grafana/public/app/plugins/datasource/prometheus/language_utils.ts

import { invert } from 'lodash';
import { Token } from 'prismjs';

import {
  AbstractLabelMatcher,
  AbstractLabelOperator,
  AbstractQuery,
  DataQuery,
  dateMath,
  DateTime,
  incrRoundDn,
  TimeRange,
} from '@grafana/data';

import { addLabelToQuery } from './add_label_to_query';
import { SUGGESTIONS_LIMIT } from './language_provider';
import { PrometheusCacheLevel, PromMetricsMetadata, PromMetricsMetadataItem } from './types';

export const processHistogramMetrics = (metrics: string[]) => {
  const resultSet: Set<string> = new Set();
  const regexp = new RegExp('_bucket($|:)');
  for (let index = 0; index < metrics.length; index++) {
    const metric = metrics[index];
    const isHistogramValue = regexp.test(metric);
    if (isHistogramValue) {
      resultSet.add(metric);
    }
  }
  return [...resultSet];
};

export function processLabels(labels: Array<{ [key: string]: string }>, withName = false) {
  // For processing we are going to use sets as they have significantly better performance than arrays
  // After we process labels, we will convert sets to arrays and return object with label values in arrays
  const valueSet: { [key: string]: Set<string> } = {};
  labels.forEach((label) => {
    const { __name__, ...rest } = label;
    if (withName) {
      valueSet['__name__'] = valueSet['__name__'] || new Set();
      if (!valueSet['__name__'].has(__name__)) {
        valueSet['__name__'].add(__name__);
      }
    }

    Object.keys(rest).forEach((key) => {
      if (!valueSet[key]) {
        valueSet[key] = new Set();
      }
      if (!valueSet[key].has(rest[key])) {
        valueSet[key].add(rest[key]);
      }
    });
  });

  // valueArray that we are going to return in the object
  const valueArray: { [key: string]: string[] } = {};
  limitSuggestions(Object.keys(valueSet)).forEach((key) => {
    valueArray[key] = limitSuggestions(Array.from(valueSet[key]));
  });

  return { values: valueArray, keys: Object.keys(valueArray) };
}

// const cleanSelectorRegexp = /\{(\w+="[^"\n]*?")(,\w+="[^"\n]*?")*\}/;
export const selectorRegexp = /\{[^}]*?(\}|$)/;
export const labelRegexp = /\b(\w+)(!?=~?)("[^"\n]*?")/g;

export function parseSelector(query: string, cursorOffset = 1): { labelKeys: any[]; selector: string } {
  if (!query.match(selectorRegexp)) {
    // Special matcher for metrics
    if (query.match(/^[A-Za-z:][\w:]*$/)) {
      return {
        selector: `{__name__="${query}"}`,
        labelKeys: ['__name__'],
      };
    }
    throw new Error('Query must contain a selector: ' + query);
  }

  // Check if inside a selector
  const prefix = query.slice(0, cursorOffset);
  const prefixOpen = prefix.lastIndexOf('{');
  const prefixClose = prefix.lastIndexOf('}');
  if (prefixOpen === -1) {
    throw new Error('Not inside selector, missing open brace: ' + prefix);
  }
  if (prefixClose > -1 && prefixClose > prefixOpen) {
    throw new Error('Not inside selector, previous selector already closed: ' + prefix);
  }
  const suffix = query.slice(cursorOffset);
  const suffixCloseIndex = suffix.indexOf('}');
  const suffixClose = suffixCloseIndex + cursorOffset;
  const suffixOpenIndex = suffix.indexOf('{');
  const suffixOpen = suffixOpenIndex + cursorOffset;
  if (suffixClose === -1) {
    throw new Error('Not inside selector, missing closing brace in suffix: ' + suffix);
  }
  if (suffixOpenIndex > -1 && suffixOpen < suffixClose) {
    throw new Error('Not inside selector, next selector opens before this one closed: ' + suffix);
  }

  // Extract clean labels to form clean selector, incomplete labels are dropped
  const selector = query.slice(prefixOpen, suffixClose);
  const labels: { [key: string]: { value: string; operator: string } } = {};
  selector.replace(labelRegexp, (label, key, operator, value) => {
    const labelOffset = query.indexOf(label);
    const valueStart = labelOffset + key.length + operator.length + 1;
    const valueEnd = labelOffset + key.length + operator.length + value.length - 1;
    // Skip label if cursor is in value
    if (cursorOffset < valueStart || cursorOffset > valueEnd) {
      labels[key] = { value, operator };
    }
    return '';
  });

  // Add metric if there is one before the selector
  const metricPrefix = query.slice(0, prefixOpen);
  const metricMatch = metricPrefix.match(/[A-Za-z:][\w:]*$/);
  if (metricMatch) {
    labels['__name__'] = { value: `"${metricMatch[0]}"`, operator: '=' };
  }

  // Build sorted selector
  const labelKeys = Object.keys(labels).sort();
  const cleanSelector = labelKeys.map((key) => `${key}${labels[key].operator}${labels[key].value}`).join(',');

  const selectorString = ['{', cleanSelector, '}'].join('');

  return { labelKeys, selector: selectorString };
}

export function expandRecordingRules(query: string, mapping: { [name: string]: string }): string {
  const ruleNames = Object.keys(mapping);
  const rulesRegex = new RegExp(`(\\s|^)(${ruleNames.join('|')})(\\s|$|\\(|\\[|\\{)`, 'ig');
  const expandedQuery = query.replace(rulesRegex, (match, pre, name, post) => `${pre}${mapping[name]}${post}`);

  // Split query into array, so if query uses operators, we can correctly add labels to each individual part.
  const queryArray = expandedQuery.split(/(\+|\-|\*|\/|\%|\^)/);

  // Regex that matches occurrences of ){ or }{ or ]{ which is a sign of incorrecly added labels.
  const invalidLabelsRegex = /(\)\{|\}\{|\]\{)/;
  const correctlyExpandedQueryArray = queryArray.map((query) => {
    return addLabelsToExpression(query, invalidLabelsRegex);
  });

  return correctlyExpandedQueryArray.join('');
}

function addLabelsToExpression(expr: string, invalidLabelsRegexp: RegExp) {
  const match = expr.match(invalidLabelsRegexp);
  if (!match) {
    return expr;
  }

  // Split query into 2 parts - before the invalidLabelsRegex match and after.
  const indexOfRegexMatch = match.index ?? 0;
  const exprBeforeRegexMatch = expr.slice(0, indexOfRegexMatch + 1);
  const exprAfterRegexMatch = expr.slice(indexOfRegexMatch + 1);

  // Create arrayOfLabelObjects with label objects that have key, operator and value.
  const arrayOfLabelObjects: Array<{ key: string; operator: string; value: string }> = [];
  exprAfterRegexMatch.replace(labelRegexp, (label, key, operator, value) => {
    arrayOfLabelObjects.push({ key, operator, value });
    return '';
  });

  // Loop through all label objects and add them to query.
  // As a starting point we have valid query without the labels.
  let result = exprBeforeRegexMatch;
  arrayOfLabelObjects.filter(Boolean).forEach((obj) => {
    // Remove extra set of quotes from obj.value
    const value = obj.value.slice(1, -1);
    result = addLabelToQuery(result, obj.key, value, obj.operator);
  });

  return result;
}

/**
 * Adds metadata for synthetic metrics for which the API does not provide metadata.
 * See https://github.com/grafana/grafana/issues/22337 for details.
 *
 * @param metadata HELP and TYPE metadata from /api/v1/metadata
 */
export function fixSummariesMetadata(metadata: { [metric: string]: PromMetricsMetadataItem[] }): PromMetricsMetadata {
  if (!metadata) {
    return metadata;
  }
  const baseMetadata: PromMetricsMetadata = {};
  const summaryMetadata: PromMetricsMetadata = {};
  for (const metric in metadata) {
    // NOTE: based on prometheus-documentation, we can receive
    // multiple metadata-entries for the given metric, it seems
    // it happens when the same metric is on multiple targets
    // and their help-text differs
    // (https://prometheus.io/docs/prometheus/latest/querying/api/#querying-metric-metadata)
    // for now we just use the first entry.
    const item = metadata[metric][0];
    baseMetadata[metric] = item;

    if (item.type === 'histogram') {
      summaryMetadata[`${metric}_bucket`] = {
        type: 'counter',
        help: `Cumulative counters for the observation buckets (${item.help})`,
      };
      summaryMetadata[`${metric}_count`] = {
        type: 'counter',
        help: `Count of events that have been observed for the histogram metric (${item.help})`,
      };
      summaryMetadata[`${metric}_sum`] = {
        type: 'counter',
        help: `Total sum of all observed values for the histogram metric (${item.help})`,
      };
    }
    if (item.type === 'summary') {
      summaryMetadata[`${metric}_count`] = {
        type: 'counter',
        help: `Count of events that have been observed for the base metric (${item.help})`,
      };
      summaryMetadata[`${metric}_sum`] = {
        type: 'counter',
        help: `Total sum of all observed values for the base metric (${item.help})`,
      };
    }
  }
  // Synthetic series
  const syntheticMetadata: PromMetricsMetadata = {};
  syntheticMetadata['ALERTS'] = {
    type: 'counter',
    help: 'Time series showing pending and firing alerts. The sample value is set to 1 as long as the alert is in the indicated active (pending or firing) state.',
  };

  return { ...baseMetadata, ...summaryMetadata, ...syntheticMetadata };
}

export function roundMsToMin(milliseconds: number): number {
  return roundSecToMin(milliseconds / 1000);
}

export function roundSecToMin(seconds: number): number {
  return Math.floor(seconds / 60);
}

// Returns number of minutes rounded up to the nearest nth minute
export function roundSecToNextMin(seconds: number, secondsToRound = 1): number {
  return Math.ceil(seconds / 60) - (Math.ceil(seconds / 60) % secondsToRound);
}

export function limitSuggestions(items: string[]) {
  return items.slice(0, SUGGESTIONS_LIMIT);
}

export function addLimitInfo(items: any[] | undefined): string {
  return items && items.length >= SUGGESTIONS_LIMIT ? `, limited to the first ${SUGGESTIONS_LIMIT} received items` : '';
}

// NOTE: the following 2 exported functions are very similar to the prometheus*Escape
// functions in datasource.ts, but they are not exactly the same algorithm, and we found
// no way to reuse one in the another or vice versa.

// Prometheus regular-expressions use the RE2 syntax (https://github.com/google/re2/wiki/Syntax),
// so every character that matches something in that list has to be escaped.
// the list of metacharacters is: *+?()|\.[]{}^$
// we make a javascript regular expression that matches those characters:
const RE2_METACHARACTERS = /[*+?()|\\.\[\]{}^$]/g;

function escapePrometheusRegexp(value: string): string {
  return value.replace(RE2_METACHARACTERS, '\\$&');
}

// based on the openmetrics-documentation, the 3 symbols we have to handle are:
// - \n ... the newline character
// - \  ... the backslash character
// - "  ... the double-quote character
export function escapeLabelValueInExactSelector(labelValue: string): string {
  return labelValue.replace(/\\/g, '\\\\').replace(/\n/g, '\\n').replace(/"/g, '\\"');
}

export function escapeLabelValueInRegexSelector(labelValue: string): string {
  return escapeLabelValueInExactSelector(escapePrometheusRegexp(labelValue));
}

const FromPromLikeMap: Record<string, AbstractLabelOperator> = {
  '=': AbstractLabelOperator.Equal,
  '!=': AbstractLabelOperator.NotEqual,
  '=~': AbstractLabelOperator.EqualRegEx,
  '!~': AbstractLabelOperator.NotEqualRegEx,
};
const ToPromLikeMap: Record<AbstractLabelOperator, string> = invert(FromPromLikeMap) as Record<
  AbstractLabelOperator,
  string
>;

export function toPromLikeExpr(labelBasedQuery: AbstractQuery): string {
  const expr = labelBasedQuery.labelMatchers
    .map((selector: AbstractLabelMatcher) => {
      const operator = ToPromLikeMap[selector.operator];
      if (operator) {
        return `${selector.name}${operator}"${selector.value}"`;
      } else {
        return '';
      }
    })
    .filter((e: string) => e !== '')
    .join(', ');

  return expr ? `{${expr}}` : '';
}

export function toPromLikeQuery(labelBasedQuery: AbstractQuery): PromLikeQuery {
  return {
    refId: labelBasedQuery.refId,
    expr: toPromLikeExpr(labelBasedQuery),
    range: true,
  };
}

export interface PromLikeQuery extends DataQuery {
  expr: string;
  range: boolean;
}

export function extractLabelMatchers(tokens: Array<string | Token>): AbstractLabelMatcher[] {
  const labelMatchers: AbstractLabelMatcher[] = [];

  for (let prop in tokens) {
    if (tokens[prop] instanceof Token) {
      let token: Token = tokens[prop] as Token;
      if (token.type === 'context-labels') {
        let labelKey = '';
        let labelValue = '';
        let labelOperator = '';
        let contentTokens: any[] = token.content as any[];
        for (let currentToken in contentTokens) {
          if (typeof contentTokens[currentToken] === 'string') {
            let currentStr: string;
            currentStr = contentTokens[currentToken] as string;
            if (currentStr === '=' || currentStr === '!=' || currentStr === '=~' || currentStr === '!~') {
              labelOperator = currentStr;
            }
          } else if (contentTokens[currentToken] instanceof Token) {
            switch (contentTokens[currentToken].type) {
              case 'label-key':
                labelKey = contentTokens[currentToken].content as string;
                break;
              case 'label-value':
                labelValue = contentTokens[currentToken].content as string;
                labelValue = labelValue.substring(1, labelValue.length - 1);
                const labelComparator = FromPromLikeMap[labelOperator];
                if (labelComparator) {
                  labelMatchers.push({ name: labelKey, operator: labelComparator, value: labelValue });
                }
                break;
            }
          }
        }
      }
    }
  }

  return labelMatchers;
}

/**
 * Calculates new interval "snapped" to the closest Nth minute, depending on cache level datasource setting
 * @param cacheLevel
 * @param range
 */
export function getRangeSnapInterval(
  cacheLevel: PrometheusCacheLevel,
  range: TimeRange
): { start: string; end: string } {
  // Don't round the range if we're not caching
  if (cacheLevel === PrometheusCacheLevel.None) {
    return {
      start: getPrometheusTime(range.from, false).toString(),
      end: getPrometheusTime(range.to, true).toString(),
    };
  }
  // Otherwise round down to the nearest nth minute for the start time
  const startTime = getPrometheusTime(range.from, false);
  // const startTimeQuantizedSeconds = roundSecToLastMin(startTime, getClientCacheDurationInMinutes(cacheLevel)) * 60;
  const startTimeQuantizedSeconds = incrRoundDn(startTime, getClientCacheDurationInMinutes(cacheLevel) * 60);

  // And round up to the nearest nth minute for the end time
  const endTime = getPrometheusTime(range.to, true);
  const endTimeQuantizedSeconds = roundSecToNextMin(endTime, getClientCacheDurationInMinutes(cacheLevel)) * 60;

  // If the interval was too short, we could have rounded both start and end to the same time, if so let's add one step to the end
  if (startTimeQuantizedSeconds === endTimeQuantizedSeconds) {
    const endTimePlusOneStep = endTimeQuantizedSeconds + getClientCacheDurationInMinutes(cacheLevel) * 60;
    return { start: startTimeQuantizedSeconds.toString(), end: endTimePlusOneStep.toString() };
  }

  const start = startTimeQuantizedSeconds.toString();
  const end = endTimeQuantizedSeconds.toString();

  return { start, end };
}

export function getClientCacheDurationInMinutes(cacheLevel: PrometheusCacheLevel) {
  switch (cacheLevel) {
    case PrometheusCacheLevel.Medium:
      return 10;
    case PrometheusCacheLevel.High:
      return 60;
    default:
      return 1;
  }
}

export function getPrometheusTime(date: string | DateTime, roundUp: boolean) {
  if (typeof date === 'string') {
    date = dateMath.parse(date, roundUp)!;
  }

  return Math.ceil(date.valueOf() / 1000);
}