Remove mapping type usage

[dreamer-89]

Is your feature request related to a problem? Please describe.

As part of #1940 we completed all breaking changes as part of 2.0 release. In this issue, we plan to remove all remaining type usages from OpenSearch engine internals.

Background

The mapping types have been deprecated in Elasticsearch (starting from 7.0) [1] with breaking changes to the index creation, put mapping, get mapping, put template, get template and get field mappings APIs and on track to be removed in 8.0 [2]. This is the debt Opensearch has inherited and the code related to mapping types support has to be removed.
[1] https://www.elastic.co/guide/en/elasticsearch/reference/current/removal-of-types.html
[2] elastic/elasticsearch#41059

Additional context

[Removal of mapping types] (https://www.elastic.co/guide/en/elasticsearch/reference/current/removal-of-types.html)
Captures remaining pending changes for Type mapping removal

Related: #1940

Pending changes

• Remove type from nested documents; where types used to store nested path information #2357
• Remove support for _type in searches #2986
• Remove pending _type usages from the codebase #3131
• [Remove] TypeFieldMapper #3196
• [Bug] Update client benchmark post type removal #3074

[elfisher]

Are we tracking this as part of a campaign to remove mapping types usage?

[dreamer-89]

Are we tracking this as part of a campaign to remove mapping types usage?

This tracks the pending internal and non-breaking remaining changes around removing mapping types.

[dreamer-89]

@reta @nknize : I listed down remaining _type references in the code base. Need your input on below.

1. Run bulk index benchmark takes user input for typeName. Code usage, defined here. type is taken as input from CLI. Is this benchmark still used today ?
2. BulkRequestParser.java has elseif check on _type field. This check should be removed as _type should be addressed as any other random word in request body; which is already handled here. See this added previously and referred in BulkRequestParserTests inside testTypesStillParsedForBulkMonitoring.
3. Can we remove type validation from MapperService & clean up tests. In 2.0 no type field is accepted and older indices would have already been validated.

Identified other references which can be removed without any issue.

1. ScriptProcessor, ConditionalProcessor. Is it possible to provide _type from curl request and get these deprecations ?
2. Multiple references of _type in _ingest/pipeline/_simulate API across multiple places. Verified that _type is ignored by engine. Reference in a. IngestRequestConverterTests.java b. IngestClientIT.java in testSimulatePipeline c. grok_120.yml yaml tests
3. ForEachProcessorTests.java need to remove fieldValue in TestProcessor constructor.
4. 60_deprecated.yml verifying deprecated type in bulk action.
5. CustomScriptPlugin adds _type field. Probably ignored on engine side and thus can be removed
6. MultiTermVectorsResponse.java has _type defined in Fields class. It is not referenced and can be removed.
7. UpdateHelper.java defines _type in ContextFields but not referenced anywhere.
8. Bunch of _type references in mocked API responses of reindex module modules/reindex/src/test/resources/responses/*.json
9. Tests for tests ClientYamlTestSuitesTests, DoSectionTests, OpenSearchClientYamlSuiteTestCaseTests have hard-coded request body.
10. HLRC TermsVectorRequest still uses _type in toXContent.

[reta]

Thanks for summarizing, @dreamer-89, regarding your questions

1. Run bulk index benchmark takes user input for typeName. Code usage, defined here. type is taken as input from CLI. Is this benchmark still used today ?

I actually don't know to be honest, but cleaning it up would be good in any case.

3. Can we remove type validation from MapperService & clean up tests.

It looks safe to be removed, 1.x does validate mapping types, so 2.0 should not run into that, @nknize wdyt?

[dreamer-89]

Thanks for summarizing, @dreamer-89, regarding your questions

1. Run bulk index benchmark takes user input for typeName. Code usage, defined here. type is taken as input from CLI. Is this benchmark still used today ?

I actually don't know to be honest, but cleaning it up would be good in any case.

Thanks @reta. Created issue #3074 to track this.

[dreamer-89]

This plugin seems to be not updated since 6.x as it uses custom typed end-point /geonames/{type}/_noop_bulk for ingestion. This type param (taken from CLI) is used only in action meta data; while hard-coded type is used in rest end-point (/geonames/type/_noop_bulk); which means for all values of this param except type; this tool was broken. With removal of type param; this PR also fixes this issue.

We should revisit if this client-benchmark-noop-api-plugin is still used. Opensearch already has benchmark tools which can probably replace this but not sure if that needs extra work to setup this plugin.

CC: @saratvemulapalli @VachaShah