This document describes how to make use of the facilities in Xapian for distributed searches.
There are two sides to the distributed searching. The client end is the program initiating the search on behalf of a user, and the server end is the program which provides a searching interface over a set of databases for the client. There can be many servers, with many clients sharing them. In theory, a server can also be a client to other servers, but this may not be very useful or efficient.
The client runs queries in the same way that it would on local databases, but with different database arguments. Instead of type, eg, "quartz", use "remote". The extra parameters describe how to connect to the server. Once the database is opened, the query process is identical to any other. Using a stub database with autobackend is a good way to wrap up access to a remote database in a neat way.
The remote backend currently support two client/server methods: prog and tcp. They both use the same protocol, although different means to contact the server.
The prog method spawns a program when the database is opened, and communicates with it over a Unix domain socket. This isn't intended for production use, but is useful for debugging and testing. The omprogsrv program, which is designed to be the program spawned, currently only uses text files indexed into inmemory databases.
From the client end, the following database parameters are used:
The tcp method uses TCP/IP sockets to connect to a running server on a remote (or indeed local) machine.
The client's database parameters are:
The arguments omtcpsrv currently knows are:
Once started, the server will run and listen for connections on the configured port, currently handling them one by one (although this will change at some point).
A remote search should behave just like the equivalent local one, with the exception of a few currently unimplemented features (e.g. match decision functors).
Exceptions are propagated across the link and thrown again at the client end.