The ActiveMQ InactivityMonitor is an active thread that checks the connection is still active and if it suspects the connection is not functioning correctly, it closes the connection.
Connections are monitored by:
- ensuring data is read from the connection during the specified time period (Max Inactivity Duration)
- writing a KeepAliveInfo message to the connection if no normal activemq traffic is sent across the connection during the specified time period
Each connection has two InactivityMonitors associated, one on each end of the connection. The InactivityMonitor expects to receive data on the connection during a specified time period. If normal activemq traffic has not been sent across the connection during that period, it expects to receive a KeepAliveInfo message sent by the InactivityMonitor on the other end of the connection.
Using the default values; if no data has been written or read from the connection for 30 seconds, the InactivityMonitor kicks in. The InactivityMonitor throws an InactivityIOException and shuts down the transport associated with the connection. This results in the following DEBUG logging:
2012-06-26 17:13:55,712 | DEBUG | 30000 ms elapsed since last read check. | org.apache.activemq.transport.AbstractInactivityMonitor
| InactivityMonitor ReadCheck
2012-06-26 17:13:55,712 | DEBUG | No message received since last read check for tcp:///127.0.0.1:52659! Throwing InactivityIOException. |
org.apache.activemq.transport.AbstractInactivityMonitor | InactivityMonitor ReadCheck
2012-06-26 17:13:55,714 | DEBUG | Transport Connection to: tcp://127.0.0.1:52659 failed:
org.apache.activemq.transport.InactivityIOException: Channel was inactive for too (>30000) long: tcp://127.0.0.1:52659 |
org.apache.activemq.broker.TransportConnection.Transport | InactivityMonitor Async Task:
org.apache.activemq.transport.InactivityIOException: Channel was inactive for too (>30000) long: tcp://127.0.0.1:52659
How can the InactivityMonitor be configured further?
There are some configuration options to further fine tune the InactivityMonitor. Note, for most use cases the default values are just fine.
|wireFormat.maxInactivityDuration||30000 msecs||Time after with the connection is closed by the broker if no data was received.|
|wireFormat.maxInactivityDurationInitalDelay||10000 msecs||Specifies the maximum delay before inactivity monitoring is started on the connection. It can be useful if a broker is under load with many connections being created concurrently.|
|transport.useInactivityMonitor||true||A value of false disables the InactivityMonitor completely and connections will never time out. By default it is enabled.|
|transport.useKeepAlive||true||Whether or not to send a KeepAliveInfo on an idle connection to prevent it from timing out. Enabled by default. Disabling the keep alive will still make connections time out if no data was received on the connection for the specified amount of time.|
These parameters can be specified directly on the client side connection URL, e.g. tcp://localhost:61616?wireFormat.maxInactivityDuration=30000 or on the broker's transport connector url.
<transportConnector name="openwire" uri="tcp://0.0.0.0:61616?wireFormat.maxInactivityDuration=30000&wireFormat.maxInactivityDurationInitalDelay=10000"/>
What happens if the maxInactivityDuration and maxInactivityDurationInitalDelay are set to different values on either side of the same connection?
At startup the InactivityMonitor negotiates the appropriate maxInactivityDuration and maxInactivityDurationInitalDelay. The shortest duration is taken for the connection.
Can the InactivityMonitor on a connection be disabled?
Setting transport.useInactivityMonitor=false will disable the InactivityMonitor. Configuring wireFormat.maxInactivityDuration=0 will achieve the same result.
Potential issues to be aware of: