kvql

A SQL-like query language on general Key-Value DB

Syntax

Basic Types:

Number: number such as integer or float

String: string around by ', ", \`,

Boolean: true or false

Select Statement:

SelectStmt ::= "SELECT" Fields "WHERE" WhereConditions ("ORDER" "BY" OrderByFields)? ("GROUP" "BY" GroupByFields)? ("LIMIT" LimitParameter)?

Fields ::= Field (, Field)* |
           "*"

Field ::= Expression ("AS" FieldName)?

FieldName ::= String

OrderByFields ::= OrderByField (, OrderByField)*

OrderByField ::= FieldName ("ASC" | "DESC")*

GroupByFields ::= FieldName (, FieldName)*

LimitParameter ::= Number "," Number |
                   Number

WhereConditions ::= "!"? Expression

Expression ::= "("? BinaryExpression | UnaryExpression ")"?

UnaryExpression ::= KeyValueField | String | Number | Boolean | FunctionCall | FieldName

BinaryExpression ::= Expression Operator Expression |
                     Expression "BETWEEN" Expression "AND" Expression |
                     Expression "IN" "(" Expression (, Expression)* ")" |
                     Expression "IN" FunctionCall

Operator ::= MathOperator | CompareOperator | AndOrOperator

AndOrOperator ::= "&" | "|" | "AND" | "OR"

MathOperator ::= "+" | "-" | "*" | "/"

CompareOperator ::= "=" | "!=" | "^=" | "~=" | ">" | ">=" | "<" | "<="

KeyValueField ::= "KEY" | "VALUE"

FunctionCall ::= FunctionName "(" FunctionArgs ")" |
                 FunctionName "(" FunctionArgs ")" FieldAccessExpression*

FunctionName ::= String

FunctionArgs ::= FunctionArg ("," FunctionArg)*

FunctionArg ::= Expression

FieldAccessExpression ::= "[" String "]" |
                          "[" Number "]"

Put Statement:

PutStmt ::= "PUT" KVPair (, KVPair)*
KVPair ::= "(" Expression, Expression ")"

Delete Statement:

DeleteStmt ::= "DELETE" "WHERE" WhereConditions ("LIMIT" LimitParameter)?

Features:

1. Scan ranger optimize: EmptyResult, PrefixScan, RangeScan, MultiGet
2. Plan support Volcano model and Batch model
3. Expression constant folding
4. Support scalar function and aggregate function
5. Support hash aggregate plan
6. Support JSON and field access expression

Known User

• c4pt0r/tcli CLI tool for TiKV

Examples:

# Simple query, get all the key-value pairs with key prefix 'k'
select * where key ^= 'k'

# Projection and complex condition
select key, int(value) + 1 where key in ('k1', 'k2', 'k3') & is_int(value)

# Aggregation query
select count(1), sum(int(value)) as sum, substr(key, 0, 2) as kprefix where key between 'k' and 'l' group by kprefix order by sum desc

# JSON access
select key, json(value)['x']['y'] where key ^= 'k' & int(json(value)['test']) >= 1
select key, json(value)['list'][1] where key ^= 'k'

# Filter by field name defined in select statement
select key, int(value) as f1 where f1 > 10
select key, split(value) as f1 where 'a' in f1
select key, value, l2_distance(list(1,2,3,4), json(value)) as l2_dis where key ^= 'embedding_json' & l2_dis > 0.6 order by l2_dis desc limit 5

# Put data
put ('k1', 'v1'), ('k2', upper('v' + key))

# Delete data by filter and limit delete rows
delete where key ^= 'prefix' and value ~= '^val_' limit 10
delete where key in ('k1', 'k2', 'k3')

How to use this library

A full example:

https://github.com/c4pt0r/kvql/blob/master/examples/memkv/memkv.go

To get better error report, you can conver the error to QueryBinder and set the origin query like below:

...
opt := kvql.NewOptimizer(query)
plan, err := opt.BuildPlan(storage)
if err != nil {
	if qerr, ok := err.(kvql.QueryBinder); ok {
		qerr.BindQuery(query)
	}
	fmt.Printf("Error: %s\n", err.Error())
}
...

After bind the query to error it will output error result like:

padding                    query
v-----vv--------------------------------------------v
Error: select * where key ^= 'asdf' and val ^= 'test'          < query line
                                        ^--                    < error position
       Syntax Error: ^= operator with invalid left expression  < error message

About padding: user can use kvql.DefaultErrorPadding to change the default left padding spaces. Or can use kvql.QueryBinder.SetPadding function to change specify error's padding. The default padding is 7 space characters (length of Error: ).

If you want to display the plan tree, like EXPLAIN statement in SQL, the kvql.FinalPlan.Explain function will return the plan tree in a string list, you can use below code to format the explain output:

...
opt := kvql.NewOptimizer(query)
plan, err := opt.BuildPlan(storage)
if err != nil {
	fatal(err)
}

output := ""
for i, plan := range plan.Explain() {
	padding := ""
	for x := 0; x < i*3; x++ {
		padding += " "
	}
	if i == 0 {
		output += fmt.Sprintf("%s%s\n", padding, plan)
	} else {
		output += fmt.Sprintf("%s`-%s\n", padding, plan)
	}
}
fmt.Println(output)

Operators and Functions

Operators

Conparation operators

• =: bytes level equals
• !=: bytes level not equals
• ^=: prefix match
• ~=: regexp match
• >: number or string greater than
• >=: number or string greater or equals than
• <: number or string less than
• <=: number or string less or equals than
• BETWEEN x AND y: great or equals than x and less or equals than y
• IN (...): in list followed by in operator

Logical operators

• &, AND: logical and
• |, OR: logical or
• !: logical not

Math operators

• +: number add or string concate
• -: number subtraction
• *: number multiply
• /: number division

Scalar Functions

Function	Description
lower(value: str): str	convert value string into lower case
upper(value: str): str	convert value string into upper case
int(value: any): int	convert value into integer, if cannot convert to integer just return error
float(value: any): float	convert value into float, if cannot convert to float just return error
str(value: any): str	convert value into string
strlen(value: any): int	convert value into string and then calculate string length
is_int(value: any): bool	return is value can be converted into integer
is_float(value: any): bool	return is value can be converted into float
substr(value: str, start: int, end: int): str	return substring of value from start position to end position
split(value: str, spliter: str): list	split value into a string list by spliter string
list(elem1: any, elem2: any...): list	convert many elements into a list, list elements' type must be same, the list type support int, str, float types
float_list(elem1: float, elem2: float...): list	convert many float elements into a list
flist(elem1: float, elem2: float...): list	same as float_list
int_list(elem1: int, elem2: int...): list	convert many integer elements into a list
ilist(elem1: int, elem2: int...): list	same as int_list
len(value: list): int	return value list length
l2_distance(left: list, right: list): float	calculate l2 distance of two list
cosine_distance(left: list, right: list): float	calculate cosine distance of two list
json(value: str): json	parse string value into json type
join(seperator: str, val1: any, val2: any...): str	join values by seperator

Aggregation Functions

Function	Description
count(value: int): int	Count value by group
sum(value: int): int	Sum value by group
avg(value: int): int	Calculate average value by group
min(value: int): int	Find the minimum value by group
max(value: int): int	Find the maxmum value by group
quantile(value: float, percent: float): float	Calculate the Quantile by group
json_arrayagg(value: any): string	Aggregate all values into a JSON array
group_concat(value: any, seperator: str): string	Join all values into a string by seperator