Use SQL to query Elasticsearch. ES V6 compatible.

• =, !=, <, >, <=, >=, <>, ()
• arithmetic operators: +, -, *, /, %, >>, <<, (), ~
• AND, OR, NOT
• AS
• LIKE, IN, REGEX, IS NULL, BETWEEN
• LIMIT, SIZE, OFFSET
• GROUP BY, ORDER BY
• GROUP_CONCAT
• AVG, MAX, MIN, SUM, COUNT
• date_histogram, histogram, date_range, range
• HAVING
• query key value macro (see usage)
• pagination (search after)
• pagination for aggregation
• UDF
• JOIN
• nested queries

• Arithmetics are allowed in SELECT and WHERE clause. They use script query, and thus are not able to utilize reverse index and can be potentially slow.
• Aggregation functions can be introduced from SELECT, ORDER BY and HAVING
• If you want to apply aggregation on some fields, they should not be in type text in ES
• COUNT(colName) will include documents w/ null values in that column in ES SQL API, while in esql we exclude null valued documents
• ES SQL API and esql do not support SELECT DISTINCT, a workaround is to query something like SELECT * FROM table GROUP BY colName
• To use regex query, the column should be keyword type, otherwise the regex is applied to all the terms produced by tokenizer from the original text rather than the original text itself
• Comparison with arithmetics can be potentially slow since it uses scripting query and thus is not able to take advantage of reverse index. For binary operators, please refer to this link on the precedence. We don't support all of them.

Please refer to code and comments in esql.go. esql.go contains all the apis that an outside user needs.

sql := `SELECT COUNT(*), MAX(colA) FROM myTable WHERE colB < 10 GROUP BY colC HAVING COUNT(*) > 20`
e := NewESql()
dsl, _, err := e.ConvertPretty(sql)    // convert sql to dsl
if err == nil {
    fmt.Println(dsl)
}

ESQL support API ProcessQueryKey to register custom policy for colName replacement. It accepts 2 functions, the first function determines whether a colName is to be replaced, the second specifies how to do the replacement. Use case: user has custom field field, but to resolve confict, server stores the field as Custom.field. ProcessQueryKey API can automatically do the conversion.

ESQL support API ProcessQueryValue to register custom policy for value processing. It accepts 2 functions, the first function determines whether a value of a colName is to be processed, the second specifies how to do the processing. Use case: user want to query time in readable format, but server stores time as an integer (unix nano). ProcessQueryValue API can automatically do the conversion.

Below shows an example.

sql := "SELECT colA FROM myTable WHERE colB < 10 AND dateTime = '2015-01-01T02:59:59Z'"
// custom policy that change colName like "col.." to "myCol.."
func myKeyFilter(colName string) bool {
    return strings.HasPrefix(colName, "col")
}
func myKeyProcess(colName string) (string, error) {
    return "myCol"+colName[3:], nil
}
// custom policy that convert formatted time string to unix nano
func myValueFilter(colName string) bool {
    return strings.Contains(colName, "Time") || strings.Contains(colName, "time")
}
func myValueProcess(timeStr string) (string, error) {
    // convert formatted time string to unix nano integer
    parsedTime, _ := time.Parse(defaultDateTimeFormat, timeStr)
    return fmt.Sprintf("%v", parsedTime.UnixNano()), nil
}
// with the 2 policies , converted dsl is equivalent to
// "SELECT myColA FROM myTable WHERE myColB < 10 AND dateTime = '1561678568048000000'
// in which the time is in unix nano format
e := NewESql()
e.ProcessQueryKey(myKeyFilter, myKeyProcess)            // set up macro for key
e.ProcessQueryValue(myValueFilter, myValueProcess)      // set up macro for value
dsl, _, err := e.ConvertPretty(sql)                     // convert sql to dsl
if err == nil {
    fmt.Println(dsl)
}

ESQL support 2 kinds of pagination: FROM keyword and ES search_after.

• FROM keyword: the same as SQL syntax. Be careful, ES only support a page smaller than 10k, if your offset is large than 10k, search_after is necessary.
• search_after: Once you know the paging tokens, just feed them to Convert or ConvertPretty API in order.

Below shows an example.

// first page
sql_page1 := "SELECT * FROM myTable ORDER BY colA, colB LIMIT 10"
e := NewESql()
dsl_page1, sortFields, err := e.ConvertPretty(sql_page1)

// second page
// 1. Use FROM to retrieve the 2nd page
sql_page2_FROM := "SELECT * FROM myTable ORDER BY colA, colB LIMIT 10 FROM 10"
dsl_page2_FROM, sortFields, err := e.ConvertPretty(sql_page2_FROM)

// 2. Use search_after to retrieve the 2nd page
// we can use sortFields and the query result from page 1 to get the page tokens
sql_page2_search_after := sql_page1
page_token_colA := "123"
page_token_colB := "bbc"
dsl_page2_search_after, sortFields, err := e.ConvertPretty(sql_page2_search_after, page_colA, page_colB)

We are using elasticsearch's SQL translate API as a reference in testing. Testing contains 3 basic steps:

• using elasticsearch's SQL translate API to translate sql to dsl
• using our library to convert sql to dsl
• query local elasticsearch server with both dsls, check the results are identical

However, since ES's SQL api is still experimental, there are many features not supported well. For such queries, testing is mannual.

Features not covered yet:

• LIKE, REGEXP keyword: ES V6.5's sql api does not support regex search but only wildcard (only support shell wildcard % and _)
• some aggregations and arithmetics are tested by manual check since ES's sql api does not support them well

To run test locally:

• download elasticsearch v6.5 (optional: kibana v6.5) and unzip them
• run chmod u+x start_service.sh test.sh
• run ./start_service.sh <elasticsearch_path> <kibana_path> to start a local elasticsearch server (by default, elasticsearch listens port 9200, kibana listens port 5600)
• run python gen_test_data.py -dmi 1 1000 20 to insert 1000 documents to the local es
• run ./test.sh TestSQL to run all the test cases in /testcases/sqls.txt
• generated dsls are stored in dslsPretty.txt for reference

To customize test cases:

• modify testcases/sqls.txt
• run python gen_test_data.py -h for guides on how to insert custom data into your lcoal es
• invalid query test cases are in testcases/sqlsInvalid.txt

This project is originated from elasticsql. Table below shows the improvement.