Lexical Structure

Take the following example:

SELECT * FROM table

This query has four tokens: SELECT, *, FROM and table. The first three, namely SELECT, * and FROM are key words meaning words that have a fixed meaning in SQL. The token table is an identifier meaning it identifies (by name) an entity inside SQL such as a table (in this case), a column, etc…​

As one can see, both key words and identifiers have the same lexical structure and thus one cannot know whether a token is one or the other without knowing the SQL language; the complete list of key words is available in the reserved appendix. Do note that key words are case-insensitive meaning the previous example can be written as:

select * fRoM table;

Identifiers however are not - as Elasticsearch is case sensitive, Elasticsearch SQL uses the received value verbatim.

To help differentiate between the two, through-out the documentation the SQL key words are upper-cased a convention we find increases readability and thus recommend to others.

Identifiers can be of two types: quoted and unquoted:

SELECT ip_address FROM "hosts-*"

This query has two identifiers, ip_address and hosts-* (an index pattern). As ip_address does not clash with any key words it can be used verbatim, hosts-* on the other hand cannot as it clashes with - (minus operation) and * hence the double quotes.

Another example:

SELECT "from" FROM "<logstash-{now/d}>"

The first identifier from needs to quoted as otherwise it clashes with the FROM key word (which is case insensitive as thus can be written as from) while the second identifier using Elasticsearch Date math support in index names would have otherwise confuse the parser.

Hence why in general, especially when dealing with user input it is highly recommended to use quotes for identifiers. It adds minimal increase to your queries and in return offers clarity and disambiguation.

Elasticsearch SQL supports two kind of implicitly-typed literals: strings and numbers.

A string literal is an arbitrary number of characters bounded by single quotes ': 'Giant Robot'. To include a single quote in the string, escape it using another single quote: 'Captain EO''s Voyage'.

Numeric literals are accepted both in decimal and scientific notation with exponent marker (e or E), starting either with a digit or decimal point .:

1969    -- integer notation
3.14    -- decimal notation
.1234   -- decimal notation starting with decimal point
4E5     -- scientific notation (with exponent marker)
1.2e-3  -- scientific notation with decimal point

Numeric literals that contain a decimal point are always interpreted as being of type double. Those without are considered integer if they fit otherwise their type is long (or BIGINT in ANSI SQL types).

When dealing with arbitrary type literal, one creates the object by casting, typically, the string representation to the desired type. This can be achieved through the dedicated cast operator and functions:

123::LONG                                   -- cast 123 to a LONG
CAST('1969-05-13T12:34:56' AS TIMESTAMP)    -- cast the given string to datetime
CONVERT('10.0.0.1', IP)                     -- cast '10.0.0.1' to an IP

Do note that Elasticsearch SQL provides functions that out of the box return popular literals (like E()) or provide dedicated parsing for certain strings.

It is worth pointing out that in SQL, single quotes ' and double quotes " have different meaning and cannot be used interchangeably. Single quotes are used to declare a string literal while double quotes for identifiers.

To wit:

SELECT "first_name" 
  FROM "musicians"  
 WHERE "last_name"  
     = 'Carroll'    

A few characters that are not alphanumeric have a dedicated meaning different from that of an operator. For completeness these are specified below:

Most operators in Elasticsearch SQL have the same precedence and are left-associative. As this is done at parsing time, parenthesis need to be used to enforce a different precedence.

The following table indicates the supported operators and their precedence (highest to lowest);

Elasticsearch SQL allows comments which are sequence of characters ignored by the parsers.

Two styles are supported:

-- single line comment
/* multi
   line
   comment
   that supports /* nested comments */
   */