Support batch ingestion in bulk API (#12457) (#13306)

* [PoC][issues-12457] Support Batch Ingestion Signed-off-by: Liyun Xiu <xiliyun@amazon.com> * Rewrite batch interface and handle error and metrics Signed-off-by: Liyun Xiu <xiliyun@amazon.com> * Remove unnecessary change Signed-off-by: Liyun Xiu <xiliyun@amazon.com> * Revert some unnecessary test change Signed-off-by: Liyun Xiu <xiliyun@amazon.com> * Keep executeBulkRequest main logic untouched Signed-off-by: Liyun Xiu <xiliyun@amazon.com> * Add UT Signed-off-by: Liyun Xiu <xiliyun@amazon.com> * Add UT & yamlRest test, fix BulkRequest se/deserialization Signed-off-by: Liyun Xiu <xiliyun@amazon.com> * Add missing java docs Signed-off-by: Liyun Xiu <xiliyun@amazon.com> * Remove Writable from BatchIngestionOption Signed-off-by: Liyun Xiu <xiliyun@amazon.com> * Add more UTs Signed-off-by: Liyun Xiu <xiliyun@amazon.com> * Fix spotlesscheck Signed-off-by: Liyun Xiu <xiliyun@amazon.com> * Rename parameter name to batch_size Signed-off-by: Liyun Xiu <xiliyun@amazon.com> * Add more rest yaml tests & update rest spec Signed-off-by: Liyun Xiu <xiliyun@amazon.com> * Remove batch_ingestion_option and only use batch_size Signed-off-by: Liyun Xiu <xiliyun@amazon.com> * Throw invalid request exception for invalid batch_size Signed-off-by: Liyun Xiu <xiliyun@amazon.com> * Update server/src/main/java/org/opensearch/action/bulk/BulkRequest.java Co-authored-by: Andriy Redko <drreta@gmail.com> Signed-off-by: Liyun Xiu <chishui2@gmail.com> * Remove version constant Signed-off-by: Liyun Xiu <xiliyun@amazon.com> --------- Signed-off-by: Liyun Xiu <xiliyun@amazon.com> Signed-off-by: Liyun Xiu <chishui2@gmail.com> Co-authored-by: Andriy Redko <drreta@gmail.com>
opensearch-project · Apr 30, 2024 · 1219c56 · 1219c56
1 parent d7e6b9c
commit 1219c56
Show file tree

Hide file tree

Showing 18 changed files with 1,486 additions and 50 deletions.
diff --git a/CHANGELOG.md b/CHANGELOG.md
@@ -28,6 +28,7 @@ The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/),
 - [Tiered Caching] Gate new stats logic behind FeatureFlags.PLUGGABLE_CACHE ([#13238](https://github.com/opensearch-project/OpenSearch/pull/13238))
 - [Tiered Caching] Add a dynamic setting to disable/enable disk cache. ([#13373](https://github.com/opensearch-project/OpenSearch/pull/13373))
 - [Remote Store] Add capability of doing refresh as determined by the translog ([#12992](https://github.com/opensearch-project/OpenSearch/pull/12992))
+- [Batch Ingestion] Add `batch_size` to `_bulk` API. ([#12457](https://github.com/opensearch-project/OpenSearch/issues/12457))
 - [Tiered caching] Make Indices Request Cache Stale Key Mgmt Threshold setting dynamic ([#12941](https://github.com/opensearch-project/OpenSearch/pull/12941))
 - Batch mode for async fetching shard information in GatewayAllocator for unassigned shards ([#8746](https://github.com/opensearch-project/OpenSearch/pull/8746))
 - [Remote Store] Add settings for remote path type and hash algorithm ([#13225](https://github.com/opensearch-project/OpenSearch/pull/13225))

diff --git a/modules/ingest-common/src/yamlRestTest/resources/rest-api-spec/test/ingest/70_bulk.yml b/modules/ingest-common/src/yamlRestTest/resources/rest-api-spec/test/ingest/70_bulk.yml
@@ -167,3 +167,90 @@ teardown:
         index: test_index
         id: test_id3
   - match: { _source: {"f1": "v2", "f2": 47, "field1": "value1"}}
+
+---
+"Test bulk API with batch enabled happy case":
+  - skip:
+      version: " - 2.13.99"
+      reason: "Added in 2.14.0"
+
+  - do:
+      bulk:
+        refresh: true
+        batch_size: 2
+        pipeline: "pipeline1"
+        body:
+          - '{"index": {"_index": "test_index", "_id": "test_id1"}}'
+          - '{"text": "text1"}'
+          - '{"index": {"_index": "test_index", "_id": "test_id2"}}'
+          - '{"text": "text2"}'
+          - '{"index": {"_index": "test_index", "_id": "test_id3"}}'
+          - '{"text": "text3"}'
+          - '{"index": {"_index": "test_index", "_id": "test_id4"}}'
+          - '{"text": "text4"}'
+          - '{"index": {"_index": "test_index", "_id": "test_id5", "pipeline": "pipeline2"}}'
+          - '{"text": "text5"}'
+          - '{"index": {"_index": "test_index", "_id": "test_id6", "pipeline": "pipeline2"}}'
+          - '{"text": "text6"}'
+
+  - match: { errors: false }
+
+  - do:
+      get:
+        index: test_index
+        id: test_id5
+  - match: { _source: {"text": "text5", "field2": "value2"}}
+
+  - do:
+      get:
+        index: test_index
+        id: test_id3
+  - match: { _source: { "text": "text3", "field1": "value1" } }
+
+---
+"Test bulk API with batch_size missing":
+  - skip:
+      version: " - 2.13.99"
+      reason: "Added in 2.14.0"
+
+  - do:
+      bulk:
+        refresh: true
+        pipeline: "pipeline1"
+        body:
+          - '{"index": {"_index": "test_index", "_id": "test_id1"}}'
+          - '{"text": "text1"}'
+          - '{"index": {"_index": "test_index", "_id": "test_id2"}}'
+          - '{"text": "text2"}'
+
+  - match: { errors: false }
+
+  - do:
+      get:
+        index: test_index
+        id: test_id1
+  - match: { _source: { "text": "text1", "field1": "value1" } }
+
+  - do:
+      get:
+        index: test_index
+        id: test_id2
+  - match: { _source: { "text": "text2", "field1": "value1" } }
+
+---
+"Test bulk API with invalid batch_size":
+  - skip:
+      version: " - 2.13.99"
+      reason: "Added in 2.14.0"
+
+  - do:
+      catch: bad_request
+      bulk:
+        refresh: true
+        batch_size: -1
+        pipeline: "pipeline1"
+        body:
+          - '{"index": {"_index": "test_index", "_id": "test_id1"}}'
+          - '{"text": "text1"}'
+          - '{"index": {"_index": "test_index", "_id": "test_id2"}}'
+          - '{"text": "text2"}'
diff --git a/rest-api-spec/src/main/resources/rest-api-spec/api/bulk.json b/rest-api-spec/src/main/resources/rest-api-spec/api/bulk.json
@@ -74,6 +74,10 @@
       "require_alias": {
         "type": "boolean",
         "description": "Sets require_alias for all incoming documents. Defaults to unset (false)"
+      },
+      "batch_size": {
+        "type": "int",
+        "description": "Sets the batch size"
       }
     },
     "body":{

diff --git a/server/src/main/java/org/opensearch/action/bulk/BulkRequest.java b/server/src/main/java/org/opensearch/action/bulk/BulkRequest.java
@@ -34,6 +34,7 @@
 
 import org.apache.lucene.util.Accountable;
 import org.apache.lucene.util.RamUsageEstimator;
+import org.opensearch.Version;
 import org.opensearch.action.ActionRequest;
 import org.opensearch.action.ActionRequestValidationException;
 import org.opensearch.action.CompositeIndicesRequest;
@@ -80,7 +81,6 @@ public class BulkRequest extends ActionRequest implements CompositeIndicesReques
     private static final long SHALLOW_SIZE = RamUsageEstimator.shallowSizeOfInstance(BulkRequest.class);
 
     private static final int REQUEST_OVERHEAD = 50;
-
     /**
      * Requests that are part of this request. It is only possible to add things that are both {@link ActionRequest}s and
      * {@link WriteRequest}s to this but java doesn't support syntax to declare that everything in the array has both types so we declare
@@ -96,6 +96,7 @@ public class BulkRequest extends ActionRequest implements CompositeIndicesReques
     private String globalRouting;
     private String globalIndex;
     private Boolean globalRequireAlias;
+    private int batchSize = 1;
 
     private long sizeInBytes = 0;
 
@@ -107,6 +108,9 @@ public BulkRequest(StreamInput in) throws IOException {
         requests.addAll(in.readList(i -> DocWriteRequest.readDocumentRequest(null, i)));
         refreshPolicy = RefreshPolicy.readFrom(in);
         timeout = in.readTimeValue();
+        if (in.getVersion().onOrAfter(Version.V_2_14_0)) {
+            batchSize = in.readInt();
+        }
     }
 
     public BulkRequest(@Nullable String globalIndex) {
@@ -346,6 +350,27 @@ public final BulkRequest timeout(TimeValue timeout) {
         return this;
     }
 
+    /**
+     * Set batch size
+     * @param size batch size from input
+     * @return {@link BulkRequest}
+     */
+    public BulkRequest batchSize(int size) {
+        if (size < 1) {
+            throw new IllegalArgumentException("batch_size must be greater than 0");
+        }
+        this.batchSize = size;
+        return this;
+    }
+
+    /**
+     * Get batch size
+     * @return batch size
+     */
+    public int batchSize() {
+        return this.batchSize;
+    }
+
     /**
      * Note for internal callers (NOT high level rest client),
      * the global parameter setting is ignored when used with:
@@ -453,6 +478,9 @@ public void writeTo(StreamOutput out) throws IOException {
         out.writeCollection(requests, DocWriteRequest::writeDocumentRequest);
         refreshPolicy.writeTo(out);
         out.writeTimeValue(timeout);
+        if (out.getVersion().onOrAfter(Version.V_2_14_0)) {
+            out.writeInt(batchSize);
+        }
     }
 
     @Override

diff --git a/server/src/main/java/org/opensearch/action/bulk/TransportBulkAction.java b/server/src/main/java/org/opensearch/action/bulk/TransportBulkAction.java
@@ -923,7 +923,8 @@ public boolean isForceExecution() {
                 }
             },
             bulkRequestModifier::markItemAsDropped,
-            executorName
+            executorName,
+            original
         );
     }
 

diff --git a/server/src/main/java/org/opensearch/common/metrics/OperationMetrics.java b/server/src/main/java/org/opensearch/common/metrics/OperationMetrics.java
@@ -37,6 +37,14 @@ public void before() {
         current.incrementAndGet();
     }
 
+    /**
+     * Invoke before the given operation begins in multiple items at the same time.
+     * @param n number of items
+     */
+    public void beforeN(int n) {
+        current.addAndGet(n);
+    }
+
     /**
      * Invoked upon completion (success or failure) of the given operation
      * @param currentTime elapsed time of the operation
@@ -46,13 +54,35 @@ public void after(long currentTime) {
         time.inc(currentTime);
     }
 
+    /**
+     * Invoked upon completion (success or failure) of the given operation for multiple items.
+     * @param n number of items completed
+     * @param currentTime elapsed time of the operation
+     */
+    public void afterN(int n, long currentTime) {
+        current.addAndGet(-n);
+        for (int i = 0; i < n; ++i) {
+            time.inc(currentTime);
+        }
+    }
+
     /**
      * Invoked upon failure of the operation.
      */
     public void failed() {
         failed.inc();
     }
 
+    /**
+     * Invoked upon failure of the operation on multiple items.
+     * @param n number of items on operation.
+     */
+    public void failedN(int n) {
+        for (int i = 0; i < n; ++i) {
+            failed.inc();
+        }
+    }
+
     public void add(OperationMetrics other) {
         // Don't try copying over current, since in-flight requests will be linked to the existing metrics instance.
         failed.inc(other.failed.count());

diff --git a/server/src/main/java/org/opensearch/ingest/CompoundProcessor.java b/server/src/main/java/org/opensearch/ingest/CompoundProcessor.java
@@ -39,10 +39,13 @@
 import java.util.ArrayList;
 import java.util.Arrays;
 import java.util.Collections;
+import java.util.HashMap;
 import java.util.List;
 import java.util.Map;
 import java.util.concurrent.TimeUnit;
+import java.util.concurrent.atomic.AtomicInteger;
 import java.util.function.BiConsumer;
+import java.util.function.Consumer;
 import java.util.function.LongSupplier;
 import java.util.stream.Collectors;
 
@@ -150,6 +153,108 @@ public void execute(IngestDocument ingestDocument, BiConsumer<IngestDocument, Ex
         innerExecute(0, ingestDocument, handler);
     }
 
+    @Override
+    public void batchExecute(List<IngestDocumentWrapper> ingestDocumentWrappers, Consumer<List<IngestDocumentWrapper>> handler) {
+        innerBatchExecute(0, ingestDocumentWrappers, handler);
+    }
+
+    /**
+     * Internal logic to process documents with current processor.
+     *
+     * @param currentProcessor index of processor to process batched documents
+     * @param ingestDocumentWrappers batched documents to be processed
+     * @param handler callback function
+     */
+    void innerBatchExecute(
+        int currentProcessor,
+        List<IngestDocumentWrapper> ingestDocumentWrappers,
+        Consumer<List<IngestDocumentWrapper>> handler
+    ) {
+        if (currentProcessor == processorsWithMetrics.size()) {
+            handler.accept(ingestDocumentWrappers);
+            return;
+        }
+        Tuple<Processor, OperationMetrics> processorWithMetric = processorsWithMetrics.get(currentProcessor);
+        final Processor processor = processorWithMetric.v1();
+        final OperationMetrics metric = processorWithMetric.v2();
+        final long startTimeInNanos = relativeTimeProvider.getAsLong();
+        int size = ingestDocumentWrappers.size();
+        metric.beforeN(size);
+        // Use synchronization to ensure batches are processed by processors in sequential order
+        AtomicInteger counter = new AtomicInteger(size);
+        List<IngestDocumentWrapper> allResults = Collections.synchronizedList(new ArrayList<>());
+        Map<Integer, IngestDocumentWrapper> slotToWrapperMap = createSlotIngestDocumentWrapperMap(ingestDocumentWrappers);
+        processor.batchExecute(ingestDocumentWrappers, results -> {
+            if (results.isEmpty()) return;
+            allResults.addAll(results);
+            // counter equals to 0 means all documents are processed and called back.
+            if (counter.addAndGet(-results.size()) == 0) {
+                long ingestTimeInMillis = TimeUnit.NANOSECONDS.toMillis(relativeTimeProvider.getAsLong() - startTimeInNanos);
+                metric.afterN(allResults.size(), ingestTimeInMillis);
+
+                List<IngestDocumentWrapper> documentsDropped = new ArrayList<>();
+                List<IngestDocumentWrapper> documentsWithException = new ArrayList<>();
+                List<IngestDocumentWrapper> documentsToContinue = new ArrayList<>();
+                int totalFailed = 0;
+                // iterate all results to categorize them to: to continue, to drop, with exception
+                for (IngestDocumentWrapper resultDocumentWrapper : allResults) {
+                    IngestDocumentWrapper originalDocumentWrapper = slotToWrapperMap.get(resultDocumentWrapper.getSlot());
+                    if (resultDocumentWrapper.getException() != null) {
+                        ++totalFailed;
+                        if (ignoreFailure) {
+                            documentsToContinue.add(originalDocumentWrapper);
+                        } else {
+                            IngestProcessorException compoundProcessorException = newCompoundProcessorException(
+                                resultDocumentWrapper.getException(),
+                                processor,
+                                originalDocumentWrapper.getIngestDocument()
+                            );
+                            documentsWithException.add(
+                                new IngestDocumentWrapper(
+                                    resultDocumentWrapper.getSlot(),
+                                    originalDocumentWrapper.getIngestDocument(),
+                                    compoundProcessorException
+                                )
+                            );
+                        }
+                    } else {
+                        if (resultDocumentWrapper.getIngestDocument() == null) {
+                            documentsDropped.add(resultDocumentWrapper);
+                        } else {
+                            documentsToContinue.add(resultDocumentWrapper);
+                        }
+                    }
+                }
+                if (totalFailed > 0) {
+                    metric.failedN(totalFailed);
+                }
+                if (!documentsDropped.isEmpty()) {
+                    handler.accept(documentsDropped);
+                }
+                if (!documentsToContinue.isEmpty()) {
+                    innerBatchExecute(currentProcessor + 1, documentsToContinue, handler);
+                }
+                if (!documentsWithException.isEmpty()) {
+                    if (onFailureProcessors.isEmpty()) {
+                        handler.accept(documentsWithException);
+                    } else {
+                        documentsWithException.forEach(
+                            doc -> executeOnFailureAsync(
+                                0,
+                                doc.getIngestDocument(),
+                                (IngestProcessorException) doc.getException(),
+                                (result, ex) -> {
+                                    handler.accept(Collections.singletonList(new IngestDocumentWrapper(doc.getSlot(), result, ex)));
+                                }
+                            )
+                        );
+                    }
+                }
+            }
+            assert counter.get() >= 0;
+        });
+    }
+
     void innerExecute(int currentProcessor, IngestDocument ingestDocument, BiConsumer<IngestDocument, Exception> handler) {
         if (currentProcessor == processorsWithMetrics.size()) {
             handler.accept(ingestDocument, null);
@@ -266,4 +371,12 @@ static IngestProcessorException newCompoundProcessorException(Exception e, Proce
         return exception;
     }
 
+    private Map<Integer, IngestDocumentWrapper> createSlotIngestDocumentWrapperMap(List<IngestDocumentWrapper> ingestDocumentWrappers) {
+        Map<Integer, IngestDocumentWrapper> slotIngestDocumentWrapperMap = new HashMap<>();
+        for (IngestDocumentWrapper ingestDocumentWrapper : ingestDocumentWrappers) {
+            slotIngestDocumentWrapperMap.put(ingestDocumentWrapper.getSlot(), ingestDocumentWrapper);
+        }
+        return slotIngestDocumentWrapperMap;
+    }
+
 }