@oneuptime/common 14.0.9 → 14.0.11

package/Tests/App/Dashboard/MessageQueueBrokerHealthReads.test.ts

@@ -0,0 +1,889 @@
+import { beforeEach, describe, expect, jest, test } from "@jest/globals";
+
+/*
+ * How a queue's broker metrics are READ, against an emulation of the
+ * aggregate endpoint that does what the server does with a request: it
+ * buckets the stored points on the interval grid (the window's own
+ * interval, or the one pinned), groups them by the request's group-by,
+ * folds each group with the request's aggregation, orders the rows newest
+ * first, applies the row limit and flags a full page `truncated`
+ * (AnalyticsDatabaseService.aggregateBy). What these pin:
+ *
+ *   - a per-period count (aggregation Sum) is never carried into an
+ *     interval it has no point in: one burst of errors counts once, and the
+ *     page reads the very query its monitor evaluates (Sum, no group-by);
+ *   - a per-series read that the row limit cut off is read again at a
+ *     coarser interval its series fit, so a partitioned topic's chart
+ *     starts at the start of the range with every partition in every
+ *     interval; where the fold composes (Max of Max, Sum of Sum) there is
+ *     one pooled read and nothing to cut off; a read still cut off drops
+ *     the interval the limit cut through rather than chart a partial total;
+ *   - a per-period count keeps whole intervals only: the one still filling
+ *     is never the tile, and a late source's recent intervals wait until
+ *     its points can have arrived.
+ */
+
+jest.mock("../../../UI/Utils/AnalyticsModelAPI/AnalyticsModelAPI", () => {
+  return {
+    __esModule: true,
+    default: {
+      aggregate: (request: unknown): Promise<unknown> => {
+        return Promise.resolve(emulateAggregate(request as AggregateRequest));
+      },
+    },
+  };
+});
+
+import {
+  MESSAGE_QUEUE_SERIES_READ_MAX_WIDENINGS,
+  MessageQueueMetricReadPlan,
+  MessageQueueTimePoint,
+  combineMessageQueueCountSeries,
+  fetchMessageQueueCatalogMetricSeries,
+  getCompleteMessageQueueCountSeries,
+  getMessageQueueCountSettleMs,
+  getMessageQueueMetricReadPlan,
+  getMessageQueueSeriesReadInterval,
+  toMessageQueueBrokerMetricResult,
+} from "../../../../App/FeatureSet/Dashboard/src/Components/MessageQueue/MessageQueueTelemetryQueries";
+import AggregatedModel from "../../../Types/BaseDatabase/AggregatedModel";
+import AggregatedResult from "../../../Types/BaseDatabase/AggregatedResult";
+import AggregationInterval from "../../../Types/BaseDatabase/AggregationInterval";
+import AggregationIntervalUtil from "../../../Types/BaseDatabase/AggregationIntervalUtil";
+import AggregationType from "../../../Types/BaseDatabase/AggregationType";
+import { LIMIT_PER_PROJECT } from "../../../Types/Database/LimitMax";
+import {
+  MESSAGE_QUEUE_METRICS,
+  MessageQueueMetricDescriptor,
+  getMessageQueueMetricDescriptorsByName,
+} from "../../../Types/MessageQueue/MessageQueueMetricCatalog";
+import {
+  MessageQueueMetricMonitorSeed,
+  getMessageQueueMetricMonitorSeed,
+} from "../../../Types/Monitor/MessageQueueAlertTemplates";
+
+interface AggregateRequest {
+  modelType: unknown;
+  aggregateBy: {
+    query: Record<string, unknown>;
+    aggregationType: AggregationType;
+    aggregationInterval?: AggregationInterval | undefined;
+    groupBy?: Record<string, unknown> | undefined;
+    groupByAttributeKeys?: Array<string> | undefined;
+    startTimestamp: Date;
+    endTimestamp: Date;
+    limit: number;
+  };
+}
+
+// One stored datapoint of the metric the test reads.
+interface StoredPoint {
+  time: number;
+  value: number;
+  attributes: Record<string, string>;
+}
+
+interface RowGroup {
+  time: number;
+  attributes: Record<string, string> | null;
+  values: Array<number>;
+}
+
+let stored: Array<StoredPoint> = [];
+let requests: Array<AggregateRequest> = [];
+// A row limit below the request's, for a test that must stay cut off.
+let serverRowLimit: number | null = null;
+
+function fold(values: Array<number>, type: AggregationType): number {
+  const total: number = values.reduce((sum: number, value: number): number => {
+    return sum + value;
+  }, 0);
+  switch (type) {
+    case AggregationType.Max:
+      return Math.max(...values);
+    case AggregationType.Min:
+      return Math.min(...values);
+    case AggregationType.Avg:
+      return total / values.length;
+    case AggregationType.Count:
+      return values.length;
+    default:
+      return total;
+  }
+}
+
+function emulateAggregate(request: AggregateRequest): AggregatedResult {
+  requests.push(request);
+  const by: AggregateRequest["aggregateBy"] = request.aggregateBy;
+  const widthMs: number = AggregationIntervalUtil.getAggregationIntervalMs(
+    AggregationIntervalUtil.getAggregationIntervalForWindow({
+      startDate: by.startTimestamp,
+      endDate: by.endTimestamp,
+      aggregationInterval: by.aggregationInterval,
+    }),
+  );
+
+  const groups: Map<string, RowGroup> = new Map<string, RowGroup>();
+  for (const point of stored) {
+    if (
+      point.time < by.startTimestamp.getTime() ||
+      point.time > by.endTimestamp.getTime()
+    ) {
+      continue;
+    }
+    // toStartOfInterval: the epoch-aligned grid.
+    const bucket: number = Math.floor(point.time / widthMs) * widthMs;
+    let attributes: Record<string, string> | null = null;
+    if (by.groupBy && by.groupBy["attributes"]) {
+      attributes = { ...point.attributes };
+    } else if (by.groupByAttributeKeys && by.groupByAttributeKeys.length > 0) {
+      attributes = {};
+      for (const key of by.groupByAttributeKeys) {
+        attributes[key] = point.attributes[key] || "";
+      }
+    }
+    const key: string = `${bucket}|${JSON.stringify(attributes)}`;
+    let group: RowGroup | undefined = groups.get(key);
+    if (!group) {
+      group = { time: bucket, attributes: attributes, values: [] };
+      groups.set(key, group);
+    }
+    group.values.push(point.value);
+  }
+
+  const rows: Array<AggregatedModel> = Array.from(groups.values())
+    .map((group: RowGroup): AggregatedModel => {
+      const row: AggregatedModel = {
+        timestamp: new Date(group.time),
+        value: fold(group.values, by.aggregationType),
+      };
+      if (group.attributes) {
+        row["attributes"] = group.attributes;
+      }
+      return row;
+    })
+    .sort((a: AggregatedModel, b: AggregatedModel): number => {
+      return b.timestamp.getTime() - a.timestamp.getTime();
+    });
+
+  const limit: number = Math.min(
+    Number(by.limit),
+    serverRowLimit === null ? Number.POSITIVE_INFINITY : serverRowLimit,
+  );
+  const data: Array<AggregatedModel> = rows.slice(0, limit);
+  return { data: data, truncated: data.length >= limit };
+}
+
+const MINUTE: number = 60 * 1000;
+const PROJECT_ID: string = "8f2a1b3c-4d5e-4f60-9a7b-1c2d3e4f5a6b";
+const KEY: string = "0123456789abcdef";
+// A window in the past, on the fifteen-minute grid: every interval closed.
+const T0: number = Date.UTC(2026, 8, 20, 6, 0, 0);
+
+function descriptorOf(
+  system: string,
+  metricName: string,
+): MessageQueueMetricDescriptor {
+  const descriptor: MessageQueueMetricDescriptor | undefined =
+    getMessageQueueMetricDescriptorsByName(metricName).find(
+      (candidate: MessageQueueMetricDescriptor): boolean => {
+        return candidate.system === system;
+      },
+    );
+  if (!descriptor) {
+    throw new Error(`No catalog entry ${system}:${metricName}`);
+  }
+  return descriptor;
+}
+
+function store(
+  time: number,
+  value: number,
+  attributes: Record<string, string> = {},
+): void {
+  stored.push({ time: time, value: value, attributes: attributes });
+}
+
+function read(data: {
+  descriptor: MessageQueueMetricDescriptor;
+  start: number;
+  end: number;
+  now?: number | undefined;
+}): Promise<Array<MessageQueueTimePoint>> {
+  return fetchMessageQueueCatalogMetricSeries({
+    projectId: PROJECT_ID,
+    keys: [KEY],
+    start: new Date(data.start),
+    end: new Date(data.end),
+    descriptor: data.descriptor,
+    now: data.now,
+  });
+}
+
+function valuesOf(series: Array<MessageQueueTimePoint>): Array<number> {
+  return series.map((point: MessageQueueTimePoint): number => {
+    return point.y;
+  });
+}
+
+function minutesOf(series: Array<MessageQueueTimePoint>): Array<number> {
+  return series.map((point: MessageQueueTimePoint): number => {
+    return (point.x.getTime() - T0) / MINUTE;
+  });
+}
+
+beforeEach(() => {
+  stored = [];
+  requests = [];
+  serverRowLimit = null;
+});
+
+describe("a per-period count is never carried into an interval it has no point in", () => {
+  test("Pub/Sub publish requests: one minute of failed requests counts in that minute only", async () => {
+    const sendRequests: MessageQueueMetricDescriptor = descriptorOf(
+      "gcp_pubsub",
+      "pubsub.googleapis.com/topic/send_request_count",
+    );
+    // A DELTA series per response code: failures only in the minutes with some.
+    for (let minute: number = 0; minute < 5; minute++) {
+      store(T0 + minute * MINUTE, 100, {
+        response_class: "success",
+        response_code: "OK",
+      });
+    }
+    store(T0 + MINUTE, 7, {
+      response_class: "deadline_exceeded",
+      response_code: "DEADLINE_EXCEEDED",
+    });
+
+    const series: Array<MessageQueueTimePoint> = await read({
+      descriptor: sendRequests,
+      start: T0,
+      end: T0 + 60 * MINUTE,
+    });
+
+    expect(valuesOf(series)).toEqual([100, 107, 100, 100, 100]);
+    expect(minutesOf(series)).toEqual([0, 1, 2, 3, 4]);
+
+    // One pooled Sum: the query the metric's monitor evaluates.
+    expect(requests).toHaveLength(1);
+    const by: AggregateRequest["aggregateBy"] = requests[0]!.aggregateBy;
+    expect(by.aggregationType).toBe(AggregationType.Sum);
+    expect(by.groupByAttributeKeys).toBeUndefined();
+    expect(by.groupBy).toBeUndefined();
+    const seed: MessageQueueMetricMonitorSeed =
+      getMessageQueueMetricMonitorSeed(sendRequests)!;
+    expect(seed.aggregationType).toBe(by.aggregationType);
+    expect(seed.groupByAttributeKeys).toEqual([]);
+  });
+
+  test("Service Bus server errors: a burst is not the newest interval's count", async () => {
+    const serverErrors: MessageQueueMetricDescriptor = descriptorOf(
+      "servicebus",
+      "azure_servererrors_total",
+    );
+    for (let minute: number = 0; minute < 4; minute++) {
+      store(T0 + minute * MINUTE, 1, {
+        metadata_operationresult: "ServerBusy",
+      });
+    }
+    store(T0 + MINUTE, 50, {
+      metadata_operationresult: "InternalServerError",
+    });
+
+    const series: Array<MessageQueueTimePoint> = await read({
+      descriptor: serverErrors,
+      start: T0,
+      end: T0 + 60 * MINUTE,
+    });
+
+    expect(valuesOf(series)).toEqual([1, 51, 1, 1]);
+    // The tile: the one server error of the newest minute.
+    expect(toMessageQueueBrokerMetricResult(serverErrors, series).value).toBe(
+      1,
+    );
+  });
+
+  test("a count read per series (the busiest one) combines only the series each interval has", async () => {
+    const serverErrors: MessageQueueMetricDescriptor = descriptorOf(
+      "servicebus",
+      "azure_servererrors_total",
+    );
+    // A catalog entry that shows the busiest operation result, not the total.
+    const busiest: MessageQueueMetricDescriptor = {
+      ...serverErrors,
+      seriesCombine: "max",
+    };
+    expect(getMessageQueueMetricReadPlan(busiest)).toEqual({
+      mode: "series",
+      aggregationType: AggregationType.Sum,
+      groupByAttributeKeys: [...serverErrors.seriesKeys!],
+      isPerPeriodCount: true,
+    });
+    store(T0, 5, { metadata_operationresult: "ServerBusy" });
+    for (let minute: number = 0; minute < 4; minute++) {
+      store(T0 + minute * MINUTE, 1, {
+        metadata_operationresult: "InternalServerError",
+      });
+    }
+
+    const series: Array<MessageQueueTimePoint> = await read({
+      descriptor: busiest,
+      start: T0,
+      end: T0 + 60 * MINUTE,
+    });
+
+    // Carried forward, the first minute's 5 would read in the next two.
+    expect(valuesOf(series)).toEqual([5, 1, 1, 1]);
+    expect(requests[0]!.aggregateBy.groupByAttributeKeys).toEqual(
+      serverErrors.seriesKeys,
+    );
+  });
+
+  test("combineMessageQueueCountSeries: the series present, summed, the busiest or averaged", () => {
+    const result: AggregatedResult = {
+      data: [
+        { timestamp: new Date(T0), value: 4, attributes: { code: "a" } },
+        { timestamp: new Date(T0), value: 6, attributes: { code: "b" } },
+        {
+          timestamp: new Date(T0 + MINUTE),
+          value: 3,
+          attributes: { code: "a" },
+        },
+        /*
+         * Unreadable rows are skipped. (Rows arrive as JSON, so a timestamp
+         * is text until it is read.)
+         */
+        { timestamp: "not a time", value: 9, attributes: { code: "a" } },
+        { timestamp: new Date(T0), value: "x", attributes: { code: "c" } },
+      ] as unknown as AggregatedResult["data"],
+    };
+
+    expect(valuesOf(combineMessageQueueCountSeries(result, "sum"))).toEqual([
+      10, 3,
+    ]);
+    expect(valuesOf(combineMessageQueueCountSeries(result, "max"))).toEqual([
+      6, 3,
+    ]);
+    expect(valuesOf(combineMessageQueueCountSeries(result, "avg"))).toEqual([
+      5, 3,
+    ]);
+    expect(combineMessageQueueCountSeries(null, "sum")).toEqual([]);
+  });
+});
+
+describe("the aggregate API's row limit", () => {
+  test("Kafka Produced over three hours for a 64-partition topic: read again in five-minute intervals, whole from the start", async () => {
+    const produced: MessageQueueMetricDescriptor = descriptorOf(
+      "kafka",
+      "kafka.partition.current_offset",
+    );
+    // Each partition's offset grows 10 messages a second.
+    for (let minute: number = 0; minute < 180; minute++) {
+      for (let partition: number = 0; partition < 64; partition++) {
+        store(T0 + minute * MINUTE, 1_000_000 + minute * 600, {
+          topic: "orders",
+          partition: String(partition),
+        });
+      }
+    }
+
+    const series: Array<MessageQueueTimePoint> = await read({
+      descriptor: produced,
+      start: T0,
+      end: T0 + 180 * MINUTE,
+    });
+
+    // 64 series × 180 minutes is over the limit: the first read is cut off.
+    expect(requests).toHaveLength(2);
+    expect(requests[0]!.aggregateBy.aggregationInterval).toBeUndefined();
+    expect(requests[0]!.aggregateBy.limit).toBe(LIMIT_PER_PROJECT);
+    expect(requests[1]!.aggregateBy.aggregationInterval).toBe(
+      AggregationInterval.FiveMinutes,
+    );
+    expect(requests[1]!.aggregateBy.groupBy).toEqual({ attributes: true });
+
+    // 640 a second from the first interval that has a predecessor on.
+    expect(minutesOf(series)[0]).toBe(5);
+    expect(series).toHaveLength(35);
+    for (const point of series) {
+      expect(point.y).toBeCloseTo(640, 6);
+    }
+  });
+
+  test("Kafka partition lag over a day: one pooled read, the laggiest partition from the first interval", async () => {
+    const partitionLag: MessageQueueMetricDescriptor = descriptorOf(
+      "kafka",
+      "kafka.consumer_group.lag",
+    );
+    // 3 groups × 40 partitions, one point each per fifteen minutes.
+    for (let interval: number = 0; interval < 96; interval++) {
+      for (let group: number = 0; group < 3; group++) {
+        for (let partition: number = 0; partition < 40; partition++) {
+          store(
+            T0 + interval * 15 * MINUTE,
+            group === 2 && partition === 39 ? 5000 : 10,
+            {
+              topic: "orders",
+              group: `g${group}`,
+              partition: String(partition),
+            },
+          );
+        }
+      }
+    }
+
+    const series: Array<MessageQueueTimePoint> = await read({
+      descriptor: partitionLag,
+      start: T0,
+      end: T0 + 96 * 15 * MINUTE,
+    });
+
+    expect(requests).toHaveLength(1);
+    expect(requests[0]!.aggregateBy.aggregationType).toBe(AggregationType.Max);
+    expect(requests[0]!.aggregateBy.groupByAttributeKeys).toBeUndefined();
+    expect(series).toHaveLength(96);
+    expect(minutesOf(series)[0]).toBe(0);
+    expect(new Set(valuesOf(series))).toEqual(new Set([5000]));
+  });
+
+  test("Pulsar topic backlog over three hours for 64 partitions: widened, every partition in every interval", async () => {
+    const backlog: MessageQueueMetricDescriptor = descriptorOf(
+      "pulsar",
+      "pulsar_msg_backlog",
+    );
+    expect(getMessageQueueMetricReadPlan(backlog).mode).toBe("series");
+    for (let minute: number = 0; minute < 180; minute++) {
+      for (let partition: number = 0; partition < 64; partition++) {
+        store(T0 + minute * MINUTE, 100, {
+          cluster: "standalone",
+          topic: `persistent://public/default/orders-partition-${partition}`,
+          partition: String(partition),
+        });
+      }
+    }
+
+    const series: Array<MessageQueueTimePoint> = await read({
+      descriptor: backlog,
+      start: T0,
+      end: T0 + 180 * MINUTE,
+    });
+
+    expect(requests).toHaveLength(2);
+    expect(requests[1]!.aggregateBy.aggregationInterval).toBe(
+      AggregationInterval.FiveMinutes,
+    );
+    expect(requests[1]!.aggregateBy.groupByAttributeKeys).toEqual(
+      backlog.seriesKeys,
+    );
+    expect(minutesOf(series)[0]).toBe(0);
+    expect(series).toHaveLength(36);
+    expect(new Set(valuesOf(series))).toEqual(new Set([6400]));
+  });
+
+  test("still cut off at the widest interval tried: the interval the limit cut through is dropped, never charted partial", async () => {
+    const backlog: MessageQueueMetricDescriptor = descriptorOf(
+      "pulsar",
+      "pulsar_msg_backlog",
+    );
+    for (let minute: number = 0; minute < 60; minute++) {
+      for (let partition: number = 0; partition < 64; partition++) {
+        store(T0 + minute * MINUTE, 100, {
+          cluster: "standalone",
+          topic: `persistent://public/default/orders-partition-${partition}`,
+          partition: String(partition),
+        });
+      }
+    }
+    // Two whole fifteen-minute intervals and part of a third fit.
+    serverRowLimit = 150;
+
+    const series: Array<MessageQueueTimePoint> = await read({
+      descriptor: backlog,
+      start: T0,
+      end: T0 + 60 * MINUTE,
+    });
+
+    expect(requests).toHaveLength(1 + MESSAGE_QUEUE_SERIES_READ_MAX_WIDENINGS);
+    expect(
+      requests.map((request: AggregateRequest): unknown => {
+        return request.aggregateBy.aggregationInterval;
+      }),
+    ).toEqual([
+      undefined,
+      AggregationInterval.FiveMinutes,
+      AggregationInterval.FifteenMinutes,
+    ]);
+    // The newest two intervals, whole; the one the limit cut through is gone.
+    expect(minutesOf(series)).toEqual([30, 45]);
+    expect(valuesOf(series)).toEqual([6400, 6400]);
+  });
+
+  test("a read that fits is read once, at the window's own interval", async () => {
+    const produced: MessageQueueMetricDescriptor = descriptorOf(
+      "kafka",
+      "kafka.partition.current_offset",
+    );
+    for (let minute: number = 0; minute < 60; minute++) {
+      for (let partition: number = 0; partition < 3; partition++) {
+        store(T0 + minute * MINUTE, minute * 60, {
+          topic: "orders",
+          partition: String(partition),
+        });
+      }
+    }
+
+    const series: Array<MessageQueueTimePoint> = await read({
+      descriptor: produced,
+      start: T0,
+      end: T0 + 60 * MINUTE,
+    });
+
+    expect(requests).toHaveLength(1);
+    expect(series).toHaveLength(59);
+    expect(series[0]!.y).toBeCloseTo(3, 6);
+  });
+
+  test("the interval a cut-off read widens to: the finest coarser one its series fit", () => {
+    const hour: { start: Date; end: Date } = {
+      start: new Date(T0),
+      end: new Date(T0 + 60 * MINUTE),
+    };
+    // 192 series (3 groups × 64 partitions): 61 minutes do not fit, 13 do.
+    expect(
+      getMessageQueueSeriesReadInterval({
+        ...hour,
+        seriesCount: 192,
+        after: AggregationInterval.Minute,
+      }),
+    ).toBe(AggregationInterval.FiveMinutes);
+    // Coarser than the interval that was cut off, even when it would fit.
+    expect(
+      getMessageQueueSeriesReadInterval({
+        ...hour,
+        seriesCount: 1,
+        after: AggregationInterval.FiveMinutes,
+      }),
+    ).toBe(AggregationInterval.FifteenMinutes);
+    /*
+     * A page exactly full reads as cut off: under the limit, not at it
+     * (100 series × 13 five-minute intervals is exactly 1,300).
+     */
+    expect(
+      getMessageQueueSeriesReadInterval({
+        ...hour,
+        seriesCount: 100,
+        after: AggregationInterval.Minute,
+        limit: 1300,
+      }),
+    ).toBe(AggregationInterval.FifteenMinutes);
+    expect(
+      getMessageQueueSeriesReadInterval({
+        ...hour,
+        seriesCount: 100,
+        after: AggregationInterval.Minute,
+        limit: 1301,
+      }),
+    ).toBe(AggregationInterval.FiveMinutes);
+    expect(
+      getMessageQueueSeriesReadInterval({
+        ...hour,
+        seriesCount: 1_000_000,
+        after: AggregationInterval.Minute,
+      }),
+    ).toBeNull();
+    expect(
+      getMessageQueueSeriesReadInterval({
+        ...hour,
+        seriesCount: 5,
+        after: AggregationInterval.Year,
+      }),
+    ).toBeNull();
+  });
+});
+
+describe("a per-period count keeps its whole intervals only", () => {
+  // 10:04, the window a day long: fifteen-minute intervals.
+  const NOW: number = Date.UTC(2026, 8, 30, 10, 4, 0);
+  const DAY_START: number = NOW - 24 * 60 * MINUTE;
+
+  test("Service Bus incoming messages over a day: the interval still filling is not the tile", async () => {
+    const incoming: MessageQueueMetricDescriptor = descriptorOf(
+      "servicebus",
+      "azure_incomingmessages_total",
+    );
+    // 600 a minute, up to 10:01: Azure's newest minutes are still on the way.
+    for (
+      let time: number = DAY_START;
+      time <= Date.UTC(2026, 8, 30, 10, 1, 0);
+      time += MINUTE
+    ) {
+      store(time, 600, {});
+    }
+
+    const series: Array<MessageQueueTimePoint> = await read({
+      descriptor: incoming,
+      start: DAY_START,
+      end: NOW,
+      now: NOW,
+    });
+
+    // The first interval began before the window, the newest is still open.
+    expect(series[0]!.x.getTime()).toBe(DAY_START - 4 * MINUTE + 15 * MINUTE);
+    expect(series[series.length - 1]!.x.getTime()).toBe(
+      Date.UTC(2026, 8, 30, 9, 45, 0),
+    );
+    expect(new Set(valuesOf(series))).toEqual(new Set([9000]));
+    expect(toMessageQueueBrokerMetricResult(incoming, series).value).toBe(9000);
+  });
+
+  test("CloudWatch's late counts: a wide interval is whole thirty minutes after it closed", async () => {
+    const sent: MessageQueueMetricDescriptor = descriptorOf(
+      "aws_sqs",
+      "amazonaws.com/aws/sqs/numberofmessagessent",
+    );
+    expect(getMessageQueueSettleMinutes(sent, DAY_START, NOW)).toBe(30);
+    // CloudWatch's points up to 09:50: the rest have not been fetched yet.
+    for (
+      let time: number = DAY_START;
+      time <= Date.UTC(2026, 8, 30, 9, 50, 0);
+      time += MINUTE
+    ) {
+      store(time, 10, { "Dimensions.QueueName": "orders" });
+    }
+
+    const series: Array<MessageQueueTimePoint> = await read({
+      descriptor: sent,
+      start: DAY_START,
+      end: NOW,
+      now: NOW,
+    });
+
+    // 09:15–09:30 closed 34 minutes ago; 09:30–09:45 only 19.
+    expect(series[series.length - 1]!.x.getTime()).toBe(
+      Date.UTC(2026, 8, 30, 9, 15, 0),
+    );
+    expect(new Set(valuesOf(series))).toEqual(new Set([150]));
+  });
+
+  test("the same count from a Metric Stream arrives live: only the interval still open waits", async () => {
+    const sent: MessageQueueMetricDescriptor = descriptorOf(
+      "aws_sqs",
+      "numberofmessagessent",
+    );
+    expect(getMessageQueueSettleMinutes(sent, DAY_START, NOW)).toBe(0);
+    for (
+      let time: number = DAY_START;
+      time <= Date.UTC(2026, 8, 30, 10, 2, 0);
+      time += MINUTE
+    ) {
+      store(time, 10, { QueueName: "orders", "resource.service.name": "SQS" });
+    }
+
+    const series: Array<MessageQueueTimePoint> = await read({
+      descriptor: sent,
+      start: DAY_START,
+      end: NOW,
+      now: NOW,
+    });
+
+    expect(series[series.length - 1]!.x.getTime()).toBe(
+      Date.UTC(2026, 8, 30, 9, 45, 0),
+    );
+    expect(new Set(valuesOf(series))).toEqual(new Set([150]));
+  });
+
+  test("an hour in one-minute intervals: a minute's point is whole once it is there, late source or not", async () => {
+    const sent: MessageQueueMetricDescriptor = descriptorOf(
+      "aws_sqs",
+      "amazonaws.com/aws/sqs/numberofmessagessent",
+    );
+    const hourStart: number = NOW - 60 * MINUTE;
+    expect(getMessageQueueSettleMinutes(sent, hourStart, NOW)).toBe(0);
+    for (
+      let time: number = hourStart;
+      time <= Date.UTC(2026, 8, 30, 9, 50, 0);
+      time += MINUTE
+    ) {
+      store(time, 10, { "Dimensions.QueueName": "orders" });
+    }
+
+    const series: Array<MessageQueueTimePoint> = await read({
+      descriptor: sent,
+      start: hourStart,
+      end: NOW,
+      now: NOW,
+    });
+
+    expect(series[series.length - 1]!.x.getTime()).toBe(
+      Date.UTC(2026, 8, 30, 9, 50, 0),
+    );
+    expect(toMessageQueueBrokerMetricResult(sent, series).value).toBe(10);
+  });
+
+  test("a level is never trimmed: its newest value is the tile", async () => {
+    const active: MessageQueueMetricDescriptor = descriptorOf(
+      "servicebus",
+      "azure_activemessages_average",
+    );
+    store(Date.UTC(2026, 8, 30, 9, 50, 0), 40, {});
+    store(Date.UTC(2026, 8, 30, 10, 3, 0), 42, {});
+
+    const series: Array<MessageQueueTimePoint> = await read({
+      descriptor: active,
+      start: DAY_START,
+      end: NOW,
+      now: NOW,
+    });
+
+    expect(valuesOf(series)).toEqual([40, 42]);
+  });
+
+  test("getCompleteMessageQueueCountSeries keeps a series whose every interval is partial", () => {
+    const sent: MessageQueueMetricDescriptor = descriptorOf(
+      "aws_sqs",
+      "amazonaws.com/aws/sqs/numberofmessagessent",
+    );
+    const partial: Array<MessageQueueTimePoint> = [
+      { x: new Date(Date.UTC(2026, 8, 30, 9, 45, 0)), y: 3 },
+    ];
+    expect(
+      getCompleteMessageQueueCountSeries(
+        partial,
+        sent,
+        { start: new Date(DAY_START), end: new Date(NOW) },
+        NOW,
+      ),
+    ).toEqual(partial);
+  });
+});
+
+// The settle time of a count over a window, in whole minutes.
+function getMessageQueueSettleMinutes(
+  descriptor: MessageQueueMetricDescriptor,
+  start: number,
+  end: number,
+): number {
+  return Math.round(
+    getMessageQueueCountSettleMs(descriptor, {
+      start: new Date(start),
+      end: new Date(end),
+    }) / MINUTE,
+  );
+}
+
+const GAUGES: Array<[string, MessageQueueMetricDescriptor]> =
+  MESSAGE_QUEUE_METRICS.filter(
+    (descriptor: MessageQueueMetricDescriptor): boolean => {
+      return descriptor.kind === "gauge";
+    },
+  ).map(
+    (
+      descriptor: MessageQueueMetricDescriptor,
+    ): [string, MessageQueueMetricDescriptor] => {
+      return [`${descriptor.system}:${descriptor.metricName}`, descriptor];
+    },
+  );
+
+describe("the read plan of every catalog metric", () => {
+  test.each(
+    MESSAGE_QUEUE_METRICS.map(
+      (
+        descriptor: MessageQueueMetricDescriptor,
+      ): [string, MessageQueueMetricDescriptor] => {
+        return [`${descriptor.system}:${descriptor.metricName}`, descriptor];
+      },
+    ),
+  )("%s", (_id: string, descriptor: MessageQueueMetricDescriptor) => {
+    const plan: MessageQueueMetricReadPlan =
+      getMessageQueueMetricReadPlan(descriptor);
+    const seriesKeys: Array<string> = [...(descriptor.seriesKeys || [])];
+
+    if (descriptor.kind === "counter") {
+      expect(plan).toEqual({
+        mode: "rate",
+        aggregationType: AggregationType.Max,
+        groupByAttributeKeys: [],
+        isPerPeriodCount: false,
+      });
+      return;
+    }
+
+    const composes: boolean =
+      (descriptor.aggregation === AggregationType.Max &&
+        descriptor.seriesCombine === "max") ||
+      (descriptor.aggregation === AggregationType.Sum &&
+        descriptor.seriesCombine === "sum");
+    expect(plan.aggregationType).toBe(descriptor.aggregation);
+    expect(plan.isPerPeriodCount).toBe(
+      descriptor.aggregation === AggregationType.Sum,
+    );
+    if (seriesKeys.length === 0 || composes) {
+      expect(plan.mode).toBe("pooled");
+      expect(plan.groupByAttributeKeys).toEqual([]);
+    } else {
+      expect(plan.mode).toBe("series");
+      expect(plan.groupByAttributeKeys).toEqual(seriesKeys);
+    }
+  });
+
+  test("every per-period count of the catalog is one pooled read, and so never cut off", () => {
+    const counts: Array<MessageQueueMetricDescriptor> =
+      MESSAGE_QUEUE_METRICS.filter(
+        (descriptor: MessageQueueMetricDescriptor): boolean => {
+          return getMessageQueueMetricReadPlan(descriptor).isPerPeriodCount;
+        },
+      );
+    expect(counts.length).toBeGreaterThan(20);
+    for (const descriptor of counts) {
+      expect([
+        descriptor.metricName,
+        getMessageQueueMetricReadPlan(descriptor).mode,
+      ]).toEqual([descriptor.metricName, "pooled"]);
+    }
+  });
+});
+
+describe("a monitor evaluates the number the page shows", () => {
+  test.each(GAUGES)(
+    "%s: a pooled read is the monitor's own query, unless the seed says otherwise",
+    (_id: string, descriptor: MessageQueueMetricDescriptor) => {
+      const plan: MessageQueueMetricReadPlan =
+        getMessageQueueMetricReadPlan(descriptor);
+      const seed: MessageQueueMetricMonitorSeed | null =
+        getMessageQueueMetricMonitorSeed(descriptor);
+      if (plan.mode !== "pooled" || !seed || seed.note !== null) {
+        return;
+      }
+      expect(seed.aggregationType).toBe(plan.aggregationType);
+      expect(seed.groupByAttributeKeys).toEqual([]);
+      expect(seed.seriesTotal).toBeNull();
+    },
+  );
+
+  test("the Kafka, Pulsar and RocketMQ lags read pooled, as their monitors do", () => {
+    for (const [system, metricName] of [
+      ["kafka", "kafka.consumer_group.lag_sum"],
+      ["kafka", "kafka.consumer_group.lag"],
+      ["pulsar", "pulsar_subscription_back_log"],
+      ["rocketmq", "rocketmq_consumer_lag_messages"],
+      ["rocketmq", "rocketmq_consumer_ready_messages"],
+    ] as Array<[string, string]>) {
+      const descriptor: MessageQueueMetricDescriptor = descriptorOf(
+        system,
+        metricName,
+      );
+      expect([
+        metricName,
+        getMessageQueueMetricReadPlan(descriptor).mode,
+      ]).toEqual([metricName, "pooled"]);
+      expect(getMessageQueueMetricMonitorSeed(descriptor)).toEqual(
+        expect.objectContaining({
+          aggregationType: AggregationType.Max,
+          groupByAttributeKeys: [],
+          note: null,
+        }),
+      );
+    }
+  });
+});