Support query pass-through for Elasticsearch connector

[kasiafi]

This PR introduces raw_query Polymorphic Table Function for full query pass-through to Elasticsearch connector.

# Elasticsearch
* Add `raw_query` table function for full query pass-through to the connector.

Documentation

Documentation issue is filed: #13007.
When documenting, also deprecate the legacy query pass-through mechanism: https://trino.io/docs/current/connector/elasticsearch.html?highlight=elasticsearch#pass-through-queries

[ebyhr]

It would be better to verify the result in my opinion and how about formatting like this?

        assertThat(
                "SELECT * FROM " +
                        "TABLE(\"" + getSession().getCatalog().orElseThrow() + "\".\"system\".\"remote_query\"(" +
                        " \"schema\" => 'tpch'," +
                        " \"index\" => 'nation'," +
                        " \"query\" => '{\"query\": {\"match\": {\"name\": \"ALGERIA\"}}}'" +
                        "))")
                .matches(...);

By the way, I expected the above SELECT query returns the result with multiple columns, but didn't. Is it intentional design? Also, I would recommend adding empt result & failure case.

[martint]

That's intentional. Pass-through queries in Elasticsearch return a single VARCHAR column as a result.

[kasiafi]

@ebyhr I added more tests including no match and failure.

[martint]

We should deprecate the old-style passthrough queries in favor of this.

[martint]

That's intentional. Pass-through queries in Elasticsearch return a single VARCHAR column as a result.

[martint]

Also, don't forget to update the docs or file an issue to do it later so we don't lose track of it.

[martint]

QueryFunctionHandle maybe?

[martint]

I don't think this is such a good idea. It masks the underlying assertion, so it's hard to tell what was different and it what way. It also doesn't scale when there are more than a couple of alternatives.

Instead, modify the query to normalize the results in some way. For instance, parse the resulting json, extract the values and sort them to get a consistent result.

[kasiafi]

@martint I applied comments. Please take a look.

[findepi]

due to the fact that the function seems to dump ES result, i would consider calling a raw_query
(especially so that we keep query name free for future)

[martint]

That's a good point. In ES parlance, queries are "search queries" (https://www.elastic.co/guide/en/elasticsearch/reference/current/search-your-data.html), and there's a Query DSL (https://www.elastic.co/guide/en/elasticsearch/reference/current/query-dsl.html) to formulate them (what we support here). They also have two other languages:

• EQL, for querying log and event data (https://www.elastic.co/guide/en/elasticsearch/reference/current/eql.html)
• SQL (https://www.elastic.co/guide/en/elasticsearch/reference/current/sql-overview.html)

[kasiafi]

@martint @findepi I renamed the function to raw_query.

I also added a test to compare the results between the legacy pass-through mechanism, and the new table function.

The old query pass-through will be deprecated in the documentation when we document the table function. @martint is there any other way we want to deprecate it than a note in the docs?

[findepi]

is there any other way we want to deprecate it than a note in the docs?

you can also mark related pieces of code as @Deprecated, with a TODO, and a link to an issue covering removal of that functionality (assuming we don't have plans to carry it forward for ever)