dist/proto/MULTI_INSTANCE_README.html

<!doctype html public "-//W3C//DTD HTML 4.01 Transitional//EN"
        "http://www.w3.org/TR/html4/loose.dtd">

<html>

<head>

<title>Managing multiple Postfix instances on a single host</title>

<meta http-equiv="Content-Type" content="text/html; charset=us-ascii">

</head>

<body>

<h1><img src="postfix-logo.jpg" width="203" height="98" ALT="">Managing
multiple Postfix instances on a single host</h1>

<hr>

<h2>Overview </h2>

<p> This document is a guide to managing multiple Postfix instances
on a single host using the postmulti(1) instance manager. Multi-instance
support is available with Postfix version 2.6 and later.  See the
postfix-wrapper(5) manual page for background on the instance
management framework, and on how to deploy a custom instance manager.
</p>

<p> Topics covered in this document: </p>

<ul>

<li><a href="#why"> Why multiple Postfix instances </a>

<li><a href="#split"> Null-client instances versus service instances </a>

<li><a href="#quick"> Multi-instance walk-through  </a>

<li><a href="#parts"> Components of a Postfix system </a>

<li><a href="#default"> The default Postfix instance </a>

<li><a href="#group"> Instance groups </a>

<li><a href="#params"> Multi-instance configuration parameters </a>

<li><a href="#how"> Using the postmulti(1) command </a>

<li><a href="#credits"> Credits </a>

</ul>

<h2><a name="why"> Why multiple Postfix instances </a></h2>

<p> Postfix is a general-purpose mail system that can be configured
to serve a variety of needs. Examples of Postfix applications are: </p>

<ul>

<li><p> Local mail submission for shell users and system processes. </p>

<li><p> Incoming (MX host) email from the Internet. </p>

<li><p> Outbound mail relay for a corporate network. </p>

<li><p> Authenticated submission for roaming users. </p>

<li><p> Before/after content-filter mail. </p>

</ul>

<p> A single Postfix configuration can provide many or all of these
services, but a complex interplay of settings may be required, for
example with master.cf options overriding main.cf settings. In this
document we take the view that multiple Postfix instances may be a
simpler way to configure a multi-function Postfix system.  With
multiple Postfix instances, each instance has its own directories
for configuration, queue and data files, but it shares all Postfix
program and documentation files with other instances. </p>

<p> Since there is no single right way to configure your system,
we recommend that you choose what makes you most comfortable. If
different Postfix services don't involve incompatible main.cf or
master.cf settings, and if they can be combined together without
complex tricks, then a single monolithic configuration may be the
simplest approach. </p>

<p> The purpose of multi-instance support in Postfix is not to force
you to create multiple Postfix instances, but rather to give you a
choice. Multiple instances give you the freedom to tune each Postfix
instance to a single task that it does well and to combine instances
into complete systems. </p>

<p> With the introduction of the postmulti(1) utility and the reduction
of the per-instance configuration footprint of a secondary Postfix
instance to just a main.cf and master.cf file (other files are now in
shared locations), we hope that multiple instances will be easier to
use than ever before. </p>

<h2><a name="split"> Null-client instances versus service instances </a></h2>

<p> In the multi-instance approach to configuring Postfix, the first
simplification is with the default local-submission Postfix instance.
</p>

<p> Most UNIX systems require support for email submission with the
sendmail(1) command so that system processes such as cron jobs can
send status reports, and so that system users can send email with
command-line utilities.  Such email can be handled with a <a
href="STANDARD_CONFIGURATION_README.html#null_client">null-client</a>
Postfix configuration that forwards all mail to a central mail hub.
The null client will typically either not run an SMTP listener at
all (master_service_disable = inet), or it will listen only on the
loopback interface (inet_interfaces = loopback-only). </p>

<p> When implementing specialized servers for inbound Internet
email, outbound MTAs, internal mail hubs, and so on, we recommend
using a null client for local submission and creating single-function
secondary Postfix instances to serve the specialized needs. </p>

<blockquote>

<p> Note: usually, you need to use different "myhostname" settings
when you run multiple instances on the same host. Otherwise, there
will be false "mail loops back to myself" alarms when one instance
tries to send mail into another instance.  Typically, the null-client
instance will use the system's hostname, and other instances will
use their own dedicated "myhostname" settings. Different names are
not needed when instances send mail to each other with a protocol
other than SMTP, or with SMTP over a TCP port other than 25 as is
usual with SMTP-based content filters.  </p>

</blockquote>

<h2><a name="quick"> Multi-instance walk-through </a></h2>

<p> Before discussing the fine details of multi-instance operation
we first show the steps for creating a border mail server. This
server has with a null-client Postfix instance for local submission,
an input Postfix instance to receive mail from the Internet, plus
an <a href="FILTER_README.html#advanced_filter">advanced</a> SMTP
content-filter and an output Postfix instance to deliver filtered
email to its internal destination. </p>

<h3>Setting up the null-client Postfix instance </h3>

<p> On a border mail hub, while mail from the Internet requires a
great deal of scrutiny, locally submitted messages are typically
limited to mail from cron jobs and other system services. In this
regard the border MTA is not different from other Unix hosts in
your environment. For this reason, it will submit locally-generated
email to the internal mail hub. We start the construction of the
border mail server with the <a href="#default_instance">default</a>
instance, which will be a local-submission <a
href="STANDARD_CONFIGURATION_README.html#null_client">null client</a>:
</p>

<blockquote>
<pre>
/etc/postfix/main.cf:
    # We are mta1.example.com
    #
    myhostname = mta1.example.com
    mydomain = example.com

    # Flat user-account namespace in example.com:
    #
    #   user@example.com not user@host.example.com
    #
    myorigin = $mydomain

    # Postfix 2.6+, disable inet services, specifically disable smtpd(8)
    #
    master_service_disable = inet

    # No local delivery:
    #
    mydestination =
    local_transport = error:5.1.1 Mailbox unavailable
    alias_database =
    alias_maps =
    local_recipient_maps =

    # Send everything to the internal mailhub
    #
    relayhost = [mailhub.example.com]

    # Indexed table macro:
    # (use "hash", ... when <a href="CDB_README.html">cdb</a> is not available)
    #
    default_database_type = cdb
    indexed = ${default_database_type}:${config_directory}/

    # Expose origin host of mail from "root", ...
    #
    smtp_generic_maps = ${indexed}generic

    # Send messages addressed to "root", ... to the MTA support team
    #
    virtual_alias_maps = ${indexed}virtual

/etc/postfix/generic:
    # The smarthost supports "+" addressing (recipient_delimiter = +).
    # Mail from "root" exposes the origin host, without replies
    # and bounces going back to the same host.
    #
    # On clustered MTAs this file is typically machine-built from
    # a template file. The build process expands the template into
    # "mtaadmin+root=mta1"
    #
    root    	mtaadmin+root=mta1

/etc/postfix/virtual:
    # Caretaker aliases:
    #
    root    	mtaadmin
    postmaster	root
</pre>
</blockquote>

<p> You would typically also add a Makefile, to automatically run
postmap(1) commands when source files change. This Makefile also
creates a "generic" database when none exists. </p>

<blockquote>
<pre>
/etc/postfix/Makefile:
    MTAADMIN=mtaadmin

    all: virtual.cdb generic.cdb

    generic: Makefile
	    @echo Creating $@
	    @rm -f $@.tmp
	    @printf '%s\t%s+root=%s\n' root $MTAADMIN `uname -n` &gt; $@.tmp
	    @mv $@.tmp generic

    %.cdb: %
	    postmap cdb:$&lt;
</pre>
</blockquote>

<p> Construct the "virtual" and "generic" databases (the latter is
created by running "make"), then start and test the null-client:
</p>

<blockquote>
<pre>
# cd /etc/postfix; make
# postfix start
# sendmail -i -f root -t &lt;&lt;EOF
From: root
To: root
Subject: test

testing
EOF
</pre>
</blockquote>

<p> The test message should be delivered the members of the "mtaadmin"
address group (or whatever address group you choose) with the
following headers: </p>

<blockquote>
<pre>
From: mtaadmin+root=mta1@example.com
To: mtadmin+root=mta1@example.com
Subject: test
</pre>
</blockquote>

<h3>Setting up the "output" Postfix instance </h3>

<p> With the null-client instance out of the way, we can create the
MTA "output" instance that will deliver filtered mail to the inside
network. We add the "output" instance first, because the output
instance needs to be up and running before the input instance can
be fully tested, and when the system boots, the "output" instance
must start before the input instance. We will put the output and
input instances into a single instance group named "mta".  </p>

<p> Just once, when adding the first secondary instance, enable
multi-instance support in the default (null-client) instance: </p>

<blockquote>
<pre>
# postmulti -e init
</pre>
</blockquote>

<p> Then create the output instance: <p>

<blockquote>
<pre>
# postmulti -I postfix-out -G mta -e create
</pre>
</blockquote>

<p> The instance configuration directory defaults to /etc/postfix-out,
more precisely, the "postfix-out" subdirectory of the parent directory
of the default-instance configuration directory. The new instance will
be created in a "disabled" state: </p>

<blockquote>
<pre>
/etc/postfix-out/main.cf
    #
    # ... "stock" main.cf settings ...
    #
    multi_instance_name = postfix-out
    queue_directory = /var/spool/postfix-out
    data_directory = /var/lib/postfix-out
    #
    multi_instance_enable = no
    master_service_disable = inet
    authorized_submit_users =
</pre>
</blockquote>

<p> This instance has a "stock" master.cf file, and its queue and
data directories, also named "postfix-out", will be located in the
same parent directories as the corresponding directories of the
default instance (e.g., /var/spool/postfix-out and /var/lib/postfix-out).
</p>

<p> While this instance is immediately safe to start, it is not yet
usefully configured. It needs to be customized to fit the role of a
post-filter re-injection SMTP service. Typical additions include: </p>

<blockquote>
<pre>
/etc/postfix-out/master.cf:
    # Replace default "smtp inet" entry with one listening on port 10026.
    127.0.0.1:10026     inet  n       -       n       -       -       smtpd

/etc/postfix-out/main.cf
    # ...

    # Comment out if you don't use IPv6 internally
    # inet_protocols = ipv4
    inet_interfaces = loopback-only
    mynetworks_style = host
    smtpd_authorized_xforward_hosts = $mynetworks

    # Don't anvil(8) control the re-injection port.
    #
    smtpd_client_connection_count_limit = 0
    smtpd_client_event_limit_exceptions = $mynetworks

    # Best practice when inet_interfaces is set, as this is not a
    # "secondary IP personality" configuration.
    #
    smtp_bind_address = 0.0.0.0

    # All header rewriting happens upstream
    #
    local_header_rewrite_clients =

    # No local delivery on border gateway
    #
    mydestination =
    alias_maps =
    alias_database =
    local_recipient_maps =
    local_transport = error:5.1.1 Mailbox unavailable

    # May need a recipient_delimiter for per-user transport lookups:
    #
    recipient_delimiter = +

    # Only one (unrestricted client)
    # With multiple instances, rarely need "-o param=value" overrides
    # in master.cf, each instance gets its own main.cf file.
    #
    # Postfix 2.10 and later: specify empty smtpd_relay_restrictions.
    smtpd_relay_restrictions =
    smtpd_recipient_restrictions = permit_mynetworks, reject

    # Tolerate occasional high latency in the  content filter.
    #
    smtpd_timeout = 1200s

    # Best when empty, with all parent domain matches explicit.
    #
    parent_domain_matches_subdomains =

    # Use the "relay" transport for inbound mail, and the default
    # "smtp" transport for outbound mail (bounces, ...). The latter
    # won't starve the former of delivery agent slots.
    #
    relay_domains = example.com, .example.com

    # With xforward, match the input instance setting, if you
    # want "yes", set both to "yes".
    #
    smtpd_client_port_logging = no

    # Transport settings ...
    # Message size limit
    # Concurrency tuning for "relay" and "smtp" transport
    # ...
</pre>
</blockquote>

<p> With the "output" configuration in place, enable and start the
instance: </p>

<blockquote>
<pre>
1 # postmulti -i postfix-out -x postconf -e \
2     "master_service_disable =" "authorized_submit_users = root"
3 # postmulti -i postfix-out -e enable
4 # postmulti -i postfix-out -p start
</pre>
</blockquote>

<p> This uses the postmulti(1) command to invoke postconf(1) in the
context (MAIL_CONFIG=/etc/postfix-out) of the output instance.  </p>

<ul>

<li> <p> Lines 1-2: With "authorized_submit_users = root", the
superuser can test the postfix-out instance with "postmulti -i
postfix-out -x sendmail -bv recipient...", but otherwise local
submission remains disabled.  </p>

<li> <p> Lines 1-2: With "master_service_disable =", the "inet"
listeners are re-enabled. </p>

<li> <p> Line 3: The output instance is enabled for multi-instance
start/stop. </p>

<li> <p> Line 4: The output instance is started. </p>

</ul>

<p> Test the output instance by submitting probe messages via "sendmail
-bv" and "telnet". For production systems, in-depth configuration tests
should be done on a lab system. The simple tests just suggested will only
confirm successful deployment of a configuration that should already be
known good. </p>

<h3> Setting up the content-filter proxy </h3>

<p> With the output instance ready, deploy your content-filter
proxy.  Most proxies will need their own /etc/rc* start/stop script.
Some proxies, however, are started on demand by the Postfix spawn(8)
service, in which case you need to add the relevant spawn(8) entry
to the output instance master.cf file.  </p>

<p> Configure the proxy to listen on 127.0.0.1:10025 and to re-inject
filtered email to 127.0.0.1:10026.  Start the proxy service if
necessary, then test the proxy via "telnet" or automated SMTP
injectors. The proxy should support the following ESMTP features:
DSN, 8BITMIME, and XFORWARD. In addition, the proxy should support
multiple mail deliveries within an SMTP session. </p>

<h3> Setting up the input Postfix instance </h3>

<p> The input Postfix instance receives mail from the network and
sends it through the content filter. Now we create the input instance,
also part of the "mta" instance group: </p>

<blockquote>
<pre>
# postmulti -I postfix-in -G mta -e create
</pre>
</blockquote>

<p> The new instance configuration directory defaults to /etc/postfix-in,
more precisely, the "postfix-in" subdirectory of the parent directory
of the default-instance configuration directory. The new instance will
be created in a "disabled" state: </p>

<blockquote>
<pre>
/etc/postfix-in/main.cf
    #
    # ... "stock" main.cf settings ...
    #
    multi_instance_name = postfix-in
    queue_directory = /var/spool/postfix-in
    data_directory = /var/lib/postfix-in
    #
    multi_instance_enable = no
    master_service_disable = inet
    authorized_submit_users =
</pre>
</blockquote>

<p> As before, make appropriate changes to main.cf and master.cf to
make the instance production ready. Consider setting "soft_bounce = yes"
during the first few hours of deployment, so you can iron-out any unexpected
"kinks". </p>

<p> Manual testing can start with:

<blockquote>
<pre>
/etc/postfix-in/main.cf
    # Accept only local traffic, but allow impersonation:
    inet_interfaces = 127.0.0.1
    smtpd_authorized_xclient_hosts = 127.0.0.1
</pre>
</blockquote>

<p> This allows you to use the Postfix-specific <a
href="XCLIENT_README.html">XCLIENT</a> SMTP command to safely
simulate connections from remote systems before any remote systems
are able to connect. If the test results look good, revert the above
settings to the required production values. Typical settings in the
pre-filter input instance include: </p>

<blockquote>
<pre>
/etc/postfix-in/main.cf
    #
    # ...
    #

    # No local delivery on border gateway
    #
    mydestination =
    alias_maps =
    alias_database =
    local_recipient_maps =
    local_transport = error:5.1.1 Mailbox unavailable

    # Don't rewrite remote headers
    #
    local_header_rewrite_clients =

    # All recipients of not yet filtered email go to the same filter together.
    #
    # With multiple instances, the content-filter is specified
    # via transport settings not the "content_filter" transport
    # switch override! Here the filter listens on local port 10025.
    #
    # If you need to route some users or recipient domains directly to the
    # output instance bypassing the filter, just define a transport table
    # with suitable entries.
    #
    default_transport = smtp:[127.0.0.1]:10025
    relay_transport = $default_transport
    virtual_transport = $default_transport
    transport_maps =

    # Pass original client log information through the filter.
    #
    smtp_send_xforward_command = yes

    # Avoid splitting the envelope and scanning messages multiple times.
    # Match the re-injection server's recipient limit.
    #
	smtp_destination_recipient_limit = 1000

    # Tolerate occasional high latency in the content filter.
    #
    smtp_data_done_timeout = 1200s

    # With xforward, match the output instance setting, if you
    # want "yes", set both to "yes".
    #
    smtpd_client_port_logging = no

    # ... Lots of settings for inbound MX host ...
</pre>
</blockquote>

<p> With the "input" instance configured, enable and start it: </p>

<blockquote>
<pre>
# postmulti -i postfix-in -x postconf -e \
    "master_service_disable =" "authorized_submit_users = root"
# postmulti -i postfix-in -e enable
# postmulti -i postfix-in -p start
</pre>
</blockquote>

<p> That's it. You now have a 3-instance configuration. A null-client
sending all locally submitted mail to the internal mail hub and a pair of
"mta" instances that receive mail from the Internet, pass it through a
content-filter, and then deliver it to the internal destination. </p>

<p> Running "postfix start" or "postfix stop" will now start/stop all
three Postfix instances. You can use "postfix -c /config/path start"
to start just one instance, or use the instance name (or instance
group name) via postmulti(1): </p>

<blockquote>
<pre>
# postmulti -i - -p stop
# postmulti -g mta -p status
# postmulti -i postfix-out -p flush
# postmulti -i postfix-in -p reload
# ...
</pre>
</blockquote>

<p> This example ends the multi-instance "walk through". The remainder
of this document provides background information on Postfix
multi-instance support features and options. </p>

<h2><a name="parts"> Components of a Postfix system </a></h2>

<p> A Postfix system consists of the following components: </p>

<p> Shared among all instances: </p>

<ul>

<li><p> Command-line utilities for administrators and users installed in
$command_directory, $sendmail_path, $mailq_path and $newaliases_path. </p>

<li><p> Daemon executables, and run-time support files installed in
$daemon_directory. </p>

<li><p> Bundled documentation, installed in $html_directory,
$manpage_directory and $readme_directory. </p>

<li><p> Entries in /etc/passwd and /etc/group for the $mail_owner user and
$setgid_group group. The $mail_owner user provides the mail system
with a protected (non-root) execution context. The $setgid_group group
is used exclusively to support the setgid postdrop(1) and postqueue(1)
utilities (it <b>must not</b> be the primary group or secondary group
of any users, including the $mail_owner user). </p>

</ul>

<p> Private to each instance: </p>

<ul>

<li><p> The main.cf, master.cf (and other optional) configuration
files in $config_directory. </p>

<li> <p> The maildrop, incoming, active, deferred and hold queues
in $queue_directory (which contains additional directories needed
by Postfix, and which optionally doubles as a chroot jail for Postfix
daemon processes). </p>

<li> <p> Various caches (TLS session, address verification, ...)
in $data_directory. </p>

</ul>

<p> The Postfix configuration parameters mentioned above are
collectively referred to as "installation parameters". Their default
values are set when the Postfix software is built from source, and
all but one may be optionally set to a non-default value via the
main.cf file.  The one parameter that (catch-22) cannot be set in
main.cf is $config_directory, as this defines the location of the
main.cf file itself. </p>

<p> Though config_directory cannot be set in main.cf, postfix(1) and
most of the other command-line Postfix utilities allow you to specify a
non-default configuration directory via a command line option (typically
<b>-c</b>) or via the MAIL_CONFIG environment variable. In this way,
it is possible to have multiple configuration directories on the same
machine, and to have multiple running master(8) daemons each with its
own configuration files, queue directory and data directory. </p>

<p> These multiple running copies of master(8) share the base Postfix
software. They do not (and cannot) share their configuration
directories, queue directories or data directories. </p>

<p> Each combination of configuration directory, together with the queue
directory and data directory (specified in the corresponding main.cf file)
make up a Postfix <b>instance</b>. </p>

<h2><a name="default"> The default Postfix instance </a></h2>

<p> One Postfix instance is special: this is the instance whose
configuration directory is the default one compiled into the Postfix
utilities. The location of the default configuration directory is
typically /etc/postfix, and can be queried via the "postconf -d
config_directory" command.  We call the instance with this configuration
directory the "default instance". </p>

<p> The default instance is responsible for local mail submission. The
setgid postdrop(1) utility is used by the sendmail(1) local submission
program to spool messages into the <b>maildrop</b> sub-directory of the
queue directory of the default instance. </p>

<p> Even in the rare case when "sendmail -C" is used to submit local mail
into a non-default Postfix instance, for security reasons, postdrop(1)
will consult the default main.cf file to check the validity of the
requested non-default configuration directory. </p>

<p> So, while in most other respects, all instances are equal, the
default instance is "more equal than others". You may choose to create
additional instances, but you must have at least the default instance,
with its configuration directory in the default compiled-in location. </p>

<h2><a name="group"> Instance groups </a></h2>

<p> The postmulti(1) multi-instance manager supports the notion of an
instance "group". Typically, the member instances of an instance group
constitute a logical service, and are expected to all be running or all
be stopped. </p>

<p> In many cases a single Postfix instance will be a complete logical
"service". You should define such instances as stand-alone instances
that are not members of any instance "group". The null-client
instance is an example of a non-group instance. </p>

<p> When a logical service consists of multiple Postfix instances,
often a pair of pre-filter and post-filter instances with a content
filter proxy between them, the related instances should be members
of a single instance group (however, the content filter usually has
its own start/stop procedure that is separate from any Postfix
instance).  </p>

<p> The default instance main.cf file's $multi_instance_directories
configuration parameter lists the configuration directories of all
secondary (non-default) instances. Together with the default instance,
these secondary instances are managed by the multi-instance manager.
Instances are started in the order listed, and stopped in the
opposite order. For instances that are members of a service "group",
you should arrange to start the service back-to-front, with the
output stages started and ready to receive mail before the input
stages are started. </p>

<h2><a name="params"> Multi-instance configuration parameters </a></h2>

<dl>

<dt> multi_instance_wrapper </dt>

<dd> <p> This default-instance configuration parameter must be set
to a suitable multi-instance manager's "wrapper" program that
controls the starting, stopping, etc. of a multi-instance Postfix
system. To use the postmulti(1) manager described in this document,
this parameter should be set with the "<a href="#init">postmulti
-e init</a>" command.  </p> </dd>

<dt> multi_instance_directories </dt>

<dd> <p> This default-instance configuration parameter specifies
an optional list of the secondary instances controlled via the
multi-instance manager. Instances are listed in their "start" order,
with the default instance always started first (if enabled). If
$multi_instance_directories is left empty, the postfix(1) command
runs with multi-instance support turned off, and none of the
multi_instance_ configuration parameters will have any effect. </p>

<p> Do not assign a non-empty list of secondary instance configuration
directories to multi_instance_directories until you have configured a
suitable multi_instance_wrapper setting! This is best accomplished via
the "<a href="#init">postmulti -e init</a>" command.
</p> </dd>

<dt> multi_instance_name </dt>

<dd> <p> Each Postfix instance may be assigned a distinct name (with
"postfix -e create/import/assign -I <i>name</i>..."). This name can
be used with the postmulti(1) command-line utility to perform tasks
on the instance by name (rather than the full pathname of its
configuration directory). Choose a name that concisely captures the
role of the instance (it must start with "postfix-").  It is an
error for two instances to have the same $multi_instance_name.  You
can leave an instance "nameless" by leaving this parameter at the
default empty setting. </p>

<p> To avoid confusion in your logs, if you don't assign each
secondary instance a non-empty (distinct) $multi_instance_name, you
should make sure that the $syslog_name setting is different for
each instance. The $syslog_name parameter defaults to $multi_instance_name
when the latter is non-empty. If at all possible, the syslog_name
should start with "postfix-", this helps log parsers to identify
log entries from secondary Postfix instances.  </p> </dd>

<dt> multi_instance_group </dt>

<dd> <p> Each Postfix instance may be assigned an "instance group"
name (with "postfix -e create/import/assign -G <i>name</i>...").
The default (empty) value of multi_instance_group parameter indicates
a stand-alone instance that is not part of any group. The group
name can be used with the postmulti(1) command-line utility to
perform a task on the members of a group by name. Choose a single-word
group name that concisely captures the role of the group.  </p>
</dd>

<dt> multi_instance_enable </dt>

<dd> <p> This parameter controls whether a Postfix instance will
be started by a Postfix multi-instance manager.  The default value
is "no". The instance can be started explicitly with "postfix -c
/path/to/config/directory"; this is useful for testing.  </p>

<p> When an instance is disabled, the postfix(1) "start" command
is replaced by "check". </p>

<p> Some postfix(1) commands (such as "stop", "flush", ...) require
a running Postfix instance, and skip instances that are disabled.
</p>

<p> Other postfix(1) commands (such as "status", "set-permissions",
"upgrade-configuration", ...) do not require a running Postfix
system, and apply to all instances whether enabled or not.  </p>
</dd>

</dl>

<p> The postmulti(1) utility can be used to create (or destroy) instances.
It can also be used to "import" or "deport" existing instances into or
from the list of managed instances. When using postmulti(1) to manage
instances, the above configuration parameters are managed for you
automatically. See below. </p>

<h2><a name="how"> Using the postmulti(1) command </a></h2>

<ul>

<li><a href="#init"> Initializing the multi-instance manager </a>

<li><a href="#list"> Listing managed instances </a>

<li><a href="#start"> Starting or stopping a multi-instance system </a>

<li><a href="#adhoc"> Ad-hoc multi-instance operations </a>

<li><a href="#create"> Creating a new Postfix instance </a>

<li><a href="#destroy"> Destroying a Postfix instance </a>

<li><a href="#import"> Importing an existing Postfix instance </a>

<li><a href="#deport"> Deporting a managed Postfix instance </a>

<li><a href="#assign"> Assigning a new name or group name </a>

<li><a href="#enable"> Enabling/disabling managed instances </a>

</ul>

<h3><a name="init"> Initializing the multi-instance manager </a></h3>

<p> Before postmulti(1) is used for the first time, you must install
it as the multi_instance_wrapper for your Postfix system and enable
multi-instance operation of the default Postfix instance. You can then
proceed to add <a href="#create">new</a> or <a href="#import">existing</a>
instances to the multi-instance configuration. This initial installation
is accomplished as follows: </p>

<blockquote>
<pre>
    # postmulti -e init
</pre>
</blockquote>

<p> This updates the default instance main.cf file as follows: </p>

<blockquote>
<pre>
    # Use postmulti(1) as a postfix-wrapper(5)
    #
    multi_instance_wrapper = ${command_directory}/postmulti -p --

    # Configure the default instance to start when in multi-instance mode
    #
    multi_instance_enable = yes
</pre>
</blockquote>

<p> If you prefer, you can make these changes by editing the default
main.cf directly, or by using "postconf -e". </p>

<h3><a name="list"> Listing managed instances </a></h3>

<p> The list of managed instances consists of the default instance and
the additional instances whose configuration directories are listed
(in start order) under the multi_instance_directories parameter of the
default main.cf configuration file.  </p>

<p> You can list selected instances, groups of instances or all
instances by specifying only the instance matching options with the
"-l" option.  The "-a" option is assumed if no other instance
selection options are specified (this behavior changes with the
"-e" option).  As a special case, even if it has an explicit name,
the default instance can always be selected via "-i -". </p>

<blockquote>
<pre>
# postmulti -l -a
# postmulti -l -g a_group
# postmulti -l -i an_instance
</pre>
</blockquote>

<p> The output is one line per instance (in "postfix start" order):
</p>

<blockquote>

<table border="1">

<tr> <th align="left">name</th> <th align="left">group</th> <th
align="left">enabled</th> <th align="left">config_directory</th>
</tr>

<tr> <td>-</td> <td>-</td> <td>yes</td> <td>/etc/postfix

<tr> <td>mta-out</td> <td>mta</td> <td>yes</td> <td>/etc/postfix/mta-out

<tr> <td>mta-in</td> <td>mta</td> <td>yes</td> <td>/etc/postfix-mta-in

<tr> <td>msa-out</td> <td>msa</td> <td>yes</td> <td>/etc/postfix-msa-out

<tr> <td>msa-in</td> <td>msa</td> <td>yes</td> <td>/etc/postfix-msa-in

<tr> <td>test</td> <td>-</td> <td>no</td> <td>/etc/postfix-test

</table>

</blockquote>

<p> The first line showing the column headings is not part of the
output. When either the instance name or the instance group is not
set, it is shown as a "-". </p>

<p> When selecting an existing instance via the "-i" option, you
can always use the full pathname of its configuration directory
instead of the instance (short) name. This is the only way to select
a non-default nameless instance. The default instance can be selected
via "-i -", whether it has a name or not. </p>

<p> To list instances in reverse start order, include the "-R"
option together with the instance selection options. </p>

<h3><a name="start"> Starting or stopping a multi-instance system
</a></h3>

<p> To start, stop, reload, etc. the complete (already configured as
above) multi-instance system just use postfix(1) as you would with a
single-instance system. The Postfix multi-instance wrapper framework
insulates Postfix init.d start and package upgrade scripts from the
details of multi-instance management! </p>

<p> The <b>-p</b> option of postmulti(1) turns on postfix(1) compatibility
mode. With this option the remaining arguments are exactly those supported
by postfix(1), but commands are applied to all instances or all enabled
instances as appropriate. As described above, this switch is required
when using postmulti(1) as the multi_instance_wrapper. </p>

<p> If you want to specify a subset of instances by name, or group name,
or run arbitrary commands (not just "postfix stop/start/etc. in the
context (MAIL_CONFIG environment variable setting) of a particular
instance or group of instances, then you can use the instance-aware
postmulti(1) utility directly. </p>

<h3><a name="adhoc"> Ad-hoc multi-instance operations </a></h3>

<p> The postmulti(1) command can be used by the administrator to run arbitrary
commands in the context of one or more Postfix instances. The most common
use-case is stopping or starting a group of Postfix instances: </p>

<blockquote>
<pre>
# postmulti -g mygroup -p start
# postmulti -g mygroup -p flush
# postmulti -g mygroup -p reload
# postmulti -g mygroup -p status
# postmulti -g mygroup -p stop
# postmulti -g mygroup -p upgrade-configuration
</pre>
</blockquote>

<p> The <b>-p</b> option is essentially a short-hand for a leading
<b>postfix</b> command argument, but with appropriate additional options
turned on depending on the first argument. In the case of "start",
disabled instances are "checked" (postfix check) rather than simply
skipped. </p>

<p> The resulting command is executed for each candidate instance with
the <b>MAIL_CONFIG</b> environment variable set to the configuration
directory of the corresponding Postfix instance. </p>

<p> The postmulti(1) utility is able to launch commands other than
postfix(1), Use the <b>-x</b> option to ask postmulti to execute an
ad-hoc command for all instances, a group of instances, or just one
instance. With ad-hoc commands the multi_instance_enable parameter
is ignored: the command is unconditionally executed for the instances
selected via -a, -g or -i. In addition to MAIL_CONFIG, the following
instance parameters are exported into the command environment: </p>

<blockquote>
<pre>
command_directory=$command_directory
daemon_directory=$daemon_directory
config_directory=$config_directory
queue_directory=$queue_directory
data_directory=$data_directory
multi_instance_name=$multi_instance_name
multi_instance_group=$multi_instance_group
multi_instance_enable=$multi_instance_enable
</pre>
</blockquote>

<p> The config_directory setting is of course the same as MAIL_CONFIG,
and is arguably redundant, but leaving it in is less surprising. If
you want to skip disabled instances, just check multi_instance_enable
environment variable and exit if it is set to "no". </p>

<p> The ability to run ad-hoc commands opens up a wealth of additional
possibilities: </p>

<ul>

<li><p> Specify an instance by name rather than configuration directory
when using sendmail(1) to send a verification probe: </p>

<blockquote>
<pre>
$ postmulti -i postfix-myinst -x sendmail -bv test@example.net
</pre>
</blockquote>

<li><p> Display non-default main.cf settings of all Postfix instances.
This uses an inline shell script to package together multiple shell
commands to execute for each instance: </p>

<blockquote>
<pre>
$ postmulti -x sh -c 'echo "-- $MAIL_CONFIG"; postconf -n'
</pre>
</blockquote>

<li><p> Put all mail in enabled member instances of a group on hold: </p>

<blockquote>
<pre>
# postmulti -g group_name -x \
    sh -c 'test $multi_instance_enable = yes &amp;&amp; postsuper -h ALL'
</pre>
</blockquote>

<li><p> Show top 10 domains in the deferred queue of all instances:
</p>

<blockquote>
<pre>
# postmulti -x sh -c 'echo "-- $MAIL_CONFIG"; qshape deferred | head -12'
</pre>
</blockquote>

</ul>

<h3><a name="create"> Creating a new Postfix instance </a></h3>

<p> The postmulti(1) command can be used to create additional Postfix
instances. New instances are created with local submission and all "inet"
services disabled via the following non-default parameter settings in
the main.cf file: </p>

<blockquote>
<pre>
authorized_submit_users =
master_service_disable = inet
</pre>
</blockquote>

<p> The above settings ensure that new instances are safe to start
immediately: they will not conflict with inet listeners in existing
Postfix instances.  They will also not accept any mail until they are
fully configured, at which point you can do away with one or both of
the above safety measures. </p>

<p> The postmulti(1) command encourages a preferred way of organizing
the configuration directories, queue directories and data directories
of non-default instances. If the default instance settings are: </p>

<blockquote>
<pre>
config_directory = /conf-path/postfix
queue_directory = /queue-path/postfix
data_directory = /data-path/postfix
</pre>
</blockquote>

<p> A newly-created instance named <i>postfix-myinst</i> will by default
have: </p>

<blockquote>
<pre>
multi_instance_enable = no
multi_instance_name = postfix-myinst
config_directory = /conf-path/postfix-myinst
queue_directory = /queue-path/postfix-myinst
data_directory = /data-path/postfix-myinst
</pre>
</blockquote>

<p> You can override any of these defaults when creating the instance,
but unless you want to spread instance queue directories over multiple
file-systems, use the default naming strategy. It keeps the multiple
instances organized in a uniform, predictable fashion. </p>

<p> When specifying the instance name later, you can refer to it
either as "postfix-myinst", or via the full path of the configuration
directory. </p>

<p> To create a new instance just use the <b>-e create</b> option: </p>

<blockquote>
<pre>
# postmulti -I postfix-myinst -e create
</pre>
</blockquote>

<p> If the new instance is to belong to a group of related instances that
implement a single logical service, assign it to a group: </p>

<blockquote>
<pre>
# postmulti -I postfix-myinst -G mygroup -e create
</pre>
</blockquote>

<p> If you want to override the conventional values of the instance
installation parameters, specify their values on the command-line: </p>

<blockquote>
<pre>
# postmulti [-I postfix-myinst] [-G mygroup] -e create \
	"config_directory = /path/to/config_directory" \
	"queue_directory = /path/to/queue_directory" \
	"data_directory = /path/to/data_directory"
</pre>
</blockquote>

<p> A note on the <b>-I</b> and <b>-G</b> options above. These are always
used to assign a name or group name to an instance, while the <b>-i</b>
and <b>-g</b> options always select existing instances.  By default,
the configuration directories of newly managed instances are appended
to the instance list. You can use the "-i" or "-g" or "-a" options to
insert the new instance before the specified instance or group, or at
the beginning of the instance list (multi_instance_directories parameter
of the default instance). </p>

<p> If you do specify a name (use "-I" with a name that is not "-")
for the new instance, you may omit any of the 3 instance installation
parameters whose instance-name based value is acceptable. Otherwise, all
three instance installation parameters are required. You should set the
"syslog_name" explicitly in the main.cf file of a "nameless" instance,
in order to avoid confusion in the mail logs when multiple instances
are in use. </p>

<h3><a name="destroy"> Destroying a Postfix instance </a></h3>

<p> If you no longer need an instance, you can destroy it via: </p>

<blockquote>
<pre>
# postmulti -i postfix-myinst -p stop
# postmulti -i postfix-myinst -e disable
# postmulti -i postfix-myinst -e destroy
</pre>
</blockquote>

<p> The instance must be stopped, disabled and have no queued messages.
This is expected to fully delete a just created instance that has never
been used. If the instance is not freshly created, files added after
the instance was created will remain in the configuration, queue or
data directories, in which case the corresponding directory may not
be fully removed and a warning to that effect will be displayed. You
can complete the destruction of the instance manually by removing any
unwanted remnants of the instance-specific "private" directories. </p>

<h3><a name="import"> Importing an existing Postfix instance </a></h3>

<p> If you already have an existing secondary Postfix instance that is
not yet managed via postmulti(1), you can "import" it into the list
of managed instances. If your instance is already using the default
configuration directory naming scheme, just specify the corresponding
instance name (the multi_instance_name parameter in its configuration
file will be adjusted to match this name if necessary): </p>

<blockquote>
<pre>
# postmulti -I postfix-myinst [-G mygroup] -e import
</pre>
</blockquote>

<p> Otherwise, you must specify the location of its configuration
directory: </p>

<blockquote>
<pre>
# postmulti [-I postfix-myinst] [-G mygroup] -e import \
	"config_directory = /path/of/config_directory"
</pre>
</blockquote>

<p> When the instance is imported, you can assign a name or a group. As
with <a href="#create">"create"</a>, you can control the placement of the
new instance in the start order by using "-i", "-g" or "-a" to prepend
before the selected instance or instances. </p>

<p> An imported instance is usually not multi-instance "enabled",
unless it was part of a multi-instance configuration at an earlier
time.  If it is fully configured and ready to run, don't forget
to <a href="#enable">enable</a> it and if necessary start it. When
other enabled instances are already running, new instances need to
be started individually when they are first created or imported.
</p>

<p> To find out what instances are running, use: </p>

<blockquote>
<pre>
# postfix status
</pre>
</blockquote>

<h3><a name="deport"> Deporting a managed Postfix instance </a></h3>

<p> You can "deport" an existing instance from the list of managed
instances.  This does not destroy the instance, rather the instance
just becomes a stand-alone Postfix instance not registered with the
multi-instance manager. postmulti(1) will refuse to "deport" an
instance that is not stopped and disabled. </p>

<blockquote>
<pre>
# postmulti -i postfix-myinst -p stop
# postmulti -i postfix-myinst -e disable
# postmulti -i postfix-myinst -e deport
</pre>
</blockquote>

<h3><a name="assign"> Assigning a new name or group name </a></h3>

<p> You can assign a new name or new group to a managed instance.
Use "-" as the new value to assign the instance to no group or make it
nameless. To specify a nameless secondary instance use the configuration
directory path instead of the old name: </p>

<blockquote>
<pre>
# postmulti -i postfix-old [-I postfix-new] [-G newgroup] -e assign
</pre>
</blockquote>

<h3><a name="enable"> Enabling/disabling managed instances </a></h3>

<p> You can enable or disable a managed instance. As documented in
postfix-wrapper(5), disabled instances are skipped with actions
that <a href="postconf.5.html#postmulti_start_commands">start</a>,
<a href="postconf.5.html#postmulti_start_commands">stop</a> or <a
href="postconf.5.html#postmulti_control_commands">control</a> running
Postfix instances. </p>

<blockquote>
<pre>
# postmulti -i postfix-myinst -e enable
# postmulti -i postfix-myinst -e disable
</pre>
</blockquote>

<h2><a name="credits"> Credits </a></h2>

<p> Wietse Venema created Postfix, designed and implemented the
multi-instance wrapper framework and provided design feedback that made
the postmulti(1) utility much more general and useful than originally
envisioned. </p>

<p> The postmulti(1) utility was developed by Victor Duchovni of Morgan
Stanley, who also wrote the initial version of this document. </p>

</body> </html>