prometheus/promql/generated_parser.y
Tobias Guggenmos 3d6cf1c289 PromQL: Make parser completely generated (#6548)
Signed-off-by: Tobias Guggenmos <tguggenm@redhat.com>
2020-01-08 11:04:47 +00:00

646 lines
21 KiB
Plaintext

// Copyright 2019 The Prometheus Authors
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
// http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.
%{
package promql
import (
"math"
"sort"
"strconv"
"time"
"github.com/prometheus/prometheus/pkg/labels"
"github.com/prometheus/prometheus/pkg/value"
)
%}
%union {
node Node
item Item
matchers []*labels.Matcher
matcher *labels.Matcher
label labels.Label
labels labels.Labels
strings []string
series []sequenceValue
uint uint64
float float64
string string
duration time.Duration
}
%token <item>
ASSIGN
BLANK
COLON
COMMA
COMMENT
DURATION
EOF
ERROR
IDENTIFIER
LEFT_BRACE
LEFT_BRACKET
LEFT_PAREN
METRIC_IDENTIFIER
NUMBER
RIGHT_BRACE
RIGHT_BRACKET
RIGHT_PAREN
SEMICOLON
SPACE
STRING
TIMES
// Operators.
%token operatorsStart
%token <item>
ADD
DIV
EQL
EQL_REGEX
GTE
GTR
LAND
LOR
LSS
LTE
LUNLESS
MOD
MUL
NEQ
NEQ_REGEX
POW
SUB
%token operatorsEnd
// Aggregators.
%token aggregatorsStart
%token <item>
AVG
BOTTOMK
COUNT
COUNT_VALUES
MAX
MIN
QUANTILE
STDDEV
STDVAR
SUM
TOPK
%token aggregatorsEnd
// Keywords.
%token keywordsStart
%token <item>
BOOL
BY
GROUP_LEFT
GROUP_RIGHT
IGNORING
OFFSET
ON
WITHOUT
%token keywordsEnd
// Start symbols for the generated parser.
%token startSymbolsStart
%token
START_METRIC
START_SERIES_DESCRIPTION
START_EXPRESSION
START_METRIC_SELECTOR
%token startSymbolsEnd
// Type definitions for grammar rules.
%type <matchers> label_match_list label_matchers
%type <matcher> label_matcher
%type <item> aggregate_op grouping_label match_op maybe_label metric_identifier unary_op
%type <labels> label_set label_set_list metric
%type <label> label_set_item
%type <strings> grouping_label_list grouping_labels maybe_grouping_labels
%type <series> series_item series_values
%type <uint> uint
%type <float> number series_value signed_number
%type <node> aggregate_expr aggregate_modifier bin_modifier binary_expr bool_modifier expr function_call function_call_args function_call_body group_modifiers matrix_selector number_literal offset_expr on_or_ignoring paren_expr string_literal subquery_expr unary_expr vector_selector
%type <string> string
%type <duration> duration maybe_duration
%start start
// Operators are listed with increasing precedence.
%left LOR
%left LAND LUNLESS
%left EQL GTE GTR LSS LTE NEQ
%left ADD SUB
%left MUL DIV MOD
%right POW
// Fake token for giving unary expressions maximal precendence
%token PREC_MAX
%nonassoc PREC_MAX
// Offset modifiers do not have associativity.
%nonassoc OFFSET
// This ensures that it is always attempted to parse range or subquery selectors when a left
// bracket is encountered.
%right LEFT_BRACKET
%%
start :
START_METRIC metric
{ yylex.(*parser).generatedParserResult = $2 }
| START_SERIES_DESCRIPTION series_description
| START_EXPRESSION /* empty */ EOF
{ yylex.(*parser).errorf("no expression found in input")}
| START_EXPRESSION expr
{ yylex.(*parser).generatedParserResult = $2 }
| START_METRIC_SELECTOR vector_selector
{ yylex.(*parser).generatedParserResult = $2 }
| start EOF
| error /* If none of the more detailed error messages are triggered, we fall back to this. */
{ yylex.(*parser).unexpected("","") }
;
expr :
aggregate_expr
| binary_expr
| function_call
| matrix_selector
| number_literal
| offset_expr
| paren_expr
| string_literal
| subquery_expr
| unary_expr
| vector_selector
;
/*
* Aggregations.
*/
aggregate_expr : aggregate_op aggregate_modifier function_call_body
{ $$ = yylex.(*parser).newAggregateExpr($1, $2, $3) }
| aggregate_op function_call_body aggregate_modifier
{ $$ = yylex.(*parser).newAggregateExpr($1, $3, $2) }
| aggregate_op function_call_body
{ $$ = yylex.(*parser).newAggregateExpr($1, &AggregateExpr{}, $2) }
| aggregate_op error
{ yylex.(*parser).unexpected("aggregation","") }
;
aggregate_modifier:
BY grouping_labels
{
$$ = &AggregateExpr{
Grouping: $2,
}
}
| WITHOUT grouping_labels
{
$$ = &AggregateExpr{
Grouping: $2,
Without: true,
}
}
;
/*
* Binary expressions.
*/
// Operator precedence only works if each of those is listed separately.
binary_expr : expr ADD bin_modifier expr { $$ = yylex.(*parser).newBinaryExpression($1, $2, $3, $4) }
| expr DIV bin_modifier expr { $$ = yylex.(*parser).newBinaryExpression($1, $2, $3, $4) }
| expr EQL bin_modifier expr { $$ = yylex.(*parser).newBinaryExpression($1, $2, $3, $4) }
| expr GTE bin_modifier expr { $$ = yylex.(*parser).newBinaryExpression($1, $2, $3, $4) }
| expr GTR bin_modifier expr { $$ = yylex.(*parser).newBinaryExpression($1, $2, $3, $4) }
| expr LAND bin_modifier expr { $$ = yylex.(*parser).newBinaryExpression($1, $2, $3, $4) }
| expr LOR bin_modifier expr { $$ = yylex.(*parser).newBinaryExpression($1, $2, $3, $4) }
| expr LSS bin_modifier expr { $$ = yylex.(*parser).newBinaryExpression($1, $2, $3, $4) }
| expr LTE bin_modifier expr { $$ = yylex.(*parser).newBinaryExpression($1, $2, $3, $4) }
| expr LUNLESS bin_modifier expr { $$ = yylex.(*parser).newBinaryExpression($1, $2, $3, $4) }
| expr MOD bin_modifier expr { $$ = yylex.(*parser).newBinaryExpression($1, $2, $3, $4) }
| expr MUL bin_modifier expr { $$ = yylex.(*parser).newBinaryExpression($1, $2, $3, $4) }
| expr NEQ bin_modifier expr { $$ = yylex.(*parser).newBinaryExpression($1, $2, $3, $4) }
| expr POW bin_modifier expr { $$ = yylex.(*parser).newBinaryExpression($1, $2, $3, $4) }
| expr SUB bin_modifier expr { $$ = yylex.(*parser).newBinaryExpression($1, $2, $3, $4) }
;
// Using left recursion for the modifier rules, helps to keep the parser stack small and
// reduces allocations
bin_modifier : group_modifiers;
bool_modifier : /* empty */
{ $$ = &BinaryExpr{
VectorMatching: &VectorMatching{Card: CardOneToOne},
}
}
| BOOL
{ $$ = &BinaryExpr{
VectorMatching: &VectorMatching{Card: CardOneToOne},
ReturnBool: true,
}
}
;
on_or_ignoring : bool_modifier IGNORING grouping_labels
{
$$ = $1
$$.(*BinaryExpr).VectorMatching.MatchingLabels = $3
}
| bool_modifier ON grouping_labels
{
$$ = $1
$$.(*BinaryExpr).VectorMatching.MatchingLabels = $3
$$.(*BinaryExpr).VectorMatching.On = true
}
;
group_modifiers: bool_modifier /* empty */
| on_or_ignoring /* empty */
| on_or_ignoring GROUP_LEFT maybe_grouping_labels
{
$$ = $1
$$.(*BinaryExpr).VectorMatching.Card = CardManyToOne
$$.(*BinaryExpr).VectorMatching.Include = $3
}
| on_or_ignoring GROUP_RIGHT maybe_grouping_labels
{
$$ = $1
$$.(*BinaryExpr).VectorMatching.Card = CardOneToMany
$$.(*BinaryExpr).VectorMatching.Include = $3
}
;
grouping_labels : LEFT_PAREN grouping_label_list RIGHT_PAREN
{ $$ = $2 }
| LEFT_PAREN grouping_label_list COMMA RIGHT_PAREN
{ $$ = $2 }
| LEFT_PAREN RIGHT_PAREN
{ $$ = []string{} }
| error
{ yylex.(*parser).unexpected("grouping opts", "\"(\"") }
;
grouping_label_list:
grouping_label_list COMMA grouping_label
{ $$ = append($1, $3.Val) }
| grouping_label
{ $$ = []string{$1.Val} }
| grouping_label_list error
{ yylex.(*parser).unexpected("grouping opts", "\",\" or \")\"") }
;
grouping_label : maybe_label
{
if !isLabel($1.Val) {
yylex.(*parser).unexpected("grouping opts", "label")
}
$$ = $1
}
| error
{ yylex.(*parser).unexpected("grouping opts", "label") }
;
/*
* Function calls.
*/
function_call : IDENTIFIER function_call_body
{
fn, exist := getFunction($1.Val)
if !exist{
yylex.(*parser).errorf("unknown function with name %q", $1.Val)
}
$$ = &Call{
Func: fn,
Args: $2.(Expressions),
}
}
;
function_call_body: LEFT_PAREN function_call_args RIGHT_PAREN
{ $$ = $2 }
| LEFT_PAREN RIGHT_PAREN
{$$ = Expressions{}}
;
function_call_args: function_call_args COMMA expr
{ $$ = append($1.(Expressions), $3.(Expr)) }
| expr
{ $$ = Expressions{$1.(Expr)} }
;
/*
* Expressions inside parentheses.
*/
paren_expr : LEFT_PAREN expr RIGHT_PAREN
{ $$ = &ParenExpr{Expr: $2.(Expr)} }
;
/*
* Offset modifiers.
*/
offset_expr: expr OFFSET duration
{
yylex.(*parser).addOffset($1, $3)
$$ = $1
}
| expr OFFSET error
{ yylex.(*parser).unexpected("offset", "duration") }
;
/*
* Subquery and range selectors.
*/
matrix_selector : expr LEFT_BRACKET duration RIGHT_BRACKET
{
vs, ok := $1.(*VectorSelector)
if !ok{
yylex.(*parser).errorf("ranges only allowed for vector selectors")
}
if vs.Offset != 0{
yylex.(*parser).errorf("no offset modifiers allowed before range")
}
$$ = &MatrixSelector{
Name: vs.Name,
Offset: vs.Offset,
LabelMatchers: vs.LabelMatchers,
Range: $3,
}
}
;
subquery_expr : expr LEFT_BRACKET duration COLON maybe_duration RIGHT_BRACKET
{
$$ = &SubqueryExpr{
Expr: $1.(Expr),
Range: $3,
Step: $5,
}
}
| expr LEFT_BRACKET duration COLON duration error
{ yylex.(*parser).unexpected("subquery selector", "\"]\"") }
| expr LEFT_BRACKET duration COLON error
{ yylex.(*parser).unexpected("subquery selector", "duration or \"]\"") }
| expr LEFT_BRACKET duration error
{ yylex.(*parser).unexpected("subquery or range", "\":\" or \"]\"") }
| expr LEFT_BRACKET error
{ yylex.(*parser).unexpected("subquery selector", "duration") }
;
/*
* Unary expressions.
*/
unary_expr :
/* gives the rule the same prec as POW. This has the effect that unary opts are always evaluated with highest precedence */
unary_op expr %prec PREC_MAX
{
if nl, ok := $2.(*NumberLiteral); ok {
if $1.Typ == SUB {
nl.Val *= -1
}
$$ = nl
} else {
$$ = &UnaryExpr{Op: $1.Typ, Expr: $2.(Expr)}
}
}
;
/*
* Vector selectors.
*/
vector_selector: metric_identifier label_matchers
{ $$ = yylex.(*parser).newVectorSelector($1.Val, $2) }
| metric_identifier
{ $$ = yylex.(*parser).newVectorSelector($1.Val, nil) }
| label_matchers
{ $$ = yylex.(*parser).newVectorSelector("", $1) }
;
label_matchers : LEFT_BRACE label_match_list RIGHT_BRACE
{ $$ = $2 }
| LEFT_BRACE label_match_list COMMA RIGHT_BRACE
{ $$ = $2 }
| LEFT_BRACE RIGHT_BRACE
{ $$ = []*labels.Matcher{} }
;
label_match_list: label_match_list COMMA label_matcher
{ $$ = append($1, $3)}
| label_matcher
{ $$ = []*labels.Matcher{$1}}
| label_match_list error
{ yylex.(*parser).unexpected("label matching", "\",\" or \"}\"") }
;
label_matcher : IDENTIFIER match_op STRING
{ $$ = yylex.(*parser).newLabelMatcher($1, $2, $3) }
| IDENTIFIER match_op error
{ yylex.(*parser).unexpected("label matching", "string")}
| IDENTIFIER error
{ yylex.(*parser).unexpected("label matching", "label matching operator") }
| error
{ yylex.(*parser).unexpected("label matching", "identifier or \"}\"")}
;
/*
* Metric descriptions.
*/
metric : metric_identifier label_set
{ $$ = append($2, labels.Label{Name: labels.MetricName, Value: $1.Val}); sort.Sort($$) }
| label_set
{$$ = $1}
;
metric_identifier: METRIC_IDENTIFIER | IDENTIFIER;
label_set : LEFT_BRACE label_set_list RIGHT_BRACE
{ $$ = labels.New($2...) }
| LEFT_BRACE label_set_list COMMA RIGHT_BRACE
{ $$ = labels.New($2...) }
| LEFT_BRACE RIGHT_BRACE
{ $$ = labels.New() }
| /* empty */
{ $$ = labels.New() }
;
label_set_list : label_set_list COMMA label_set_item
{ $$ = append($1, $3) }
| label_set_item
{ $$ = []labels.Label{$1} }
| label_set_list error
{ yylex.(*parser).unexpected("label set", "\",\" or \"}\"", ) }
;
label_set_item : IDENTIFIER EQL STRING
{ $$ = labels.Label{Name: $1.Val, Value: yylex.(*parser).unquoteString($3.Val) } }
| IDENTIFIER EQL error
{ yylex.(*parser).unexpected("label set", "string")}
| IDENTIFIER error
{ yylex.(*parser).unexpected("label set", "\"=\"")}
| error
{ yylex.(*parser).unexpected("label set", "identifier or \"}\"") }
;
/*
* Series descriptions (only used by unit tests).
*/
series_description: metric series_values
{
yylex.(*parser).generatedParserResult = &seriesDescription{
labels: $1,
values: $2,
}
}
;
series_values : /*empty*/
{ $$ = []sequenceValue{} }
| series_values SPACE series_item
{ $$ = append($1, $3...) }
| series_values SPACE
{ $$ = $1 }
| error
{ yylex.(*parser).unexpected("series values", "") }
;
series_item : BLANK
{ $$ = []sequenceValue{{omitted: true}}}
| BLANK TIMES uint
{
$$ = []sequenceValue{}
for i:=uint64(0); i < $3; i++{
$$ = append($$, sequenceValue{omitted: true})
}
}
| series_value
{ $$ = []sequenceValue{{value: $1}}}
| series_value TIMES uint
{
$$ = []sequenceValue{}
for i:=uint64(0); i <= $3; i++{
$$ = append($$, sequenceValue{value: $1})
}
}
| series_value signed_number TIMES uint
{
$$ = []sequenceValue{}
for i:=uint64(0); i <= $4; i++{
$$ = append($$, sequenceValue{value: $1})
$1 += $2
}
}
;
series_value : IDENTIFIER
{
if $1.Val != "stale" {
yylex.(*parser).unexpected("series values", "number or \"stale\"")
}
$$ = math.Float64frombits(value.StaleNaN)
}
| number
| signed_number
;
/*
* Keyword lists.
*/
aggregate_op : AVG | BOTTOMK | COUNT | COUNT_VALUES | MAX | MIN | QUANTILE | STDDEV | STDVAR | SUM | TOPK ;
// inside of grouping options label names can be recognized as keywords by the lexer. This is a list of keywords that could also be a label name.
maybe_label : AVG | BOOL | BOTTOMK | BY | COUNT | COUNT_VALUES | GROUP_LEFT | GROUP_RIGHT | IDENTIFIER | IGNORING | LAND | LOR | LUNLESS | MAX | METRIC_IDENTIFIER | MIN | OFFSET | ON | QUANTILE | STDDEV | STDVAR | SUM | TOPK;
unary_op : ADD | SUB;
match_op : EQL | NEQ | EQL_REGEX | NEQ_REGEX ;
/*
* Literals.
*/
number_literal : number { $$ = &NumberLiteral{$1}} ;
number : NUMBER { $$ = yylex.(*parser).number($1.Val) } ;
signed_number : ADD number { $$ = $2 }
| SUB number { $$ = -$2 }
;
uint : NUMBER
{
var err error
$$, err = strconv.ParseUint($1.Val, 10, 64)
if err != nil {
yylex.(*parser).errorf("invalid repetition in series values: %s", err)
}
}
;
duration : DURATION
{
var err error
$$, err = parseDuration($1.Val)
if err != nil {
yylex.(*parser).error(err)
}
}
;
string_literal : string { $$ = &StringLiteral{$1}} ;
string : STRING { $$ = yylex.(*parser).unquoteString($1.Val) } ;
/*
* Wrappers for optional arguments.
*/
maybe_duration : /* empty */
{$$ = 0}
| duration
;
maybe_grouping_labels: /* empty */ { $$ = nil }
| grouping_labels
;
%%