Running the Dataman Server

The dataman server consists of three executable programs. The programs are called dataman, dataman_con, and dataman_srv. Dataman is the parent of the other two, and it's job is to start and monitor the other two programs. Dataman_con is the connection manager. It manages the connections that clients make to the database server. When running, you will see one instance of the connection manager for each connection in addition to the main connection manager. The dataman_srv process is the actual database server.

If the dataman process catches the termination of either of the child processes it will send a mail to the dataman user. As a result you will need to create either a dataman user, or alias a user to receive mail destined for dataman. After sending a mail the terminated program gets re-started.

The connection server listens on port 8758. So, you should make sure nothing else is running on this port. It is also a good idea to make an entry in /etc/services so that netstat and others can see that dataman is listening.

The package as distributed contains an example RC script that shows a simple way to start the server processes. The server usage is:

    dataman [-D|q|s|t] [-m size] [ -n num]
        -D start with debugging turned on
        -m size where size is the number of Kilobytes to
            use for shared memory segments (1,2,3...)
        -n num where num is the numer of worker threads
            in the database server
        -q query status of dataman
        -s controlled shutdown of dataman processes
        -t forced termination of dataman processes
        -v print version number and exit

Debugging will produce copious amounds of information in logfiles located in the /tmp directory. Only turn this on if you know what you're looking for, and will use it for only short runs.

Querying the status of dataman will tell you if dataman is running or not and the status of the pid lock files. If dataman terminated abnormally there typically are files left named .dataman*.pid in the /tmp directory. These will be automatically cleaned up on the restart of dataman. If dataman is running the return from the query is a 1, and a 2 if it isn't running (good for use in scripts).

The controlled shutdown of dataman is a nice way to shut down the server processes. If any clients are attached to the server at the time the shutdown command is issued, the connection server stops accepting new connections, and the processes wait until all of clients terminate. Then the servers close down and the processes terminate.

The forced shutdown, however, is not as nice. This forces down all of the server processes running. Any clients connected at the time are forcibly disconnected and the system terminates. Any 'in-process' operations may not be completed.

The -m and -n switches are new with version 3.3.0

There is a shared memory segment that is used for communication between each active connection and the server. By default this segment is 1 Kilobyte (1024 bytes). The idea is that most data records are smaller than 1K. The -m switch lets you set the size of this segment up to 4K in 1K increments. The values may be 1, 2, 3, or 4. If data records are larger than the defined size, the server and connection manager take turns using the memory segment to copy data in and out. Obviously, if you have large data records this can slow down the communications. This should not be much of a problem though, because then you start impacting the TCP/IP connection between client and server, and that will be much more noticable than the memory operations.

You may configure the number of worker threads for the database server with the -n switch. It is a good idea to strike a balance between the number of simultaneous connections you expect to handle and the number of CPUs you have on your system. Most connections stand idle most of the time, so having a thread for each connection will waste resources. If you have only one CPU the only time you will get thread switches is during a long syscall or on a timeslice. Many threads on a single CPU system thus doesn't buy you much. The default number of worker threads is 10.