postmaster

Name

postmaster -- Run the Postgres multi-user backend

postmaster [ -B nBuffers ] [ -D DataDir ] [ -N maxBackends ] [ -S ] [ -d DebugLevel ] [ -i ] [ -l ] [ -o BackendOptions ] [ -p port ] [ -n | -s ]

Inputs

postmaster accepts the following command line arguments:

-B nBuffers

Sets the number of shared-memory disk buffers for the postmaster to allocate for use by the backend server processes that it starts. This value defaults to 64 buffers, where each buffer is 8k bytes (or whatever BLCKSZ is set to in src/include/config.h).

-D DataDir

Specifies the directory to use as the root of the tree of database directories. If -D is not given, the default data directory name is the value of the environment variable PGDATA. If PGDATA is not set, then the directory used is $POSTGRESHOME/data. If neither environment variable is set and this command-line option is not specified, the default directory that was set at compile-time is used.

-N maxBackends

Sets the maximum number of backend server processes that this postmaster is allowed to start. By default, this value is 32, but it can be set as high as 1024 if your system will support that many processes. (Note that -B is required to be at least twice -N, so you'll need to increase -B if you increase -N.) Both the default and upper limit values for -N can be altered when building Postgres (see src/include/config.h).

-S

Specifies that the postmaster process should start up in silent mode. That is, it will disassociate from the user's (controlling) tty, start its own process group, and redirect its standard output and standard error to /dev/null.

Note that using this switch makes it very difficult to troubleshoot problems, since all tracing and logging output that would normally be generated by this postmaster and its child backends will be discarded.

-d DebugLevel

Determines the amount of debugging output the backend servers will produce. If DebugLevel is one, the postmaster will trace all connection traffic. Levels two and higher turn on increasing amounts of debug output from the backend processes, and the postmaster displays more information including the backend environment and process traffic. Note that unless the postmaster's standard output and standard error are redirected into a log file, all this output will appear on the controlling tty of the postmaster.

-i

Allows clients to connect via TCP/IP (Internet domain) connections. Without this option, only local Unix domain socket connections are accepted.

-l

Enables secure connections using SSL. The -i option is also required. You must have compiled with SSL enabled to use this option.

-o BackendOptions

The postgres option(s) specified in BackendOptions are passed to all backend server processes started by this postmaster. If the option string contains any spaces, the entire string must be quoted.

-p port

Specifies the TCP/IP port or local Unix domain socket file extension on which the postmaster is to listen for connections from frontend applications. Defaults to the value of the PGPORT environment variable, or if PGPORT is not set, then defaults to the value established when Postgres was compiled (normally 5432). If you specify a port other than the default port, then all frontend applications (including psql) must specify the same port using either command-line options or PGPORT.

Two additional command line options are available for debugging problems that cause a backend to die abnormally. These options control the behavior of the postmaster in this situation, and neither option is intended for use in ordinary operation.

The ordinary strategy for this situation is to notify all other backends that they must terminate and then reinitialize the shared memory and semaphores. This is because an errant backend could have corrupted some shared state before terminating.

These special-case options are:

-n

postmaster will not reinitialize shared data structures. A knowledgeable system programmer can then use a debugger to examine shared memory and semaphore state.

-s

postmaster will stop all other backend processes by sending the signal SIGSTOP, but will not cause them to terminate. This permits system programmers to collect core dumps from all backend processes by hand.

Outputs

semget: No space left on device

If you see this message, you should run the ipcclean command. After doing so, try starting postmaster again. If this still doesn't work, you probably need to configure your kernel for shared memory and semaphores as described in the installation notes. If you run multiple instances of postmaster on a single host, or have a kernel with particularly small shared memory and/or semaphore limits, you may have to reconfigure your kernel to increase its shared memory or semaphore parameters. Tip: You may be able to postpone reconfiguring your kernel by decreasing -B to reduce Postgres' shared memory consumption, and/or by reducing -N to reduce Postgres' semaphore consumption.

StreamServerPort: cannot bind to port

If you see this message, you should make certain that there is no other postmaster process already running on the same port number. The easiest way to determine this is by using the command $ ps -ax | grep postmaster on BSD-based systems, or $ ps -e | grep postmast for System V-like or POSIX-compliant systems such as HP-UX.

If you are sure that no other postmaster processes are running and you still get this error, try specifying a different port using the -p option. You may also get this error if you terminate the postmaster and immediately restart it using the same port; in this case, you must simply wait a few seconds until the operating system closes the port before trying again. Finally, you may get this error if you specify a port number that your operating system considers to be reserved. For example, many versions of Unix consider port numbers under 1024 to be trusted and only permit the Unix superuser to access them.

IpcMemoryAttach: shmat() failed: Permission denied

A likely explanation is that another user attempted to start a postmaster process on the same port which acquired shared resources and then died. Since Postgres shared memory keys are based on the port number assigned to the postmaster, such conflicts are likely if there is more than one installation on a single host. If there are no other postmaster processes currently running (see above), run ipcclean and try again. If other postmaster images are running, you will have to find the owners of those processes to coordinate the assignment of port numbers and/or removal of unused shared memory segments.

Description

postmaster manages the communication between frontend and backend processes, as well as allocating the shared buffer pool and SysV semaphores (on machines without a test-and-set instruction). postmaster does not itself interact with the user and should be started as a background process.

Only one postmaster should be running at a time in a given Postgres installation. Here, an installation means a database directory and postmaster port number. You can run more than one postmaster on a machine only if each one has a separate directory and port number.

Notes

If at all possible, do not use SIGKILL when killing the postmaster. SIGHUP, SIGINT, or SIGTERM (the default signal for kill(1))" should be used instead. Using $ kill -KILL or its alternative form $ kill -9 will prevent postmaster from freeing the system resources (e.g., shared memory and semaphores) that it holds before dying. Use SIGTERM instead to avoid having to clean up manually (as described earlier).

Useful utilities for dealing with shared memory problems include ipcs(1), ipcrm(1), and ipcclean(1).

Usage

To start postmaster using default values, type:

 

        $ nohup postmaster > logfile 2>&1   

This command will start up postmaster on the default port (5432). This is the simplest and most common way to start the postmaster.

To start postmaster with a specific port:

 

        $ nohup postmaster -p 1234 &   

This command will start up postmaster communicating through the port 1234. In order to connect to this postmaster using psql, you would need to run it as:

 

        $ psql -p 1234   

or set the environment variable PGPORT:

 

        $ export PGPORT 1234 
        $ psql