Make peer recovery send file chunks async

server/src/main/java/org/elasticsearch/indices/recovery/MultiFileTransfer.java

@@ -0,0 +1,210 @@
+/*
+ * Licensed to Elasticsearch under one or more contributor
+ * license agreements. See the NOTICE file distributed with
+ * this work for additional information regarding copyright
+ * ownership. Elasticsearch 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.elasticsearch.indices.recovery;
+
+import org.apache.logging.log4j.Logger;
+import org.apache.logging.log4j.message.ParameterizedMessage;
+import org.elasticsearch.Assertions;
+import org.elasticsearch.action.ActionListener;
+import org.elasticsearch.common.collect.Tuple;
+import org.elasticsearch.common.util.concurrent.AsyncIOProcessor;
+import org.elasticsearch.common.util.concurrent.ThreadContext;
+import org.elasticsearch.core.internal.io.IOUtils;
+import org.elasticsearch.index.seqno.LocalCheckpointTracker;
+import org.elasticsearch.index.store.StoreFileMetaData;
+
+import java.io.Closeable;
+import java.io.IOException;
+import java.util.Iterator;
+import java.util.List;
+import java.util.function.Consumer;
+
+import static org.elasticsearch.index.seqno.SequenceNumbers.NO_OPS_PERFORMED;
+import static org.elasticsearch.index.seqno.SequenceNumbers.UNASSIGNED_SEQ_NO;
+
+/**
+ * File chunks are sent/requested sequentially by at most one thread at any time. However, the sender/requestor won't wait for the response
+ * before processing the next file chunk request to reduce the recovery time especially on secure/compressed or high latency communication.
+ * <p>
+ * The sender/requestor can send up to {@code maxConcurrentFileChunks} file chunk requests without waiting for responses. Since the recovery
+ * target can receive file chunks out of order, it has to buffer those file chunks in memory and only flush to disk when there's no gap.
+ * To ensure the recover target never buffers more than {@code maxConcurrentFileChunks} file chunks, we allow the sender/requestor to send
+ * only up to {@code maxConcurrentFileChunks} file chunk requests from the last flushed (and acknowledged) file chunk. We leverage the local
+ * checkpoint tracker for this purpose. We generate a new sequence number and assign it to each file chunk request before sending; then mark
+ * that sequence number as processed when we receive a response for the corresponding file chunk request. With the local checkpoint tracker,
+ * we know the last acknowledged-flushed file-chunk is a file chunk whose {@code requestSeqId} equals to the local checkpoint because the
+ * recover target can flush all file chunks up to the local checkpoint.
+ * <p>
+ * When the number of un-replied file chunk requests reaches the limit (i.e. the gap between the max_seq_no and the local checkpoint is
+ * greater than {@code maxConcurrentFileChunks}), the sending/requesting thread will abort its execution. That process will be resumed by
+ * one of the networking threads which receive/handle the responses of the current pending file chunk requests. This process will continue
+ * until all chunk requests are sent/responded.
+ */
+abstract class MultiFileTransfer<Request extends MultiFileTransfer.ChunkRequest> implements Closeable {
+    private Status status = Status.PROCESSING;
+    private final Logger logger;
+    private final ActionListener<Void> listener;
+    private final LocalCheckpointTracker requestSeqIdTracker = new LocalCheckpointTracker(NO_OPS_PERFORMED, NO_OPS_PERFORMED);
+    private final AsyncIOProcessor<FileChunkResponseItem> processor;
+    private final int maxConcurrentFileChunks;
+    private StoreFileMetaData currentFile = null;
+    private final Iterator<StoreFileMetaData> remainingFiles;
+    private Tuple<StoreFileMetaData, Request> readAheadRequest = null;
+
+    protected MultiFileTransfer(Logger logger, ThreadContext threadContext, ActionListener<Void> listener,
+                                int maxConcurrentFileChunks, List<StoreFileMetaData> files) {
+        this.logger = logger;
+        this.maxConcurrentFileChunks = maxConcurrentFileChunks;
+        this.listener = listener;
+        this.processor = new AsyncIOProcessor<>(logger, maxConcurrentFileChunks, threadContext) {
+            @Override
+            protected void write(List<Tuple<FileChunkResponseItem, Consumer<Exception>>> items) {
+                handleItems(items);
+            }
+        };
+        this.remainingFiles = files.iterator();
+    }
+
+    public final void start() {
+        addItem(UNASSIGNED_SEQ_NO, null, null); // put a dummy item to start the processor
+    }
+
+    private void addItem(long requestSeqId, StoreFileMetaData md, Exception failure) {
+        processor.put(new FileChunkResponseItem(requestSeqId, md, failure), e -> { assert e == null : e; });
+    }
+
+    private void handleItems(List<Tuple<FileChunkResponseItem, Consumer<Exception>>> items) {
+        if (status != Status.PROCESSING) {
+            assert status == Status.FAILED : "must not receive any response after the transfer was completed";
+            // These exceptions will be ignored as we record only the first failure, log them for debugging purpose.
+            items.stream().filter(item -> item.v1().failure != null).forEach(item ->
+                logger.debug(new ParameterizedMessage("failed to transfer a file chunk request {}", item.v1().md), item.v1().failure));
+            return;
+        }
+        try {
+            for (Tuple<FileChunkResponseItem, Consumer<Exception>> item : items) {
+                final FileChunkResponseItem resp = item.v1();
+                if (resp.requestSeqId == UNASSIGNED_SEQ_NO) {
+                    continue; // not an actual item
+                }
+                requestSeqIdTracker.markSeqNoAsProcessed(resp.requestSeqId);
+                if (resp.failure != null) {
+                    handleError(resp.md, resp.failure);
+                    throw resp.failure;
+                }
+            }
+            while (requestSeqIdTracker.getMaxSeqNo() - requestSeqIdTracker.getProcessedCheckpoint() < maxConcurrentFileChunks) {
+                final Tuple<StoreFileMetaData, Request> request = readAheadRequest != null ? readAheadRequest : getNextRequest();
+                readAheadRequest = null;
+                if (request == null) {
+                    assert currentFile == null && remainingFiles.hasNext() == false;
+                    if (requestSeqIdTracker.getMaxSeqNo() == requestSeqIdTracker.getProcessedCheckpoint()) {
+                        onCompleted(null);
+                    }
+                    return;
+                }
+                final long requestSeqId = requestSeqIdTracker.generateSeqNo();
+                sendChunkRequest(request.v2(), ActionListener.wrap(
+                    r -> addItem(requestSeqId, request.v1(), null),
+                    e -> addItem(requestSeqId, request.v1(), e)));
+            }
+            // While we are waiting for the responses, we can prepare the next request in advance
+            // so we can send it immediately when the responses arrive to reduce the transfer time.
+            if (readAheadRequest == null) {
+                readAheadRequest = getNextRequest();
+            }
+        } catch (Exception e) {
+            onCompleted(e);
+        }
+    }
+
+    private void onCompleted(Exception failure) {
+        if (Assertions.ENABLED && status != Status.PROCESSING) {
+            throw new AssertionError("invalid status: expected [" + Status.PROCESSING + "] actual [" + status + "]", failure);
+        }
+        status = failure == null ? Status.SUCCESS : Status.FAILED;
+        try {
+            IOUtils.close(failure, this);
+        } catch (Exception e) {
+            listener.onFailure(e);
+            return;
+        }
+        listener.onResponse(null);
+    }
+
+    private Tuple<StoreFileMetaData, Request> getNextRequest() throws Exception {
+        try {
+            if (currentFile == null) {
+                if (remainingFiles.hasNext()) {
+                    currentFile = remainingFiles.next();
+                    onNewFile(currentFile);
+                } else {
+                    return null;
+                }
+            }
+            final StoreFileMetaData md = currentFile;
+            final Request request = nextChunkRequest(md);
+            if (request.lastChunk()) {
+                currentFile = null;
+            }
+            return Tuple.tuple(md, request);
+        } catch (Exception e) {
+            handleError(currentFile, e);
+            throw e;
+        }
+    }
+
+    /**
+     * This method is called when starting sending/requesting a new file. Subclasses should override
+     * this method to reset the file offset or close the previous file and open a new file if needed.
+     */
+    protected abstract void onNewFile(StoreFileMetaData md) throws IOException;
+
+    protected abstract Request nextChunkRequest(StoreFileMetaData md) throws IOException;
+
+    protected abstract void sendChunkRequest(Request request, ActionListener<Void> listener);
+
+    protected abstract void handleError(StoreFileMetaData md, Exception e) throws Exception;
+
+    private static class FileChunkResponseItem {
+        final long requestSeqId;
+        final StoreFileMetaData md;
+        final Exception failure;
+
+        FileChunkResponseItem(long requestSeqId, StoreFileMetaData md, Exception failure) {
+            this.requestSeqId = requestSeqId;
+            this.md = md;
+            this.failure = failure;
+        }
+    }
+
+    protected interface ChunkRequest {
+        /**
+         * @return {@code true} if this chunk request is the last chunk of the current file
+         */
+        boolean lastChunk();
+    }
+
+    private enum Status {
+        PROCESSING,
+        SUCCESS,
+        FAILED
+    }
+}

server/src/main/java/org/elasticsearch/indices/recovery/PeerRecoverySourceService.java

@@ -176,7 +176,7 @@ private RecoverySourceHandler createRecoverySourceHandler(StartRecoveryRequest r
                 final RemoteRecoveryTargetHandler recoveryTarget =
                     new RemoteRecoveryTargetHandler(request.recoveryId(), request.shardId(), transportService,
                         request.targetNode(), recoverySettings, throttleTime -> shard.recoveryStats().addThrottleTime(throttleTime));
-                handler = new RecoverySourceHandler(shard, recoveryTarget, request,
+                handler = new RecoverySourceHandler(shard, recoveryTarget, shard.getThreadPool(), request,
                     Math.toIntExact(recoverySettings.getChunkSize().getBytes()), recoverySettings.getMaxConcurrentFileChunks());
                 return handler;
             }