pflogsumm.pl1

Usage

Overview

    Pflogsumm is a log analyzer/summarizer for the Postfix MTA.  It is
    designed to provide an over-view of Postfix activity, with just enough
    detail to give the administrator a "heads up" for potential trouble
    spots.
    
    Pflogsumm generates summaries and, in some cases, detailed reports of
    mail server traffic volumes, rejected and bounced email, and server
    warnings, errors and panics.

Options

    --bounce_detail <cnt>
                   Limit detailed bounce reports to the top <cnt>.  0
                   to suppress entirely.
    -d today       generate report for just today
    -d yesterday   generate report for just "yesterday"
    --deferral_detail <cnt>
                   Limit detailed deferral reports to the top <cnt>.  0
                   to suppress entirely.
    --detail <cnt>
    
                   Sets all --*_detail, -h and -u to <cnt>.  Is
                   over-ridden by individual settings.  --detail 0
                   suppresses *all* detail.
    -e             extended (extreme? excessive?) detail
                   Emit detailed reports.  At present, this includes
                   only a per-message report, sorted by sender domain,
                   then user-in-domain, then by queue i.d.
                   WARNING: the data built to generate this report can
                   quickly consume very large amounts of memory if a
                   lot of log entries are processed!
    -h <cnt>       top <cnt> to display in host/domain reports.
    
                   0 = none.
                   See also: "-u" and "--*_detail" options for further
                             report-limiting options.
    --help         Emit short usage message and bail out.
    
                   (By happy coincidence, "-h" alone does much the same,
                   being as it requires a numeric argument :-).  Yeah, I
                   know: lame.)
    -i
    --ignore_case  Handle complete email address in a case-insensitive
                   manner.
                   
                   Normally pflogsumm lower-cases only the host and
                   domain parts, leaving the user part alone.  This
                   option causes the entire email address to be lower-
                   cased.
    --iso_date_time
                   For summaries that contain date or time information,
                   use ISO 8601 standard formats (CCYY-MM-DD and HH:MM),
                   rather than "Mon DD CCYY" and "HHMM".
    -m             modify (mung?) UUCP-style bang-paths
    --uucp_mung
                   This is for use when you have a mix of Internet-style
                   domain addresses and UUCP-style bang-paths in the log.
                   Upstream UUCP feeds sometimes mung Internet domain
                   style address into bang-paths.  This option can
                   sometimes undo the "damage".  For example:
                   "somehost.dom!username@foo" (where "foo" is the next
                   host upstream and "somehost.dom" was whence the email
                   originated) will get converted to
                   "[email protected]".  This also affects the
                   extended detail report (-e), to help ensure that by-
                    domain-by-name sorting is more accurate.
    --mailq        Run "mailq" command at end of report.
    
                   Merely a convenience feature.  (Assumes that "mailq"
                   is in $PATH.  See "$mailqCmd" variable to path thisi
                   if desired.)
    --no_bounce_detail
    --no_deferral_detail
    --no_reject_detail
                   These switches are depreciated in favour of
                   --bounce_detail, --deferral_detail and
                   --reject_detail, respectively.
                   Suppresses the printing of the following detailed
                   reports, respectively:
                        message bounce detail (by relay)
                        message deferral detail
                        message reject detail
                   See also: "-u" and "-h" for further report-limiting
                             options.
    --no_no_msg_size
                    Do not emit report on "Messages with no size data".
                    Message size is reported only by the queue manager.
                    The message may be delivered long-enough after the
                    (last) qmgr log entry that the information is not in
                    the log(s) processed by a particular run of
                    pflogsumm.pl.  This throws off "Recipients by message
                    size" and the total for "bytes delivered." These are
                    normally reported by pflogsumm as "Messages with no
                    size data."
    --no_smtpd_warnings
                   This switch is depreciated in favour of
                   smtpd_warning_detail
                    On a busy mail server, say at an ISP, SMTPD warnings
                    can result in a rather sizeable report.  This option
                    turns reporting them off.
    --problems_first
                   Emit "problems" reports (bounces, defers, warnings,
                   etc.) before "normal" stats.
    --rej_add_from
                   For those reject reports that list IP addresses or
                   host/domain names: append the email from address to
                   each listing.  (Does not apply to "Improper use of
                   SMTP command pipelining" report.)
    -q             quiet - don't print headings for empty reports
    
                   note: headings for warning, fatal, and "master"
                   messages will always be printed.
    --reject_detail <cnt>
                   Limit detailed smtpd reject, warn, hold and discard
                   reports to the top <cnt>.  0 to suppress entirely.
    --smtp_detail <cnt>
                   Limit detailed smtp delivery reports to the top <cnt>.
                   0 to suppress entirely.
    --smtpd_stats
                   Generate smtpd connection statistics.
                   The "per-day" report is not generated for single-day
                   reports.  For multiple-day reports: "per-hour" numbers
                   are daily averages (reflected in the report heading).
    --smtpd_warning_detail <cnt>
                   Limit detailed smtpd warnings reports to the top <cnt>.
                   0 to suppress entirely.
    --syslog_name=name
                   Set syslog_name to look for for Postfix log entries.
                   By default, pflogsumm looks for entries in logfiles
                   with a syslog name of "postfix," the default.
                   If you've set a non-default "syslog_name" parameter
                   in your Postfix configuration, use this option to
                   tell pflogsumm what that is.
                   See the discussion about the use of this option under
                   "NOTES," below.
    -u <cnt>       top <cnt> to display in user reports. 0 == none.
                   See also: "-h" and "--*_detail" options for further
                             report-limiting options.
    --verbose_msg_detail
                   For the message deferral, bounce and reject summaries:
                   display the full "reason", rather than a truncated one.
                   Note: this can result in quite long lines in the report.
    --verp_mung    do "VERP" generated address (?) munging.  Convert
    --verp_mung=2  sender addresses of the form
                   "[email protected]"
                    to
                      "[email protected]"
                    In other words: replace the numeric value with "ID".
                   By specifying the optional "=2" (second form), the
                   munging is more "aggressive", converting the address
                   to something like:
                        "[email protected]"
                   Actually: specifying anything less than 2 does the
                   "simple" munging and anything greater than 1 results
                   in the more "aggressive" hack being applied.
                   See "NOTES" regarding this option.
    --version      Print program name and version and bail out.
    --zero_fill    "Zero-fill" certain arrays so reports come out with
                   data in columns that that might otherwise be blank.

Return value

    Pflogsumm doesn't return anything of interest to the shell.

Errors

    Error messages are emitted to stderr.

Examples

    Produce a report of previous day's activities:
        pflogsumm.pl -d yesterday /var/log/maillog
    A report of prior week's activities (after logs rotated):
        pflogsumm.pl /var/log/maillog.0
    What's happened so far today:
        pflogsumm.pl -d today /var/log/maillog
    Crontab entry to generate a report of the previous day's activity
    at 10 minutes after midnight.
        10 0 * * * /usr/local/sbin/pflogsumm -d yesterday /var/log/maillog
        2>&1 |/usr/bin/mailx -s "`uname -n` daily mail stats" postmaster
    Crontab entry to generate a report for the prior week's activity.
    (This example assumes one rotates ones mail logs weekly, some time
    before 4:10 a.m. on Sunday.)
        10 4 * * 0   /usr/local/sbin/pflogsumm /var/log/maillog.0
        2>&1 |/usr/bin/mailx -s "`uname -n` weekly mail stats" postmaster
    The two crontab examples, above, must actually be a single line
    each.  They're broken-up into two-or-more lines due to page
    formatting issues.

Notes

    Pflogsumm makes no attempt to catch/parse non-Postfix log
    entries.  Unless it has "postfix/" in the log entry, it will be
    ignored.
    It's important that the logs are presented to pflogsumm in
    chronological order so that message sizes are available when
    needed.
    For display purposes: integer values are munged into "kilo" and
    "mega" notation as they exceed certain values.  I chose the
    admittedly arbitrary boundaries of 512k and 512m as the points at
    which to do this--my thinking being 512x was the largest number
    (of digits) that most folks can comfortably grok at-a-glance.
    These are "computer" "k" and "m", not 1000 and 1,000,000.  You
    can easily change all of this with some constants near the
    beginning of the program.
    "Items-per-day" reports are not generated for single-day
    reports.  For multiple-day reports: "Items-per-hour" numbers are
    daily averages (reflected in the report headings).
    Message rejects, reject warnings, holds and discards are all
    reported under the "rejects" column for the Per-Hour and Per-Day
    traffic summaries.
    Verp munging may not always result in correct address and
    address-count reduction.
    Verp munging is always in a state of experimentation.  The use
    of this option may result in inaccurate statistics with regards
    to the "senders" count.
    UUCP-style bang-path handling needs more work.  Particularly if
    Postfix is not being run with "swap_bangpath = yes" and/or *is* being
    run with "append_dot_mydomain = yes", the detailed by-message report
    may not be sorted correctly by-domain-by-user.  (Also depends on
    upstream MTA, I suspect.)
    The "percent rejected" and "percent discarded" figures are only
    approximations.  They are calculated as follows (example is for
    "percent rejected"):
        percent rejected =
        
            (rejected / (delivered + rejected + discarded)) * 100
    There are some issues with the use of --syslog_name.  The problem is
    that, even with $syslog_name set, Postfix will sometimes still log
    things with "postfix" as the syslog_name.  This is noted in
    /etc/postfix/sample-misc.cf:
        # Beware: a non-default syslog_name setting takes effect only
        # after process initialization. Some initialization errors will be
        # logged with the default name, especially errors while parsing
        # the command line and errors while accessing the Postfix main.cf
        # configuration file.
    As a consequence, pflogsumm must always look for "postfix," in logs,
    as well as whatever is supplied for syslog_name.
    Where this becomes an issue is where people are running two or more
    instances of Postfix, logging to the same file.  In such a case:
        . Neither instance may use the default "postfix" syslog name
          and...
        . Log entries that fall victim to what's described in
          sample-misc.cf will be reported under "postfix", so that if
          you're running pflogsumm twice, once for each syslog_name, such
          log entries will show up in each report.
    The Pflogsumm Home Page is at:
        http://jimsun.LinxNet.com/postfix_contrib.html

Requirements

    For certain options (e.g.: --smtpd_stats), Pflogsumm requires the
    Date::Calc module, which can be obtained from CPAN at
    http://www.perl.com.
    Pflogsumm is currently written and tested under Perl 5.8.3.
    As of version 19990413-02, pflogsumm worked with Perl 5.003, but
    future compatibility is not guaranteed.