2025-04-01 10:38:02 +09:00

186 lines
5.4 KiB
TypeScript

import { FieldType, toDataFrame } from '@grafana/data';
import { config } from '@grafana/runtime';
import { addExtractedFields } from './extractFields';
import { fieldExtractors } from './fieldExtractors';
import { ExtractFieldsOptions, FieldExtractorID } from './types';
describe('Extract fields from text', () => {
it('JSON extractor', async () => {
const extractor = fieldExtractors.get(FieldExtractorID.JSON);
const out = extractor.getParser({})('{"a":"148.1672","av":41923755,"c":148.25}');
expect(out).toMatchInlineSnapshot(`
{
"a": "148.1672",
"av": 41923755,
"c": 148.25,
}
`);
});
it('Test key-values with single/double quotes', async () => {
const extractor = fieldExtractors.get(FieldExtractorID.KeyValues);
const out = extractor.getParser({})('a="1", "b"=\'2\',c=3 x:y ;\r\nz="d and 4"');
expect(out).toMatchInlineSnapshot(`
{
"a": "1",
"b": "2",
"c": "3",
"x": "y",
"z": "d and 4",
}
`);
});
it('Test key-values with nested single/double quotes', async () => {
const extractor = fieldExtractors.get(FieldExtractorID.KeyValues);
const out = extractor.getParser({})(
`a="1", "b"=\'2\',c=3 x:y ;\r\nz="dbl_quotes=\\"Double Quotes\\" sgl_quotes='Single Quotes'"`
);
expect(out).toMatchInlineSnapshot(`
{
"a": "1",
"b": "2",
"c": "3",
"x": "y",
"z": "dbl_quotes="Double Quotes" sgl_quotes='Single Quotes'",
}
`);
});
it('Test key-values with nested separator characters', async () => {
const extractor = fieldExtractors.get(FieldExtractorID.KeyValues);
const out = extractor.getParser({})(`a="1", "b"=\'2\',c=3 x:y ;\r\nz="This is; testing& validating, 1=:2"`);
expect(out).toMatchInlineSnapshot(`
{
"a": "1",
"b": "2",
"c": "3",
"x": "y",
"z": "This is; testing& validating, 1=:2",
}
`);
});
it('Test key-values where some values are null', async () => {
const extractor = fieldExtractors.get(FieldExtractorID.KeyValues);
const out = extractor.getParser({})(`a=, "b"=\'2\',c=3 x: `);
expect(out).toMatchInlineSnapshot(`
{
"a": "",
"b": "2",
"c": "3",
"x": "",
}
`);
});
it('Split key+values', async () => {
const extractor = fieldExtractors.get(FieldExtractorID.KeyValues);
const out = extractor.getParser({})('a="1", "b"=\'2\',c=3 x:y ;\r\nz="7"');
expect(out).toMatchInlineSnapshot(`
{
"a": "1",
"b": "2",
"c": "3",
"x": "y",
"z": "7",
}
`);
});
it('Split URL style parameters', async () => {
const extractor = fieldExtractors.get(FieldExtractorID.KeyValues);
const out = extractor.getParser({})('a=b&c=d&x=123');
expect(out).toMatchInlineSnapshot(`
{
"a": "b",
"c": "d",
"x": "123",
}
`);
});
it('Prometheus labels style (not really supported)', async () => {
const extractor = fieldExtractors.get(FieldExtractorID.KeyValues);
const out = extractor.getParser({})('{foo="bar", baz="42"}');
expect(out).toMatchInlineSnapshot(`
{
"baz": "42",
"foo": "bar",
}
`);
});
it('deduplicates names', async () => {
const frame = toDataFrame({
fields: [{ name: 'foo', type: FieldType.string, values: ['{"foo":"extracedValue1"}'] }],
});
config.featureToggles.extractFieldsNameDeduplication = true;
const newFrame = addExtractedFields(frame, { format: FieldExtractorID.JSON, source: 'foo' });
config.featureToggles.extractFieldsNameDeduplication = false;
expect(newFrame.fields.length).toBe(2);
expect(newFrame.fields[1].name).toBe('foo 1');
});
it('keeps correct names when deduplication is active', async () => {
const frame = toDataFrame({
fields: [{ name: 'foo', type: FieldType.string, values: ['{"bar":"extracedValue1"}'] }],
});
config.featureToggles.extractFieldsNameDeduplication = true;
const newFrame = addExtractedFields(frame, { format: FieldExtractorID.JSON, source: 'foo' });
config.featureToggles.extractFieldsNameDeduplication = false;
expect(newFrame.fields.length).toBe(2);
expect(newFrame.fields[1].name).toBe('bar');
});
it('splits by regexp', async () => {
const extractor = fieldExtractors.get(FieldExtractorID.RegExp);
const opts: ExtractFieldsOptions = { regExp: '/^(?<FieldA>\\w+)[^\\w]+(?<FieldB>\\w+)$/' };
const parse = extractor.getParser(opts);
const out = parse('abc - re30z');
expect(out).toMatchInlineSnapshot(`
{
"FieldA": "abc",
"FieldB": "re30z",
}
`);
});
describe('Delimiter', () => {
it('splits by comma', async () => {
const extractor = fieldExtractors.get(FieldExtractorID.Delimiter);
const parse = extractor.getParser({});
const out = parse('a,b,c');
expect(out).toMatchInlineSnapshot(`
{
"a": 1,
"b": 1,
"c": 1,
}
`);
});
it('trims whitespace', async () => {
const extractor = fieldExtractors.get(FieldExtractorID.Delimiter);
const parse = extractor.getParser({});
const out = parse(` a, b,c, d `);
expect(out).toMatchInlineSnapshot(`
{
"a": 1,
"b": 1,
"c": 1,
"d": 1,
}
`);
});
});
});