[Task Manager] Log at different levels based on the state (elastic#10…

…1751) (elastic#102402)

* Log at different levels based on the state

* Fix types and add tests

* Remove unnecessary code

* Add more descriptive message

* Partially fix failing tests

* Move into separate function

* Get rid of customStatus in favor of moving the logging logic to a separate, mockable function

* Remove debug logging

* Do not log as an error if the stats are empty

* PR feedback

* Add docker whitelist

* alpha order

* English is hard

* Removing extra newline

* PR feedback around ignoring capacity estimation

* Move json utils

docs/settings/task-manager-settings.asciidoc

@@ -28,6 +28,9 @@ Task Manager runs background tasks by polling for work on an interval.  You can
   | `xpack.task_manager.max_workers`
   | The maximum number of tasks that this Kibana instance will run simultaneously.  Defaults to 10.
     Starting in 8.0, it will not be possible to set the value greater than 100.
+
+  | `xpack.task_manager.monitored_stats_warn_delayed_task_start_in_seconds`
+  | The amount of seconds we allow a task to delay before printing a warning server log.  Defaults to 60.
 |===
 
 [float]

src/dev/build/tasks/os_packages/docker_generator/resources/base/bin/kibana-docker

@@ -327,6 +327,7 @@ kibana_vars=(
     xpack.task_manager.monitored_aggregated_stats_refresh_rate
     xpack.task_manager.monitored_stats_required_freshness
     xpack.task_manager.monitored_stats_running_average_window
+    xpack.task_manager.monitored_stats_warn_delayed_task_start_in_seconds
     xpack.task_manager.monitored_task_execution_thresholds
     xpack.task_manager.poll_interval
     xpack.task_manager.request_capacity

x-pack/plugins/task_manager/server/config.test.ts

@@ -20,6 +20,7 @@ describe('config validation', () => {
         "monitored_aggregated_stats_refresh_rate": 60000,
         "monitored_stats_required_freshness": 4000,
         "monitored_stats_running_average_window": 50,
+        "monitored_stats_warn_delayed_task_start_in_seconds": 60,
         "monitored_task_execution_thresholds": Object {
           "custom": Object {},
           "default": Object {
@@ -68,6 +69,7 @@ describe('config validation', () => {
         "monitored_aggregated_stats_refresh_rate": 60000,
         "monitored_stats_required_freshness": 4000,
         "monitored_stats_running_average_window": 50,
+        "monitored_stats_warn_delayed_task_start_in_seconds": 60,
         "monitored_task_execution_thresholds": Object {
           "custom": Object {},
           "default": Object {
@@ -103,6 +105,7 @@ describe('config validation', () => {
         "monitored_aggregated_stats_refresh_rate": 60000,
         "monitored_stats_required_freshness": 4000,
         "monitored_stats_running_average_window": 50,
+        "monitored_stats_warn_delayed_task_start_in_seconds": 60,
         "monitored_task_execution_thresholds": Object {
           "custom": Object {
             "alerting:always-fires": Object {

x-pack/plugins/task_manager/server/config.ts

@@ -18,6 +18,7 @@ export const DEFAULT_VERSION_CONFLICT_THRESHOLD = 80;
 // Refresh aggregated monitored stats at a default rate of once a minute
 export const DEFAULT_MONITORING_REFRESH_RATE = 60 * 1000;
 export const DEFAULT_MONITORING_STATS_RUNNING_AVERGAE_WINDOW = 50;
+export const DEFAULT_MONITORING_STATS_WARN_DELAYED_TASK_START_IN_SECONDS = 60;
 
 export const taskExecutionFailureThresholdSchema = schema.object(
   {
@@ -109,6 +110,10 @@ export const configSchema = schema.object(
         defaultValue: {},
       }),
     }),
+    /* The amount of seconds we allow a task to delay before printing a warning server log */
+    monitored_stats_warn_delayed_task_start_in_seconds: schema.number({
+      defaultValue: DEFAULT_MONITORING_STATS_WARN_DELAYED_TASK_START_IN_SECONDS,
+    }),
   },
   {
     validate: (config) => {

x-pack/plugins/task_manager/server/integration_tests/managed_configuration.test.ts

@@ -37,6 +37,7 @@ describe('managed configuration', () => {
       version_conflict_threshold: 80,
       max_poll_inactivity_cycles: 10,
       monitored_aggregated_stats_refresh_rate: 60000,
+      monitored_stats_warn_delayed_task_start_in_seconds: 60,
       monitored_stats_required_freshness: 4000,
       monitored_stats_running_average_window: 50,
       request_capacity: 1000,

x-pack/plugins/task_manager/server/lib/calculate_health_status.mock.ts

@@ -0,0 +1,14 @@
+/*
+ * Copyright Elasticsearch B.V. and/or licensed to Elasticsearch B.V. under one
+ * or more contributor license agreements. Licensed under the Elastic License
+ * 2.0; you may not use this file except in compliance with the Elastic License
+ * 2.0.
+ */
+
+const createCalculateHealthStatusMock = () => {
+  return jest.fn();
+};
+
+export const calculateHealthStatusMock = {
+  create: createCalculateHealthStatusMock,
+};

x-pack/plugins/task_manager/server/lib/calculate_health_status.ts

@@ -0,0 +1,79 @@
+/*
+ * Copyright Elasticsearch B.V. and/or licensed to Elasticsearch B.V. under one
+ * or more contributor license agreements. Licensed under the Elastic License
+ * 2.0; you may not use this file except in compliance with the Elastic License
+ * 2.0.
+ */
+
+import { isString } from 'lodash';
+import { JsonValue } from '@kbn/common-utils';
+import { HealthStatus, RawMonitoringStats } from '../monitoring';
+import { TaskManagerConfig } from '../config';
+
+export function calculateHealthStatus(
+  summarizedStats: RawMonitoringStats,
+  config: TaskManagerConfig
+): HealthStatus {
+  const now = Date.now();
+
+  // if "hot" health stats are any more stale than monitored_stats_required_freshness (pollInterval +1s buffer by default)
+  // consider the system unhealthy
+  const requiredHotStatsFreshness: number = config.monitored_stats_required_freshness;
+
+  // if "cold" health stats are any more stale than the configured refresh (+ a buffer), consider the system unhealthy
+  const requiredColdStatsFreshness: number = config.monitored_aggregated_stats_refresh_rate * 1.5;
+
+  /**
+   * If the monitored stats aren't fresh, return a red status
+   */
+  const healthStatus =
+    hasStatus(summarizedStats.stats, HealthStatus.Error) ||
+    hasExpiredHotTimestamps(summarizedStats, now, requiredHotStatsFreshness) ||
+    hasExpiredColdTimestamps(summarizedStats, now, requiredColdStatsFreshness)
+      ? HealthStatus.Error
+      : hasStatus(summarizedStats.stats, HealthStatus.Warning)
+      ? HealthStatus.Warning
+      : HealthStatus.OK;
+  return healthStatus;
+}
+
+function hasStatus(stats: RawMonitoringStats['stats'], status: HealthStatus): boolean {
+  return Object.values(stats)
+    .map((stat) => stat?.status === status)
+    .includes(true);
+}
+
+/**
+ * If certain "hot" stats are not fresh, then the _health api will should return a Red status
+ * @param monitoringStats The monitored stats
+ * @param now The time to compare against
+ * @param requiredFreshness How fresh should these stats be
+ */
+function hasExpiredHotTimestamps(
+  monitoringStats: RawMonitoringStats,
+  now: number,
+  requiredFreshness: number
+): boolean {
+  const diff =
+    now -
+    getOldestTimestamp(
+      monitoringStats.last_update,
+      monitoringStats.stats.runtime?.value.polling.last_successful_poll
+    );
+  return diff > requiredFreshness;
+}
+
+function hasExpiredColdTimestamps(
+  monitoringStats: RawMonitoringStats,
+  now: number,
+  requiredFreshness: number
+): boolean {
+  return now - getOldestTimestamp(monitoringStats.stats.workload?.timestamp) > requiredFreshness;
+}
+
+function getOldestTimestamp(...timestamps: Array<JsonValue | undefined>): number {
+  const validTimestamps = timestamps
+    .map((timestamp) => (isString(timestamp) ? Date.parse(timestamp) : NaN))
+    .filter((timestamp) => !isNaN(timestamp));
+  return validTimestamps.length ? Math.min(...validTimestamps) : 0;
+}

x-pack/plugins/task_manager/server/lib/log_health_metrics.mock.ts

@@ -0,0 +1,14 @@
+/*
+ * Copyright Elasticsearch B.V. and/or licensed to Elasticsearch B.V. under one
+ * or more contributor license agreements. Licensed under the Elastic License
+ * 2.0; you may not use this file except in compliance with the Elastic License
+ * 2.0.
+ */
+
+const createLogHealthMetricsMock = () => {
+  return jest.fn();
+};
+
+export const logHealthMetricsMock = {
+  create: createLogHealthMetricsMock,
+};