Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Support ML Inference Search Processor Writing to Search Extension #3061

Merged
merged 4 commits into from
Oct 17, 2024
Merged

Support ML Inference Search Processor Writing to Search Extension #3061

Show file tree
Hide file tree
Changes from 3 commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
8381354
add MLInferenceSearchResponse
mingshl Oct 2, 2024
ccf9c61
use stream method
mingshl Oct 16, 2024
81e6fe8
add more tests
mingshl Oct 17, 2024
f7682e0
check if response is in type of MLInferenceSearchResponse
mingshl Oct 17, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
57 changes: 57 additions & 0 deletions plugin/src/main/java/org/opensearch/ml/processor/MLInferenceSearchResponse.java
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,57 @@
/*
* Copyright OpenSearch Contributors
* SPDX-License-Identifier: Apache-2.0
*/
package org.opensearch.ml.processor;

import java.io.IOException;
import java.util.Map;

import org.opensearch.action.search.SearchResponse;
import org.opensearch.action.search.SearchResponseSections;
import org.opensearch.action.search.ShardSearchFailure;
import org.opensearch.core.xcontent.XContentBuilder;

public class MLInferenceSearchResponse extends SearchResponse {
mingshl marked this conversation as resolved.
Show resolved Hide resolved
private static final String EXT_SECTION_NAME = "ext";

private Map<String, Object> params;

public MLInferenceSearchResponse(
Map<String, Object> params,
SearchResponseSections internalResponse,
String scrollId,
int totalShards,
int successfulShards,
int skippedShards,
long tookInMillis,
ShardSearchFailure[] shardFailures,
Clusters clusters
) {
super(internalResponse, scrollId, totalShards, successfulShards, skippedShards, tookInMillis, shardFailures, clusters);
this.params = params;
}

public void setParams(Map<String, Object> params) {
this.params = params;
}

public Map<String, Object> getParams() {
return this.params;
}
mingshl marked this conversation as resolved.
Show resolved Hide resolved

@Override
public XContentBuilder toXContent(XContentBuilder builder, Params params) throws IOException {
builder.startObject();
innerToXContent(builder, params);
mingshl marked this conversation as resolved.
Show resolved Hide resolved

if (this.params != null) {
builder.startObject(EXT_SECTION_NAME);
builder.field(MLInferenceSearchResponseProcessor.TYPE, this.params);

builder.endObject();
}
builder.endObject();
return builder;
}
}
79 changes: 58 additions & 21 deletions plugin/src/main/java/org/opensearch/ml/processor/MLInferenceSearchResponseProcessor.java
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -14,13 +14,7 @@
import static org.opensearch.ml.processor.MLInferenceIngestProcessor.OVERRIDE;

import java.io.IOException;
import java.util.ArrayList;
import java.util.Collection;
import java.util.HashMap;
import java.util.HashSet;
import java.util.List;
import java.util.Map;
import java.util.Set;
import java.util.*;
mingshl marked this conversation as resolved.
Show resolved Hide resolved
import java.util.concurrent.atomic.AtomicBoolean;

import org.apache.logging.log4j.LogManager;
Expand Down Expand Up @@ -84,6 +78,9 @@ public class MLInferenceSearchResponseProcessor extends AbstractProcessor implem
// it can be overwritten using max_prediction_tasks when creating processor
public static final int DEFAULT_MAX_PREDICTION_TASKS = 10;
public static final String DEFAULT_OUTPUT_FIELD_NAME = "inference_results";
// allow to write to the extension of the search response, the path to point to search extension
// is prefix with ext.ml_inference
public static final String EXTENSION_PREFIX = "ext.ml_inference";
mingshl marked this conversation as resolved.
Show resolved Hide resolved

protected MLInferenceSearchResponseProcessor(
String modelId,
Expand Down Expand Up @@ -158,7 +155,19 @@ public void processResponseAsync(

// if many to one, run rewriteResponseDocuments
if (!oneToOne) {
rewriteResponseDocuments(response, responseListener);
// use MLInferenceSearchResponseProcessor to allow writing to extension
MLInferenceSearchResponse mLInferenceSearchResponse = new MLInferenceSearchResponse(
null,
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

In your SearchResponse, I noticed:

        if (this.params != null) {
            builder.startObject(EXT_SECTION_NAME);
            builder.field(MLInferenceSearchResponseProcessor.TYPE, this.params);

            builder.endObject();
        }
        ```
        
        But we are sending null from here?

Copy link
Collaborator Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

if a search response doesn't have a search extension, we don't add null params to the search extension,

for example, a search response with null params looks like this:

{
  "took": 1,
  "timed_out": false,
  "_shards": {
    "total": 1,
    "successful": 1,
    "skipped": 0,
    "failed": 0
  },
  "hits": {
    "total": {
      "value": 3,
      "relation": "eq"
    },
    "max_score": 1,
    "hits": [
      {
        "_index": "review_string_index",
        "_id": "1",
        "_score": 1,
        "_source": {
          "review": "Dr. Eric Goldberg is a fantastic doctor who has correctly diagnosed every issue that my wife and I have had. Unlike many of my past doctors, Dr. Goldberg is very accessible and we have been able to schedule appointments with him and his staff very quickly. We are happy to have him in the neighborhood and look forward to being his patients for many years to come.",
          "label": "5 stars"
        }
      }
}

for example, a search response with some params looks like this:

{
  "took": 1,
  "timed_out": false,
  "_shards": {
    "total": 1,
    "successful": 1,
    "skipped": 0,
    "failed": 0
  },
  "hits": {
    "total": {
      "value": 3,
      "relation": "eq"
    },
    "max_score": 1,
    "hits": [
      {
        "_index": "review_string_index",
        "_id": "1",
        "_score": 1,
        "_source": {
          "review": "Dr. Eric Goldberg is a fantastic doctor who has correctly diagnosed every issue that my wife and I have had. Unlike many of my past doctors, Dr. Goldberg is very accessible and we have been able to schedule appointments with him and his staff very quickly. We are happy to have him in the neighborhood and look forward to being his patients for many years to come.",
          "label": "5 stars"
        }
      }
  },
  "ext": {
    "ml_inference": {
      "llm_response": """ Based on the context provided:

- The first document is a positive review of Dr. Eric Goldberg from a patient. It praises Dr. Goldberg for correctly diagnosing issues for the patient and their wife. It also notes that Dr. Goldberg is very accessible and appointments can be scheduled quickly with him and his staff. The patient expresses happiness that Dr. Goldberg is in their neighborhood and looks forward to being his patient for many years.

- The second document just says "happy visit". 

- The third document says "sad place".

- In summary, the first document positively reviews a doctor, Dr. Eric Goldberg. The other two documents don't provide much context on their own, just mentioning a "happy visit" and "sad place"."""
    }
  }
}

Copy link
Collaborator Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

following up in this discussion, I added the check to see if the processing response is in the type of MLInferenceSearchResponse, initiate it when it's not a MLInferenceSearchResponse . f7682e0

response.getInternalResponse(),
response.getScrollId(),
response.getTotalShards(),
response.getSuccessfulShards(),
response.getSkippedShards(),
response.getSuccessfulShards(),
response.getShardFailures(),
response.getClusters()
);
rewriteResponseDocuments(mLInferenceSearchResponse, responseListener);
} else {
// if one to one, make one hit search response and run rewriteResponseDocuments
GroupedActionListener<SearchResponse> combineResponseListener = getCombineResponseGroupedActionListener(
Expand Down Expand Up @@ -545,22 +554,37 @@ public void onResponse(Map<Integer, MLOutput> multipleMLOutputs) {
} else {
modelOutputValuePerDoc = modelOutputValue;
}

if (sourceAsMap.containsKey(newDocumentFieldName)) {
if (override) {
sourceAsMapWithInference.remove(newDocumentFieldName);
sourceAsMapWithInference.put(newDocumentFieldName, modelOutputValuePerDoc);
// writing to search response extension
if (newDocumentFieldName.startsWith(EXTENSION_PREFIX)) {
Map<String, Object> params = ((MLInferenceSearchResponse) response).getParams();
String paramsName = newDocumentFieldName.replaceFirst(EXTENSION_PREFIX + ".", "");

if (params != null) {
params.put(paramsName, modelOutputValuePerDoc);
((MLInferenceSearchResponse) response).setParams(params);
} else {
logger
.debug(
"{} already exists in the search response hit. Skip processing this field.",
newDocumentFieldName
);
// TODO when the response has the same field name, should it throw exception? currently,
// ingest processor quietly skip it
Map<String, Object> newParams = new HashMap<>();
newParams.put(paramsName, modelOutputValuePerDoc);
((MLInferenceSearchResponse) response).setParams(newParams);
}
} else {
sourceAsMapWithInference.put(newDocumentFieldName, modelOutputValuePerDoc);
// writing to search response hits
if (sourceAsMap.containsKey(newDocumentFieldName)) {
if (override) {
sourceAsMapWithInference.remove(newDocumentFieldName);
sourceAsMapWithInference.put(newDocumentFieldName, modelOutputValuePerDoc);
} else {
logger
.debug(
"{} already exists in the search response hit. Skip processing this field.",
newDocumentFieldName
);
// TODO when the response has the same field name, should it throw exception? currently,
// ingest processor quietly skip it
mingshl marked this conversation as resolved.
Show resolved Hide resolved
}
} else {
sourceAsMapWithInference.put(newDocumentFieldName, modelOutputValuePerDoc);
}
}
}
}
Expand Down Expand Up @@ -774,6 +798,19 @@ public MLInferenceSearchResponseProcessor create(
+ ". Please adjust mappings."
);
}
boolean writeToSearchExtension = false;
mingshl marked this conversation as resolved.
Show resolved Hide resolved

if (outputMaps != null) {
writeToSearchExtension = outputMaps
.stream()
.filter(Objects::nonNull) // To avoid potential NullPointerExceptions from null outputMaps
.flatMap(outputMap -> outputMap.keySet().stream())
.anyMatch(key -> key.startsWith(EXTENSION_PREFIX));
}

if (writeToSearchExtension & oneToOne) {
throw new IllegalArgumentException("Write model response to search extension does not support when one_to_one is true.");
}

return new MLInferenceSearchResponseProcessor(
modelId,
Expand Down
Loading
Loading