expire.8cn

Langue: en

Version: 19 Oct 1994 (ubuntu - 08/07/09)

Section: 8 (Commandes administrateur)

NAME

expire, doexpire, expireiflow - expire old news
upact - update news active file

SYNOPSIS

/usr/lib/news/expire/expire [ -a archdir ] [ -p ] [ -s ] [ -F c ] [ -c ] [ -n nnnnn ] [ -t ] [ -l ] [ -v ] [ -d ] [ -r ] [ -g ] [ -h ] [ -H historydir ] [ controlfile ]
/usr/lib/news/expire/doexpire [ expireoptions ] [ -f ] [ -e ]
/usr/lib/news/expire/expireiflow minimum expireoptions
/usr/lib/news/expire/upact [ -b ] [ -p ] [ -s ] [ -# ]

DESCRIPTION

Expire expires old news, removing it from the current-news directories and (if asked to) archiving it elsewhere. It updates news's history file to match. Expire should normally be run nightly, typically by using doexpire (see below).

Expire's operations are controlled by a control file (which can be named or supplied on standard input), which is not optional---there is no default behavior. Each line of the control file (except for empty lines and lines starting with `#', which are ignored) should have four white-space-separated fields, as follows.

The first field is a newsgroup pattern list (containing no spaces!); partial specifications are acceptable (e.g. `comp' specifies all groups with that prefix). See newssys(5) for full details.

The second field is one letter, `m', `u', or `x', specifying that the line applies only to moderated groups, only to unmoderated groups, or to both, respectively.

The third field specifies the expiry period in days. The most general form is three numbers separated by dashes. The units are days, decimal fractions are permitted, and ``never'' is shorthand for an extremely large number. The first number gives the retention period: how long must pass after an article's arrival before it is a candidate for expiry. The third number gives the purge period: how long must pass after arrival before the article will be expired unconditionally. The middle number gives the expiry period: how long after an article's arrival it is expired by default. An explicit expiry date in the article will override the expiry period but not the retention period or the purge period. If the field contains only two numbers with a dash separating them, the retention period defaults to 0. If the field contains only a number, the retention period defaults to 0 and the purge period defaults to `never'. (But see below.) The retention period must be less than the purge period, and the expiry period must lie between them.

The fourth field is an archiving directory, or `@' which indicates that the default archiving directory (see -a) should be used, or `-' which suppresses archiving. An explicit archiving directory (not `@') prefixed with `=' means that articles should be archived into that directory itself; normally they go into subdirectories under it by newsgroup name, as in the current-news directory tree. (E.g., article 123 of comp.pc.drivel being archived into archive directory /exp would normally become /exp/comp/pc/drivel/123, but if the archiving directory was given as `=/exp' rather than `/exp', it would become /exp/123.) Expire creates subdirectories under an archiving directory automatically, but will not create the archiving directory itself. Archiving directories must be given as full pathnames.

The first line of the control file which applies to a given article is used to control its expiry. It is an error for no line to apply; the last line should be something like `all x 7 -' to ensure that at least one line is always applicable. Cross-posted articles are treated as if they were independently posted to each group.

The retention and purge defaults can be overridden by including a bounds line, one with the special first field /bounds/. The retention and purge defaults for following lines will be those of the bounds line. The defaults ``stretch'' as necessary to ensure that the purge period is never less than the expiry period and the retention period is never greater than the expiry period. The other fields of a bounds line are ignored but must be present.

Entries in the history file can be retained after article expiry, to stop a late-arriving copy of the article from being taken as a new article. To arrange this, include a line with the special first field /expired/; this line then controls the expiry of history lines after the corresponding articles expire. Dates are still measured from article arrival, not expiry. The other fields of such a line are ignored but must be present. It is strongly recommended that such a line be included, and that it specify as long a time as practical.

Command-line options are:

-a dir
dir is the default archiving directory; if no default is given, the control file may not contain any `@' archive-directory fields.
-p
print an `index' line for each archived article, containing its pathname, message ID, date received, and `Subject:' line.
-s
space is tight; optimize error recovery to minimize space consumed rather than to leave as much evidence as possible.
-F c
the subfield separator character in the middle history field is c rather than the normal `~'.
-c
check the format and consistency of the control file and the active file, but do not do any expiring.
-n nnnnn
set expire's idea of the time to nnnnn (for testing).
-t
print (on standard error) a shell-script-like description of what would be done, but don't do it. In the absence of archiving, all output lines will be of the form ``remove name'', where name is a pathname relative to /var/spool/news. If an article is to be archived, this will be preceded (on the same line) by ``copy name dir ; '', where name is as in remove and dir is an archiving directory (including any `=' prefix) as specified by the control file or the -a option.
-l
consider first filename in a history line to be the leader of its line, to be expired only after all others have expired. (Meant for use on obnoxious systems like VMS which don't support real links.)
-r
suppress history rebuild. Mostly for emergencies. (This leaves the history file out of date and larger than necessary, but improves speed and eliminates the need for several megabytes of temporary storage.)
-h
do not expire any article which would be archived if it were expired. Mostly for emergencies, so that expire can be run (to delete articles in non-archived groups) even if space is short in archiving areas.
-H historydir
the history file and its associated database files are in historydir rather than in /var/lib/news. (Useful because expire needs to do operations that can't be done through symbolic links.)
-v
verbose: report some statistics after termination.
-g
report expiry dates that getindate(3) does not like. Expire ignores such dates, treating the article as if it had no explicit expiry date.
-d
turn on (voluminous and cryptic) debugging output.

Expire considers the middle field of a history line to consist of one or more subfields separated by `~'. The first is the arrival date, which can be either an Internet-format human-readable date or a decimal seconds count; expire leaves this field unchanged. The second---if present, non-null, and not `-'---is an explicit expiry date for the file, again in either format, which expire will convert to a decimal seconds count as it regenerates the history file. Subsequent fields are preserved but ignored.

Doexpire checks whether another doexpire is running, checks that there is enough disk space, invokes expire with any expireoptions given and with /var/lib/news/explist as the control file, then runs upact and expov (see newsoverview(8CN)), and reports any difficulties by sending mail via report. This is usually better than just running expire directly. If space is not adequate for archiving, doexpire reports this and invokes expire with the -h option. If -r is not among the expireoptions, and disk space is persistently inadequate for the temporaries needed for history rebuilding, doexpire reports this and invokes expire with the -r option anyway. -f suppresses this, forcing expire to be run without -r regardless of the space situation. -e suppresses running of upact and expov, restricting doexpire to running expire only. -r implies -e.

Expireiflow checks whether there are at least minimum megabytes available for articles, and invokes doexpire with -r and the expireoptions (if any) if not. This may be useful on systems which run close to the edge on disk space.

Upact updates the active file to match the articles in /var/spool/news (for various reasons, expire does not do this). Normally, it updates the third field, and makes sure the second field (which relaynews updates in place and cannot expand) has a `0' on the front. The -p option suppresses all manipulation of the second field. The -b option causes both second and third fields to be rebuilt based on the articles, for disaster recovery. The -s option invokes a slower but more robust version of upact's internal machinery, which may be needed if the output of the system's ls command is broken in some way. Upact builds a new version of the active file in active.tmp, and normally renames this to active at the end; the -# option suppresses the renaming, leaving the result in active.tmp for debugging or inspection.

Upact forces the third field of the active file to be at least five digits, for backward compatibility, but otherwise just makes it as large as necessary. This field is never updated in place, so extending it to a larger number of digits is unnecessary.

The new implementation of upact supersedes the old recovact and updatemin commands, and is much faster than the old upact.

FILES

 /var/lib/news/historyhistory file
 /var/lib/news/history.pagdbm database for history file
 /var/lib/news/history.dirdbm database for history file
 /var/lib/news/explistexpiry control file
 /var/lib/news/history.ohistory file as of last expiry
 /var/lib/news/history.n*new history file and dbm files abuilding
 /var/lib/news/LOCKexpiredoexpire's lock file
 
 

SEE ALSO

inews(1CN), dbm(3), newssys(5), relaynews(8CN)

HISTORY

Written at U of Toronto by Henry Spencer, with contributions by Geoff Collyer. The idea for the fast algorithm in upact came from Bernd Felsche of MetaPro Systems, although the final code is rather different from his.

BUGS

Archiving is always done by copying, never by linking. This has the side effect that cross-posted articles are archived as several independent copies.

The -p subject-finder botches continued header lines, although such lines are rare.

One cannot put more than one newsgroup into a single archiving directory with the `=' feature, since the article numbers will collide with each other and expire doesn't do anything about this. Note that archiving a newsgroup which has subgroups into an `=' directory puts all the subgroups in the same directory as the parent! (Specifying the group as `foo.bar,!foo.bar.all' will avoid this.)

Expire uses access(2) to test for the presence of archiving directories, which can cause anomalies if it is run setuid (normally it's not).

Expire startup time is proportional to the product of the number of entries in the control file and the number of lines in active. It can be significant for complex control files.