Adding the SearchPhaseResultsProcessor interface in Search Pipeline (opensearch-project#7283)

* Initial code for adding the SearchPhaseInjectorProcessor interface in Search Pipeline

Signed-off-by: Navneet Verma <navneev@amazon.com>

* Pass PipelinedRequest to SearchAsyncActions

We should resolve a search pipeline once at the start of a search
request and then propagate that pipeline through the async actions.

When completing a search phase, we will then use that pipeline to inject
behavior (if applicable).

Signed-off-by: Michael Froh <froh@amazon.com>

* Renamed SearchPhaseInjectorProcessor to SearchPhaseResultsProcessor and fixed the comments

Signed-off-by: Navneet Verma <navneev@amazon.com>

* Make PipelinedSearchRequest extend SearchRequest

Rather than wrapping a SearchRequest in a PipelinedSearchRequest,
changes are less intrusive if we say that a PipelinedSearchRequest
"is a" SearchRequest.

Signed-off-by: Michael Froh <froh@amazon.com>

* Revert code change from merge conflict

Signed-off-by: Michael Froh <froh@amazon.com>

* Updated the changelog with more appropiate wording for the change.

Signed-off-by: Navneet Verma <navneev@amazon.com>

* Fixed Typos in the code

Signed-off-by: Navneet Verma <navneev@amazon.com>

* Fixing comments relating to return of SearchPhaseResults from processor

Signed-off-by: Navneet Verma <navneev@amazon.com>

* Moved SearchPhaseName enum in separate class and fixed comments.

Signed-off-by: Navneet Verma <navneev@amazon.com>

* Resolve remaining merge conflict

Signed-off-by: Michael Froh <froh@amazon.com>

---------

Signed-off-by: Navneet Verma <navneev@amazon.com>
Signed-off-by: Michael Froh <froh@amazon.com>
Co-authored-by: Michael Froh <froh@amazon.com>
Co-authored-by: Andrew Ross <andrross@amazon.com>

Author: 3 people

CHANGELOG.md

@@ -80,6 +80,7 @@ The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/),
 
 ## [Unreleased 2.x]
 ### Added
+- [SearchPipeline] Add new search pipeline processor type, SearchPhaseResultsProcessor, that can modify the result of one search phase before starting the next phase.([#7283](https://github.com/opensearch-project/OpenSearch/pull/7283))
 - Add task cancellation monitoring service ([#7642](https://github.com/opensearch-project/OpenSearch/pull/7642))
 - Add TokenManager Interface ([#7452](https://github.com/opensearch-project/OpenSearch/pull/7452))
 - Add Remote store as a segment replication source ([#7653](https://github.com/opensearch-project/OpenSearch/pull/7653))

modules/search-pipeline-common/src/yamlRestTest/resources/rest-api-spec/test/search_pipeline/50_script_processor.yml

@@ -39,7 +39,7 @@ teardown:
               {
                 "script" : {
                   "lang" : "painless",
-                  "source" : "ctx._source['size'] += 10; ctx._source['from'] -= 1; ctx._source['explain'] = !ctx._source['explain']; ctx._source['version'] = !ctx._source['version']; ctx._source['seq_no_primary_term'] = !ctx._source['seq_no_primary_term']; ctx._source['track_scores'] = !ctx._source['track_scores']; ctx._source['track_total_hits'] = 1; ctx._source['min_score'] -= 0.9; ctx._source['terminate_after'] += 2; ctx._source['profile'] = !ctx._source['profile'];"
+                  "source" : "ctx._source['size'] += 10; ctx._source['from'] = ctx._source['from'] <= 0 ? ctx._source['from'] : ctx._source['from'] - 1 ; ctx._source['explain'] = !ctx._source['explain']; ctx._source['version'] = !ctx._source['version']; ctx._source['seq_no_primary_term'] = !ctx._source['seq_no_primary_term']; ctx._source['track_scores'] = !ctx._source['track_scores']; ctx._source['track_total_hits'] = 1; ctx._source['min_score'] -= 0.9; ctx._source['terminate_after'] += 2; ctx._source['profile'] = !ctx._source['profile'];"
                 }
               }
             ]

server/src/main/java/org/opensearch/action/search/AbstractSearchAsyncAction.java

@@ -57,6 +57,7 @@
 import org.opensearch.search.internal.InternalSearchResponse;
 import org.opensearch.search.internal.SearchContext;
 import org.opensearch.search.internal.ShardSearchRequest;
+import org.opensearch.search.pipeline.PipelinedRequest;
 import org.opensearch.transport.Transport;
 
 import java.util.ArrayDeque;
@@ -696,7 +697,11 @@ private void raisePhaseFailure(SearchPhaseExecutionException exception) {
      * @see #onShardResult(SearchPhaseResult, SearchShardIterator)
      */
     final void onPhaseDone() {  // as a tribute to @kimchy aka. finishHim()
-        executeNextPhase(this, getNextPhase(results, this));
+        final SearchPhase nextPhase = getNextPhase(results, this);
+        if (request instanceof PipelinedRequest && nextPhase != null) {
+            ((PipelinedRequest) request).transformSearchPhaseResults(results, this, this.getName(), nextPhase.getName());
+        }
+        executeNextPhase(this, nextPhase);
     }
 
     @Override

server/src/main/java/org/opensearch/action/search/ArraySearchPhaseResults.java

@@ -66,7 +66,7 @@ boolean hasResult(int shardIndex) {
     }
 
     @Override
-    AtomicArray<Result> getAtomicArray() {
+    public AtomicArray<Result> getAtomicArray() {
         return results;
     }
 }

server/src/main/java/org/opensearch/action/search/CanMatchPreFilterSearchPhase.java

@@ -94,7 +94,7 @@ final class CanMatchPreFilterSearchPhase extends AbstractSearchAsyncAction<CanMa
     ) {
         // We set max concurrent shard requests to the number of shards so no throttling happens for can_match requests
         super(
-            "can_match",
+            SearchPhaseName.CAN_MATCH.getName(),
             logger,
             searchTransportService,
             nodeIdToConnection,

server/src/main/java/org/opensearch/action/search/DfsQueryPhase.java

@@ -69,7 +69,7 @@ final class DfsQueryPhase extends SearchPhase {
         Function<ArraySearchPhaseResults<SearchPhaseResult>, SearchPhase> nextPhaseFactory,
         SearchPhaseContext context
     ) {
-        super("dfs_query");
+        super(SearchPhaseName.DFS_QUERY.getName());
         this.progressListener = context.getTask().getProgressListener();
         this.queryResult = queryResult;
         this.searchResults = searchResults;

server/src/main/java/org/opensearch/action/search/ExpandSearchPhase.java

@@ -62,7 +62,7 @@ final class ExpandSearchPhase extends SearchPhase {
     private final AtomicArray<SearchPhaseResult> queryResults;
 
     ExpandSearchPhase(SearchPhaseContext context, InternalSearchResponse searchResponse, AtomicArray<SearchPhaseResult> queryResults) {
-        super("expand");
+        super(SearchPhaseName.EXPAND.getName());
         this.context = context;
         this.searchResponse = searchResponse;
         this.queryResults = queryResults;

server/src/main/java/org/opensearch/action/search/FetchSearchPhase.java

@@ -92,7 +92,7 @@ final class FetchSearchPhase extends SearchPhase {
         SearchPhaseContext context,
         BiFunction<InternalSearchResponse, AtomicArray<SearchPhaseResult>, SearchPhase> nextPhaseFactory
     ) {
-        super("fetch");
+        super(SearchPhaseName.FETCH.getName());
         if (context.getNumShards() != resultConsumer.getNumShards()) {
             throw new IllegalStateException(
                 "number of shards must match the length of the query results but doesn't:"

server/src/main/java/org/opensearch/action/search/SearchPhase.java

@@ -34,6 +34,7 @@
 import org.opensearch.common.CheckedRunnable;
 
 import java.io.IOException;
+import java.util.Locale;
 import java.util.Objects;
 
 /**
@@ -54,4 +55,13 @@ protected SearchPhase(String name) {
     public String getName() {
         return name;
     }
+
+    /**
+     * Returns the SearchPhase name as {@link SearchPhaseName}. Exception will come if SearchPhase name is not defined
+     * in {@link SearchPhaseName}
+     * @return {@link SearchPhaseName}
+     */
+    public SearchPhaseName getSearchPhaseName() {
+        return SearchPhaseName.valueOf(name.toUpperCase(Locale.ROOT));
+    }
 }

server/src/main/java/org/opensearch/action/search/SearchPhaseContext.java

@@ -50,7 +50,7 @@
  *
  * @opensearch.internal
  */
-interface SearchPhaseContext extends Executor {
+public interface SearchPhaseContext extends Executor {
     // TODO maybe we can make this concrete later - for now we just implement this in the base class for all initial phases
 
     /**

server/src/main/java/org/opensearch/action/search/SearchPhaseName.java

@@ -0,0 +1,31 @@
+/*
+ * SPDX-License-Identifier: Apache-2.0
+ *
+ * The OpenSearch Contributors require contributions made to
+ * this file be licensed under the Apache-2.0 license or a
+ * compatible open source license.
+ */
+
+package org.opensearch.action.search;
+
+/**
+ * Enum for different Search Phases in OpenSearch
+ * @opensearch.internal
+ */
+public enum SearchPhaseName {
+    QUERY("query"),
+    FETCH("fetch"),
+    DFS_QUERY("dfs_query"),
+    EXPAND("expand"),
+    CAN_MATCH("can_match");
+
+    private final String name;
+
+    SearchPhaseName(final String name) {
+        this.name = name;
+    }
+
+    public String getName() {
+        return name;
+    }
+}

server/src/main/java/org/opensearch/action/search/SearchPhaseResults.java

@@ -42,7 +42,7 @@
  *
  * @opensearch.internal
  */
-abstract class SearchPhaseResults<Result extends SearchPhaseResult> {
+public abstract class SearchPhaseResults<Result extends SearchPhaseResult> {
     private final int numShards;
 
     SearchPhaseResults(int numShards) {
@@ -75,7 +75,13 @@ final int getNumShards() {
 
     void consumeShardFailure(int shardIndex) {}
 
-    AtomicArray<Result> getAtomicArray() {
+    /**
+     * Returns an {@link AtomicArray} of {@link Result}, which are nothing but the SearchPhaseResults
+     * for shards. The {@link Result} are of type {@link SearchPhaseResult}
+     *
+     * @return an {@link AtomicArray} of {@link Result}
+     */
+    public AtomicArray<Result> getAtomicArray() {
         throw new UnsupportedOperationException();
     }
 

server/src/main/java/org/opensearch/action/search/SearchScrollAsyncAction.java

@@ -266,7 +266,7 @@ protected SearchPhase sendResponsePhase(
         SearchPhaseController.ReducedQueryPhase queryPhase,
         final AtomicArray<? extends SearchPhaseResult> fetchResults
     ) {
-        return new SearchPhase("fetch") {
+        return new SearchPhase(SearchPhaseName.FETCH.getName()) {
             @Override
             public void run() throws IOException {
                 sendResponse(queryPhase, fetchResults);

server/src/main/java/org/opensearch/action/search/SearchScrollQueryThenFetchAsyncAction.java

@@ -92,7 +92,7 @@ protected void executeInitialPhase(
 
     @Override
     protected SearchPhase moveToNextPhase(BiFunction<String, String, DiscoveryNode> clusterNodeLookup) {
-        return new SearchPhase("fetch") {
+        return new SearchPhase(SearchPhaseName.FETCH.getName()) {
             @Override
             public void run() {
                 final SearchPhaseController.ReducedQueryPhase reducedQueryPhase = searchPhaseController.reducedScrollQueryPhase(

server/src/main/java/org/opensearch/action/search/TransportSearchAction.java

@@ -390,13 +390,12 @@ private void executeRequest(
             relativeStartNanos,
             System::nanoTime
         );
-        SearchRequest searchRequest;
+        PipelinedRequest searchRequest;
         ActionListener<SearchResponse> listener;
         try {
-            PipelinedRequest pipelinedRequest = searchPipelineService.resolvePipeline(originalSearchRequest);
-            searchRequest = pipelinedRequest.transformedRequest();
+            searchRequest = searchPipelineService.resolvePipeline(originalSearchRequest);
             listener = ActionListener.wrap(
-                r -> originalListener.onResponse(pipelinedRequest.transformResponse(r)),
+                r -> originalListener.onResponse(searchRequest.transformResponse(r)),
                 originalListener::onFailure
             );
         } catch (Exception e) {

server/src/main/java/org/opensearch/plugins/SearchPipelinePlugin.java

@@ -9,6 +9,7 @@
 package org.opensearch.plugins;
 
 import org.opensearch.search.pipeline.Processor;
+import org.opensearch.search.pipeline.SearchPhaseResultsProcessor;
 import org.opensearch.search.pipeline.SearchRequestProcessor;
 import org.opensearch.search.pipeline.SearchResponseProcessor;
 
@@ -42,4 +43,15 @@ default Map<String, Processor.Factory<SearchRequestProcessor>> getRequestProcess
     default Map<String, Processor.Factory<SearchResponseProcessor>> getResponseProcessors(Processor.Parameters parameters) {
         return Collections.emptyMap();
     }
+
+    /**
+     * Returns additional search pipeline search phase results processor types added by this plugin.
+     *
+     * The key of the returned {@link Map} is the unique name for the processor which is specified
+     * in pipeline configurations, and the value is a {@link org.opensearch.search.pipeline.Processor.Factory}
+     * to create the processor from a given pipeline configuration.
+     */
+    default Map<String, Processor.Factory<SearchPhaseResultsProcessor>> getSearchPhaseResultsProcessors(Processor.Parameters parameters) {
+        return Collections.emptyMap();
+    }
 }

server/src/main/java/org/opensearch/search/pipeline/Pipeline.java

@@ -8,13 +8,16 @@
 
 package org.opensearch.search.pipeline;
 
+import org.opensearch.action.search.SearchPhaseContext;
+import org.opensearch.action.search.SearchPhaseResults;
 import org.opensearch.action.search.SearchRequest;
 import org.opensearch.action.search.SearchResponse;
 import org.opensearch.common.Nullable;
 import org.opensearch.common.io.stream.BytesStreamOutput;
 import org.opensearch.common.io.stream.NamedWriteableAwareStreamInput;
 import org.opensearch.common.io.stream.NamedWriteableRegistry;
 import org.opensearch.common.io.stream.StreamInput;
+import org.opensearch.search.SearchPhaseResult;
 
 import java.util.Collections;
 import java.util.List;
@@ -28,6 +31,7 @@ class Pipeline {
 
     public static final String REQUEST_PROCESSORS_KEY = "request_processors";
     public static final String RESPONSE_PROCESSORS_KEY = "response_processors";
+    public static final String PHASE_PROCESSORS_KEY = "phase_results_processors";
     private final String id;
     private final String description;
     private final Integer version;
@@ -36,7 +40,7 @@ class Pipeline {
     // Then these can be CompoundProcessors instead of lists.
     private final List<SearchRequestProcessor> searchRequestProcessors;
     private final List<SearchResponseProcessor> searchResponseProcessors;
-
+    private final List<SearchPhaseResultsProcessor> searchPhaseResultsProcessors;
     private final NamedWriteableRegistry namedWriteableRegistry;
     private final LongSupplier relativeTimeSupplier;
 
@@ -46,6 +50,7 @@ class Pipeline {
         @Nullable Integer version,
         List<SearchRequestProcessor> requestProcessors,
         List<SearchResponseProcessor> responseProcessors,
+        List<SearchPhaseResultsProcessor> phaseResultsProcessors,
         NamedWriteableRegistry namedWriteableRegistry,
         LongSupplier relativeTimeSupplier
     ) {
@@ -54,6 +59,7 @@ class Pipeline {
         this.version = version;
         this.searchRequestProcessors = Collections.unmodifiableList(requestProcessors);
         this.searchResponseProcessors = Collections.unmodifiableList(responseProcessors);
+        this.searchPhaseResultsProcessors = Collections.unmodifiableList(phaseResultsProcessors);
         this.namedWriteableRegistry = namedWriteableRegistry;
         this.relativeTimeSupplier = relativeTimeSupplier;
     }
@@ -78,6 +84,10 @@ List<SearchResponseProcessor> getSearchResponseProcessors() {
         return searchResponseProcessors;
     }
 
+    List<SearchPhaseResultsProcessor> getSearchPhaseResultsProcessors() {
+        return searchPhaseResultsProcessors;
+    }
+
     protected void beforeTransformRequest() {}
 
     protected void afterTransformRequest(long timeInNanos) {}
@@ -168,14 +178,33 @@ SearchResponse transformResponse(SearchRequest request, SearchResponse response)
         return response;
     }
 
+    <Result extends SearchPhaseResult> void runSearchPhaseResultsTransformer(
+        SearchPhaseResults<Result> searchPhaseResult,
+        SearchPhaseContext context,
+        String currentPhase,
+        String nextPhase
+    ) throws SearchPipelineProcessingException {
+
+        try {
+            for (SearchPhaseResultsProcessor searchPhaseResultsProcessor : searchPhaseResultsProcessors) {
+                if (currentPhase.equals(searchPhaseResultsProcessor.getBeforePhase().getName())
+                    && nextPhase.equals(searchPhaseResultsProcessor.getAfterPhase().getName())) {
+                    searchPhaseResultsProcessor.process(searchPhaseResult, context);
+                }
+            }
+        } catch (RuntimeException e) {
+            throw new SearchPipelineProcessingException(e);
+        }
+    }
+
     static final Pipeline NO_OP_PIPELINE = new Pipeline(
         SearchPipelineService.NOOP_PIPELINE_ID,
         "Pipeline that does not transform anything",
         0,
         Collections.emptyList(),
         Collections.emptyList(),
+        Collections.emptyList(),
         null,
         () -> 0L
     );
-
 }

server/src/main/java/org/opensearch/search/pipeline/PipelineWithMetrics.java

@@ -43,12 +43,22 @@ class PipelineWithMetrics extends Pipeline {
         Integer version,
         List<SearchRequestProcessor> requestProcessors,
         List<SearchResponseProcessor> responseProcessors,
+        List<SearchPhaseResultsProcessor> phaseResultsProcessors,
         NamedWriteableRegistry namedWriteableRegistry,
         OperationMetrics totalRequestMetrics,
         OperationMetrics totalResponseMetrics,
         LongSupplier relativeTimeSupplier
     ) {
-        super(id, description, version, requestProcessors, responseProcessors, namedWriteableRegistry, relativeTimeSupplier);
+        super(
+            id,
+            description,
+            version,
+            requestProcessors,
+            responseProcessors,
+            phaseResultsProcessors,
+            namedWriteableRegistry,
+            relativeTimeSupplier
+        );
         this.totalRequestMetrics = totalRequestMetrics;
         this.totalResponseMetrics = totalResponseMetrics;
         for (Processor requestProcessor : getSearchRequestProcessors()) {
@@ -64,6 +74,7 @@ static PipelineWithMetrics create(
         Map<String, Object> config,
         Map<String, Processor.Factory<SearchRequestProcessor>> requestProcessorFactories,
         Map<String, Processor.Factory<SearchResponseProcessor>> responseProcessorFactories,
+        Map<String, Processor.Factory<SearchPhaseResultsProcessor>> phaseResultsProcessorFactories,
         NamedWriteableRegistry namedWriteableRegistry,
         OperationMetrics totalRequestProcessingMetrics,
         OperationMetrics totalResponseProcessingMetrics
@@ -79,6 +90,16 @@ static PipelineWithMetrics create(
             RESPONSE_PROCESSORS_KEY
         );
         List<SearchResponseProcessor> responseProcessors = readProcessors(responseProcessorFactories, responseProcessorConfigs);
+        List<Map<String, Object>> phaseResultsProcessorConfigs = ConfigurationUtils.readOptionalList(
+            null,
+            null,
+            config,
+            PHASE_PROCESSORS_KEY
+        );
+        List<SearchPhaseResultsProcessor> phaseResultsProcessors = readProcessors(
+            phaseResultsProcessorFactories,
+            phaseResultsProcessorConfigs
+        );
         if (config.isEmpty() == false) {
             throw new OpenSearchParseException(
                 "pipeline ["
@@ -93,6 +114,7 @@ static PipelineWithMetrics create(
             version,
             requestProcessors,
             responseProcessors,
+            phaseResultsProcessors,
             namedWriteableRegistry,
             totalRequestProcessingMetrics,
             totalResponseProcessingMetrics,