Optimizations in s3 async upload flow (opensearch-project#11327)

Signed-off-by: vikasvb90 <vikasvb@amazon.com>

plugins/repository-s3/src/main/java/org/opensearch/repositories/s3/S3AsyncService.java

@@ -374,7 +374,7 @@ private static IrsaCredentials buildFromEnvironment(IrsaCredentials defaults) {
         return new IrsaCredentials(webIdentityTokenFile, roleArn, roleSessionName);
     }
 
-    private synchronized void releaseCachedClients() {
+    public synchronized void releaseCachedClients() {
         // the clients will shutdown when they will not be used anymore
         for (final AmazonAsyncS3Reference clientReference : clientsCache.values()) {
             clientReference.decRef();

plugins/repository-s3/src/main/java/org/opensearch/repositories/s3/S3BlobContainer.java

@@ -91,6 +91,7 @@
 import org.opensearch.repositories.s3.async.UploadRequest;
 import org.opensearch.repositories.s3.utils.HttpRangeUtils;
 
+import java.io.BufferedInputStream;
 import java.io.ByteArrayInputStream;
 import java.io.IOException;
 import java.io.InputStream;
@@ -188,10 +189,38 @@ public void asyncBlobUpload(WriteContext writeContext, ActionListener<Void> comp
             writeContext.getWritePriority(),
             writeContext.getUploadFinalizer(),
             writeContext.doRemoteDataIntegrityCheck(),
-            writeContext.getExpectedChecksum()
+            writeContext.getExpectedChecksum(),
+            blobStore.isUploadRetryEnabled()
         );
         try {
-            long partSize = blobStore.getAsyncTransferManager().calculateOptimalPartSize(writeContext.getFileSize());
+            if (uploadRequest.getContentLength() > ByteSizeUnit.GB.toBytes(10) && blobStore.isRedirectLargeUploads()) {
+                StreamContext streamContext = SocketAccess.doPrivileged(
+                    () -> writeContext.getStreamProvider(uploadRequest.getContentLength())
+                );
+                InputStreamContainer inputStream = streamContext.provideStream(0);
+                try {
+                    executeMultipartUpload(
+                        blobStore,
+                        uploadRequest.getKey(),
+                        inputStream.getInputStream(),
+                        uploadRequest.getContentLength()
+                    );
+                    completionListener.onResponse(null);
+                } catch (Exception ex) {
+                    logger.error(
+                        () -> new ParameterizedMessage(
+                            "Failed to upload large file {} of size {} ",
+                            uploadRequest.getKey(),
+                            uploadRequest.getContentLength()
+                        ),
+                        ex
+                    );
+                    completionListener.onFailure(ex);
+                }
+                return;
+            }
+            long partSize = blobStore.getAsyncTransferManager()
+                .calculateOptimalPartSize(writeContext.getFileSize(), writeContext.getWritePriority(), blobStore.isUploadRetryEnabled());
             StreamContext streamContext = SocketAccess.doPrivileged(() -> writeContext.getStreamProvider(partSize));
             try (AmazonAsyncS3Reference amazonS3Reference = SocketAccess.doPrivileged(blobStore::asyncClientReference)) {
 
@@ -537,8 +566,14 @@ void executeSingleUpload(final S3BlobStore blobStore, final String blobName, fin
 
         PutObjectRequest putObjectRequest = putObjectRequestBuilder.build();
         try (AmazonS3Reference clientReference = blobStore.clientReference()) {
+            final InputStream requestInputStream;
+            if (blobStore.isUploadRetryEnabled()) {
+                requestInputStream = new BufferedInputStream(input, (int) (blobSize + 1));
+            } else {
+                requestInputStream = input;
+            }
             SocketAccess.doPrivilegedVoid(
-                () -> clientReference.get().putObject(putObjectRequest, RequestBody.fromInputStream(input, blobSize))
+                () -> clientReference.get().putObject(putObjectRequest, RequestBody.fromInputStream(requestInputStream, blobSize))
             );
         } catch (final SdkException e) {
             throw new IOException("Unable to upload object [" + blobName + "] using a single upload", e);
@@ -578,6 +613,13 @@ void executeMultipartUpload(final S3BlobStore blobStore, final String blobName,
             createMultipartUploadRequestBuilder.serverSideEncryption(ServerSideEncryption.AES256);
         }
 
+        final InputStream requestInputStream;
+        if (blobStore.isUploadRetryEnabled()) {
+            requestInputStream = new BufferedInputStream(input, (int) (partSize + 1));
+        } else {
+            requestInputStream = input;
+        }
+
         CreateMultipartUploadRequest createMultipartUploadRequest = createMultipartUploadRequestBuilder.build();
         try (AmazonS3Reference clientReference = blobStore.clientReference()) {
             uploadId.set(
@@ -601,10 +643,9 @@ void executeMultipartUpload(final S3BlobStore blobStore, final String blobName,
                     .build();
 
                 bytesCount += uploadPartRequest.contentLength();
-
                 final UploadPartResponse uploadResponse = SocketAccess.doPrivileged(
                     () -> clientReference.get()
-                        .uploadPart(uploadPartRequest, RequestBody.fromInputStream(input, uploadPartRequest.contentLength()))
+                        .uploadPart(uploadPartRequest, RequestBody.fromInputStream(requestInputStream, uploadPartRequest.contentLength()))
                 );
                 parts.add(CompletedPart.builder().partNumber(uploadPartRequest.partNumber()).eTag(uploadResponse.eTag()).build());
             }

plugins/repository-s3/src/main/java/org/opensearch/repositories/s3/S3BlobStore.java

@@ -56,8 +56,10 @@
 import static org.opensearch.repositories.s3.S3Repository.BUFFER_SIZE_SETTING;
 import static org.opensearch.repositories.s3.S3Repository.BULK_DELETE_SIZE;
 import static org.opensearch.repositories.s3.S3Repository.CANNED_ACL_SETTING;
+import static org.opensearch.repositories.s3.S3Repository.REDIRECT_LARGE_S3_UPLOAD;
 import static org.opensearch.repositories.s3.S3Repository.SERVER_SIDE_ENCRYPTION_SETTING;
 import static org.opensearch.repositories.s3.S3Repository.STORAGE_CLASS_SETTING;
+import static org.opensearch.repositories.s3.S3Repository.UPLOAD_RETRY_ENABLED;
 
 class S3BlobStore implements BlobStore {
 
@@ -71,6 +73,10 @@ class S3BlobStore implements BlobStore {
 
     private volatile ByteSizeValue bufferSize;
 
+    private volatile boolean redirectLargeUploads;
+
+    private volatile boolean uploadRetryEnabled;
+
     private volatile boolean serverSideEncryption;
 
     private volatile ObjectCannedACL cannedACL;
@@ -119,6 +125,9 @@ class S3BlobStore implements BlobStore {
         this.normalExecutorBuilder = normalExecutorBuilder;
         this.priorityExecutorBuilder = priorityExecutorBuilder;
         this.urgentExecutorBuilder = urgentExecutorBuilder;
+        // Settings to initialize blobstore with.
+        this.redirectLargeUploads = REDIRECT_LARGE_S3_UPLOAD.get(repositoryMetadata.settings());
+        this.uploadRetryEnabled = UPLOAD_RETRY_ENABLED.get(repositoryMetadata.settings());
     }
 
     @Override
@@ -130,6 +139,8 @@ public void reload(RepositoryMetadata repositoryMetadata) {
         this.cannedACL = initCannedACL(CANNED_ACL_SETTING.get(repositoryMetadata.settings()));
         this.storageClass = initStorageClass(STORAGE_CLASS_SETTING.get(repositoryMetadata.settings()));
         this.bulkDeletesSize = BULK_DELETE_SIZE.get(repositoryMetadata.settings());
+        this.redirectLargeUploads = REDIRECT_LARGE_S3_UPLOAD.get(repositoryMetadata.settings());
+        this.uploadRetryEnabled = UPLOAD_RETRY_ENABLED.get(repositoryMetadata.settings());
     }
 
     @Override
@@ -149,6 +160,14 @@ int getMaxRetries() {
         return service.settings(repositoryMetadata).maxRetries;
     }
 
+    public boolean isRedirectLargeUploads() {
+        return redirectLargeUploads;
+    }
+
+    public boolean isUploadRetryEnabled() {
+        return uploadRetryEnabled;
+    }
+
     public String bucket() {
         return bucket;
     }

plugins/repository-s3/src/main/java/org/opensearch/repositories/s3/S3Repository.java

@@ -147,6 +147,20 @@ class S3Repository extends MeteredBlobStoreRepository {
      */
     static final ByteSizeValue MAX_FILE_SIZE_USING_MULTIPART = new ByteSizeValue(5, ByteSizeUnit.TB);
 
+    /**
+     * Whether large uploads need to be redirected to slow sync s3 client.
+     */
+    static final Setting<Boolean> REDIRECT_LARGE_S3_UPLOAD = Setting.boolSetting(
+        "redirect_large_s3_upload",
+        true,
+        Setting.Property.NodeScope
+    );
+
+    /**
+     * Whether retry on uploads are enabled. This setting wraps inputstream with buffered stream to enable retries.
+     */
+    static final Setting<Boolean> UPLOAD_RETRY_ENABLED = Setting.boolSetting("s3_upload_retry_enabled", true, Setting.Property.NodeScope);
+
     /**
      * Minimum threshold below which the chunk is uploaded using a single request. Beyond this threshold,
      * the S3 repository will use the AWS Multipart Upload API to split the chunk into several parts, each of buffer_size length, and
@@ -391,7 +405,9 @@ public void reload(RepositoryMetadata newRepositoryMetadata) {
 
         // Reload configs for S3RepositoryPlugin
         service.settings(metadata);
+        service.releaseCachedClients();
         s3AsyncService.settings(metadata);
+        s3AsyncService.releaseCachedClients();
 
         // Reload configs for S3BlobStore
         BlobStore blobStore = getBlobStore();

plugins/repository-s3/src/main/java/org/opensearch/repositories/s3/S3RepositoryPlugin.java

@@ -261,7 +261,9 @@ public List<Setting<?>> getSettings() {
             S3ClientSettings.IDENTITY_TOKEN_FILE_SETTING,
             S3ClientSettings.ROLE_SESSION_NAME_SETTING,
             S3Repository.PARALLEL_MULTIPART_UPLOAD_MINIMUM_PART_SIZE_SETTING,
-            S3Repository.PARALLEL_MULTIPART_UPLOAD_ENABLED_SETTING
+            S3Repository.PARALLEL_MULTIPART_UPLOAD_ENABLED_SETTING,
+            S3Repository.REDIRECT_LARGE_S3_UPLOAD,
+            S3Repository.UPLOAD_RETRY_ENABLED
         );
     }
 

plugins/repository-s3/src/main/java/org/opensearch/repositories/s3/S3Service.java

@@ -438,7 +438,7 @@ private static IrsaCredentials buildFromEnviroment(IrsaCredentials defaults) {
         return new IrsaCredentials(webIdentityTokenFile, roleArn, roleSessionName);
     }
 
-    private synchronized void releaseCachedClients() {
+    public synchronized void releaseCachedClients() {
         // the clients will shutdown when they will not be used anymore
         for (final AmazonS3Reference clientReference : clientsCache.values()) {
             clientReference.decRef();

plugins/repository-s3/src/main/java/org/opensearch/repositories/s3/async/AsyncPartsHandler.java

@@ -23,7 +23,6 @@
 import org.opensearch.common.StreamContext;
 import org.opensearch.common.blobstore.stream.write.WritePriority;
 import org.opensearch.common.io.InputStreamContainer;
-import org.opensearch.core.common.unit.ByteSizeUnit;
 import org.opensearch.repositories.s3.SocketAccess;
 import org.opensearch.repositories.s3.StatsMetricPublisher;
 import org.opensearch.repositories.s3.io.CheckedContainer;
@@ -55,8 +54,8 @@ public class AsyncPartsHandler {
      * @param uploadId Upload Id against which multi-part is being performed
      * @param completedParts Reference of completed parts
      * @param inputStreamContainers Checksum containers
-     * @return list of completable futures
      * @param statsMetricPublisher sdk metric publisher
+     * @return list of completable futures
      * @throws IOException thrown in case of an IO error
      */
     public static List<CompletableFuture<CompletedPart>> uploadParts(
@@ -69,7 +68,8 @@ public static List<CompletableFuture<CompletedPart>> uploadParts(
         String uploadId,
         AtomicReferenceArray<CompletedPart> completedParts,
         AtomicReferenceArray<CheckedContainer> inputStreamContainers,
-        StatsMetricPublisher statsMetricPublisher
+        StatsMetricPublisher statsMetricPublisher,
+        boolean uploadRetryEnabled
     ) throws IOException {
         List<CompletableFuture<CompletedPart>> futures = new ArrayList<>();
         for (int partIdx = 0; partIdx < streamContext.getNumberOfParts(); partIdx++) {
@@ -95,7 +95,8 @@ public static List<CompletableFuture<CompletedPart>> uploadParts(
                 futures,
                 uploadPartRequestBuilder.build(),
                 inputStreamContainer,
-                uploadRequest
+                uploadRequest,
+                uploadRetryEnabled
             );
         }
 
@@ -132,6 +133,18 @@ public static void cleanUpParts(S3AsyncClient s3AsyncClient, UploadRequest uploa
         }));
     }
 
+    public static InputStream maybeRetryInputStream(
+        InputStream inputStream,
+        WritePriority writePriority,
+        boolean uploadRetryEnabled,
+        long contentLength
+    ) {
+        if (uploadRetryEnabled == true && (writePriority == WritePriority.HIGH || writePriority == WritePriority.URGENT)) {
+            return new BufferedInputStream(inputStream, (int) (contentLength + 1));
+        }
+        return inputStream;
+    }
+
     private static void uploadPart(
         S3AsyncClient s3AsyncClient,
         ExecutorService executorService,
@@ -142,7 +155,8 @@ private static void uploadPart(
         List<CompletableFuture<CompletedPart>> futures,
         UploadPartRequest uploadPartRequest,
         InputStreamContainer inputStreamContainer,
-        UploadRequest uploadRequest
+        UploadRequest uploadRequest,
+        boolean uploadRetryEnabled
     ) {
         Integer partNumber = uploadPartRequest.partNumber();
 
@@ -154,9 +168,13 @@ private static void uploadPart(
         } else {
             streamReadExecutor = executorService;
         }
-        // Buffered stream is needed to allow mark and reset ops during IO errors so that only buffered
-        // data can be retried instead of retrying whole file by the application.
-        InputStream inputStream = new BufferedInputStream(inputStreamContainer.getInputStream(), (int) (ByteSizeUnit.MB.toBytes(1) + 1));
+
+        InputStream inputStream = maybeRetryInputStream(
+            inputStreamContainer.getInputStream(),
+            uploadRequest.getWritePriority(),
+            uploadRetryEnabled,
+            uploadPartRequest.contentLength()
+        );
         CompletableFuture<UploadPartResponse> uploadPartResponseFuture = SocketAccess.doPrivileged(
             () -> s3AsyncClient.uploadPart(
                 uploadPartRequest,

plugins/repository-s3/src/main/java/org/opensearch/repositories/s3/async/AsyncTransferManager.java

@@ -34,11 +34,11 @@
 import org.opensearch.common.io.InputStreamContainer;
 import org.opensearch.common.util.ByteUtils;
 import org.opensearch.core.common.unit.ByteSizeUnit;
+import org.opensearch.core.common.unit.ByteSizeValue;
 import org.opensearch.repositories.s3.SocketAccess;
 import org.opensearch.repositories.s3.StatsMetricPublisher;
 import org.opensearch.repositories.s3.io.CheckedContainer;
 
-import java.io.BufferedInputStream;
 import java.io.IOException;
 import java.io.InputStream;
 import java.util.Arrays;
@@ -183,7 +183,8 @@ private void doUploadInParts(
                 uploadId,
                 completedParts,
                 inputStreamContainers,
-                statsMetricPublisher
+                statsMetricPublisher,
+                uploadRequest.isUploadRetryEnabled()
             );
         } catch (Exception ex) {
             try {
@@ -302,10 +303,13 @@ private static void handleException(CompletableFuture<Void> returnFuture, Suppli
     /**
      * Calculates the optimal part size of each part request if the upload operation is carried out as multipart upload.
      */
-    public long calculateOptimalPartSize(long contentLengthOfSource) {
+    public long calculateOptimalPartSize(long contentLengthOfSource, WritePriority writePriority, boolean uploadRetryEnabled) {
         if (contentLengthOfSource < ByteSizeUnit.MB.toBytes(5)) {
             return contentLengthOfSource;
         }
+        if (uploadRetryEnabled && (writePriority == WritePriority.HIGH || writePriority == WritePriority.URGENT)) {
+            return new ByteSizeValue(5, ByteSizeUnit.MB).getBytes();
+        }
         double optimalPartSize = contentLengthOfSource / (double) MAX_UPLOAD_PARTS;
         optimalPartSize = Math.ceil(optimalPartSize);
         return (long) Math.max(optimalPartSize, minimumPartSize);
@@ -335,9 +339,13 @@ private void uploadInOneChunk(
         } else {
             streamReadExecutor = executorService;
         }
-        // Buffered stream is needed to allow mark and reset ops during IO errors so that only buffered
-        // data can be retried instead of retrying whole file by the application.
-        InputStream inputStream = new BufferedInputStream(inputStreamContainer.getInputStream(), (int) (ByteSizeUnit.MB.toBytes(1) + 1));
+
+        InputStream inputStream = AsyncPartsHandler.maybeRetryInputStream(
+            inputStreamContainer.getInputStream(),
+            uploadRequest.getWritePriority(),
+            uploadRequest.isUploadRetryEnabled(),
+            uploadRequest.getContentLength()
+        );
         CompletableFuture<Void> putObjectFuture = SocketAccess.doPrivileged(
             () -> s3AsyncClient.putObject(
                 putObjectRequestBuilder.build(),

plugins/repository-s3/src/main/java/org/opensearch/repositories/s3/async/UploadRequest.java

@@ -25,6 +25,8 @@ public class UploadRequest {
     private final boolean doRemoteDataIntegrityCheck;
     private final Long expectedChecksum;
 
+    private boolean uploadRetryEnabled;
+
     /**
      * Construct a new UploadRequest object
      *
@@ -43,7 +45,8 @@ public UploadRequest(
         WritePriority writePriority,
         CheckedConsumer<Boolean, IOException> uploadFinalizer,
         boolean doRemoteDataIntegrityCheck,
-        Long expectedChecksum
+        Long expectedChecksum,
+        boolean uploadRetryEnabled
     ) {
         this.bucket = bucket;
         this.key = key;
@@ -52,6 +55,7 @@ public UploadRequest(
         this.uploadFinalizer = uploadFinalizer;
         this.doRemoteDataIntegrityCheck = doRemoteDataIntegrityCheck;
         this.expectedChecksum = expectedChecksum;
+        this.uploadRetryEnabled = uploadRetryEnabled;
     }
 
     public String getBucket() {
@@ -81,4 +85,8 @@ public boolean doRemoteDataIntegrityCheck() {
     public Long getExpectedChecksum() {
         return expectedChecksum;
     }
+
+    public boolean isUploadRetryEnabled() {
+        return uploadRetryEnabled;
+    }
 }