Initial code for plugin based on opensearch-plugin-template-java #3

kevinawskendra · 2022-07-11T16:50:13Z

Description

Initial code for plugin based on opensearch-plugin-template-java

Copied files from https://github.com/opensearch-project/opensearch-plugin-template-java (excluding CONTRIBUTING.md, LICENSE.txt, NOTICE.txt, and README.md) then followed README.
Renamed LICENSE and NOTICE to LICENSE.txt, NOTICE.txt, respectively, in order for ./gradlew check to succeed.

Issues Resolved

N/A

By submitting this pull request, I confirm that my contribution is made under the terms of the Apache 2.0 license.
For more information on following Developer Certificate of Origin and signing off your commits, please check here.

ps48 · 2022-07-11T18:54:01Z

@kevinawskendra Thanks for the PR. Below are some recommended changes:

Please use commit signoffs otherwise the DCO workflow would fail.
Before adding the plugin template code. Let's add an RFC design document created for this new plugin. You may checkout an example here.

Signed-off-by: 🍕Kevin Li <kevlim@amazon.com>

macohen · 2022-10-07T16:10:59Z

Mahita, we can review this request after you resolve the DCO issue in your commits.

Signed-off-by: Mahita Mahesh <mahitam@amazon.com>

…tep to split document source content into passages Signed-off-by: Mahita Mahesh <mahitam@amazon.com>

Signed-off-by: 🍕Kevin Li <kevlim@amazon.com> Signed-off-by: Mahita Mahesh <mahitam@amazon.com>

…stener. Signed-off-by: 🍕Kevin Li <kevlim@amazon.com> Signed-off-by: Mahita Mahesh <mahitam@amazon.com>

…r hits based on response, add plugin security policy. Signed-off-by: 🍕Kevin Li <kevlim@amazon.com> Signed-off-by: Mahita Mahesh <mahitam@amazon.com>

…t setting. Signed-off-by: 🍕Kevin Li <kevlim@amazon.com> Signed-off-by: Mahita Mahesh <mahitam@amazon.com>

Signed-off-by: 🍕Kevin Li <kevlim@amazon.com> Signed-off-by: Mahita Mahesh <mahitam@amazon.com>

…ield from the OS query. Signed-off-by: 🍕Kevin Li <kevlim@amazon.com> Signed-off-by: Mahita Mahesh <mahitam@amazon.com>

Signed-off-by: Mahita Mahesh <mahitam@amazon.com>

…tId. Signed-off-by: 🍕Kevin Li <kevlim@amazon.com> Signed-off-by: Mahita Mahesh <mahitam@amazon.com>

Signed-off-by: 🍕Kevin Li <kevlim@amazon.com> Signed-off-by: Mahita Mahesh <mahitam@amazon.com>

…ex level setting Signed-off-by: Mahita Mahesh <mahitam@amazon.com>

Signed-off-by: Mahita Mahesh <mahitam@amazon.com>

msfroh · 2022-10-07T16:26:18Z

...nker/src/main/java/org/opensearch/search/relevance/preprocess/SlidingWindowTextSplitter.java

+  /**
+   * The minimum step size by which to move the sliding window after a split. Step size can be larger
+   * in practice because of respecting sentence boundaries.
+   */
+  private Integer stepSize;
+
+  /**
+   * The maximum number of passages to be extracted from the input text
+   */
+  private Integer maximumPassages;


Why are these Integer rather than int?

The contract of setSlidingWindow implicitly says that they can't be null.

Updated to int

msfroh · 2022-10-07T16:29:29Z

semantic-reranker/src/main/java/org/opensearch/search/relevance/client/KendraHttpClient.java

+        System.out.println(rerankRequest.getRerankingEndpointId());
+        System.out.println(rerankRequest.getDocuments());
+        System.out.println(rerankRequest.getQueryText());
+        Request<Void> request = new DefaultRequest<>(aws4Signer.getServiceName());
+        request.setHttpMethod(HttpMethodName.POST);
+        request.setEndpoint(URI.create(serviceEndpoint));
+        request.setHeaders(Map.of("Content-Type", "application/x-amz-json-1.0", "X-Amz-Target", "AWSKendraRerankingFrontendService.Rerank"));
+        System.out.println("objectMapper: " + objectMapper.writeValueAsString(rerankRequest));


We'll need to remove these System.out.println before merging.

Left by accident. Removed now.

msfroh · 2022-10-07T17:12:24Z

semantic-reranker/src/main/java/org/opensearch/search/relevance/client/KendraHttpClient.java

+        request.setEndpoint(URI.create(serviceEndpoint));
+        request.setHeaders(Map.of("Content-Type", "application/x-amz-json-1.0", "X-Amz-Target", "AWSKendraRerankingFrontendService.Rerank"));
+        System.out.println("objectMapper: " + objectMapper.writeValueAsString(rerankRequest));
+        request.setContent(new ByteArrayInputStream(objectMapper.writeValueAsString(rerankRequest).getBytes(StandardCharsets.UTF_8)));


There will be a dedicated Kendra reranking client in the AWS SDK, right?

I'm assuming this code (and the whole dto package) are temporary until you have added the client to the SDK.

Yes, the service is currently under development so we don't have an external SDK. We will update this code once it is ready.

msfroh · 2022-10-07T17:32:38Z

...-reranker/src/main/java/org/opensearch/search/relevance/actionfilter/SearchActionFilter.java

+
+          final SearchHits newHits = doSemanticRerank(queryParserResult, hits, pluginEnabledFetchSource);
+
+          LOGGER.info("Extracting remaining response fields");


I feel like this section can (and should) be extracted to support any external reranker. Anyone who is trying to manipulate the SearchHits of a SearchResponse would need to jump through these same hoops.

More generally, how hard would it be to split out the generic reranking logic from the Kendra-specific implementation?

I think the generic logic is probably:

Call an abstract shouldRerank method based on the search request, and if true, create a generic response listener with a Reranker.

The generic response listener extracts the SearchHits from the response, passes them to the reranker, gets back new SearchHits and uses them (and the old SearchResponse) to create a new SearchResponse.

IMO, that would also help for breaking up this class (which is a little big right now).

Thanks for the suggestion, I've refactored in the latest commit.

msfroh · 2022-10-07T17:50:38Z

...-reranker/src/main/java/org/opensearch/search/relevance/actionfilter/SearchActionFilter.java

+        if (pluginEnabledFetchSource) {
+          // User disabled fetching document source, so remove from response
+          searchHit.sourceRef(null);
+        }


It took me a while to understand that pluginEnabledFetchSource meant "The user did not request the document source, the plugin did."

Maybe name this suppressSourceOnResponse or something and comment when you initialize it?

Thanks for the naming suggestion, I've updated in the latest commit and tried to re-word the comment to make the purpose clearer. Let me know what you think.

msfroh · 2022-10-07T18:01:08Z

semantic-reranker/src/main/java/org/opensearch/search/relevance/model/dto/Document.java

+public class Document {
+
+  private String id;
+  private String title;


The existence of the title field looks like it's just part of the Kendra reranking API, and this plugin always passes null.

The only fields that matter for this plugin are id, tokenizedBodyPassages, and originalScore, right?

(I'm also assuming that these classes will be replaced with a proper SDK client model.)

We are still figuring out the user experience for configuring a mapping between fields in their document and the title and body expected by the ranker, so I will address this in a follow up.

msfroh · 2022-10-07T18:04:07Z

semantic-reranker/src/main/java/org/opensearch/search/relevance/preprocess/TextTokenizer.java

+  }
+
+  public List<String> tokenize(String text) {
+    String[] tokens = text.split("[\\p{Punct}\\s]+");


Written like this, the tokenizer will compile the regex on every invocation.

You should use define a constant Pattern field, initialized with Pattern.compile to capture the regex, and use the Pattern's split method to split strings.

Thank you for catching this. Updated.

macohen · 2022-10-10T18:31:14Z

When this PR is complete, will we be able to close out #4?

Signed-off-by: Mahita Mahesh <mahitam@amazon.com>

mahitamahesh · 2022-10-10T22:13:57Z

@macohen Yes, I think so. We are looking to make some updates to the document pre-processing logic in a follow up PR, but that won't contradict anything mentioned in #4.

…nd remove unused KendraClient. Signed-off-by: 🍕Kevin Li <kevlim@amazon.com>

Signed-off-by: Mahita Mahesh <mahitam@amazon.com>

…ot default to false Signed-off-by: Mahita Mahesh <mahitam@amazon.com>

msfroh · 2022-10-12T23:21:14Z

...c-reranker/src/main/java/org/opensearch/search/relevance/control/KendraSearchExtBuilder.java

  private static final ObjectParser<KendraSearchExtBuilder, Void> PARSER;
-  private static final ParseField RANKER_ENABLED = new ParseField("enabled");
+  private static final ParseField RANKER_ENABLED = new ParseField(Constants.ENABLED);
+  private static final ParseField BODY_FIELD = new ParseField(Constants.BODY_FIELD);


I was chatting with some folks about the long-term strategy where we'll support multiple rerankers, and trying to make sure that users of this plugin will be able to send the same queries in the future.

The proposed spec is described in #12.

Could we change the ext parameters to something more like:

{ "ext": { "rankers": [ { "name" : "kendra_intelligent_ranking", "properties" : { "body_field": [ "field1", "field2" ] } } ] } }

I want to propose a slightly modified structure

{ "ext": { "relevancy_config": { "rankers": [ { "name" : "kendra_intelligent_ranking", "enabled": true "properties" : { "body_field": [ "field1", "field2" ] } } ] } } }

We need an ExtBuilder NAME to identify the ext config specified by this plugin, under which the inputs will be nested (I’m using relevancy_config used as a placeholder). See corresponding code and usage from LTR as an example.

We need to have a flag to enable/disable rankers if we are to support index-level and request-level configuration. For example:

PUT my-index-000001/_settings { "index" : { "kendra_intelligent_ranking" : { "enabled" : "true", "body_field": "text" } } }

will enable the Kendra ranker for my-index-000001, but issuing a query without an ext entry like below, will override this and disable the Kendra ranker for this request, when in fact, it should be enabled.

GET my-index-000001/_search { "query": { "match": { "text": { "query": "research" } } } }

mahitamahesh · 2022-10-13T21:19:14Z

On another note, we have implemented plugin controls via index settings, whereas #12 makes use of the _meta field under my-index-000001/_mapping

Has anyone built a proof of concept with _meta to test that the integration works for this use case? If we need to make the change from settings to mapping, can we base our implementation off of this PoC?
If we support purely UI-based configuration in the future, we can make the API change from settings to mapping under the hood and not break customer workflows. Else I think this would be a breaking change.

mahitamahesh · 2022-10-14T23:47:38Z

@msfroh Summarizing our offline discussion here:

I. Transformer Interfaces

The plugin will invoke specific transformers at different points during the query flow (e.g. rewriters have to be applied upfront, whereas rankers will be invoked just before response, once all documents have been fetched). Further, the interface for each transformer type will differ based on the parameters it operates upon (e.g. rewriters operate on the query, rankers operate on SearchHits).

Launch proposal:

Support one transformer type: result_transformer (request_transformer, etc. to be added in the future)
Have a generic ResultTransformer interface with the methods below, and with a concrete implementation for KendraIntelligentRanking.
- boolean shouldApplyTransformer (SearchRequest)
- SearchHits apply (SearchHits)
Modify plugin logic to invoke a list of ResultTransformers (we will launch with a list of size 1), each of which operates on the SearchHits from the previous stage
We will not support passing context from one transformer to the next. This can be added in the future.

II. Search Configuration (for Index and Request level)

Support index-level configuration via index settings (not _mapping, as proposed in #12
Support request-level configuration via ext field in the request DSL.

Sample configuration based on #12 is at the end of this comment.

Launch proposal:

Maintain a list of supported transformer types (currently ["result_transformer"]) to perform validation
Have a generic transformer configuration with attributes below. Each implemented transformer will define its properties and logic for parsing + validating them
- name
- properties
Add unit tests with dummy operations to test that a chain of transformers works

Sample Configuration

"search_configuration": {
  "request_transformers": [
    {
      "name": "querqy",
      "properties": {
        "querqy": {
          "matching_query": {
            "must_match": {
              "query": "rambo"
            },
            "multi_match": {
              "query": "rambo",
              "fields": [
                "field1",
                "field2"
              ]
            }
          },
          "query_fields": [
            "title^3.0",
            "brand^2.1",
            "shortSummary"
          ]
        }
      }
    }
  ],
  "result_transformers": [
    {
      "name": "kendra_intelligent_ranking",
      "properties": {
        "title_fields": [
          "title"
        ],
        "body_fields": [
          "published",
          "description"
        ]
      }
    },
    {
      "name": "xyz_highlighting_service",
      "properties": {
        "prop1": "val1"
      }
    }
  ]
}

msfroh · 2022-10-18T00:08:02Z

Thanks @mahitamahesh ! I really like the API as described above.

I think it satisfies the needs of the Kendra-based reranker calls, but will also make it easy to integrate with other rerankers (and other result transformers) soon.

… bug. Signed-off-by: 🍕Kevin Li <kevlim@amazon.com>

Signed-off-by: 🍕Kevin Li <kevlim@amazon.com>

Signed-off-by: Mahita Mahesh <mahitam@amazon.com>

mahitamahesh · 2022-10-25T16:46:04Z

Minor tweaks compared to the proposal above:

I. Result Transformer Interface

List<Setting<?>> getTransformerSettings();

boolean shouldTransform(final SearchRequest request, final ResultTransformerConfiguration configuration);

SearchHits transform(final SearchHits hits, final SearchRequest request, final ResultTransformerConfiguration configuration);

II. Configuration

Index settings

Index settings does not support lists of objects Error: only value lists are allowed in serialized settings
Naming following OpenSearch plugin conventions, hence index.plugin.searchrelevance.*

PUT my-index/_settings

{
  "index": {
    "plugin": {
      "searchrelevance": {
        "result_transformers": {
          "kendra_intelligent_ranking": {
            "order": 2,
            "properties": {
              "title_fields": [
                "title"
              ],
              "body_fields": [
                "published",
                "description"
              ]
            }
          },
          "transformer2": {
            "order": 1,
            "properties": {
              "prop1": "val1"
            }
          }
        }
      }
    }
  }
}

Request-level configuration

Moved result_transformer directly under ext
We will add a new SearchExtBuilder class for each transformer type that is added

GET my-index/_search
{
  "query": {
    "match": {
      "text": {
        "query": "research"
      }
    }
  },
  "ext": {
    "result_transformer" : {
      "kendra_intelligent_ranking": {
        "order": 2,
        "properties": {
          "title_fields": [
            "title"
          ],
          "body_fields": [
            "published",
            "description"
          ]
        }
      },
      "transformer2": {
        "order": 1,
        "properties": {
          "prop1": "val1"
        }
      }
    }
  }
}

Signed-off-by: Mahita Mahesh <mahitam@amazon.com>

msfroh · 2022-10-25T21:40:24Z

Moved result_transformer directly under ext

What was the motivation for that change? (I don't think it's an irreversible change, but I'm curious about why it was needed.)

msfroh

Adding a few nitpicky comments

msfroh · 2022-10-26T18:17:09Z

...pensearch/search/relevance/transformer/kendraintelligentranking/KendraIntelligentRanker.java

+  private final OpenSearchClient openSearchClient;
+  private final KendraHttpClient kendraClient;
+  private final TextTokenizer textTokenizer;
+  private final CloseableHttpClient httpClient;


This field is no longer used, and can be removed.

I though that we might also be able to remove the Apache HttpClient/HttpCore dependencies, but they need to be included so the AwsHttpClient still works.

msfroh · 2022-10-26T18:25:20Z

...ance/transformer/kendraintelligentranking/configuration/KendraIntelligentRankerSettings.java

+  /**
+   * The session token for connecting to Kendra.
+   */
+  public static final Setting<SecureString> SESSION_TOKEN_SETTING = SecureSetting.secureString("kendra_intelligent_ranking.aws.session_token", null);


Does it ever make sense to specify a session token in the settings?

AFAIK, session tokens are always short-lived, so it never makes sense to include one in the opensearch.yml.

Leaving this here since using session key is completely optional. People could use it for testing if long lived credentials are not set up yet.

msfroh · 2022-10-26T18:28:40Z

semantic-reranker/src/main/plugin-metadata/plugin-security.policy

+
+  permission java.net.SocketPermission "*", "connect,resolve";
+  permission java.lang.RuntimePermission "getClassLoader";
+  permission java.util.PropertyPermission "*", "read,write";


Can this permission be scoped down?

In particular, I think we can limit it to kendra_intelligent_ranking.* properties.

Removed entirely

msfroh · 2022-10-26T18:33:41Z

semantic-reranker/config/external_service.yml

@@ -0,0 +1 @@
+# possible location for external configs


Can we delete this file?

msfroh · 2022-10-26T19:10:21Z

semantic-reranker/build.gradle

+    implementation 'commons-logging:commons-logging:1.2'
+    implementation 'com.amazonaws:aws-java-sdk-sts:1.12.300'
+    implementation 'com.amazonaws:aws-java-sdk-core:1.12.300'
+    implementation 'joda-time:joda-time:2.10.12'


Looks like we can remove the dependency on joda-time.

msfroh · 2022-10-26T19:17:53Z

semantic-reranker/build.gradle

+}
+
+dependencies {
+    implementation 'com.google.guava:guava:21.0'


The uses of Guava in SlidingWindowTextSplitter and SearchRelevancePlugin seem trivial enough to implement with built-in Java classes, so I would drop the Guava dependency.

The replacements needed are:

Lists.newArrayList() -> new ArrayList<>() ImmutableList.of(...) -> Arrays.asList(...)

Removed Guava. Will likely remove lang3 in a follow up PR too.

…date service name. Signed-off-by: Mahita Mahesh <mahitam@amazon.com>

msfroh · 2022-10-27T20:53:43Z

@mahitamahesh -- how would you feel about us merging this PR, and doing cleanup in subsequent PRs?

IMO, we're functionally complete and the general code structure is good. We have the cleanup items I listed above (which I'm happy to do) and maybe some minor tweaks here and there, but I think piling more on to this PR will make it harder to follow. What do you think?

mahitamahesh · 2022-10-27T21:39:28Z

@msfroh I have a commit ready addressing all of the cleanup items above, so I'll squeeze that in here if that's alright, and then we can call this PR done.

…bles, dependencies, permissions Signed-off-by: Mahita Mahesh <mahitam@amazon.com>

kevinawskendra · 2022-11-04T17:52:14Z

LGTM

kevinawskendra added 2 commits July 18, 2022 11:17

Initial code for plugin based on opensearch-plugin-template-java

8ae48d3

Signed-off-by: 🍕Kevin Li <kevlim@amazon.com>

Update developer url

95d226f

Signed-off-by: 🍕Kevin Li <kevlim@amazon.com>

kevinawskendra force-pushed the main branch from 5e1b53a to 95d226f Compare July 18, 2022 15:18

Restructure files and moved semantic-reranker code to its own folder.

2757d32

Signed-off-by: 🍕Kevin Li <kevlim@amazon.com>

kevinawskendra force-pushed the main branch from 69dae30 to 2757d32 Compare July 18, 2022 15:30

Mahita Mahesh and others added 18 commits October 7, 2022 12:28

Rename files related to semantic-reranker plugin

2bea455

Signed-off-by: Mahita Mahesh <mahitam@amazon.com>

Create structure for semantic reranker plugin and add preprocessing s…

64ac44c

…tep to split document source content into passages Signed-off-by: Mahita Mahesh <mahitam@amazon.com>

Implement BM25 util class for scoring passages.

dd06ec5

Signed-off-by: 🍕Kevin Li <kevlim@amazon.com> Signed-off-by: Mahita Mahesh <mahitam@amazon.com>

Implement logic to get top k passages and pass queryText to action li…

3cabab2

…stener. Signed-off-by: 🍕Kevin Li <kevlim@amazon.com> Signed-off-by: Mahita Mahesh <mahitam@amazon.com>

Implement logic to call external service with top passages and reorde…

4d0c9f5

…r hits based on response, add plugin security policy. Signed-off-by: 🍕Kevin Li <kevlim@amazon.com> Signed-off-by: Mahita Mahesh <mahitam@amazon.com>

Add custom plugin setting for enabling/disabling and logic to use tha…

1d833c7

…t setting. Signed-off-by: 🍕Kevin Li <kevlim@amazon.com> Signed-off-by: Mahita Mahesh <mahitam@amazon.com>

Add initial aws sdk integration with credentials.

9bb94df

Signed-off-by: 🍕Kevin Li <kevlim@amazon.com> Signed-off-by: Mahita Mahesh <mahitam@amazon.com>

Add aws sdk dependencies missed in previous commit.

0bf39f1

Signed-off-by: 🍕Kevin Li <kevlim@amazon.com> Signed-off-by: Mahita Mahesh <mahitam@amazon.com>

Implement basic query extraction logic to extract queryText and bodyF…

cb411f5

…ield from the OS query. Signed-off-by: 🍕Kevin Li <kevlim@amazon.com> Signed-off-by: Mahita Mahesh <mahitam@amazon.com>

Temporarily disable license, third party, logger usage checks

0c12baf

Signed-off-by: Mahita Mahesh <mahitam@amazon.com>

Add unit tests for text splitter

ec8266d

Signed-off-by: Mahita Mahesh <mahitam@amazon.com>

Limit maximum number of passages generated by text splitter

de9d1a3

Signed-off-by: Mahita Mahesh <mahitam@amazon.com>

Fetch source for reranking only, if disabled in request

8906fe5

Signed-off-by: Mahita Mahesh <mahitam@amazon.com>

Handle edge case pagination/scrolling and add new setting for endpoin…

f2d6f77

…tId. Signed-off-by: 🍕Kevin Li <kevlim@amazon.com> Signed-off-by: Mahita Mahesh <mahitam@amazon.com>

Implement http call to AWS service.

98619f5

Signed-off-by: 🍕Kevin Li <kevlim@amazon.com> Signed-off-by: Mahita Mahesh <mahitam@amazon.com>

Support request-level enable/disable of reranking call overriding ind…

9fc77d3

…ex level setting Signed-off-by: Mahita Mahesh <mahitam@amazon.com>

Use tokenized passages in request

8255fdb

Signed-off-by: Mahita Mahesh <mahitam@amazon.com>

Update naming convention to Kendra Intelligent Ranking

6d6a5c1

Signed-off-by: Mahita Mahesh <mahitam@amazon.com>

mahitamahesh force-pushed the main branch from 99eee30 to 6d6a5c1 Compare October 7, 2022 16:36

msfroh requested changes Oct 7, 2022

View reviewed changes

Address comments, update licenses and javadoc

22638a7

Signed-off-by: Mahita Mahesh <mahitam@amazon.com>

Add AssumeRole support when calling Kendra, rename client settings, a…

8fa316d

…nd remove unused KendraClient. Signed-off-by: 🍕Kevin Li <kevlim@amazon.com>

kevinawskendra force-pushed the main branch from bc3950c to 8fa316d Compare October 11, 2022 15:19

Mahita Mahesh added 2 commits October 11, 2022 23:43

Support Kendra body field configuration at index and request level

eefd761

Signed-off-by: Mahita Mahesh <mahitam@amazon.com>

Bug fix: Make enabled field in ext Boolean type to support null and n…

d136343

…ot default to false Signed-off-by: Mahita Mahesh <mahitam@amazon.com>

msfroh reviewed Oct 12, 2022

View reviewed changes

Update API request structure, call rescore API, fix passage splitting…

acde59c

… bug. Signed-off-by: 🍕Kevin Li <kevlim@amazon.com>

kevinawskendra force-pushed the main branch from df5d4c5 to acde59c Compare October 18, 2022 14:37

Update to use raw HTTP URI for Kendra.

85f5e26

Signed-off-by: 🍕Kevin Li <kevlim@amazon.com>

macohen mentioned this pull request Oct 21, 2022

[On-boarding]: search-relevance plugin for reranking results opensearch-project/opensearch-build#2784

Closed

Mahita Mahesh added 2 commits October 25, 2022 09:43

Remove test file with duplicate of BM25ScorerTests to fix failing build

d2d8719

Signed-off-by: Mahita Mahesh <mahitam@amazon.com>

Refactor to generic, extensible Transformer interface and rename plugin

e88b69a

Signed-off-by: Mahita Mahesh <mahitam@amazon.com>

Use rescore execution plan ID in raw HTTP URI for Kendra

e4f1859

Signed-off-by: Mahita Mahesh <mahitam@amazon.com>

msfroh reviewed Oct 26, 2022

View reviewed changes

Expand kendra query parser to support multi-match query type. Also up…

6d02ea1

…date service name. Signed-off-by: Mahita Mahesh <mahitam@amazon.com>

Address PR comments: Use single SearchExtBuilder, remove unused varia…

eb67efc

…bles, dependencies, permissions Signed-off-by: Mahita Mahesh <mahitam@amazon.com>

msfroh approved these changes Nov 2, 2022

View reviewed changes

kevinawskendra closed this Nov 4, 2022

kevinawskendra reopened this Nov 4, 2022

msfroh merged commit 9c3994b into opensearch-project:main Nov 4, 2022


		final SearchHits newHits = doSemanticRerank(queryParserResult, hits, pluginEnabledFetchSource);

		LOGGER.info("Extracting remaining response fields");

Initial code for plugin based on opensearch-plugin-template-java #3

Initial code for plugin based on opensearch-plugin-template-java #3

Conversation

kevinawskendra commented Jul 11, 2022

Description

Issues Resolved

ps48 commented Jul 11, 2022

macohen commented Oct 7, 2022

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

macohen commented Oct 10, 2022

mahitamahesh commented Oct 10, 2022

Choose a reason for hiding this comment

Choose a reason for hiding this comment

mahitamahesh commented Oct 13, 2022

mahitamahesh commented Oct 14, 2022

I. Transformer Interfaces

II. Search Configuration (for Index and Request level)

Sample Configuration

msfroh commented Oct 18, 2022

mahitamahesh commented Oct 25, 2022

I. Result Transformer Interface

II. Configuration

Index settings

Request-level configuration

msfroh commented Oct 25, 2022

msfroh left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

msfroh commented Oct 27, 2022

mahitamahesh commented Oct 27, 2022

kevinawskendra commented Nov 4, 2022