;
import static java.util.stream.Collectors.joining;
import static java.util.stream.Collectors.toList;

/**
 * An implementation of {@link ActiveFs} which operates on other partitions as a cluster.
 * Contains some redundancy and fail-safety capabilities.
 * 

 * This implementation inherits the most strict limitations of all the file systems in cluster,
 * as well as defines several limitations over those specified in {@link ActiveFs} interface:
 * 
    
 *     
  • Uploaded files should be immutable
    • 
 *     
  • Deletion of files is not guaranteed
    • 
 *     
  • Based on previous limitation, moving a file also does not guarantees that source file will be deleted
    • 
 *     
  • Paths should not contain existing filenames as part of the path
    • 
 * 

 */
public final class ClusterActiveFs implements ActiveFs, WithInitializer, EventloopService, EventloopJmxBeanWithStats {
	private static final Logger logger = LoggerFactory.getLogger(ClusterActiveFs.class);

	private final FsPartitions partitions;

	/**
	 * Maximum allowed number of dead partitions, if there are more dead partitions than this number,
	 * the cluster is considered malformed.
	 */
	private int deadPartitionsThreshold = 0;

	/**
	 * Minimum number of uploads that have to succeed in order for upload to be considered successful.
	 */
	private int uploadTargetsMin = 1;

	/**
	 * Initial number of uploads that are initiated.
	 */
	private int uploadTargetsMax = 1;

	// region JMX
	private final PromiseStats uploadStartPromise = PromiseStats.create(Duration.ofMinutes(5));
	private final PromiseStats uploadFinishPromise = PromiseStats.create(Duration.ofMinutes(5));
	private final PromiseStats appendStartPromise = PromiseStats.create(Duration.ofMinutes(5));
	private final PromiseStats appendFinishPromise = PromiseStats.create(Duration.ofMinutes(5));
	private final PromiseStats downloadStartPromise = PromiseStats.create(Duration.ofMinutes(5));
	private final PromiseStats downloadFinishPromise = PromiseStats.create(Duration.ofMinutes(5));
	private final PromiseStats listPromise = PromiseStats.create(Duration.ofMinutes(5));
	private final PromiseStats infoPromise = PromiseStats.create(Duration.ofMinutes(5));
	private final PromiseStats infoAllPromise = PromiseStats.create(Duration.ofMinutes(5));
	private final PromiseStats copyPromise = PromiseStats.create(Duration.ofMinutes(5));
	private final PromiseStats copyAllPromise = PromiseStats.create(Duration.ofMinutes(5));
	private final PromiseStats movePromise = PromiseStats.create(Duration.ofMinutes(5));
	private final PromiseStats moveAllPromise = PromiseStats.create(Duration.ofMinutes(5));
	private final PromiseStats deletePromise = PromiseStats.create(Duration.ofMinutes(5));
	private final PromiseStats deleteAllPromise = PromiseStats.create(Duration.ofMinutes(5));
	// endregion

	// region creators
	private ClusterActiveFs(FsPartitions partitions) {
		this.partitions = partitions;
	}

	public static ClusterActiveFs create(FsPartitions partitions) {
		return new ClusterActiveFs(partitions);
	}

	/**
	 * Sets the replication count that determines how many copies of the file should persist over the cluster.
	 */
	public ClusterActiveFs withReplicationCount(int replicationCount) {
		checkArgument(1 <= replicationCount, "Replication count cannot be less than one");
		this.deadPartitionsThreshold = replicationCount - 1;
		this.uploadTargetsMin = replicationCount;
		this.uploadTargetsMax = replicationCount;
		return this;
	}

	/**
	 * Sets the replication count as well as number of upload targets that determines the number of server where file will be uploaded.
	 */
	@SuppressWarnings("UnusedReturnValue")
	public ClusterActiveFs withPersistenceOptions(int deadPartitionsThreshold, int uploadTargetsMin, int uploadTargetsMax) {
		checkArgument(0 <= deadPartitionsThreshold,
				"Dead partitions threshold cannot be less than zero");
		checkArgument(0 <= uploadTargetsMin,
				"Minimum number of upload targets should not be less than zero");
		checkArgument(0 < uploadTargetsMax && uploadTargetsMin <= uploadTargetsMax,
				"Maximum number of upload targets should be greater than zero, " +
						"should not be less than minimum number of upload targets");
		this.deadPartitionsThreshold = deadPartitionsThreshold;
		this.uploadTargetsMin = uploadTargetsMin;
		this.uploadTargetsMax = uploadTargetsMax;
		return this;
	}
	// endregion

	@Override
	public @NotNull Eventloop getEventloop() {
		return partitions.getEventloop();
	}

	@Override
	public Promise> upload(@NotNull String name) {
		return doUpload(name, fs -> fs.upload(name), identity(), uploadStartPromise, uploadFinishPromise);
	}

	@Override
	public Promise> upload(@NotNull String name, long size) {
		return doUpload(name, fs -> fs.upload(name, size), ofFixedSize(size), uploadStartPromise, uploadFinishPromise);
	}

	@Override
	public Promise> append(@NotNull String name, long offset) {
		return doUpload(name, fs -> fs.append(name, offset), identity(), appendStartPromise, appendFinishPromise);
	}

	@Override
	public Promise> download(@NotNull String name, long offset, long limit) {
		return broadcast(
				(id, fs) -> {
					logger.trace("downloading file {} from {}", name, id);
					return fs.download(name, offset, limit)
							.whenException(e -> logger.warn("Failed to connect to a server with key " + id + " to download file " + name, e))
							.map(supplier -> supplier
									.withEndOfStream(eos -> eos
											.whenException(partitions.wrapDeathFn(id))));
				},
				AsyncCloseable::close)
				.map(filterErrorsFn(() -> {
					throw new FsIOException("Could not download file '" + name + "' from any server");
				}))
				.then(suppliers -> {
					ChannelByteCombiner combiner = ChannelByteCombiner.create();
					for (ChannelSupplier supplier : suppliers) {
						combiner.addInput().set(supplier);
					}
					return Promise.of(combiner.getOutput().getSupplier());
				})
				.whenComplete(downloadStartPromise.recordStats());
	}

	@Override
	public Promise copy(@NotNull String name, @NotNull String target) {
		return ActiveFs.super.copy(name, target)
				.whenComplete(copyPromise.recordStats());
	}

	@Override
	public Promise copyAll(Map sourceToTarget) {
		return ActiveFs.super.copyAll(sourceToTarget)
				.whenComplete(copyAllPromise.recordStats());
	}

	@Override
	public Promise move(@NotNull String name, @NotNull String target) {
		return ActiveFs.super.move(name, target)
				.whenComplete(movePromise.recordStats());
	}

	@Override
	public Promise moveAll(Map sourceToTarget) {
		return ActiveFs.super.moveAll(sourceToTarget)
				.whenComplete(moveAllPromise.recordStats());
	}

	@Override
	public Promise delete(@NotNull String name) {
		return broadcast(fs -> fs.delete(name))
				.whenComplete(deletePromise.recordStats())
				.toVoid();
	}

	@Override
	public Promise deleteAll(Set toDelete) {
		return broadcast(fs -> fs.deleteAll(toDelete))
				.whenComplete(deleteAllPromise.recordStats())
				.toVoid();
	}

	@Override
	public Promise> list(@NotNull String glob) {
		return broadcast(fs -> fs.list(glob))
				.map(filterErrorsFn())
				.map(maps -> FileMetadata.flatten(maps.stream()))
				.whenComplete(listPromise.recordStats());
	}

	@Override
	public Promise<@Nullable FileMetadata> info(@NotNull String name) {
		return broadcast(fs -> fs.info(name))
				.map(filterErrorsFn())
				.map(meta -> meta.stream().max(FileMetadata.COMPARATOR).orElse(null))
				.whenComplete(infoPromise.recordStats());
	}

	@Override
	public Promise> infoAll(@NotNull Set names) {
		if (names.isEmpty()) return Promise.of(emptyMap());

		return broadcast(fs -> fs.infoAll(names))
				.map(filterErrorsFn())
				.map(maps -> FileMetadata.flatten(maps.stream()))
				.whenComplete(infoAllPromise.recordStats());
	}

	@Override
	public Promise ping() {
		return partitions.checkAllPartitions()
				.then(this::ensureIsAlive);
	}

	@Override
	public @NotNull Promise start() {
		checkArgument(deadPartitionsThreshold < partitions.getPartitions().size(),
				"Dead partitions threshold should be less than number of partitions");
		checkArgument(uploadTargetsMax <= partitions.getPartitions().size(),
				"Maximum number of upload targets should not exceed total number of partitions");

		return ping();
	}

	@Override
	public @NotNull Promise stop() {
		return Promise.complete();
	}

	@Override
	public String toString() {
		return "ClusterActiveFs{partitions=" + partitions + '}';
	}

	public boolean isAlive() {
		return partitions.getDeadPartitions().size() <= deadPartitionsThreshold;
	}

	private Promise ensureIsAlive() {
		return isAlive() ?
				Promise.complete() :
				Promise.ofException(new FsIOException("There are more dead partitions than allowed(" +
						partitions.getDeadPartitions().size() + " dead, threshold is " + deadPartitionsThreshold + "), aborting"));
	}

	private Promise> doUpload(
			String name,
			AsyncFunction> action,
			ChannelConsumerTransformer> transformer,
			PromiseStats startStats,
			PromiseStats finishStats) {
		return ensureIsAlive()
				.then(() -> collect(name, action))
				.then(containers -> {
					ChannelByteSplitter splitter = ChannelByteSplitter.create(uploadTargetsMin);
					for (Container> container : containers) {
						splitter.addOutput().set(container.value);
					}

					if (logger.isTraceEnabled()) {
						logger.trace("uploading file {} to {}, {}", name, containers.stream()
								.map(container -> container.id.toString())
								.collect(joining(", ", "[", "]")), this);
					}

					return Promise.of(splitter.getInput().getConsumer()
									.transformWith(transformer))
							.whenComplete(finishStats.recordStats());
				})
				.whenComplete(startStats.recordStats());
	}

	private Promise>>> collect(
			String name,
			AsyncFunction> action
	) {
		Iterator idIterator = partitions.select(name).iterator();
		Set> consumers = new HashSet<>();
		RefBoolean failed = new RefBoolean(false);
		return Promises.toList(
						Stream.generate(() -> first(
										transformIterator(idIterator,
												id -> call(id, action)
														.whenResult(consumer -> {
															if (failed.get()) {
																consumer.close();
															} else {
																consumers.add(consumer);
															}
														})
														.map(consumer -> new Container<>(id, consumer.withAcknowledgement(ack -> ack.whenException(partitions.wrapDeathFn(id))))))))
								.limit(uploadTargetsMax))
				.whenException(() -> {
					consumers.forEach(AsyncCloseable::close);
					failed.set(true);
					throw new FsIOException("Didn't connect to enough partitions to upload '" + name + '\'');
				});
	}

	private  Promise call(Object id, AsyncFunction action) {
		return call(id, ($, fs) -> action.apply(fs));
	}

	private  Promise call(Object id, AsyncBiFunction action) {
		ActiveFs fs = partitions.get(id);
		if (fs == null) {  // marked as dead already by somebody
			return Promise.ofException(new FsIOException("Partition '" + id + "' is not alive"));
		}
		return action.apply(id, fs)
				.whenException(partitions.wrapDeathFn(id));
	}

	private  Promise>> broadcast(AsyncBiFunction action, Consumer cleanup) {
		return ensureIsAlive()
				.then(() -> Promise.ofCallback(cb ->
						Promises.toList(partitions.getAlivePartitions().entrySet().stream()
										.map(entry -> action.apply(entry.getKey(), entry.getValue())
												.whenException(partitions.wrapDeathFn(entry.getKey()))
												.whenResult($ -> cb.isComplete(), cleanup::accept)
												.toTry()
												.then(aTry -> ensureIsAlive()
														.map($ -> aTry)
														.whenException(e -> aTry.ifSuccess(cleanup)))))
								.run(cb)));
	}

	private static  FunctionEx>, List> filterErrorsFn() {
		return filterErrorsFn(Collections::emptyList);
	}

	private static  FunctionEx>, List> filterErrorsFn(SupplierEx> fallback) {
		return tries -> {
			List successes = tries.stream().filter(Try::isSuccess).map(Try::get).collect(toList());
			if (!successes.isEmpty()) {
				return successes;
			}

			List exceptions = tries.stream().filter(Try::isException).map(Try::getException).collect(toList());
			if (!exceptions.isEmpty()) {
				Exception exception = exceptions.get(0);
				if (exceptions.stream().skip(1).allMatch(e -> e == exception)) {
					throw exception;
				}
			}
			return fallback.get();
		};
	}

	private  Promise>> broadcast(AsyncFunction action) {
		return broadcast(($, fs) -> action.apply(fs), $ -> {});
	}

	private static class Container {
		final Object id;
		final T value;

		Container(Object id, T value) {
			this.id = id;
			this.value = value;
		}
	}

	// region JMX
	@JmxAttribute
	public int getDeadPartitionsThreshold() {
		return deadPartitionsThreshold;
	}

	@JmxAttribute
	public int getUploadTargetsMin() {
		return uploadTargetsMin;
	}

	@JmxAttribute
	public int getUploadTargetsMax() {
		return uploadTargetsMax;
	}

	@JmxOperation
	public void setReplicationCount(int replicationCount) {
		withReplicationCount(replicationCount);
	}

	@JmxAttribute
	public void setDeadPartitionsThreshold(int deadPartitionsThreshold) {
		withPersistenceOptions(deadPartitionsThreshold, uploadTargetsMin, uploadTargetsMax);
	}

	@JmxAttribute
	public void setUploadTargetsMin(int uploadTargetsMin) {
		withPersistenceOptions(deadPartitionsThreshold, uploadTargetsMin, uploadTargetsMax);
	}

	@JmxAttribute
	public void setUploadTargetsMax(int uploadTargetsMax) {
		withPersistenceOptions(deadPartitionsThreshold, uploadTargetsMin, uploadTargetsMax);
	}

	@JmxAttribute
	public int getAlivePartitionCount() {
		return partitions.getAlivePartitions().size();
	}

	@JmxAttribute
	public int getDeadPartitionCount() {
		return partitions.getDeadPartitions().size();
	}

	@JmxAttribute
	public String[] getAlivePartitions() {
		return partitions.getAlivePartitions().keySet().stream()
				.map(Object::toString)
				.toArray(String[]::new);
	}

	@JmxAttribute
	public String[] getDeadPartitions() {
		return partitions.getDeadPartitions().keySet().stream()
				.map(Object::toString)
				.toArray(String[]::new);
	}

	@JmxAttribute
	public PromiseStats getUploadStartPromise() {
		return uploadStartPromise;
	}

	@JmxAttribute
	public PromiseStats getUploadFinishPromise() {
		return uploadFinishPromise;
	}

	@JmxAttribute
	public PromiseStats getAppendStartPromise() {
		return appendStartPromise;
	}

	@JmxAttribute
	public PromiseStats getAppendFinishPromise() {
		return appendFinishPromise;
	}

	@JmxAttribute
	public PromiseStats getDownloadStartPromise() {
		return downloadStartPromise;
	}

	@JmxAttribute
	public PromiseStats getDownloadFinishPromise() {
		return downloadFinishPromise;
	}

	@JmxAttribute
	public PromiseStats getListPromise() {
		return listPromise;
	}

	@JmxAttribute
	public PromiseStats getInfoPromise() {
		return infoPromise;
	}

	@JmxAttribute
	public PromiseStats getInfoAllPromise() {
		return infoAllPromise;
	}

	@JmxAttribute
	public PromiseStats getDeletePromise() {
		return deletePromise;
	}

	@JmxAttribute
	public PromiseStats getDeleteAllPromise() {
		return deleteAllPromise;
	}

	@JmxAttribute
	public PromiseStats getCopyPromise() {
		return copyPromise;
	}

	@JmxAttribute
	public PromiseStats getCopyAllPromise() {
		return copyAllPromise;
	}

	@JmxAttribute
	public PromiseStats getMovePromise() {
		return movePromise;
	}

	@JmxAttribute
	public PromiseStats getMoveAllPromise() {
		return moveAllPromise;
	}

	@JmxAttribute(name = "")
	public FsPartitions getPartitions() {
		return partitions;
	}
	// endregion
}
    

    

    

            

    

            



    
    






    
© 2015 - 2024 Weber Informatics LLC | Privacy Policy