[FEATURE] Make OpenSearch function registered dynamically

[dai-chen]

Is your feature request related to a problem?
Currently all OpenSearch functions are "hardcoded" in ANTLR grammar file. As external storage like S3 and Prometheus added, this will cause problem. For example, in current implementation, query like "... FROM s3.logs WHERE QUERY_STRING(...)" can pass ANTLR parser but fail at execution time due to missing QUERY_STRING function support in S3 storage.

What solution would you like?
One approach is each storage engine registers specific function dynamically when plugin bootstraps. The ANTLR grammar only includes common ANSI SQL functions that has in-memory implementation along with a new udf rule that match unrecognized functions as User-Defined Function. Later analyzer can verify if the UDF is valid or not. This approach naturally provides UDF support for future.

What alternatives have you considered?
N/A

Do you have any additional context?
N/A

[MaxKsyunz]

@dai-chen, v1 engine supports this syntax for relevance queries -- SELECT * FROM index WHERE field = match("string").

If we add it to v2 engine, this feature would need to support this syntax as well. Especially given #787.

[dai-chen]

@dai-chen, v1 engine supports this syntax for relevance queries -- SELECT * FROM index WHERE field = match("string").

If we add it to v2 engine, this feature would need to support this syntax as well. Especially given #787.

Sure, we need to figure out a general way and allow each storage engine to register specific data type and function dynamically. Earlier our storage engine interface was referencing Apache Calcite. Probably we can explore further here: https://github.com/apache/calcite/blob/main/core/src/main/java/org/apache/calcite/schema/Schema.java

[biggoon]

Another think to add: we need to move the analyzers relevant to the user-defined functions (such as the HighlightAnalyzer) out of core and inject them into the Analyzer dynamically.
see: https://github.com/opensearch-project/sql/blob/2.x/core/src/main/java/org/opensearch/sql/analysis/Analyzer.java#L364

[dai-chen]

Another think to add: we need to move the analyzers relevant to the user-defined functions (such as the HighlightAnalyzer) out of core and inject them into the Analyzer dynamically. see: https://github.com/opensearch-project/sql/blob/2.x/core/src/main/java/org/opensearch/sql/analysis/Analyzer.java#L364

That'd great be if we can do that for highlight as well. I've met this problem with highlight operator when refactor optimizer: https://github.com/opensearch-project/sql/pull/1091/files#diff-3334dc7821335d1c35cb35e3da00960106b6ac04c4ca092937c51c076056d272R103. I have to add a pushDownHighlight rule, however there is no fallback if push down not happen on other storage like S3/Prometheus. There is no in-memory implementation and highlight maybe only make sense for OpenSearch storage.