CONNECTION_CACHE_README.html [plain text]
<!doctype html public "-//W3C//DTD HTML 4.01 Transitional//EN"
"http://www.w3.org/TR/html4/loose.dtd">
<html>
<head>
<title>Postfix Connection Cache </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="">Postfix Connection Cache </h1>
<hr>
<h2>Introduction</h2>
<p> This document describes the Postfix connection cache implementation,
which is available with Postfix version 2.2 and later. </p>
<p> Topics covered in this document: </p>
<ul>
<li><a href="#summary"> What SMTP connection caching can do for you</a>
<li><a href="#implementation"> Connection cache implementation</a>
<li><a href="#configuration"> Connection cache configuration</a>
<li><a href="#safety">Connection cache safety mechanisms </a>
<li><a href="#limitations">Connection cache limitations</a>
<li><a href="#statistics">Connection cache statistics</a>
</ul>
<h2><a name="summary">What SMTP connection caching can do for
you</a></h2>
<p> With SMTP connection caching, Postfix can deliver multiple
messages over the same SMTP connection. By default, Postfix 2.2
reuses an SMTP connection automatically when a destination has
high volume of mail in the active queue. </p>
<p> SMTP Connection caching is a performance feature. Whether or not
it actually improves performance depends on the conditions: </p>
<ul>
<li> <p> SMTP Connection caching can greatly improve performance
when delivering mail to a destination with multiple mail servers,
because it can help Postfix to skip over a non-responding server.
</p>
<li> <p> Otherwise, the benefits of SMTP connection caching are
minor: it eliminates the latency of the TCP handshake (SYN, SYN+ACK,
ACK), plus the latency of the SMTP initial handshake (220 greeting,
EHLO command, EHLO response).
<li> <p> SMTP Connection caching gives no gains with respect to
SMTP session tear-down. The Postfix smtp(8) client normally does
not wait for the server's reply to the QUIT command, and it never
waits for the TCP final handshake to complete. </p>
<li> <p> SMTP Connection caching introduces some overhead: the
client needs to send an RSET command to find out if a connection
is still usable, before it can send the next MAIL FROM command.
</p>
</ul>
<p> For other potential issues with SMTP connection caching, see
the discussion of <a href="#limitations">limitations</a> at the end
of this document. </p>
<h2><a name="implementation">Connection cache implementation</a></h2>
<p> For an overview of how Postfix delivers mail, see the Postfix
architecture OVERVIEW document. </p>
<p> The Postfix connection cache is shared among Postfix mail
delivering processes. This maximizes the opportunity to reuse an
open connection. Other MTAs such as Sendmail or exim have a
non-shared connection cache. Here, a connection can be reused only
by the mail delivering process that creates the connection. To get
the same performance improvement as with a shared connection cache,
non-shared connections need to be kept open for a longer time. </p>
<p> The scache(8) server, introduced with Postfix version 2.2,
maintains the shared connection cache. With Postfix version 2.2,
only the smtp(8) client has support to access this cache. </p>
<blockquote>
<table>
<tr> <td> </td> <td> <tt> /-- </tt> </td> <td align="center"
colspan="3" bgcolor="#f0f0ff"> smtp(8) </td> <td colspan="2"> <tt>
--> </tt> Internet </td> </tr>
<tr> <td align="center" bgcolor="#f0f0ff"> qmgr(8) </td> <td> </td>
<td align="center" rowspan="3"> </td> <td align="center"
rowspan="3"><tt>|<br>|<br>|<br>|<br>v</tt></td> <td> </td>
</tr>
<tr> <td> </td> <td> <tt> \-- </tt> </td> <td align="center"
colspan="2" bgcolor="#f0f0ff"> smtp(8) </td> <td align="left"> <tt>
--> </tt> Internet </td> </tr>
<tr> <td colspan="3"> </td> <td align="center"><tt>^<br>|</tt></td>
<td> </td> </tr>
<tr> <td colspan="3"> </td> <td align="center" colspan="3"
bgcolor="#f0f0ff"> scache(8) </td> </tr>
</table>
</blockquote>
<p> When SMTP connection caching is enabled (see next section), the
smtp(8) client does not disconnect after a mail transaction, but
gives the connection to the scache(8) server which keeps the
connection open for a limited amount of time. </p>
<p> After handing over the open connection to the scache(8) server,
the smtp(8) client continues with some other mail delivery request.
Meanwhile, any smtp(8) client process can ask the scache(8) server
for that cached connection and reuse it for mail delivery. </p>
<p> The connection cache can be searched by destination domain name
(the right-hand side of the recipient address) and by the IP address
of the host at the other end of the connection. This allows Postfix
to reuse a connection even when the remote host is mail server for
domains with different names. </p>
<h2><a name="configuration">Connection cache configuration </a></h2>
<p> The Postfix smtp(8) client supports two connection caching
strategies: </p>
<ul>
<li> <p> On-demand connection caching. This is enabled by default,
and is controlled with the smtp_connection_cache_on_demand configuration
parameter. When this feature is enabled, the Postfix smtp(8) client
automatically saves a connection to the connection cache when a
destination has a high volume of mail in the active queue. </p>
<p> Example: </p>
<blockquote>
<pre>
/etc/postfix/main.cf:
smtp_connection_cache_on_demand = yes
</pre>
</blockquote>
<li> <p> Per-destination connection caching. This is enabled by
explicitly listing specific destinations with the
smtp_connection_cache_destinations configuration parameter. After
completing delivery to a selected destination, the Postfix smtp(8)
client <i>always</i> saves the connection to the connection cache.
</p>
<p> Specify a comma or white space separated list of destinations
or pseudo-destinations: </p>
<ul>
<li> <p> if mail is sent without a relay host: a domain name (the
right-hand side of an email address, without the [] around a numeric
IP address), </p>
<li> <p> if mail is sent via a relay host: a relay host name (without
the [] or non-default TCP port), as specified in main.cf or in the
transport map, </p>
<li> <p> a /file/name with domain names and/or relay host names as
defined above, </p>
<li> <p> a "type:table" with domain names and/or relay host names
on the left-hand side. The right-hand side result from "type:table"
lookups is ignored. </p>
</ul>
<p> Examples: </p>
<blockquote>
<pre>
/etc/postfix/main.cf:
smtp_connection_cache_destinations = $relayhost
smtp_connection_cache_destinations = hotmail.com, ...
smtp_connection_cache_destinations = static:all (<i>not recommended</i>)
</pre>
</blockquote>
</ul>
<h2><a name="safety">Connection cache safety mechanisms </a></h2>
<p> Connection caching must be used wisely. It is anti-social to
keep an unused SMTP connection open for a significant amount of
time, and it is unwise to send huge numbers of messages through
the same connection. In order to avoid problems with SMTP connection
caching, Postfix implements the following safety mechanisms: </p>
<ul>
<li> <p> The Postfix scache(8) server keeps a connection open for
only a limited time. The time limit is specified with the
smtp_connection_cache_time_limit and with the connection_cache_ttl_limit
configuration parameters. This prevents anti-social behavior. </p>
<li> <p> The Postfix smtp(8) client reuses a session for only a
limited number of times. This avoids triggering bugs in implementations
that do not correctly handle multiple deliveries per session. </p>
<p> With Postfix 2.2 the use count is limited with the
smtp_connection_cache_reuse_limit configuration parameter. With
Postfix 2.3 this is replaced by a time limit which is specified
with the smtp_connection_reuse_time_limit parameter. In addition,
Postfix 2.3 logs the use count of multiply-used connections,
as shown in the following example: </p>
<blockquote>
<pre>
Nov 3 16:04:31 myname postfix/smtp[30840]: 19B6B2900FE:
to=<wietse@test.example.com>, orig_to=<wietse@test>,
relay=mail.example.com[1.2.3.4], <b>conn_use=2</b>, delay=0.22,
delays=0.04/0.01/0.05/0.1, dsn=2.0.0, status=sent (250 2.0.0 Ok)
</pre>
</blockquote>
<li> <p> The connection cache explicitly labels each cached connection
with destination domain and IP address information. A connection
cache lookup succeeds only when the correct information is specified.
This prevents mis-delivery of mail. </p>
</ul>
<h2><a name="limitations">Connection cache limitations</a></h2>
<p> Postfix SMTP connection caching conflicts with certain applications:
</p>
<ul>
<li> <p> The Postfix shared connection cache cannot be used with
TLS, because saved TLS session information can be used only when a
new connection is created (this limitation does not exist in
connection caching implementations that reuse a connection only in
the process that creates it). For this reason, the Postfix smtp(8)
client always closes the connection after completing an attempt to
deliver mail over TLS. </p>
<li> <p> Postfix connection caching currently does not support
multiple SASL accounts per mail server. Specifically, Postfix
connection caching assumes that a SASL credential is valid for all
hostnames or domain names that deliver via the same mail server IP
address and TCP port, and assumes that the SASL credential does not
depend on the message originator. </p>
</ul>
<h2><a name="statistics">Connection cache statistics </a></h2>
<p> The scache(8) connection cache server logs statistics about the
peak cache size and the cache hit rates. This information is logged
every connection_cache_status_update_time seconds, when the process
terminates after the maximal idle time is exceeded, or when Postfix
is reloaded. </p>
<ul>
<li> <p> Hit rates for connection cache lookups by domain will tell
you how useful connection caching is. </p>
<li> <p> Connection cache lookups by network address will always
fail, unless you're sending mail to different domains that share
the same MX hosts. </p>
<li> <p> No statistics are logged when no attempts are made to
access the connection cache. </p>
</ul>
</body>
</html>