mirror of
https://git.mirrors.martin98.com/https://github.com/SigNoz/signoz
synced 2025-08-17 08:05:57 +08:00
chore: move clickhousereader filter suggestion methods to dedicated file (#6061)
This commit is contained in:
Raj Kamal Singh
GitHub
parent
0feab5aa93
commit
708158f50f
270
pkg/query-service/app/clickhouseReader/filter_suggestions.go
Normal file
270
pkg/query-service/app/clickhouseReader/filter_suggestions.go
Normal file
@ -0,0 +1,270 @@
|
||||
// Clickhouse reader methods for powering QB filter suggestions
|
||||
package clickhouseReader
|
||||
|
||||
import (
|
||||
"context"
|
||||
"database/sql"
|
||||
"fmt"
|
||||
"slices"
|
||||
"strings"
|
||||
|
||||
"go.signoz.io/signoz/pkg/query-service/model"
|
||||
v3 "go.signoz.io/signoz/pkg/query-service/model/v3"
|
||||
"go.uber.org/zap"
|
||||
)
|
||||
|
||||
func (r *ClickHouseReader) GetQBFilterSuggestionsForLogs(
|
||||
ctx context.Context,
|
||||
req *v3.QBFilterSuggestionsRequest,
|
||||
) (*v3.QBFilterSuggestionsResponse, *model.ApiError) {
|
||||
suggestions := v3.QBFilterSuggestionsResponse{
|
||||
AttributeKeys: []v3.AttributeKey{},
|
||||
ExampleQueries: []v3.FilterSet{},
|
||||
}
|
||||
|
||||
// Use existing autocomplete logic for generating attribute suggestions
|
||||
attribKeysResp, err := r.GetLogAttributeKeys(
|
||||
ctx, &v3.FilterAttributeKeyRequest{
|
||||
SearchText: req.SearchText,
|
||||
DataSource: v3.DataSourceLogs,
|
||||
Limit: int(req.AttributesLimit),
|
||||
})
|
||||
if err != nil {
|
||||
return nil, model.InternalError(fmt.Errorf("couldn't get attribute keys: %w", err))
|
||||
}
|
||||
|
||||
suggestions.AttributeKeys = attribKeysResp.AttributeKeys
|
||||
|
||||
// Rank suggested attributes
|
||||
slices.SortFunc(suggestions.AttributeKeys, func(a v3.AttributeKey, b v3.AttributeKey) int {
|
||||
|
||||
// Higher score => higher rank
|
||||
attribKeyScore := func(a v3.AttributeKey) int {
|
||||
|
||||
// Scoring criteria is expected to get more sophisticated in follow up changes
|
||||
if a.Type == v3.AttributeKeyTypeResource {
|
||||
return 2
|
||||
}
|
||||
|
||||
if a.Type == v3.AttributeKeyTypeTag {
|
||||
return 1
|
||||
}
|
||||
|
||||
return 0
|
||||
}
|
||||
|
||||
// To sort in descending order of score the return value must be negative when a > b
|
||||
return attribKeyScore(b) - attribKeyScore(a)
|
||||
})
|
||||
|
||||
// Put together suggested example queries.
|
||||
|
||||
newExampleQuery := func() v3.FilterSet {
|
||||
// Include existing filter in example query if specified.
|
||||
if req.ExistingFilter != nil {
|
||||
return *req.ExistingFilter
|
||||
}
|
||||
|
||||
return v3.FilterSet{
|
||||
Operator: "AND",
|
||||
Items: []v3.FilterItem{},
|
||||
}
|
||||
}
|
||||
|
||||
// Suggest example queries for top suggested log attributes and resource attributes
|
||||
exampleAttribs := []v3.AttributeKey{}
|
||||
for _, attrib := range suggestions.AttributeKeys {
|
||||
isAttributeOrResource := slices.Contains([]v3.AttributeKeyType{
|
||||
v3.AttributeKeyTypeResource, v3.AttributeKeyTypeTag,
|
||||
}, attrib.Type)
|
||||
|
||||
isNumOrStringType := slices.Contains([]v3.AttributeKeyDataType{
|
||||
v3.AttributeKeyDataTypeInt64, v3.AttributeKeyDataTypeFloat64, v3.AttributeKeyDataTypeString,
|
||||
}, attrib.DataType)
|
||||
|
||||
if isAttributeOrResource && isNumOrStringType {
|
||||
exampleAttribs = append(exampleAttribs, attrib)
|
||||
}
|
||||
|
||||
if len(exampleAttribs) >= int(req.ExamplesLimit) {
|
||||
break
|
||||
}
|
||||
}
|
||||
|
||||
if len(exampleAttribs) > 0 {
|
||||
exampleAttribValues, err := r.getValuesForLogAttributes(
|
||||
ctx, exampleAttribs, req.ExamplesLimit,
|
||||
)
|
||||
if err != nil {
|
||||
// Do not fail the entire request if only example query generation fails
|
||||
zap.L().Error("could not find attribute values for creating example query", zap.Error(err))
|
||||
} else {
|
||||
|
||||
// add example queries for as many attributes as possible.
|
||||
// suggest 1st value for 1st attrib, followed by 1st value for second attrib and so on
|
||||
// and if there is still room, suggest 2nd value for 1st attrib, 2nd value for 2nd attrib and so on
|
||||
for valueIdx := 0; valueIdx < int(req.ExamplesLimit); valueIdx++ {
|
||||
for attrIdx, attr := range exampleAttribs {
|
||||
needMoreExamples := len(suggestions.ExampleQueries) < int(req.ExamplesLimit)
|
||||
|
||||
if needMoreExamples && valueIdx < len(exampleAttribValues[attrIdx]) {
|
||||
exampleQuery := newExampleQuery()
|
||||
exampleQuery.Items = append(exampleQuery.Items, v3.FilterItem{
|
||||
Key: attr,
|
||||
Operator: "=",
|
||||
Value: exampleAttribValues[attrIdx][valueIdx],
|
||||
})
|
||||
|
||||
suggestions.ExampleQueries = append(
|
||||
suggestions.ExampleQueries, exampleQuery,
|
||||
)
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
// Suggest static example queries for standard log attributes if needed.
|
||||
if len(suggestions.ExampleQueries) < int(req.ExamplesLimit) {
|
||||
exampleQuery := newExampleQuery()
|
||||
exampleQuery.Items = append(exampleQuery.Items, v3.FilterItem{
|
||||
Key: v3.AttributeKey{
|
||||
Key: "body",
|
||||
DataType: v3.AttributeKeyDataTypeString,
|
||||
Type: v3.AttributeKeyTypeUnspecified,
|
||||
IsColumn: true,
|
||||
},
|
||||
Operator: "contains",
|
||||
Value: "error",
|
||||
})
|
||||
suggestions.ExampleQueries = append(suggestions.ExampleQueries, exampleQuery)
|
||||
}
|
||||
|
||||
return &suggestions, nil
|
||||
}
|
||||
|
||||
// Get up to `limit` values seen for each attribute in `attributes`
|
||||
// Returns a slice of slices where the ith slice has values for ith entry in `attributes`
|
||||
func (r *ClickHouseReader) getValuesForLogAttributes(
|
||||
ctx context.Context, attributes []v3.AttributeKey, limit uint64,
|
||||
) ([][]any, *model.ApiError) {
|
||||
/*
|
||||
The query used here needs to be as cheap as possible, and while uncommon, it is possible for
|
||||
a tag to have 100s of millions of values (eg: message, request_id)
|
||||
|
||||
Construct a query to UNION the result of querying first `limit` values for each attribute. For example:
|
||||
```
|
||||
select * from (
|
||||
(
|
||||
select tagKey, stringTagValue, int64TagValue, float64TagValue
|
||||
from signoz_logs.distributed_tag_attributes
|
||||
where tagKey = $1 and (
|
||||
stringTagValue != '' or int64TagValue is not null or float64TagValue is not null
|
||||
)
|
||||
limit 2
|
||||
) UNION DISTINCT (
|
||||
select tagKey, stringTagValue, int64TagValue, float64TagValue
|
||||
from signoz_logs.distributed_tag_attributes
|
||||
where tagKey = $2 and (
|
||||
stringTagValue != '' or int64TagValue is not null or float64TagValue is not null
|
||||
)
|
||||
limit 2
|
||||
)
|
||||
) settings max_threads=2
|
||||
```
|
||||
Since tag_attributes table uses ReplacingMergeTree, the values would be distinct and no order by
|
||||
is being used to ensure the `limit` clause minimizes the amount of data scanned.
|
||||
|
||||
This query scanned ~30k rows per attribute on fiscalnote-v2 for attributes like `message` and `time`
|
||||
that had >~110M values each
|
||||
*/
|
||||
|
||||
if len(attributes) > 10 {
|
||||
zap.L().Error(
|
||||
"log attribute values requested for too many attributes. This can lead to slow and costly queries",
|
||||
zap.Int("count", len(attributes)),
|
||||
)
|
||||
attributes = attributes[:10]
|
||||
}
|
||||
|
||||
tagQueries := []string{}
|
||||
tagKeyQueryArgs := []any{}
|
||||
for idx, attrib := range attributes {
|
||||
tagQueries = append(tagQueries, fmt.Sprintf(`(
|
||||
select tagKey, stringTagValue, int64TagValue, float64TagValue
|
||||
from %s.%s
|
||||
where tagKey = $%d and (
|
||||
stringTagValue != '' or int64TagValue is not null or float64TagValue is not null
|
||||
)
|
||||
limit %d
|
||||
)`, r.logsDB, r.logsTagAttributeTable, idx+1, limit))
|
||||
|
||||
tagKeyQueryArgs = append(tagKeyQueryArgs, attrib.Key)
|
||||
}
|
||||
|
||||
query := fmt.Sprintf(`select * from (
|
||||
%s
|
||||
) settings max_threads=2`, strings.Join(tagQueries, " UNION DISTINCT "))
|
||||
|
||||
rows, err := r.db.Query(ctx, query, tagKeyQueryArgs...)
|
||||
if err != nil {
|
||||
zap.L().Error("couldn't query attrib values for suggestions", zap.Error(err))
|
||||
return nil, model.InternalError(fmt.Errorf(
|
||||
"couldn't query attrib values for suggestions: %w", err,
|
||||
))
|
||||
}
|
||||
defer rows.Close()
|
||||
|
||||
result := make([][]any, len(attributes))
|
||||
|
||||
// Helper for getting hold of the result slice to append to for each scanned row
|
||||
resultIdxForAttrib := func(key string, dataType v3.AttributeKeyDataType) int {
|
||||
return slices.IndexFunc(attributes, func(attrib v3.AttributeKey) bool {
|
||||
return attrib.Key == key && attrib.DataType == dataType
|
||||
})
|
||||
}
|
||||
|
||||
// Scan rows and append to result
|
||||
for rows.Next() {
|
||||
var tagKey string
|
||||
var stringValue string
|
||||
var float64Value sql.NullFloat64
|
||||
var int64Value sql.NullInt64
|
||||
|
||||
err := rows.Scan(
|
||||
&tagKey, &stringValue, &int64Value, &float64Value,
|
||||
)
|
||||
if err != nil {
|
||||
return nil, model.InternalError(fmt.Errorf(
|
||||
"couldn't scan attrib value rows: %w", err,
|
||||
))
|
||||
}
|
||||
|
||||
if len(stringValue) > 0 {
|
||||
attrResultIdx := resultIdxForAttrib(tagKey, v3.AttributeKeyDataTypeString)
|
||||
if attrResultIdx >= 0 {
|
||||
result[attrResultIdx] = append(result[attrResultIdx], stringValue)
|
||||
}
|
||||
|
||||
} else if int64Value.Valid {
|
||||
attrResultIdx := resultIdxForAttrib(tagKey, v3.AttributeKeyDataTypeInt64)
|
||||
if attrResultIdx >= 0 {
|
||||
result[attrResultIdx] = append(result[attrResultIdx], int64Value.Int64)
|
||||
}
|
||||
|
||||
} else if float64Value.Valid {
|
||||
attrResultIdx := resultIdxForAttrib(tagKey, v3.AttributeKeyDataTypeFloat64)
|
||||
if attrResultIdx >= 0 {
|
||||
result[attrResultIdx] = append(result[attrResultIdx], float64Value.Float64)
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
if err := rows.Err(); err != nil {
|
||||
return nil, model.InternalError(fmt.Errorf(
|
||||
"couldn't scan attrib value rows: %w", err,
|
||||
))
|
||||
}
|
||||
|
||||
return result, nil
|
||||
}
|
||||
@ -10,7 +10,6 @@ import (
|
||||
"os"
|
||||
"reflect"
|
||||
"regexp"
|
||||
"slices"
|
||||
"sort"
|
||||
"strconv"
|
||||
"strings"
|
||||
@ -4069,262 +4068,6 @@ func (r *ClickHouseReader) GetLogAttributeValues(ctx context.Context, req *v3.Fi
|
||||
|
||||
}
|
||||
|
||||
func (r *ClickHouseReader) GetQBFilterSuggestionsForLogs(
|
||||
ctx context.Context,
|
||||
req *v3.QBFilterSuggestionsRequest,
|
||||
) (*v3.QBFilterSuggestionsResponse, *model.ApiError) {
|
||||
suggestions := v3.QBFilterSuggestionsResponse{
|
||||
AttributeKeys: []v3.AttributeKey{},
|
||||
ExampleQueries: []v3.FilterSet{},
|
||||
}
|
||||
|
||||
// Use existing autocomplete logic for generating attribute suggestions
|
||||
attribKeysResp, err := r.GetLogAttributeKeys(
|
||||
ctx, &v3.FilterAttributeKeyRequest{
|
||||
SearchText: req.SearchText,
|
||||
DataSource: v3.DataSourceLogs,
|
||||
Limit: int(req.AttributesLimit),
|
||||
})
|
||||
if err != nil {
|
||||
return nil, model.InternalError(fmt.Errorf("couldn't get attribute keys: %w", err))
|
||||
}
|
||||
|
||||
suggestions.AttributeKeys = attribKeysResp.AttributeKeys
|
||||
|
||||
// Rank suggested attributes
|
||||
slices.SortFunc(suggestions.AttributeKeys, func(a v3.AttributeKey, b v3.AttributeKey) int {
|
||||
|
||||
// Higher score => higher rank
|
||||
attribKeyScore := func(a v3.AttributeKey) int {
|
||||
|
||||
// Scoring criteria is expected to get more sophisticated in follow up changes
|
||||
if a.Type == v3.AttributeKeyTypeResource {
|
||||
return 2
|
||||
}
|
||||
|
||||
if a.Type == v3.AttributeKeyTypeTag {
|
||||
return 1
|
||||
}
|
||||
|
||||
return 0
|
||||
}
|
||||
|
||||
// To sort in descending order of score the return value must be negative when a > b
|
||||
return attribKeyScore(b) - attribKeyScore(a)
|
||||
})
|
||||
|
||||
// Put together suggested example queries.
|
||||
|
||||
newExampleQuery := func() v3.FilterSet {
|
||||
// Include existing filter in example query if specified.
|
||||
if req.ExistingFilter != nil {
|
||||
return *req.ExistingFilter
|
||||
}
|
||||
|
||||
return v3.FilterSet{
|
||||
Operator: "AND",
|
||||
Items: []v3.FilterItem{},
|
||||
}
|
||||
}
|
||||
|
||||
// Suggest example queries for top suggested log attributes and resource attributes
|
||||
exampleAttribs := []v3.AttributeKey{}
|
||||
for _, attrib := range suggestions.AttributeKeys {
|
||||
isAttributeOrResource := slices.Contains([]v3.AttributeKeyType{
|
||||
v3.AttributeKeyTypeResource, v3.AttributeKeyTypeTag,
|
||||
}, attrib.Type)
|
||||
|
||||
isNumOrStringType := slices.Contains([]v3.AttributeKeyDataType{
|
||||
v3.AttributeKeyDataTypeInt64, v3.AttributeKeyDataTypeFloat64, v3.AttributeKeyDataTypeString,
|
||||
}, attrib.DataType)
|
||||
|
||||
if isAttributeOrResource && isNumOrStringType {
|
||||
exampleAttribs = append(exampleAttribs, attrib)
|
||||
}
|
||||
|
||||
if len(exampleAttribs) >= int(req.ExamplesLimit) {
|
||||
break
|
||||
}
|
||||
}
|
||||
|
||||
if len(exampleAttribs) > 0 {
|
||||
exampleAttribValues, err := r.getValuesForLogAttributes(
|
||||
ctx, exampleAttribs, req.ExamplesLimit,
|
||||
)
|
||||
if err != nil {
|
||||
// Do not fail the entire request if only example query generation fails
|
||||
zap.L().Error("could not find attribute values for creating example query", zap.Error(err))
|
||||
} else {
|
||||
|
||||
// add example queries for as many attributes as possible.
|
||||
// suggest 1st value for 1st attrib, followed by 1st value for second attrib and so on
|
||||
// and if there is still room, suggest 2nd value for 1st attrib, 2nd value for 2nd attrib and so on
|
||||
for valueIdx := 0; valueIdx < int(req.ExamplesLimit); valueIdx++ {
|
||||
for attrIdx, attr := range exampleAttribs {
|
||||
needMoreExamples := len(suggestions.ExampleQueries) < int(req.ExamplesLimit)
|
||||
|
||||
if needMoreExamples && valueIdx < len(exampleAttribValues[attrIdx]) {
|
||||
exampleQuery := newExampleQuery()
|
||||
exampleQuery.Items = append(exampleQuery.Items, v3.FilterItem{
|
||||
Key: attr,
|
||||
Operator: "=",
|
||||
Value: exampleAttribValues[attrIdx][valueIdx],
|
||||
})
|
||||
|
||||
suggestions.ExampleQueries = append(
|
||||
suggestions.ExampleQueries, exampleQuery,
|
||||
)
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
// Suggest static example queries for standard log attributes if needed.
|
||||
if len(suggestions.ExampleQueries) < int(req.ExamplesLimit) {
|
||||
exampleQuery := newExampleQuery()
|
||||
exampleQuery.Items = append(exampleQuery.Items, v3.FilterItem{
|
||||
Key: v3.AttributeKey{
|
||||
Key: "body",
|
||||
DataType: v3.AttributeKeyDataTypeString,
|
||||
Type: v3.AttributeKeyTypeUnspecified,
|
||||
IsColumn: true,
|
||||
},
|
||||
Operator: "contains",
|
||||
Value: "error",
|
||||
})
|
||||
suggestions.ExampleQueries = append(suggestions.ExampleQueries, exampleQuery)
|
||||
}
|
||||
|
||||
return &suggestions, nil
|
||||
}
|
||||
|
||||
// Get up to `limit` values seen for each attribute in `attributes`
|
||||
// Returns a slice of slices where the ith slice has values for ith entry in `attributes`
|
||||
func (r *ClickHouseReader) getValuesForLogAttributes(
|
||||
ctx context.Context, attributes []v3.AttributeKey, limit uint64,
|
||||
) ([][]any, *model.ApiError) {
|
||||
/*
|
||||
The query used here needs to be as cheap as possible, and while uncommon, it is possible for
|
||||
a tag to have 100s of millions of values (eg: message, request_id)
|
||||
|
||||
Construct a query to UNION the result of querying first `limit` values for each attribute. For example:
|
||||
```
|
||||
select * from (
|
||||
(
|
||||
select tagKey, stringTagValue, int64TagValue, float64TagValue
|
||||
from signoz_logs.distributed_tag_attributes
|
||||
where tagKey = $1 and (
|
||||
stringTagValue != '' or int64TagValue is not null or float64TagValue is not null
|
||||
)
|
||||
limit 2
|
||||
) UNION DISTINCT (
|
||||
select tagKey, stringTagValue, int64TagValue, float64TagValue
|
||||
from signoz_logs.distributed_tag_attributes
|
||||
where tagKey = $2 and (
|
||||
stringTagValue != '' or int64TagValue is not null or float64TagValue is not null
|
||||
)
|
||||
limit 2
|
||||
)
|
||||
) settings max_threads=2
|
||||
```
|
||||
Since tag_attributes table uses ReplacingMergeTree, the values would be distinct and no order by
|
||||
is being used to ensure the `limit` clause minimizes the amount of data scanned.
|
||||
|
||||
This query scanned ~30k rows per attribute on fiscalnote-v2 for attributes like `message` and `time`
|
||||
that had >~110M values each
|
||||
*/
|
||||
|
||||
if len(attributes) > 10 {
|
||||
zap.L().Error(
|
||||
"log attribute values requested for too many attributes. This can lead to slow and costly queries",
|
||||
zap.Int("count", len(attributes)),
|
||||
)
|
||||
attributes = attributes[:10]
|
||||
}
|
||||
|
||||
tagQueries := []string{}
|
||||
tagKeyQueryArgs := []any{}
|
||||
for idx, attrib := range attributes {
|
||||
tagQueries = append(tagQueries, fmt.Sprintf(`(
|
||||
select tagKey, stringTagValue, int64TagValue, float64TagValue
|
||||
from %s.%s
|
||||
where tagKey = $%d and (
|
||||
stringTagValue != '' or int64TagValue is not null or float64TagValue is not null
|
||||
)
|
||||
limit %d
|
||||
)`, r.logsDB, r.logsTagAttributeTable, idx+1, limit))
|
||||
|
||||
tagKeyQueryArgs = append(tagKeyQueryArgs, attrib.Key)
|
||||
}
|
||||
|
||||
query := fmt.Sprintf(`select * from (
|
||||
%s
|
||||
) settings max_threads=2`, strings.Join(tagQueries, " UNION DISTINCT "))
|
||||
|
||||
rows, err := r.db.Query(ctx, query, tagKeyQueryArgs...)
|
||||
if err != nil {
|
||||
zap.L().Error("couldn't query attrib values for suggestions", zap.Error(err))
|
||||
return nil, model.InternalError(fmt.Errorf(
|
||||
"couldn't query attrib values for suggestions: %w", err,
|
||||
))
|
||||
}
|
||||
defer rows.Close()
|
||||
|
||||
result := make([][]any, len(attributes))
|
||||
|
||||
// Helper for getting hold of the result slice to append to for each scanned row
|
||||
resultIdxForAttrib := func(key string, dataType v3.AttributeKeyDataType) int {
|
||||
return slices.IndexFunc(attributes, func(attrib v3.AttributeKey) bool {
|
||||
return attrib.Key == key && attrib.DataType == dataType
|
||||
})
|
||||
}
|
||||
|
||||
// Scan rows and append to result
|
||||
for rows.Next() {
|
||||
var tagKey string
|
||||
var stringValue string
|
||||
var float64Value sql.NullFloat64
|
||||
var int64Value sql.NullInt64
|
||||
|
||||
err := rows.Scan(
|
||||
&tagKey, &stringValue, &int64Value, &float64Value,
|
||||
)
|
||||
if err != nil {
|
||||
return nil, model.InternalError(fmt.Errorf(
|
||||
"couldn't scan attrib value rows: %w", err,
|
||||
))
|
||||
}
|
||||
|
||||
if len(stringValue) > 0 {
|
||||
attrResultIdx := resultIdxForAttrib(tagKey, v3.AttributeKeyDataTypeString)
|
||||
if attrResultIdx >= 0 {
|
||||
result[attrResultIdx] = append(result[attrResultIdx], stringValue)
|
||||
}
|
||||
|
||||
} else if int64Value.Valid {
|
||||
attrResultIdx := resultIdxForAttrib(tagKey, v3.AttributeKeyDataTypeInt64)
|
||||
if attrResultIdx >= 0 {
|
||||
result[attrResultIdx] = append(result[attrResultIdx], int64Value.Int64)
|
||||
}
|
||||
|
||||
} else if float64Value.Valid {
|
||||
attrResultIdx := resultIdxForAttrib(tagKey, v3.AttributeKeyDataTypeFloat64)
|
||||
if attrResultIdx >= 0 {
|
||||
result[attrResultIdx] = append(result[attrResultIdx], float64Value.Float64)
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
if err := rows.Err(); err != nil {
|
||||
return nil, model.InternalError(fmt.Errorf(
|
||||
"couldn't scan attrib value rows: %w", err,
|
||||
))
|
||||
}
|
||||
|
||||
return result, nil
|
||||
}
|
||||
|
||||
func readRow(vars []interface{}, columnNames []string, countOfNumberCols int) ([]string, map[string]string, []map[string]string, *v3.Point) {
|
||||
// Each row will have a value and a timestamp, and an optional list of label values
|
||||
// example: {Timestamp: ..., Value: ...}
|
||||
|
||||
Loading…
x
Reference in New Issue
Block a user