@oneuptime/common 12.0.5 → 12.0.7

package/Tests/Server/Utils/InstanceHealth/TelemetryIngestion.test.ts

@@ -0,0 +1,960 @@
+import {
+  ClickhouseAppInstance,
+  ClickhouseClient,
+} from "../../../../Server/Infrastructure/ClickhouseDatabase";
+import {
+  attachProjectNames,
+  buildProjectIngestionQuery,
+  buildSignalIngestionQuery,
+  buildTelemetryStorageSizeQuery,
+  getTelemetryIngestionByProject,
+  getTelemetryIngestionBySignal,
+  MAX_PROJECTS_PER_SIGNAL,
+  mergeProjectIngestionRows,
+  sumCountsOrNull,
+  TELEMETRY_INGESTION_SIGNALS,
+  TelemetryIngestionSignal,
+  TelemetryIngestionSignalSpec,
+  TelemetryProjectIngestion,
+  TelemetryProjectIngestionResult,
+  TelemetryProjectSignalIngestion,
+  TelemetrySignalIngestion,
+  TelemetrySignalIngestionResult,
+  TelemetrySignalProjectRows,
+  toCountOrNull,
+} from "../../../../Server/Utils/InstanceHealth/TelemetryIngestion";
+import AnalyticsTableName from "../../../../Types/AnalyticsDatabase/AnalyticsTableName";
+import { JSONObject } from "../../../../Types/JSON";
+import { afterEach, describe, expect, jest, test } from "@jest/globals";
+import "../../TestingUtils/Init";
+
+/*
+ * The telemetry ingestion probes behind the master-admin Telemetry page.
+ *
+ * Three things are worth a test here and nothing else really is:
+ *
+ *  1. The SQL. These queries run against multi-billion-row tables, and the only
+ *     reason they are affordable is that they filter on the column each table is
+ *     PARTITIONED and primary-key ordered by, and group by projectId — the
+ *     leading primary-key column. Swap `time` for `createdAt`, or group by any
+ *     other column, and the page still renders correct numbers in development
+ *     while melting a production cluster. The assertions below pin the shape
+ *     that keeps them cheap.
+ *
+ *  2. null vs 0. A count we could not read must never render as a confident
+ *     zero: an operator reads "0 traces" as "this tenant stopped sending
+ *     traces", which is a very different incident from "we could not measure".
+ *     Every merge and sum path is checked for that distinction.
+ *
+ *  3. Independent degradation. One missing table, one failed query or an absent
+ *     ClickHouse must not blank the signals that DID answer.
+ */
+
+function makeClient(
+  query: (options: { query: string }) => Promise<{ data: Array<JSONObject> }>,
+): ClickhouseClient {
+  return {
+    query: async (options: { query: string }): Promise<unknown> => {
+      const result: { data: Array<JSONObject> } = await query(options);
+
+      return {
+        json: (): Promise<{ data: Array<JSONObject> }> => {
+          return Promise.resolve(result);
+        },
+      };
+    },
+  } as unknown as ClickhouseClient;
+}
+
+function specFor(
+  telemetryType: TelemetryIngestionSignal,
+): TelemetryIngestionSignalSpec {
+  const spec: TelemetryIngestionSignalSpec | undefined =
+    TELEMETRY_INGESTION_SIGNALS.find(
+      (candidate: TelemetryIngestionSignalSpec): boolean => {
+        return candidate.telemetryType === telemetryType;
+      },
+    );
+
+  if (!spec) {
+    throw new Error(`No telemetry signal registered for ${telemetryType}`);
+  }
+
+  return spec;
+}
+
+function projectRows(data: {
+  telemetryType: string;
+  available?: boolean | undefined;
+  truncated?: boolean | undefined;
+  rows: Array<{
+    projectId: string;
+    lastMinute: number | null;
+    lastHour: number | null;
+    lastDay: number | null;
+  }>;
+}): TelemetrySignalProjectRows {
+  return {
+    telemetryType: data.telemetryType,
+    available: data.available ?? true,
+    truncated: data.truncated ?? false,
+    rows: data.rows,
+  };
+}
+
+function signalOf(
+  project: TelemetryProjectIngestion,
+  telemetryType: string,
+): TelemetryProjectSignalIngestion {
+  const signal: TelemetryProjectSignalIngestion | undefined =
+    project.signals.find(
+      (candidate: TelemetryProjectSignalIngestion): boolean => {
+        return candidate.telemetryType === telemetryType;
+      },
+    );
+
+  if (!signal) {
+    throw new Error(`Project ${project.projectId} has no ${telemetryType} row`);
+  }
+
+  return signal;
+}
+
+describe("telemetry signal registry", () => {
+  test("covers exactly the three signals OneUptime ingests", () => {
+    expect(
+      TELEMETRY_INGESTION_SIGNALS.map(
+        (spec: TelemetryIngestionSignalSpec): string => {
+          return spec.telemetryType;
+        },
+      ),
+    ).toEqual(["Logs", "Metrics", "Traces"]);
+  });
+
+  test("points each signal at its own analytics table", () => {
+    expect(specFor(TelemetryIngestionSignal.Logs).table).toBe(
+      AnalyticsTableName.Log,
+    );
+    expect(specFor(TelemetryIngestionSignal.Metrics).table).toBe(
+      AnalyticsTableName.Metric,
+    );
+    expect(specFor(TelemetryIngestionSignal.Traces).table).toBe(
+      AnalyticsTableName.Span,
+    );
+  });
+
+  /*
+   * Spans are ordered on startTime, not `time` — a copy-paste of the log spec
+   * would filter on a column the table does not have and drop traces entirely.
+   */
+  test("counts each signal on the event-time column its table is ordered by", () => {
+    expect(specFor(TelemetryIngestionSignal.Logs).timeColumn).toBe("time");
+    expect(specFor(TelemetryIngestionSignal.Metrics).timeColumn).toBe("time");
+    expect(specFor(TelemetryIngestionSignal.Traces).timeColumn).toBe(
+      "startTime",
+    );
+  });
+
+  /*
+   * `createdAt` is the true write time but is unindexed and unpartitioned on
+   * these tables, so filtering on it forces a full-table scan. No spec may name
+   * it, no matter how tempting "when did we actually write this" sounds.
+   */
+  test("never counts on the unindexed write-time column", () => {
+    for (const spec of TELEMETRY_INGESTION_SIGNALS) {
+      expect(spec.timeColumn).not.toBe("createdAt");
+    }
+  });
+});
+
+describe("buildSignalIngestionQuery", () => {
+  const query: string = buildSignalIngestionQuery({
+    spec: specFor(TelemetryIngestionSignal.Logs),
+    databaseName: "oneuptime",
+  });
+
+  test("bounds the scan to the last day so only the last day's partitions are read", () => {
+    expect(query).toContain("WHERE `time` >= now() - INTERVAL 1 DAY");
+  });
+
+  // A future-dated event would otherwise inflate every window it lands in.
+  test("bounds the scan at the present too", () => {
+    expect(query).toContain("`time` <= now()");
+  });
+
+  /*
+   * The smaller windows are countIf projections of the SAME scan. Issuing them
+   * as separate queries would triple the read for identical numbers.
+   */
+  test("derives the minute and hour windows from the one day-bounded scan", () => {
+    expect(query).toContain(
+      "countIf(`time` >= now() - INTERVAL 1 MINUTE) AS last_minute",
+    );
+    expect(query).toContain(
+      "countIf(`time` >= now() - INTERVAL 1 HOUR) AS last_hour",
+    );
+    expect(query).toContain("count() AS last_day");
+    expect(query.match(/FROM/g)).toHaveLength(1);
+  });
+
+  test("reads the app-facing distributed table, which spans every shard", () => {
+    expect(query).toContain("FROM `oneuptime`.`LogItemV3`");
+    expect(query).not.toContain("LogItemV3Local");
+  });
+
+  test("uses the span's own start-time column for traces", () => {
+    const traceQuery: string = buildSignalIngestionQuery({
+      spec: specFor(TelemetryIngestionSignal.Traces),
+      databaseName: "oneuptime",
+    });
+
+    expect(traceQuery).toContain(
+      "countIf(`startTime` >= now() - INTERVAL 1 MINUTE)",
+    );
+    expect(traceQuery).toContain("FROM `oneuptime`.`SpanItemV3`");
+  });
+
+  test("counts rows only — no telemetry row data is selected", () => {
+    expect(query).not.toContain("SELECT *");
+    expect(query).not.toContain("body");
+    expect(query).not.toContain("attributes");
+  });
+});
+
+describe("buildProjectIngestionQuery", () => {
+  const query: string = buildProjectIngestionQuery({
+    spec: specFor(TelemetryIngestionSignal.Logs),
+    databaseName: "oneuptime",
+    limit: 500,
+  });
+
+  /*
+   * projectId is the LEADING primary-key column of all three telemetry tables.
+   * It is the one dimension these tables can be grouped on without a full scan
+   * of the day's partitions — grouping by service, host or attribute would not
+   * be affordable here.
+   */
+  test("groups on the leading primary-key column", () => {
+    expect(query).toContain("GROUP BY projectId");
+    expect(query).toContain("projectId AS project_id");
+  });
+
+  test("keeps the same day-bounded single scan as the by-signal query", () => {
+    expect(query).toContain("WHERE `time` >= now() - INTERVAL 1 DAY");
+    expect(query).toContain("`time` <= now()");
+    expect(query).toContain(
+      "countIf(`time` >= now() - INTERVAL 1 MINUTE) AS last_minute",
+    );
+    expect(query.match(/FROM/g)).toHaveLength(1);
+  });
+
+  /*
+   * If the cap ever bites it must drop the SMALLEST tenants, not an arbitrary
+   * slice — and projectId as the tie-break keeps the list from reshuffling
+   * between two polls that return identical numbers.
+   */
+  test("orders by the widest window so a cap drops the smallest tenants", () => {
+    expect(query).toContain("ORDER BY last_day DESC, project_id ASC");
+  });
+
+  test("caps the number of tenants returned", () => {
+    expect(query).toContain("LIMIT 500");
+  });
+
+  test("never emits a fractional or zero limit", () => {
+    expect(
+      buildProjectIngestionQuery({
+        spec: specFor(TelemetryIngestionSignal.Logs),
+        databaseName: "oneuptime",
+        limit: 10.7,
+      }),
+    ).toContain("LIMIT 10");
+
+    expect(
+      buildProjectIngestionQuery({
+        spec: specFor(TelemetryIngestionSignal.Logs),
+        databaseName: "oneuptime",
+        limit: 0,
+      }),
+    ).toContain("LIMIT 1");
+  });
+
+  test("selects no telemetry row data, only the tenant id and counts", () => {
+    expect(query).not.toContain("SELECT *");
+    expect(query).not.toContain("body");
+  });
+});
+
+describe("buildTelemetryStorageSizeQuery", () => {
+  const query: string = buildTelemetryStorageSizeQuery({
+    clusterName: "oneuptime",
+    databaseName: "oneuptime",
+    storageTableNames: ["LogItemV3Local", "MetricItemV3Local"],
+  });
+
+  /*
+   * The app-facing names are Distributed wrappers that hold no parts of their
+   * own. Sizing them would match nothing and leave every "Actual size" blank.
+   */
+  test("sizes the local storage tables, not the distributed wrappers", () => {
+    expect(query).toContain("'LogItemV3Local', 'MetricItemV3Local'");
+  });
+
+  /*
+   * cluster() hits ONE replica per shard. clusterAllReplicas would sum the same
+   * shard once per replica and overreport the footprint by the replication
+   * factor.
+   */
+  test("reads one replica per shard so replicas are not double-counted", () => {
+    expect(query).toContain("cluster('oneuptime', system.parts)");
+    expect(query).not.toContain("clusterAllReplicas");
+  });
+
+  test("counts only active parts, so replaced parts are not added twice", () => {
+    expect(query).toContain("WHERE active");
+  });
+
+  /*
+   * data_uncompressed_bytes is the actual data volume the page claims to show;
+   * bytes_on_disk is the compressed footprint and is a different number.
+   */
+  test("reports the uncompressed volume the page labels 'actual size'", () => {
+    expect(query).toContain(
+      "sum(data_uncompressed_bytes) AS uncompressed_bytes",
+    );
+    expect(query).not.toContain("bytes_on_disk");
+  });
+
+  test("escapes quotes in the cluster and database names", () => {
+    expect(
+      buildTelemetryStorageSizeQuery({
+        clusterName: "it's",
+        databaseName: "db's",
+        storageTableNames: ["a'b"],
+      }),
+    ).toContain("cluster('it''s', system.parts)");
+    expect(
+      buildTelemetryStorageSizeQuery({
+        clusterName: "it's",
+        databaseName: "db's",
+        storageTableNames: ["a'b"],
+      }),
+    ).toContain("database = 'db''s'");
+    expect(
+      buildTelemetryStorageSizeQuery({
+        clusterName: "it's",
+        databaseName: "db's",
+        storageTableNames: ["a'b"],
+      }),
+    ).toContain("'a''b'");
+  });
+});
+
+describe("toCountOrNull", () => {
+  test("reads ClickHouse's stringified UInt64 counts as numbers", () => {
+    expect(toCountOrNull("4210")).toBe(4210);
+    expect(toCountOrNull(0)).toBe(0);
+  });
+
+  // Absent is not zero: the page prints "—" for one and "0" for the other.
+  test("keeps an unreadable count unknown rather than calling it zero", () => {
+    expect(toCountOrNull(null)).toBeNull();
+    expect(toCountOrNull(undefined)).toBeNull();
+    expect(toCountOrNull("")).toBeNull();
+    expect(toCountOrNull("not-a-number")).toBeNull();
+    expect(toCountOrNull(Infinity)).toBeNull();
+  });
+});
+
+describe("sumCountsOrNull", () => {
+  test("adds the counts it knows", () => {
+    expect(sumCountsOrNull([1, 2, 3])).toBe(6);
+  });
+
+  /*
+   * A partial total is still the most useful number available — dropping the
+   * whole row because one signal is unreadable hides the two that answered.
+   */
+  test("totals what is known when only some inputs are missing", () => {
+    expect(sumCountsOrNull([5, null, 7])).toBe(12);
+  });
+
+  test("stays unknown when nothing is known", () => {
+    expect(sumCountsOrNull([null, null])).toBeNull();
+    expect(sumCountsOrNull([])).toBeNull();
+  });
+
+  // Zero is a real total and must not be mistaken for "nothing was known".
+  test("reports a genuine zero as zero", () => {
+    expect(sumCountsOrNull([0, 0])).toBe(0);
+  });
+});
+
+describe("mergeProjectIngestionRows", () => {
+  test("folds the three signals into one row per project", () => {
+    const projects: Array<TelemetryProjectIngestion> =
+      mergeProjectIngestionRows([
+        projectRows({
+          telemetryType: "Logs",
+          rows: [
+            { projectId: "p1", lastMinute: 1, lastHour: 10, lastDay: 100 },
+          ],
+        }),
+        projectRows({
+          telemetryType: "Metrics",
+          rows: [
+            { projectId: "p1", lastMinute: 2, lastHour: 20, lastDay: 200 },
+          ],
+        }),
+        projectRows({
+          telemetryType: "Traces",
+          rows: [
+            { projectId: "p1", lastMinute: 3, lastHour: 30, lastDay: 300 },
+          ],
+        }),
+      ]);
+
+    expect(projects).toHaveLength(1);
+    expect(projects[0]!.projectId).toBe("p1");
+    expect(projects[0]!.lastMinute).toBe(6);
+    expect(projects[0]!.lastHour).toBe(60);
+    expect(projects[0]!.lastDay).toBe(600);
+  });
+
+  /*
+   * A tenant sending only logs really did send zero metrics. That is a fact, not
+   * a gap, and the row has to say 0 — otherwise every logs-only instance shows a
+   * page full of dashes.
+   */
+  test("a project missing from a successful signal ingested none of it", () => {
+    const projects: Array<TelemetryProjectIngestion> =
+      mergeProjectIngestionRows([
+        projectRows({
+          telemetryType: "Logs",
+          rows: [{ projectId: "p1", lastMinute: 1, lastHour: 1, lastDay: 1 }],
+        }),
+        projectRows({ telemetryType: "Metrics", rows: [] }),
+        projectRows({ telemetryType: "Traces", rows: [] }),
+      ]);
+
+    expect(signalOf(projects[0]!, "Metrics").lastDay).toBe(0);
+    expect(signalOf(projects[0]!, "Traces").lastDay).toBe(0);
+    expect(projects[0]!.lastDay).toBe(1);
+  });
+
+  /*
+   * The opposite case, and the one that matters most: when the QUERY failed we
+   * know nothing about that signal for anybody. Reporting 0 there would read as
+   * "every tenant stopped sending traces" — an incident that never happened.
+   */
+  test("a signal whose query failed stays unknown for every project", () => {
+    const projects: Array<TelemetryProjectIngestion> =
+      mergeProjectIngestionRows([
+        projectRows({
+          telemetryType: "Logs",
+          rows: [{ projectId: "p1", lastMinute: 1, lastHour: 1, lastDay: 1 }],
+        }),
+        projectRows({
+          telemetryType: "Traces",
+          available: false,
+          rows: [],
+        }),
+      ]);
+
+    expect(signalOf(projects[0]!, "Traces").lastDay).toBeNull();
+    expect(signalOf(projects[0]!, "Traces").lastHour).toBeNull();
+    expect(signalOf(projects[0]!, "Traces").lastMinute).toBeNull();
+    // The total still reports the signals that did answer.
+    expect(projects[0]!.lastDay).toBe(1);
+  });
+
+  /*
+   * Rows returned by a signal that reported unavailable are not trustworthy
+   * (a partial read before a timeout), so they must not create project rows.
+   */
+  test("ignores rows carried by a signal that reported itself unavailable", () => {
+    const projects: Array<TelemetryProjectIngestion> =
+      mergeProjectIngestionRows([
+        projectRows({
+          telemetryType: "Logs",
+          available: false,
+          rows: [{ projectId: "p1", lastMinute: 1, lastHour: 1, lastDay: 1 }],
+        }),
+      ]);
+
+    expect(projects).toEqual([]);
+  });
+
+  test("orders the biggest tenant first", () => {
+    const projects: Array<TelemetryProjectIngestion> =
+      mergeProjectIngestionRows([
+        projectRows({
+          telemetryType: "Logs",
+          rows: [
+            { projectId: "small", lastMinute: 0, lastHour: 1, lastDay: 5 },
+            { projectId: "big", lastMinute: 0, lastHour: 9, lastDay: 900 },
+            { projectId: "medium", lastMinute: 0, lastHour: 4, lastDay: 40 },
+          ],
+        }),
+      ]);
+
+    expect(
+      projects.map((project: TelemetryProjectIngestion): string => {
+        return project.projectId;
+      }),
+    ).toEqual(["big", "medium", "small"]);
+  });
+
+  /*
+   * Two tenants with identical volumes must not swap places between polls —
+   * a table that reshuffles under a cursor is unusable.
+   */
+  test("breaks ties on project id so the order is stable between refreshes", () => {
+    const projects: Array<TelemetryProjectIngestion> =
+      mergeProjectIngestionRows([
+        projectRows({
+          telemetryType: "Logs",
+          rows: [
+            { projectId: "b", lastMinute: 0, lastHour: 0, lastDay: 10 },
+            { projectId: "a", lastMinute: 0, lastHour: 0, lastDay: 10 },
+          ],
+        }),
+      ]);
+
+    expect(
+      projects.map((project: TelemetryProjectIngestion): string => {
+        return project.projectId;
+      }),
+    ).toEqual(["a", "b"]);
+  });
+
+  test("every project carries one entry per signal, in registry order", () => {
+    const projects: Array<TelemetryProjectIngestion> =
+      mergeProjectIngestionRows([
+        projectRows({
+          telemetryType: "Logs",
+          rows: [{ projectId: "p1", lastMinute: 0, lastHour: 0, lastDay: 1 }],
+        }),
+        projectRows({ telemetryType: "Metrics", rows: [] }),
+        projectRows({ telemetryType: "Traces", rows: [] }),
+      ]);
+
+    expect(
+      projects[0]!.signals.map(
+        (signal: TelemetryProjectSignalIngestion): string => {
+          return signal.telemetryType;
+        },
+      ),
+    ).toEqual(["Logs", "Metrics", "Traces"]);
+  });
+});
+
+describe("attachProjectNames", () => {
+  const projects: Array<TelemetryProjectIngestion> = mergeProjectIngestionRows([
+    projectRows({
+      telemetryType: "Logs",
+      rows: [
+        { projectId: "known", lastMinute: 0, lastHour: 0, lastDay: 20 },
+        { projectId: "deleted", lastMinute: 0, lastHour: 0, lastDay: 10 },
+      ],
+    }),
+  ]);
+
+  test("labels a project with its name", () => {
+    const named: Array<TelemetryProjectIngestion> = attachProjectNames(
+      projects,
+      new Map<string, string>([["known", "Acme"]]),
+    );
+
+    expect(named[0]!.projectName).toBe("Acme");
+  });
+
+  /*
+   * Telemetry outlives its project: retention keeps the rows for days after the
+   * project row is gone. That volume is real and still has to be attributable,
+   * so an unresolved project keeps a null name rather than being dropped.
+   */
+  test("keeps a project whose name no longer resolves", () => {
+    const named: Array<TelemetryProjectIngestion> = attachProjectNames(
+      projects,
+      new Map<string, string>([["known", "Acme"]]),
+    );
+
+    expect(named).toHaveLength(2);
+    expect(named[1]!.projectId).toBe("deleted");
+    expect(named[1]!.projectName).toBeNull();
+    expect(named[1]!.lastDay).toBe(10);
+  });
+
+  test("leaves the counts untouched", () => {
+    const named: Array<TelemetryProjectIngestion> = attachProjectNames(
+      projects,
+      new Map<string, string>(),
+    );
+
+    expect(named[0]!.lastDay).toBe(20);
+    expect(named[0]!.signals).toEqual(projects[0]!.signals);
+  });
+});
+
+describe("getTelemetryIngestionBySignal", () => {
+  afterEach(() => {
+    jest.restoreAllMocks();
+  });
+
+  test("reports not connected when ClickHouse is absent, without throwing", async () => {
+    jest
+      .spyOn(ClickhouseAppInstance, "getDataSource")
+      .mockReturnValue(null as unknown as ClickhouseClient);
+
+    const result: TelemetrySignalIngestionResult =
+      await getTelemetryIngestionBySignal();
+
+    expect(result.connected).toBe(false);
+    expect(result.tables).toEqual([]);
+  });
+
+  test("reads each signal's counts and footprint", async () => {
+    jest.spyOn(ClickhouseAppInstance, "getDataSource").mockReturnValue(
+      makeClient(
+        async (options: {
+          query: string;
+        }): Promise<{ data: Array<JSONObject> }> => {
+          if (options.query.includes("system.parts")) {
+            return {
+              data: [{ table: "LogItemV3Local", uncompressed_bytes: "2048" }],
+            };
+          }
+
+          return {
+            data: [{ last_minute: "1", last_hour: "60", last_day: "1440" }],
+          };
+        },
+      ),
+    );
+
+    const result: TelemetrySignalIngestionResult =
+      await getTelemetryIngestionBySignal();
+
+    expect(result.connected).toBe(true);
+    expect(result.tables).toHaveLength(3);
+
+    const logs: TelemetrySignalIngestion = result.tables[0]!;
+    expect(logs.telemetryType).toBe("Logs");
+    expect(logs.lastMinute).toBe(1);
+    expect(logs.lastHour).toBe(60);
+    expect(logs.lastDay).toBe(1440);
+    expect(logs.uncompressedBytes).toBe(2048);
+    expect(logs.available).toBe(true);
+  });
+
+  /*
+   * An instance that only ingests logs has no metric or span table. That must
+   * cost the missing signal its row's numbers and nothing else.
+   */
+  test("a failing table leaves the signals that answered intact", async () => {
+    jest.spyOn(ClickhouseAppInstance, "getDataSource").mockReturnValue(
+      makeClient(
+        async (options: {
+          query: string;
+        }): Promise<{ data: Array<JSONObject> }> => {
+          if (options.query.includes("SpanItemV3")) {
+            throw new Error("UNKNOWN_TABLE");
+          }
+
+          if (options.query.includes("system.parts")) {
+            return { data: [] };
+          }
+
+          return {
+            data: [{ last_minute: "1", last_hour: "2", last_day: "3" }],
+          };
+        },
+      ),
+    );
+
+    const result: TelemetrySignalIngestionResult =
+      await getTelemetryIngestionBySignal();
+
+    const traces: TelemetrySignalIngestion = result.tables[2]!;
+    expect(traces.telemetryType).toBe("Traces");
+    expect(traces.available).toBe(false);
+    expect(traces.lastDay).toBeNull();
+    expect(result.tables[0]!.lastDay).toBe(3);
+  });
+
+  // The counts are the point; a blank size column is a far smaller loss.
+  test("a failing size query still reports the ingestion counts", async () => {
+    jest.spyOn(ClickhouseAppInstance, "getDataSource").mockReturnValue(
+      makeClient(
+        async (options: {
+          query: string;
+        }): Promise<{ data: Array<JSONObject> }> => {
+          if (options.query.includes("system.parts")) {
+            throw new Error("ACCESS_DENIED");
+          }
+
+          return {
+            data: [{ last_minute: "1", last_hour: "2", last_day: "3" }],
+          };
+        },
+      ),
+    );
+
+    const result: TelemetrySignalIngestionResult =
+      await getTelemetryIngestionBySignal();
+
+    expect(result.tables[0]!.uncompressedBytes).toBeNull();
+    expect(result.tables[0]!.lastDay).toBe(3);
+  });
+});
+
+describe("getTelemetryIngestionByProject", () => {
+  afterEach(() => {
+    jest.restoreAllMocks();
+  });
+
+  test("reports not connected when ClickHouse is absent, without throwing", async () => {
+    jest
+      .spyOn(ClickhouseAppInstance, "getDataSource")
+      .mockReturnValue(null as unknown as ClickhouseClient);
+
+    const result: TelemetryProjectIngestionResult =
+      await getTelemetryIngestionByProject();
+
+    expect(result.connected).toBe(false);
+    expect(result.projects).toEqual([]);
+  });
+
+  test("merges the per-signal groupings into per-project rows", async () => {
+    jest.spyOn(ClickhouseAppInstance, "getDataSource").mockReturnValue(
+      makeClient(
+        async (options: {
+          query: string;
+        }): Promise<{ data: Array<JSONObject> }> => {
+          if (options.query.includes("LogItemV3")) {
+            return {
+              data: [
+                {
+                  project_id: "p1",
+                  last_minute: "1",
+                  last_hour: "5",
+                  last_day: "50",
+                },
+              ],
+            };
+          }
+
+          if (options.query.includes("MetricItemV3")) {
+            return {
+              data: [
+                {
+                  project_id: "p1",
+                  last_minute: "2",
+                  last_hour: "6",
+                  last_day: "60",
+                },
+                {
+                  project_id: "p2",
+                  last_minute: "0",
+                  last_hour: "0",
+                  last_day: "5",
+                },
+              ],
+            };
+          }
+
+          return { data: [] };
+        },
+      ),
+    );
+
+    const result: TelemetryProjectIngestionResult =
+      await getTelemetryIngestionByProject();
+
+    expect(result.connected).toBe(true);
+    expect(
+      result.projects.map((project: TelemetryProjectIngestion): string => {
+        return project.projectId;
+      }),
+    ).toEqual(["p1", "p2"]);
+    expect(result.projects[0]!.lastDay).toBe(110);
+    // p2 sent metrics only, so its logs really are zero.
+    expect(signalOf(result.projects[1]!, "Logs").lastDay).toBe(0);
+  });
+
+  test("names are left for the caller to resolve", async () => {
+    jest.spyOn(ClickhouseAppInstance, "getDataSource").mockReturnValue(
+      makeClient(async (): Promise<{ data: Array<JSONObject> }> => {
+        return {
+          data: [
+            {
+              project_id: "p1",
+              last_minute: "0",
+              last_hour: "0",
+              last_day: "1",
+            },
+          ],
+        };
+      }),
+    );
+
+    const result: TelemetryProjectIngestionResult =
+      await getTelemetryIngestionByProject();
+
+    expect(result.projects[0]!.projectName).toBeNull();
+  });
+
+  test("a failing signal is reported as unavailable, not as zero", async () => {
+    jest.spyOn(ClickhouseAppInstance, "getDataSource").mockReturnValue(
+      makeClient(
+        async (options: {
+          query: string;
+        }): Promise<{ data: Array<JSONObject> }> => {
+          if (options.query.includes("SpanItemV3")) {
+            throw new Error("UNKNOWN_TABLE");
+          }
+
+          return {
+            data: [
+              {
+                project_id: "p1",
+                last_minute: "0",
+                last_hour: "0",
+                last_day: "1",
+              },
+            ],
+          };
+        },
+      ),
+    );
+
+    const result: TelemetryProjectIngestionResult =
+      await getTelemetryIngestionByProject();
+
+    expect(result.signals).toEqual([
+      { telemetryType: "Logs", available: true, truncated: false },
+      { telemetryType: "Metrics", available: true, truncated: false },
+      { telemetryType: "Traces", available: false, truncated: false },
+    ]);
+    expect(signalOf(result.projects[0]!, "Traces").lastDay).toBeNull();
+  });
+
+  /*
+   * Silently dropping tenants off the bottom of a list an operator reads as
+   * "everyone who is ingesting" is the failure this flag exists to prevent.
+   */
+  test("flags truncation when a signal fills the cap", async () => {
+    jest.spyOn(ClickhouseAppInstance, "getDataSource").mockReturnValue(
+      makeClient(
+        async (options: {
+          query: string;
+        }): Promise<{ data: Array<JSONObject> }> => {
+          if (!options.query.includes("LogItemV3")) {
+            return { data: [] };
+          }
+
+          return {
+            data: Array.from(
+              { length: MAX_PROJECTS_PER_SIGNAL },
+              (_unused: unknown, index: number): JSONObject => {
+                return {
+                  project_id: `p${index}`,
+                  last_minute: "0",
+                  last_hour: "0",
+                  last_day: "1",
+                };
+              },
+            ),
+          };
+        },
+      ),
+    );
+
+    const result: TelemetryProjectIngestionResult =
+      await getTelemetryIngestionByProject();
+
+    expect(result.truncated).toBe(true);
+    expect(result.projects).toHaveLength(MAX_PROJECTS_PER_SIGNAL);
+  });
+
+  test("does not flag truncation when every signal came back under the cap", async () => {
+    jest.spyOn(ClickhouseAppInstance, "getDataSource").mockReturnValue(
+      makeClient(async (): Promise<{ data: Array<JSONObject> }> => {
+        return {
+          data: [
+            {
+              project_id: "p1",
+              last_minute: "0",
+              last_hour: "0",
+              last_day: "1",
+            },
+          ],
+        };
+      }),
+    );
+
+    const result: TelemetryProjectIngestionResult =
+      await getTelemetryIngestionByProject();
+
+    expect(result.truncated).toBe(false);
+  });
+
+  test("survives ClickHouse failing entirely", async () => {
+    jest.spyOn(ClickhouseAppInstance, "getDataSource").mockReturnValue(
+      makeClient(async (): Promise<{ data: Array<JSONObject> }> => {
+        throw new Error("connection refused");
+      }),
+    );
+
+    const result: TelemetryProjectIngestionResult =
+      await getTelemetryIngestionByProject();
+
+    expect(result.connected).toBe(true);
+    expect(result.projects).toEqual([]);
+    expect(
+      result.signals.every((signal: { available: boolean }): boolean => {
+        return !signal.available;
+      }),
+    ).toBe(true);
+  });
+});
+
+describe("query timeouts", () => {
+  afterEach(() => {
+    jest.restoreAllMocks();
+  });
+
+  /*
+   * These are dashboard polls against the largest tables in the system. Without
+   * a server-side cap, one refresh on a struggling cluster can hold a query slot
+   * for as long as the HTTP client will wait.
+   */
+  test("every probe query is capped with max_execution_time", async () => {
+    const settings: Array<unknown> = [];
+
+    jest.spyOn(ClickhouseAppInstance, "getDataSource").mockReturnValue({
+      query: (options: {
+        clickhouse_settings?: { max_execution_time?: number };
+      }): Promise<unknown> => {
+        settings.push(options.clickhouse_settings?.max_execution_time);
+
+        return Promise.resolve({
+          json: (): Promise<{ data: Array<JSONObject> }> => {
+            return Promise.resolve({ data: [] });
+          },
+        });
+      },
+    } as unknown as ClickhouseClient);
+
+    await getTelemetryIngestionBySignal();
+    await getTelemetryIngestionByProject();
+
+    expect(settings.length).toBeGreaterThan(0);
+    expect(
+      settings.every((value: unknown): boolean => {
+        return typeof value === "number" && value > 0;
+      }),
+    ).toBe(true);
+  });
+});