The Cluster Membership object

Table of Contents

Introduction

The membership component in the Apache Tribes Channel is responsible for dynamic discovery of other members(nodes) in the cluster.

Default Implementation

The default implementation of the cluster group notification is built on top of multicast heartbeats sent using UDP packets to a multicast IP address. Cluster members are grouped together by using the same multicast address/port combination. Each member sends out a heartbeat with a given interval (frequency), and this heartbeat is used for dynamic discovery. In a similar fashion, if a heartbeat has not been received in a timeframe specified by dropTime ms. a member is considered suspect and the channel and any membership listener will be notified.

Attributes

Multicast Attributes

Attribute Description
className

The default value is org.apache.catalina.tribes.membership.McastService and is currently the only implementation. This implementation uses multicast heartbeats for member discovery.

address

The multicast address that the membership will broadcast its presence and listen for other heartbeats on. The default value is 228.0.0.4 Make sure your network is enabled for multicast traffic.
The multicast address, in conjunction with the port is what creates a cluster group. To divide up your farm into several different group, or to split up QA from production, change the port or the address
Previously known as mcastAddr.

port

The multicast port, the default value is 45564
The multicast port, in conjunction with the address is what creates a cluster group. To divide up your farm into several different group, or to split up QA from production, change the port or the address

frequency

The frequency in milliseconds in which heartbeats are sent out. The default value is 500 ms.
In most cases the default value is sufficient. Changing this value, simply changes the interval in between heartbeats.

dropTime

The membership component will time out members and notify the Channel if a member fails to send a heartbeat within a give time. The default value is 3000 ms. This means, that if a heartbeat is not received from a member in that timeframe, the membership component will notify the cluster of this.
On a high latency network you may wish to increase this value, to protect against false positives.
Apache Tribes also provides a TcpFailureDetector that will verify a timeout using a TCP connection when a heartbeat timeout has occurred. This protects against false positives.

bind

Use this attribute if you wish to bind your multicast traffic to a specific network interface. By default, or when this attribute is unset, it tries to bind to 0.0.0.0 and sometimes on multihomed hosts this becomes a problem.

ttl

The time-to-live setting for the multicast heartbeats. This setting should be a value between 0 and 255. The default value is VM implementation specific.

domain

Apache Tribes has the ability to logically group members into domains, by using this domain attribute. The org.apache.catalina.tribes.Member.getDomain() method returns the value specified here.

soTimeout

The sending and receiving of heartbeats is done on a single thread, hence to avoid blocking this thread forever, you can control the SO_TIMEOUT value on this socket.
If a value smaller or equal to 0 is presented, the code will default this value to frequency

recoveryEnabled

In case of a network failure, Java multicast socket don't transparently fail over, instead the socket will continuously throw IOException upon each receive request. When recoveryEnabled is set to true, this will close the multicast socket and open a new socket with the same properties as defined above.
The default is true.

recoveryCounter

When recoveryEnabled==true this value indicates how many times an error has to occur before recovery is attempted. The default is 10.

recoverySleepTime

When recoveryEnabled==true this value indicates how long time (in milliseconds) the system will sleep in between recovery attempts, until it recovers successfully. The default is 5000 (5 seconds).

localLoopbackDisabled

Membership uses multicast, it will call java.net.MulticastSocket.setLoopbackMode(localLoopbackDisabled). When localLoopbackDisabled==true multicast messages will not reach other nodes on the same local machine. The default is false.