Running SQL Relay

Starting Everything at Once:

The sqlr-start program is provided to simplify the startup process.

sqlr-start -id ID -config CONFIGFILE

The sqlr-start program starts up one instance of sqlr-listener, one instance of sqlr-scaler and the number of sqlr-connection daemons specified in the config file for the given id.

The sqlr-start program also checks to see if a sqlr-cachemanager is running. If one is then it leaves it running. If not then it starts one.

See the section about sqlr-cachemanager below for an important note about it.

Debugging and Tracing:

The instance tag of the CONFIGFILE contains an optional debug attribute. When it is set, the sqlr-start program starts up debugging versions of the sqlr-listener and/or sqlr-connection daemons. These daemons output debugging information to files in /usr/local/firstworks/var/sqlrelay/debug. See Configuring SQL Relay for more information on the debug attribute.

The sqlr-start program also takes an optional -strace parameter. When run as root, if the -strace parameter is used, sqlr-start starts the sqlr-connection daemons using strace -ff -o sqlr-connection-strace causing files to be written to the root directory. See the strace man page for more information.

Starting Daemons Individually:

If for some reason you don't want to use sqlr-start, you can start the daemons individually.

Setting Up a Listener:

The first step in is setting up a listener. The job of the listener is to connect to a TCP port and wait for queries. When it gets one, it waits for an available connection daemon and hands off the client to it.

Run the sqlr-listener daemon to establish a listener.

sqlr-listener -id ID -config CONFIGFILE

CONFIGFILE is the filename of the configuration file to use. The default configuration file is /usr/local/firstworks/etc/sqlrelay.conf

ID is an identifier that associates a listener with a pool of connections.

Only one listener needs to be established for a given ID.

Setting Up a Connection:

Connecting to a database is the next step. To estabish a connection, run one or more of the sqlr-connection daemons.

sqlr-connection-"dbase" -id ID -config CONFIGFILE -connectionid CONNECTIONID

Where "dbase" is one of oracle7, oracle8, mysql, msql, postgresql, sqlite, lago, odbc, db2, interbase, sybase or freetds.

CONFIGFILE is the configuration file to use when starting the daemon. The default configuration file is /usr/local/firstworks/etc/sqlrelay.conf

ID is just an identifier. Starting multiple connections with the same ID make them available as a pool of connections. The more connections you start, the more queries can be executed in parallel. In addition, the ID argument specifies which config file entry to use.

CONNECTIONID is another identifier. When using replicated or clustered databases, within a pool of connections, some connections will connected to 1 machine and some will connect to another. This argument tells the connection daemon which machine to connect to. When not using clustered or replicated databases, this id will be the same for every connection in the pool.

Setting Up the Scaler:

The third step is running the scaler. You can start as many sqlr-connection daemons as you like. If you start fewer than the maximum number defined for the ID, the sqlr-scaler daemon will fire up new connections on demand. The new connections will time out after a period of inactivity and shut down on their own.

Run the sqlr-scaler daemon to establish a scaler.

sqlr-scaler -id ID -config CONFIGFILE

CONFIGFILE is the filename of the configuration file to use. The default configuration file is /usr/local/firstworks/etc/sqlrelay.conf

ID is the same as the ID argument in the sqlr-connection daemon and associates a scaler with a pool of connections.

Only one scaler needs to be established for a given ID.

Setting Up the Cache Manager:

The fourth step is running the cache manager. Clients can cache result sets with a time-to-live on them. The cache manager goes through the cached result sets periodically and removes the ones that have expired.

Run the sqlr-cachemanager daemon to establish the cache manager.

sqlr-cachemanager [INTERVAL]

The optional INTERVAL argument specifies in seconds how often the cache manager scans the result sets. Each scan is scheduled from the end of the previous scan. If the argument is left off, the cache manager scans at a default interval of 30 seconds.

Only one cache manager needs to be started per machine.

Important Note: Since cache managers clean up after SQL Relay clients, not servers, they need to be run on machines which run clients that could cache result sets. These may not be the same machines that run the sqlr-listener and sqlr-connection daemons. If there is no SQL Relay installation on those machines, you should create the cache directory (/usr/local/firstworks/var/sqlrelay/cache unless changed at compile time), install the sqlr-cachemanager program by itself and set it up to run at boot time.

Extending, Killing and Restarting:

Once the daemons are up and running, clients can use them. In the event that more connections are required, additional sqlr-connection-"dbase" daemons may be started from the command line using the same ID and CONFIGFILE and will be immediately available to clients.

Connection daemons should not be killed once they have been started without restarting everything with that ID.

The sqlr-stop command is provided to kill running daemons. It accepts a single argument, the id of the instance to kill. Running it kills anything with "sqlr-" and that id in it's name. It uses grep, so it's not perfect. If you have id's like "web" and "webster" and run "sqlr-stop web" then it will kill both of them. Running it with no arguments will kill all SQL Relay processes.

Using the Command Line Interfaces:

Four command line utilities are provided for use with SQL Relay: query, fields, backupschema and sqlrsh. The syntax for each is:





The HOST, PORT, SOCKET, USER and PASSWORD arguments specify which listener to connect to and what authentication to use.

The query command executes the query specified in the QUERY parameter and returns the result set as a quote/comma/return seperated value list to standard output.

The fields command returns a comma seperated list of the fields in the table specified by the TABLE argument to standard output.

The backupschema command only works against oracle databases and uses the "select table_name from user_tables" query in combination with the query and fields commands to generate Oracle SQL Loader control files for all tables owned by the user a particular connection is logged in as. The command is useful for backing up an entire schema of data to a readily re-importable format.

sqlrsh is an interactive query tool similar to (though not as powerful as) SQL Plus, mysql, psql, isql and the like. Run it and type help; at the prompt for detailed usage instructions. When sqlrsh starts up, it reads and executes two rc files, the system rc file (most likely /usr/local/firstworks/etc/sqlrshrc) and a personal rc file .sqlrshrc in your home directory. These files should contain sqlrsh commands, ending in semicolons, seperated by carraige returns. Any commands may be used in the rc files, even queries.

Common Problems:

  • FreeTDS or Sybase connection fails with error: Connection refused.

    • The FreeTDS and Sybase libraries try to connect to port 4000 (or 1433 if FreeTDS is compiled with version 7.0 (MS SQL Server) support) by default. The environment variables TDSPORT and DBLIB_PORT must both be set to override the default port. As of version 0.24, the FreeTDS and Sybase connections' connectstring now accept a port argument which instruct the connections to set those variables.

      Sybase ASE 11.9.2 runs on port 4100 by default and MS SQL Server runs on port 1433 by default. If you try to connect to an instance of one of them and get an error, you can either reconfigure the instance to run on port 4000 or set the port argument of the connectstring in CONFIGFILE to the port that the instance is actually running on.

  • bind failed

    • The most common cause of this problem is configuring SQL Relay to listen on the same port that the database is listening on. For example, if your database is listening on port 4100 and you run SQL Relay on the same machine, you can't configure SQL Relay to listen on port 4100 or it will issue "bind failed" when the listener tries to run.

    • A slightly less common cause of this problem is configuring SQL Relay to listen on a port that some other service is already listening on. For example, web and cache servers often listen on port 8080 and IRC servers often listen on port 7000. You can see if some other service is listening on the port you want to SQL Relay to listen on by running netstat -ap | grep PORT where PORT is replaced with the port number that you'd like SQL Relay to listen on. If you get anything back from that command, then there is another service already listening on that port.

    • If you kill and restart the daemons and get the message: "bind failed." as the listener is starting, then there are 2 possibilities. First, all the daemons may not have been killed. In this case, kill them all and make sure they are dead by using ps -efal | grep sqlr- before restarting them. The second possibility is that the port the listener was listening on didn't get closed. Executing netstat -a | grep PORTNUMBER will reveal any connections still lingering on the port. If all the daemons are dead but the connections are still lingering, wait 2 minutes or so before restarting the daemons. The lingering connections should have timed out by then.

  • Everything starts fine but clients can't execute queries.

    • The most common cause of this problem is telling the client to connect to the port that the database is listening on rather than the port that SQL Relay is listening on. For example, if the database is listening on port 4100 and you have an instance of SQL Relay connected to it and have configured SQL Relay to listen on port 9000, then a common mistake would be to try to connect an SQL Relay client to port 4100 instead of port 9000.

    • SQL Relay daemons use semaphores and shared memory segments. If a daemon crashes unexpectedly, even if you kill all the other daemons, a semaphore or shared memory segment may still be hanging around. These can interfere with future attempts to start up daemons with the same ID. You can use the ipcs command to inspect the shared memory segments and semaphores and the ipcrm command to remove any lingering ones.

    • The SQL Relay daemons also use temporary files, usually located in /usr/local/firstworks/var/sqlrelay/tmp. That directory should have 777 permissions, but sometimes it doesn't. The sockseq file in that directory should have 666 permissions, but sometimes it doesn't. The files named ID and ID-CONNECTIONID owned by the user that started the connections in that directory should get removed by sqlr-stop, but sometimes they don't. Sometimes sqlr-stop isn't even used to stop the daemons.