You are viewing an old version of this page. View the current version.

Compare with Current View Page History

« Previous Version 4 Next »

Failover

On each request to the server, the client will send the version number associated with the list of servers in the cluster it is aware of. Initially this version will be zero and the list will be empty. Only when the server sees the client has an old list will the server send the updated list. This is an important distinction as the list is not transmitted back and forth on every request, only on change. If the membership of the cluster is stable there is essentially no clustering overhead to the protocol – 8 byte overhead to each request and 1 byte on each response – so you will not see an exponential slowdown in response times the more members are added to the cluster. This new list takes affect for all proxies that share the same connection.

When a server shuts down, more connections are refused, existing connections not in mid-request are closed, any remaining connections are closed immediately after completion of the request in progress and clients can failover gracefully to the next server in the list. If a server crashes requests are retried on the next server in the list (or depending on the ConnectionStrategy). This failover pattern is followed until there are no more servers in the list at which point the client attempts a final multicast search (if it was created with a multicast PROVIDER_URL) before abandoning the request and throwing an exception to the caller.

By default, the failover is ordered but random selection is supported. The multicast discovery aspect of the client adds a nice randomness to the selection of the first server.

Discovery

Each discoverable service has a URI which is broadcast as a heartbeat to other servers in the cluster. This URI advertises the service's type, its cluster group, and its location in the format of 'group:type:location'. Say for example "cluster1:ejb:ejbd://thehost:4201". The URI is sent out repeatedly in a pulse and its presence on the network indicates its availability and its absence indicates the service is no longer available.

The sending of this pulse (the heartbeat) can be done via UDP or TCP: multicast and "multipoint" respectively. More on that in the following section. The rate at which the heartbeat is pulsed to the network can be specified via the 'heart_rate' property. The default is 500 milliseconds. This rate is also used when listening for services on the network. If a service goes missing for the duration of 'heart_rate' multiplied by 'max_missed_heartbeats', then the service is considered dead.

The 'group' property, cluster1 in the example, is used to dissect the servers on the network into smaller logical clusters. A given server will broadcast all it's services with the group prefixed in the URI, as well it will ignore any services it sees broadcast if they do not share the same group name.

Multicast

Multicast is the preferred way to broadcast the heartbeat on the network. The simple technique of broadcasting a non-changing service URI on the network has specific advantages to multicast. The URI itself is essentially stateless and there is no "i'm alive" URI or an "i'm dead" URI.

In this way the issues with UDP being unordered and unreliable melt away as state is no longer a concern and packet sizes are always small. Complicated libraries that ride atop UDP and attempt to offer reliability (retransmission) and ordering on UDP can be avoided. As well the advantages UDP has over TCP are retained as there are no java layers attempting to force UDP communication to be more TCP-like. The simple design means UDP/Multicast is only used for discovery and from there on out critical information is transmitted over TCP/IP which is obviously going to do a better job at ensuring reliability and ordering.

The multicast functionality is not just for servers to find each other in a cluster, it can also be used for EJB clients to discover a server. A special "multicast://" URL can be used in the InitialContext properties to signify that multicast should be used to seed the connection process. Such as:

   Properties properties = new Properties();
   properties.setProperty(Context.INITIAL_CONTEXT_FACTORY, "org.apache.openejb.client.RemoteInitialContextFactory");
   properties.setProperty(Context.PROVIDER_URL, "multicast://239.255.2.3:6142");
   InitialContext remoteContext = new InitialContext(properties);

The URL has optional query parameters such as "schemes" and "group" and "timeout" which allow you to zero in on a particular type of service of a particular cluster group as well as set how long you are willing to wait in the discovery process till finally giving up. The first matching service that it sees "flowing" around on the UDP stream is the one it picks and sticks to for that and subsequent requests, ensuring UDP is only used when there are no other servers to talk to.

Note that EJB clients do not need to use multicast to find a server. If the client knows the URL of a server in the cluster, it may use it and connect directly to that server, at which point that server will share the full list of its peers.

Multipoint

Multipoint simply refers to servers in the cluster joining together to create a fully connected mesh topology. This is done with TCP and one connection open between each server in the cluster. The handling of these sockets is all asynchronous Java NIO code which allows the server to handle many connections (all of them) with one thread. From a pure threading perspective, the option is extremely efficient with just one thread to listen and broadcast to many peers.

A peer can join the cluster by simply connecting to any one of the other servers in the cluster. When that happens, then two servers will exchange their peer lists and each will begin initiating connections to any peers that were not previously known. It is possible in this process that two servers learn of each other at the same time and each attempts to connect to the other, resulting in two connections between the same two servers. When this happens both servers will detect the extra connection and one of the connections will be dropped and one will be kept. In practice this race condition rarely happens and can be avoided almost entirely by fanning out server startup by as little as 100 milliseconds.

The general disadvantage of this topology is the number of connections required. The number of connections for the network of servers is equal to "(n * n - n) / 2 ", where n is the number of servers. For example, with 5 servers you need 10 connections, with 10 servers you need 45 connections, and with 50 servers you need 1225 connections. This is of course the number of connections across the entire network, each individual server only needs "n - 1" connections.

  • No labels