Metacat has built in replication to allow different Metacat servers to share data between themselves. The replication scheme that Metacat uses is both push and pull. There are several triggers that can start a replication mechanism.

• Delta-T monitoring. At a set time interval a server checks each of the other servers in its list for updated documents
• INSERT trigger. Whenever a document is inserted, the server notifies the remote hosts in its list that it has a new file available.
• UPDATE trigger. Whenever a document is updated, the server notifies each server in its list of the update.
• File locking. When a local user tries to alter a document on a local server that belongs to a remote server, the local server must first obtain a lock on that file. Once the lock is obtained, the file can be updated, then it is force replicated out to each server in the list. The lock ensures that the remote copy is up to date and that an older file does not overwrite a newer one. Only a documents home server can give a lock for that file to be altered.

Each server contains a list of servers to which it can replicate. One-way replication is enabled by the 'replicate' flag in the list. The server list may look like the following.

The server list is kept in a table in the database called xml_replication. Localhost must always be the first entry in the table and have a serverid of 1. The server field must always point to the other server's replication servlet, hence the servlet/replication on the end of both of the sample servers. Note that any port numbers (if your servlet engine is not running on port 80) must also be included. The replicate flag is set to 1 if you want this server to copy data TO the remote host. (Note that both servers (the local host and the remote host) must have each other in their respective tables or replication will not take place.)

Our three servers, snoopy, alpha and dev are all set up to replicate between themselves. Snoopy is a one way replicator. Meaning that it only pushes data to alpha and dev but does not pull back from them. This is achieved by dev and alpha setting snoopy's 'replicate' value to 0 indicating that they do not want to send their files to snoopy. Alpha and dev have a two-way replication agreement since each of them have a 1 in their 'replicate' value for the other.

There is an html control panel for controling replication. After installing Metacat, you can access it by going through the Metacat servlet context you have setup and calling up replControl.html. For instance, if you setup a Metacat servlet instance called 'Metacat' you would probably type http://server.domain.com:8080/Metacat/replControl.html. The control panel is an easy interface for adding/removing/altering servers and starting the delta-T handler. It will also allow you to 'force replicate' your server list. This is useful if you want to initialize the state of one Metacat server from an existing state of another (i.e. copy all of the data from an existing server).