| POSTFIX-WRAPPER(5) | File Formats Manual | POSTFIX-WRAPPER(5) |
This document describes how the familiar "postfix start" etc. user interface can be used to manage one or multiple Postfix instances, and gives details of an API to coordinate activities between the postfix(1) command and a multi-instance manager program.
With multi-instance support, the default Postfix instance is always required. This instance is identified by the config_directory parameter's default value.
Even with multiple Postfix instances, you can keep using the same postfix commands in boot scripts, upgrade procedures, and other places. The commands do more work, but humans are not forced to learn new tricks.
For example, to start all Postfix instances, use:
Other postfix(1) commands also work as expected. For example, to find out what Postfix instances exist in a multi-instance configuration, use:
This enumerates the status of all Postfix instances within a multi-instance configuration.
Alternatively, the postfix(1) command accepts the instance's configuration directory via the MAIL_CONFIG environment variable (the -c command-line option has higher precedence).
Otherwise, the postfix(1) command will operate on all Postfix instances.
When the postfix(1) command operates in multi-instance mode as discussed below, the command needs to execute start, stop, etc. commands for each Postfix instance. This multiplication of commands is handled by a multi-instance manager program.
Turning on postfix(1) multi-instance mode goes as follows: in the default Postfix instance's main.cf file, 1) specify the pathname of a multi-instance manager program with the multi_instance_wrapper parameter; 2) populate the multi_instance_directories parameter with the configuration directory pathnames of additional Postfix instances. For example:
The $daemon_directory/postfix-wrapper file implements a simple manager and contains instructions for creating Postfix instances by hand. The postmulti(1) command provides a more extensive implementation including support for life-cycle management.
The multi_instance_directories and other main.cf parameters are listed below in the CONFIGURATION PARAMETERS section.
In multi-instance mode, the postfix(1) command invokes the $multi_instance_wrapper command instead of the postfix-script file. This multi-instance manager in turn executes the postfix(1) command in single-instance mode for each Postfix instance.
To illustrate the main ideas behind multi-instance operation, below is an example of a simple but useful multi-instance manager implementation:
: ${command_directory?"do not invoke this command directly"}
POSTCONF=$command_directory/postconf
POSTFIX=$command_directory/postfix
instance_dirs=̀$POSTCONF -h multi_instance_directories |
sed 's/,/ /'̀ || exit 1
err=0
for dir in $config_directory $instance_dirs
do
case "$1" in
stop|abort|flush|reload|drain)
test "̀$POSTCONF -c $dir -h multi_instance_enablè" \
= yes || continue;;
start)
test "̀$POSTCONF -c $dir -h multi_instance_enablè" \
= yes || {
$POSTFIX -c $dir check || err=$?
continue
};;
esac
$POSTFIX -c $dir "$@" || err=$?
done
exit $err
The setting "multi_instance_enable = yes" allows the multi-instance manager to start (stop, etc.) the corresponding Postfix instance. For safety reasons, this setting is not the default.
The default setting "multi_instance_enable = no" is useful for manual testing with "postfix -c /path/name start" etc. The multi-instance manager will not start such an instance, and it will skip commands such as "stop" or "flush" that require a running Postfix instance. The multi-instance manager will execute commands such as "check", "set-permissions" or "upgrade-configuration", and it will replace "start" by "check" so that problems will be reported even when the instance is disabled.
Before Postfix multi-instance support was implemented, the executables, manpages, etc., have always been maintained as part of the default Postfix instance.
With multi-instance support, we simply continue to do this. Specifically, a Postfix instance will not check or update shared files when that instance's config_directory value is listed with the default main.cf file's multi_instance_directories parameter.
The consequence of this approach is that the default Postfix instance should be checked and updated before any other instances.
The postfix(1) command operates on only one Postfix instance when the -c option is specified, or when MAIL_CONFIG is present in the process environment. This is necessary to terminate recursion.
Otherwise, when the multi_instance_directories parameter value is non-empty, the postfix(1) command executes the command specified with the multi_instance_wrapper parameter, instead of executing the commands in postfix-script.
The multi-instance manager skips commands such as "stop" or "reload" that require a running Postfix instance, when an instance does not have "multi_instance_enable = yes". This avoids false error messages.
The multi-instance manager replaces a "start" command by "check" when a Postfix instance's main.cf file does not have "multi_instance_enable = yes". This substitution ensures that problems will be reported even when the instance is disabled.
No Postfix command or script will update or check shared files when its config_directory value is listed in the default main.cf's multi_instance_directories parameter value. Therefore, the default instance should be checked and updated before any Postfix instances that depend on it.
Set-gid commands such as postdrop(1) and postqueue(1) effectively append the multi_instance_directories parameter value to the legacy alternate_config_directories parameter value. The commands use this information to determine whether a -c option or MAIL_CONFIG environment setting specifies a legitimate value.
The legacy alternate_config_directories parameter remains necessary for non-default Postfix instances that are running different versions of Postfix, or that are not managed together with the default Postfix instance.