Task Queue Metrics to identify the cause for Overload Protection
- Harsh Bansal
HiveMQ Overload protection uses the total number of tasks to ascertain when to become active and throttle incoming work.
While growing task queues are generally a sign of overload on a broker, short spikes are normal and not an immediate reason to worry.
HiveMQ uses task queues to manage and execute asynchronous tasks efficiently. It receives incoming work, splits it into tasks and adds them to a queue which is processed one by one in a controlled and orderly manner. Each task is typically a discrete unit of work and can be executed independently.
The tasks can be grouped into the following 4 categories:
Persistence tasks
These tasks relate to disk reads and writes.
Total tasks
com.hivemq.persistence-executor.queued
com.hivemq.persistence.executor.total.tasksConnection establishment tasks
com.hivemq.persistence_executor.client-events.tasks
com.hivemq.persistence_executor.client-session.tasksMessaging tasks
com.hivemq.persistence_executor.incoming-message-flow.tasks
com.hivemq.persistence_executor.queued.messages.tasksSingle-writer-callback tasks
These tasks relate to data replication between individual cluster members
com.hivemq.single-writer-callback-executor.*.queued
Netty tasks
These tasks are related to Network Input-Output throughput
Extension tasks
These tasks denote SDK calls made by extensions
com.hivemq.extension.task.executors.queued
How to query these metrics?