Task Manager health monitoring
editTask Manager health monitoring
editThis functionality is in technical preview and may be changed or removed in a future release. Elastic will work to fix any issues, but features in technical preview are not subject to the support SLA of official GA features.
The Task Manager has an internal monitoring mechanism to keep track of a variety of metrics, which can be consumed with either the health monitoring API or the Kibana server log.
The health monitoring API provides a reliable endpoint that can be monitored. Consuming this endpoint doesn’t cause additional load, but rather returns the latest health checks made by the system. This design enables consumption by external monitoring services at a regular cadence without additional load to the system.
Each Kibana instance exposes its own endpoint at:
$ curl -X GET api/task_manager/_health
Monitoring the _health
endpoint of each Kibana instance in the cluster is the recommended method of ensuring confidence in mission critical services such as Alerting, Actions, and Reporting.
Configuring the monitored health statistics
editThe health monitoring API monitors the performance of Task Manager out of the box. However, certain performance considerations are deployment specific and you can configure them.
A health threshold is the threshold for failed task executions. Once a task exceeds this threshold, a status of warn
or error
is set on the task type execution. To configure a health threshold, use the xpack.task_manager.monitored_task_execution_thresholds
setting. You can apply this this setting to all task types in the system, or to a custom task type.
By default, this setting marks the health of every task type as warning
when it exceeds 80% failed executions, and as error
at 90%.
Set this value to a number between 0 to 100. The threshold is hit when the value exceeds this number.
To avoid a status of error
, set the threshold at 100. To hit error
the moment any task fails, set the threshold to 0.
Create a custom configuration to set lower thresholds for task types you consider critical, such as alerting tasks that you want to detect sooner in an external monitoring service.
xpack.task_manager.monitored_task_execution_thresholds: default: error_threshold: 70 warn_threshold: 50 custom: "alerting:.index-threshold": error_threshold: 50 warn_threshold: 0
A default configuration that sets the system-wide |
|
A custom configuration for the |
Consuming health stats
editThe health API is best consumed by via the /api/task_manager/_health
endpoint.
Additionally, there are two ways to consume these metrics:
Debug logging
The metrics are logged in the Kibana DEBUG
logger at a regular cadence.
To enable Task Manager debug logging in your Kibana instance, add the following to your kibana.yml
:
logging: loggers: - context: plugins.taskManager appenders: [console] level: debug
These stats are logged based on the number of milliseconds set in your xpack.task_manager.poll_interval
setting, which could add substantial noise to your logs. Only enable this level of logging temporarily.
Automatic logging
By default, the health API runs at a regular cadence, and each time it runs, it attempts to self evaluate its performance. If this self evaluation yields a potential problem,
a message will log to the Kibana server log. In addition, the health API will look at how long tasks have waited to start (from when they were scheduled to start). If this number exceeds a configurable threshold (xpack.task_manager.monitored_stats_health_verbose_log.warn_delayed_task_start_in_seconds
), the same message as above will log to the Kibana server log.
This message looks like:
Detected potential performance issue with Task Manager. Set 'xpack.task_manager.monitored_stats_health_verbose_log.enabled: true' in your Kibana.yml to enable debug logging`
If this message appears, set xpack.task_manager.monitored_stats_health_verbose_log.enabled
to true
in your kibana.yml
. This will start logging the health metrics at either a warn
or error
log level, depending on the detected severity of the potential problem.
Making sense of Task Manager health stats
editThe health monitoring API exposes three sections: configuration
, workload
and runtime
:
Configuration |
This section summarizes the current configuration of Task Manager. This includes dynamic configurations that change over time, such as |
Workload |
This section summarizes the work load across the cluster, including the tasks in the system, their types, and current status. |
Runtime |
This section tracks execution performance of Task Manager, tracking task drift, worker load, and execution stats broken down by type, including duration and execution results. |
Capacity Estimation |
This section provides a rough estimate about the sufficiency of its capacity. As the name suggests, these are estimates based on historical data and should not be used as predictions. Use these estimations when following the Task Manager Scaling guidance. |
Each section has a timestamp
and a status
that indicates when the last update to this section took place and whether the health of this section was evaluated as OK
, Warning
or Error
.
The root status
indicates the status
of the system overall.
The Runtime status
indicates whether task executions have exceeded any of the configured health thresholds. An OK
status means none of the threshold have been exceeded. A Warning
status means that at least one warning threshold has been exceeded. An Error
status means that at least one error threshold has been exceeded.
Some tasks (such as connectors) will incorrectly report their status as successful even if the task failed. The runtime and workload block will return data about success and failures and will not take this into consideration.
To get a better sense of action failures, please refer to the Event log index for more accurate context into failures and successes.
The Capacity Estimation status
indicates the sufficiency of the observed capacity. An OK
status means capacity is sufficient. A Warning
status means that capacity is sufficient for the scheduled recurring tasks, but non-recurring tasks often cause the cluster to exceed capacity. An Error
status means that there is insufficient capacity across all types of tasks.
By monitoring the status
of the system overall, and the status
of specific task types of interest, you can evaluate the health of the Kibana Task Management system.