@oneuptime/common 12.0.5 → 12.0.7

package/Server/Utils/DataSource/Connectors/LokiConnector.ts

@@ -0,0 +1,606 @@
+import AggregatedModel from "../../../../Types/BaseDatabase/AggregatedModel";
+import AggregatedResult from "../../../../Types/BaseDatabase/AggregatedResult";
+import {
+  DATA_SOURCE_CONNECT_TIMEOUT_IN_MS,
+  DATA_SOURCE_MAX_SERIES,
+  DATA_SOURCE_MAX_TABLE_ROWS,
+  DATA_SOURCE_MAX_TIME_SERIES_ROWS,
+  DATA_SOURCE_QUERY_TIMEOUT_IN_MS,
+  DataSourceLimitsUtil,
+} from "../../../../Types/DataSource/DataSourceLimits";
+import DataSourceTableResult from "../../../../Types/DataSource/DataSourceTableResult";
+import DataSourceType from "../../../../Types/DataSource/DataSourceType";
+import BadDataException from "../../../../Types/Exception/BadDataException";
+import Dictionary from "../../../../Types/Dictionary";
+import DataSourceHttpFetch, {
+  DataSourceHttpRequest,
+  DataSourceHttpResponse,
+} from "../HttpFetch";
+import SqlResultShaper from "../SqlResultShaper";
+import {
+  DataSourceConnectionSettings,
+  DataSourceConnector,
+  DataSourceQueryWindow,
+} from "../Types";
+
+/*
+ * Connector for Grafana Loki (and Loki-compatible endpoints such as
+ * GrafanaCloud Logs). Queries are LogQL, executed through the
+ * /loki/api/v1/query_range endpoint:
+ *
+ *  - LogQL METRIC queries (rate(), count_over_time(), ...) come back as a
+ *    Prometheus-style "matrix" and map to chartable time series;
+ *  - raw LOG queries come back as "streams" and are only representable as a
+ *    table (one row per log line).
+ *
+ * Loki's HTTP API takes timestamps in UNIX NANOSECONDS. Millisecond epochs
+ * exceed Number.MAX_SAFE_INTEGER once multiplied by 1e6, so all
+ * nanosecond math in this file goes through BigInt — float arithmetic would
+ * silently corrupt the low digits of the window bounds.
+ */
+
+export type LokiFetchFunction = (
+  request: DataSourceHttpRequest,
+) => Promise<DataSourceHttpResponse>;
+
+export interface LokiQueryRangeData {
+  resultType: string;
+  result: Array<unknown>;
+}
+
+export interface LokiParsedPoint {
+  timestamp: Date;
+  value: number;
+}
+
+const NANOSECONDS_PER_MILLISECOND: bigint = BigInt(1000000);
+
+// Optionally-signed base-10 integer string, the shape BigInt() accepts.
+const INTEGER_STRING_REGEX: RegExp = /^-?\d+$/;
+
+/*
+ * Exact unix-nanosecond string for a Date. BigInt math keeps every digit:
+ * 1699999999999 ms * 1e6 is above 2^53, where float multiplication would
+ * round the tail off.
+ */
+export function toLokiNanoseconds(date: Date): string {
+  return (BigInt(date.getTime()) * NANOSECONDS_PER_MILLISECOND).toString();
+}
+
+/*
+ * Parse a Loki nanosecond timestamp (string in stream entries; tolerate a
+ * number for robustness). Returns null when unparseable. Strings are kept
+ * out of Number() entirely — 19-digit nanosecond epochs do not fit in a
+ * float.
+ */
+export function parseLokiNanoseconds(raw: unknown): bigint | null {
+  if (typeof raw === "number" && Number.isFinite(raw)) {
+    return BigInt(Math.trunc(raw));
+  }
+  if (typeof raw !== "string") {
+    return null;
+  }
+  const trimmed: string = raw.trim();
+  if (!INTEGER_STRING_REGEX.test(trimmed)) {
+    return null;
+  }
+  return BigInt(trimmed);
+}
+
+/*
+ * Nanosecond epoch -> Date (millisecond precision; sub-millisecond digits
+ * are truncated). Division happens in BigInt space BEFORE converting to a
+ * Number so the millisecond value is exact.
+ */
+export function lokiNanosecondsToDate(raw: unknown): Date | null {
+  const nanoseconds: bigint | null = parseLokiNanoseconds(raw);
+  if (nanoseconds === null) {
+    return null;
+  }
+  const date: Date = new Date(
+    Number(nanoseconds / NANOSECONDS_PER_MILLISECOND),
+  );
+  return isNaN(date.getTime()) ? null : date;
+}
+
+/*
+ * Validate the Loki response envelope ({ status: "success", data:
+ * { resultType, result } }) and return the data section. Throws
+ * operator-actionable BadDataExceptions for every malformed shape.
+ */
+export function parseLokiResponseData(bodyJson: unknown): LokiQueryRangeData {
+  if (
+    bodyJson === null ||
+    bodyJson === undefined ||
+    typeof bodyJson !== "object" ||
+    Array.isArray(bodyJson)
+  ) {
+    throw new BadDataException(
+      "Loki did not return a JSON object. Check that the data source URL points at a Grafana Loki endpoint.",
+    );
+  }
+
+  const body: Record<string, unknown> = bodyJson as Record<string, unknown>;
+
+  if (body["status"] !== "success") {
+    const errorDetail: string =
+      typeof body["error"] === "string" ? `: ${body["error"]}` : ".";
+    throw new BadDataException(
+      `Loki query failed with status "${String(body["status"])}"${errorDetail}`,
+    );
+  }
+
+  const dataRaw: unknown = body["data"];
+  if (
+    dataRaw === null ||
+    dataRaw === undefined ||
+    typeof dataRaw !== "object" ||
+    Array.isArray(dataRaw)
+  ) {
+    throw new BadDataException(
+      "Loki response is missing the data object. Check that the data source URL points at a Grafana Loki endpoint.",
+    );
+  }
+
+  const data: Record<string, unknown> = dataRaw as Record<string, unknown>;
+  const resultType: unknown = data["resultType"];
+  const result: unknown = data["result"];
+
+  if (typeof resultType !== "string" || !Array.isArray(result)) {
+    throw new BadDataException(
+      "Loki response is missing data.resultType or data.result.",
+    );
+  }
+
+  return { resultType: resultType, result: result };
+}
+
+/*
+ * Extract string labels from a metric/stream label object, skipping the
+ * given keys. Non-object inputs yield an empty label set.
+ */
+function extractLabels(
+  raw: unknown,
+  excludeKeys: Array<string>,
+): Record<string, string> {
+  const labels: Record<string, string> = {};
+  if (
+    raw === null ||
+    raw === undefined ||
+    typeof raw !== "object" ||
+    Array.isArray(raw)
+  ) {
+    return labels;
+  }
+  const rawRecord: Record<string, unknown> = raw as Record<string, unknown>;
+  for (const key of Object.keys(rawRecord)) {
+    if (excludeKeys.includes(key)) {
+      continue;
+    }
+    const value: unknown = rawRecord[key];
+    if (value === null || value === undefined) {
+      continue;
+    }
+    labels[key] = String(value);
+  }
+  return labels;
+}
+
+/*
+ * One matrix sample: [unixSeconds (may be fractional), "value"]. Returns
+ * null for malformed or unplottable (NaN/Inf) samples so single bad points
+ * do not fail the whole chart.
+ */
+export function parseLokiMatrixPoint(raw: unknown): LokiParsedPoint | null {
+  if (!Array.isArray(raw) || raw.length < 2) {
+    return null;
+  }
+
+  const timestampRaw: unknown = raw[0];
+  const valueRaw: unknown = raw[1];
+
+  let seconds: number | null = null;
+  if (typeof timestampRaw === "number" && Number.isFinite(timestampRaw)) {
+    seconds = timestampRaw;
+  } else if (
+    typeof timestampRaw === "string" &&
+    timestampRaw.trim() !== "" &&
+    Number.isFinite(Number(timestampRaw))
+  ) {
+    seconds = Number(timestampRaw);
+  }
+  if (seconds === null || seconds <= 0) {
+    return null;
+  }
+
+  let value: number | null = null;
+  if (typeof valueRaw === "number") {
+    value = valueRaw;
+  } else if (typeof valueRaw === "string" && valueRaw.trim() !== "") {
+    value = Number(valueRaw);
+  }
+  if (value === null || !Number.isFinite(value)) {
+    return null;
+  }
+
+  return { timestamp: new Date(Math.round(seconds * 1000)), value: value };
+}
+
+/*
+ * "matrix" result -> AggregatedResult. Labels (minus __name__) become the
+ * per-point `attributes` object the chart layer groups series on; series
+ * without labels omit `attributes` entirely (single unnamed series).
+ * Applies the series cap and the total-point cap, flagging `truncated`.
+ */
+export function lokiMatrixToTimeSeries(
+  result: Array<unknown>,
+): AggregatedResult {
+  const data: Array<AggregatedModel> = [];
+  let truncated: boolean = false;
+  let seriesUsed: number = 0;
+  let rowCapHit: boolean = false;
+
+  for (const rawSeries of result) {
+    if (
+      rawSeries === null ||
+      typeof rawSeries !== "object" ||
+      Array.isArray(rawSeries)
+    ) {
+      continue;
+    }
+
+    if (seriesUsed >= DATA_SOURCE_MAX_SERIES) {
+      truncated = true;
+      break;
+    }
+    seriesUsed += 1;
+
+    const series: Record<string, unknown> = rawSeries as Record<
+      string,
+      unknown
+    >;
+    const labels: Record<string, string> = extractLabels(series["metric"], [
+      "__name__",
+    ]);
+    const values: unknown = series["values"];
+    if (!Array.isArray(values)) {
+      continue;
+    }
+
+    for (const rawPoint of values) {
+      const parsed: LokiParsedPoint | null = parseLokiMatrixPoint(rawPoint);
+      if (!parsed) {
+        continue;
+      }
+      if (data.length >= DATA_SOURCE_MAX_TIME_SERIES_ROWS) {
+        truncated = true;
+        rowCapHit = true;
+        break;
+      }
+      const point: AggregatedModel = {
+        timestamp: parsed.timestamp,
+        value: parsed.value,
+      } as AggregatedModel;
+      if (Object.keys(labels).length > 0) {
+        (point as Record<string, unknown>)["attributes"] = { ...labels };
+      }
+      data.push(point);
+    }
+
+    if (rowCapHit) {
+      break;
+    }
+  }
+
+  return { data: data, truncated: truncated };
+}
+
+interface FlatLogEntry {
+  order: bigint;
+  row: Record<string, unknown>;
+}
+
+/*
+ * "streams" result -> flat table rows { timestamp: Date, line, ...stream
+ * labels }, sorted newest-first (interleaved across streams, the way a log
+ * table reads). The `timestamp` and `line` cells always hold the entry's
+ * real values — stream labels that happen to use those names are dropped
+ * rather than allowed to clobber them. Rows are NOT capped here; the caller
+ * clamps via SqlResultShaper so cap+1 fetches surface `truncated`.
+ */
+export function lokiStreamsToTableRows(
+  result: Array<unknown>,
+): Array<Record<string, unknown>> {
+  const entries: Array<FlatLogEntry> = [];
+
+  for (const rawStream of result) {
+    if (
+      rawStream === null ||
+      typeof rawStream !== "object" ||
+      Array.isArray(rawStream)
+    ) {
+      continue;
+    }
+    const stream: Record<string, unknown> = rawStream as Record<
+      string,
+      unknown
+    >;
+    const labels: Record<string, string> = extractLabels(stream["stream"], []);
+    const values: unknown = stream["values"];
+    if (!Array.isArray(values)) {
+      continue;
+    }
+
+    for (const rawEntry of values) {
+      if (!Array.isArray(rawEntry) || rawEntry.length < 2) {
+        continue;
+      }
+      const nanoseconds: bigint | null = parseLokiNanoseconds(rawEntry[0]);
+      if (nanoseconds === null) {
+        continue;
+      }
+      const timestamp: Date = new Date(
+        Number(nanoseconds / NANOSECONDS_PER_MILLISECOND),
+      );
+      if (isNaN(timestamp.getTime())) {
+        continue;
+      }
+      const lineRaw: unknown = rawEntry[1];
+      const row: Record<string, unknown> = {
+        timestamp: timestamp,
+        line: typeof lineRaw === "string" ? lineRaw : String(lineRaw ?? ""),
+      };
+      for (const key of Object.keys(labels)) {
+        if (key !== "timestamp" && key !== "line") {
+          row[key] = labels[key];
+        }
+      }
+      entries.push({ order: nanoseconds, row: row });
+    }
+  }
+
+  entries.sort((a: FlatLogEntry, b: FlatLogEntry): number => {
+    if (a.order === b.order) {
+      return 0;
+    }
+    return a.order < b.order ? 1 : -1;
+  });
+
+  return entries.map((entry: FlatLogEntry): Record<string, unknown> => {
+    return entry.row;
+  });
+}
+
+/*
+ * "matrix" result -> flat table rows { timestamp: Date, value, ...labels }.
+ * As with streams, the reserved `timestamp`/`value` cells win over labels
+ * with the same name, and capping is left to the caller's SqlResultShaper
+ * pass.
+ */
+export function lokiMatrixToTableRows(
+  result: Array<unknown>,
+): Array<Record<string, unknown>> {
+  const rows: Array<Record<string, unknown>> = [];
+
+  for (const rawSeries of result) {
+    if (
+      rawSeries === null ||
+      typeof rawSeries !== "object" ||
+      Array.isArray(rawSeries)
+    ) {
+      continue;
+    }
+    const series: Record<string, unknown> = rawSeries as Record<
+      string,
+      unknown
+    >;
+    const labels: Record<string, string> = extractLabels(series["metric"], [
+      "__name__",
+    ]);
+    const values: unknown = series["values"];
+    if (!Array.isArray(values)) {
+      continue;
+    }
+
+    for (const rawPoint of values) {
+      const parsed: LokiParsedPoint | null = parseLokiMatrixPoint(rawPoint);
+      if (!parsed) {
+        continue;
+      }
+      const row: Record<string, unknown> = {
+        timestamp: parsed.timestamp,
+        value: parsed.value,
+      };
+      for (const key of Object.keys(labels)) {
+        if (key !== "timestamp" && key !== "value") {
+          row[key] = labels[key];
+        }
+      }
+      rows.push(row);
+    }
+  }
+
+  return rows;
+}
+
+export default class LokiConnector implements DataSourceConnector {
+  public dataSourceType: DataSourceType = DataSourceType.Loki;
+
+  private fetchFunction: LokiFetchFunction;
+
+  /*
+   * The optional fetchFunction is a DI seam for tests. The default wrapper
+   * resolves DataSourceHttpFetch.fetch at CALL time so jest.spyOn on the
+   * static method also intercepts connectors constructed earlier.
+   */
+  public constructor(fetchFunction?: LokiFetchFunction | undefined) {
+    this.fetchFunction =
+      fetchFunction ||
+      ((request: DataSourceHttpRequest): Promise<DataSourceHttpResponse> => {
+        return DataSourceHttpFetch.fetch(request);
+      });
+  }
+
+  private getBaseUrl(settings: DataSourceConnectionSettings): string {
+    if (!settings.url || settings.url.trim() === "") {
+      throw new BadDataException(
+        "Loki data source requires a URL (e.g. http://loki.example.com:3100).",
+      );
+    }
+    return settings.url.trim().replace(/\/+$/, "");
+  }
+
+  /*
+   * All egress goes through DataSourceHttpFetch, which runs the SSRF egress
+   * guard and pins validated addresses before dialing. Errors are
+   * sanitized so credentials can never surface in an API response.
+   */
+  private async fetchSanitized(
+    settings: DataSourceConnectionSettings,
+    request: DataSourceHttpRequest,
+  ): Promise<DataSourceHttpResponse> {
+    try {
+      return await this.fetchFunction(request);
+    } catch (error) {
+      throw new BadDataException(
+        SqlResultShaper.sanitizeError(
+          error,
+          SqlResultShaper.collectSecrets(settings),
+        ),
+      );
+    }
+  }
+
+  public async testConnection(
+    settings: DataSourceConnectionSettings,
+  ): Promise<void> {
+    const baseUrl: string = this.getBaseUrl(settings);
+
+    const endDate: Date = new Date();
+    const startDate: Date = new Date(endDate.getTime() - 60 * 60 * 1000);
+
+    const params: URLSearchParams = new URLSearchParams();
+    params.set("start", toLokiNanoseconds(startDate));
+    params.set("end", toLokiNanoseconds(endDate));
+
+    const headers: Dictionary<string> =
+      DataSourceHttpFetch.buildAuthHeaders(settings);
+
+    const response: DataSourceHttpResponse = await this.fetchSanitized(
+      settings,
+      {
+        method: "GET",
+        url: `${baseUrl}/loki/api/v1/labels?${params.toString()}`,
+        headers: headers,
+        timeoutInMs: DATA_SOURCE_CONNECT_TIMEOUT_IN_MS,
+      },
+    );
+
+    const body: unknown = response.bodyJson;
+    if (
+      body === null ||
+      body === undefined ||
+      typeof body !== "object" ||
+      Array.isArray(body) ||
+      (body as Record<string, unknown>)["status"] !== "success"
+    ) {
+      throw new BadDataException(
+        'Connected to the host, but it did not respond like a Grafana Loki API (expected { "status": "success" } from /loki/api/v1/labels). Check the URL — it should be the Loki base URL, e.g. http://loki.example.com:3100.',
+      );
+    }
+  }
+
+  private async runRangeQuery(
+    settings: DataSourceConnectionSettings,
+    query: string,
+    window: DataSourceQueryWindow,
+    limit: number,
+  ): Promise<LokiQueryRangeData> {
+    if (!query || query.trim() === "") {
+      throw new BadDataException("Query is required.");
+    }
+
+    const baseUrl: string = this.getBaseUrl(settings);
+
+    const stepInSeconds: number =
+      window.stepInSeconds ??
+      DataSourceLimitsUtil.getStepInSeconds(window.startDate, window.endDate);
+
+    const params: URLSearchParams = new URLSearchParams();
+    params.set("query", query);
+    params.set("start", toLokiNanoseconds(window.startDate));
+    params.set("end", toLokiNanoseconds(window.endDate));
+    params.set("step", `${stepInSeconds}s`);
+    params.set("limit", limit.toString());
+
+    const headers: Dictionary<string> =
+      DataSourceHttpFetch.buildAuthHeaders(settings);
+
+    const response: DataSourceHttpResponse = await this.fetchSanitized(
+      settings,
+      {
+        method: "GET",
+        url: `${baseUrl}/loki/api/v1/query_range?${params.toString()}`,
+        headers: headers,
+        timeoutInMs: DATA_SOURCE_QUERY_TIMEOUT_IN_MS,
+      },
+    );
+
+    return parseLokiResponseData(response.bodyJson);
+  }
+
+  public async queryTimeSeries(
+    settings: DataSourceConnectionSettings,
+    query: string,
+    window: DataSourceQueryWindow,
+  ): Promise<AggregatedResult> {
+    const data: LokiQueryRangeData = await this.runRangeQuery(
+      settings,
+      query,
+      window,
+      DATA_SOURCE_MAX_TIME_SERIES_ROWS + 1,
+    );
+
+    if (data.resultType === "streams") {
+      throw new BadDataException(
+        'This LogQL query returns raw log streams, which cannot be charted. Wrap it in a LogQL metric function such as rate() or count_over_time() — e.g. sum(rate({app="api"} [5m])) — or use the Table widget to view the log lines.',
+      );
+    }
+
+    if (data.resultType !== "matrix") {
+      throw new BadDataException(
+        `Loki returned an unexpected result type "${data.resultType}" for a range query. Charts require a LogQL metric query that produces a matrix result.`,
+      );
+    }
+
+    return lokiMatrixToTimeSeries(data.result);
+  }
+
+  public async queryTable(
+    settings: DataSourceConnectionSettings,
+    query: string,
+    window: DataSourceQueryWindow,
+  ): Promise<DataSourceTableResult> {
+    const data: LokiQueryRangeData = await this.runRangeQuery(
+      settings,
+      query,
+      window,
+      DATA_SOURCE_MAX_TABLE_ROWS + 1,
+    );
+
+    let rows: Array<Record<string, unknown>>;
+    if (data.resultType === "streams") {
+      rows = lokiStreamsToTableRows(data.result);
+    } else if (data.resultType === "matrix") {
+      rows = lokiMatrixToTableRows(data.result);
+    } else {
+      throw new BadDataException(
+        `Loki returned an unexpected result type "${data.resultType}" for a range query. Expected "streams" (raw logs) or "matrix" (metric query).`,
+      );
+    }
+
+    return SqlResultShaper.rowsToTable(rows, {
+      maxRows: DATA_SOURCE_MAX_TABLE_ROWS,
+    });
+  }
+}