| < readers.conf | Russ Allbery > Software > INN > INN CURRENT Documentation | subscriptions > |
(Configuration file for storage manager)
The file pathetc/storage.conf contains the rules to be used in assigning articles to different storage methods. These rules determine where incoming articles will be stored.
The storage manager is a unified interface between INN and a variety of different storage methods, allowing the news administrator to choose between different storage methods with different trade-offs (or even use several at the same time for different newsgroups, or articles of different sizes). The rest of INN need not care what type of storage method was used for a given article; the storage manager will figure this out automatically when that article is retrieved via the storage API. Note that you may also want to see the options provided in inn.conf(5) regarding article storage.
The storage.conf file consists of a series of storage method entries.
Blank lines and lines beginning with a number sign (#) are ignored.
The maximum number of characters in each line is 255. The order of
entries in this file is important, see below.
Each entry specifies a storage method and a set of rules. Articles which match all of the rules of a storage method entry will be stored using that storage method; if an article matches multiple storage method entries, the first one will be used. Each entry is formatted as follows:
method <methodname> {
class: <storage_class>
newsgroups: <wildmat>
size: <minsize>[,<maxsize>]
expires: <mintime>[,<maxtime>]
options: <options>
exactmatch: <bool>
}
If spaces or tabs are included in a value, that value must be enclosed in
double quotes (""). If either a number sign (#) or a double quote
are meant to be included verbatim in a value, they should be escaped
with \.
<methodname> is the name of a storage method to use for articles which match the rules of this entry. The currently available storage methods are:
cnfs
timecaf
timehash
tradspool
trash
See the STORAGE METHODS section below for more details.
The meanings of the keys in each storage method entry are as follows:
An identifier for this storage method entry. <storage_class> should be a
number between 0 and 255. It should be unique across all of the entries
in this file. It is mainly used for specifying expiration times by storage
class as described in expire.ctl(5); timehash and timecaf will also set
the top-level directory in which articles accepted by this storage class are
stored. The assignment of a particular number to a storage class is arbitrary
but permanent (since it is used in storage tokens). Storage classes can be
for instance numbered sequentially in storage.conf.
What newsgroups are stored using this storage method. <wildmat> is a
uwildmat pattern which is matched against the newsgroups an article
is posted to. If storeonxref in inn.conf is true, this pattern
will be matched against the newsgroup names in the Xref header field
body; otherwise, it will be matched against the newsgroup names in the
Newsgroups header field body (see inn.conf(5) for discussion of the
differences between these possibilities). Poison wildmat expressions
(expressions starting with @) are allowed and can be used to exclude
certain group patterns: articles crossposted to poisoned newsgroups
will not be stored using this storage method. The <wildmat> pattern
is matched in order.
There is no default newsgroups pattern; if an entry should match all
newsgroups, use an explicit newsgroups: *.
A range of article sizes (in bytes) which should be stored using this storage
method. If <maxsize> is 0 or not given, the upper size of articles is
limited only by maxartsize in inn.conf. The size: field is optional
and may be omitted entirely if you want articles of any size to be stored
in this storage method (if, of course, these articles fulfill all the other
requirements of this storage method entry). By default, <minsize> is set
to 0.
A range of article expiration times which should be stored using this storage method. Be careful; this is less useful than it may appear at first. This is based only on the Expires header field of the article, not on any local expiration policies or anything in expire.ctl! If <mintime> is non-zero, then this entry will not match any article without an Expires header field. This key is therefore only really useful for assigning articles with requested longer expire times to a separate storage method. Articles only match if the time until expiration (that is to say, the amount of time into the future that the Expires header field of the article requests that it remain around) falls in the interval specified by <mintime> and <maxtime>.
The format of these parameters is 0d0h0m0s (days, hours, minutes, and
seconds into the future). If <maxtime> is 0s or is not specified, there is
no upper bound on expire times falling into this entry (note that this key has
no effect on when the article will actually be expired, but only on whether
or not the article will be stored using this storage method). This field is
also optional and may be omitted entirely if you do not want to store articles
according to their Expires header field, if any.
A <mintime> value greater than 0s implies that this storage method won't
match any article without an Expires header field.
This key is for passing special options to storage methods that require them
(currently only cnfs). See the STORAGE METHODS section below for
a description of its use.
If this key is set to true, all the newsgroups in the Newsgroups header
field body of incoming articles will be examined to see if they match
newsgroups patterns. (Normally, any non-zero number of matching
newsgroups is sufficient, provided no newsgroup matches a poison
wildmat as described above.) This is a boolean value; true, yes
and on are usable to enable this key. The case of these values is
not significant. The default is false.
If an article matches all of the constraints of an entry, it is stored via that storage method and is associated with that <storage_class>. This file is scanned in order and the first matching entry is used to store the article.
If an article does not match any entry, either by being posted to a newsgroup which does not match any of the <wildmat> patterns or by being outside the size and expires ranges of all entries whose newsgroups pattern it does match, the article is not stored and is rejected by innd. When this happens, the error message:
cant store article: no matching entry in storage.conf
is logged to syslog. If you want to silently drop articles matching certain
newsgroup patterns or size or expires ranges, assign them to the trash
storage method rather than having them not match any storage method entry.
Currently, there are five storage methods available. Each method has its pros and cons; you can choose any mixture of them as is suitable for your environment. Note that each method has an attribute EXPENSIVESTAT which indicates whether checking the existence of an article is expensive or not. This is used to run expireover(8).
The cnfs storage method stores articles in large cyclic buffers (CNFS
stands for Cyclic News File System). Articles are stored in CNFS buffers
in arrival order, and when the buffer fills, it wraps around to the beginning
and stores new articles over the top of the oldest articles in the buffer. The
expire time of articles stored in CNFS buffers is therefore entirely
determined by how long it takes the buffer to wrap around, which depends
on how quickly data is being stored in it. (This method is therefore said
to have self-expire functionality. It also means that when an article is
cancelled, the cycbuff doesn't go back and use space until it rolls over and
the whole cycbuff starts being reused.) EXPENSIVESTAT is false for this
method.
CNFS has its own configuration file, cycbuff.conf, which describes some
subtleties to the basic description given above. Storage method entries
for the cnfs storage method must have an options: field specifying
the metacycbuff into which articles matching that entry should be stored;
see cycbuff.conf(5) for details on metacycbuffs.
Advantages: By far the fastest of all storage methods (except for trash),
since it eliminates the overhead of dealing with a file system and creating
new files. Unlike all other storage methods, it does not require manual
article expiration. With CNFS, the server will never throttle itself
due to a full spool disk, and groups are restricted to just the buffer
files given so that they can never use more than the amount of disk space
allocated to them.
Disadvantages: Article retention times are more difficult to control because old articles are overwritten automatically. Attacks on Usenet, such as flooding or massive amounts of spam, can result in wanted articles expiring much faster than intended (with no warning).
This method stores multiple articles in one file, whose name is based on the article's arrival time and the storage class. The file name will be:
<patharticles>/timecaf-nn/bb/aacc.CF
where nn is the hexadecimal value of <storage_class>, bb and aacc
are the hexadecimal components of the arrival time, and CF is a
hardcoded extension. (The arrival time, in seconds since the epoch,
is converted to hexadecimal and interpreted as 0xaabbccdd, with
aa, bb, and cc used to build the path.) This method does not
have self-expire functionality (meaning expire has to run periodically
to delete old articles, as well as cancelled articles if immediatecancel
is not set to true in inn.conf). EXPENSIVESTAT is false for this method.
Advantages: It is roughly four times faster than timehash for article
writes, since much of the file system overhead is bypassed, while still
retaining the same fine control over article retention time.
Disadvantages: Using this method means giving up all but the most careful
manually fiddling with the article spool; in this aspect, it looks like
cnfs. As one of the newer and least widely used storage types,
timecaf has not been as thoroughly tested as the other methods.
This method is very similar to timecaf except that each article is
stored in a separate file. The name of the file for a given article
will be:
<patharticles>/time-nn/bb/cc/yyyy-aadd
where nn is the hexadecimal value of <storage_class>, yyyy is a
hexadecimal sequence number, and bb, cc, and aadd are components
of the arrival time in hexadecimal (the arrival time is interpreted as
documented above under timecaf). This method does not have self-expire
functionality. Cancelled articles are removed immediately. EXPENSIVESTAT
is true for this method.
Advantages: Heavy traffic groups do not cause bottlenecks, and a fine control of article retention time is still possible.
Disadvantages: The ability to easily find all articles in a given newsgroup and manually fiddle with the article spool is lost, and INN still suffers from speed degradation due to file system overhead (creating and deleting individual files is a slow operation).
Traditional spool, or tradspool, is the traditional news article storage
format. Each article is stored in an individual text file named:
<patharticles>/news/group/name/nnnnn
where news/group/name is the name of the newsgroup to which the article
was posted with each period changed to a slash, and nnnnn is the
sequence number of the article in that newsgroup. For crossposted articles,
the article is linked into each newsgroup to which it is crossposted
(using either hard or symbolic links). This is the way versions of INN
prior to 2.0 stored all articles, as well as being the article storage format
used by C News and earlier news systems. This method does not have
self-expire functionality. Cancelled articles are removed immediately.
EXPENSIVESTAT is true for this method.
Advantages: It is widely used and well-understood; it can read article spools written by older versions of INN and it is compatible with all third-party INN add-ons. This storage mechanism provides easy and direct access to the articles stored on the server and makes writing programs that fiddle with the news spool very easy, and gives fine control over article retention times.
Disadvantages: It takes a very fast file system and I/O system to keep up with current Usenet traffic volumes due to file system overhead. Groups with heavy traffic tend to create a bottleneck because of inefficiencies in storing large numbers of article files in a single directory. It requires a nightly expire program to delete old articles out of the news spool, a process that can slow down the server for several hours or more.
This method silently discards all articles stored in it. Its only real uses are for testing and for silently discarding articles matching a particular storage method entry (for whatever reason). Articles stored in this method take up no disk space and can never be retrieved, so this method has self-expire functionality of a sort. EXPENSIVESTAT is false for this method.
The following sample storage.conf file would store all articles
posted to alt.binaries.* in the BINARIES CNFS metacycbuff,
all articles over roughly 50 KB in any other hierarchy in the LARGE CNFS
metacycbuff, all other articles in alt.* in one timehash class, and all
other articles in any newsgroups in a second timehash class, except for
the internal.* hierarchy which is stored in traditional spool format.
method tradspool {
class: 1
newsgroups: internal.*
}
method cnfs {
class: 2
newsgroups: alt.binaries.*
options: BINARIES
}
method cnfs {
class: 3
newsgroups: *
size: 50000
options: LARGE
}
method timehash {
class: 4
newsgroups: alt.*
}
method timehash {
class: 5
newsgroups: *
}
Notice that the last storage method entry will catch everything. This is a good habit to get into; make sure that you have at least one catch-all entry just in case something you did not expect falls through the cracks. Notice also that the special rule for the internal.* hierarchy is first, so it will catch even articles crossposted to alt.binaries.* or over 50 KB in size.
As for poison wildmat expressions, if you have for instance an article crossposted between misc.foo and misc.bar, the pattern:
misc.*,!misc.bar
will match that article whereas the pattern:
misc.*,@misc.bar
will not match that article. An article posted only to misc.bar will fail to match either pattern.
Usually, high-volume groups and groups whose articles do not need to be kept
around very long (binaries groups, *.jobs*, news.lists.filters, etc.) are
stored in CNFS buffers. Use the other methods (or CNFS buffers again) for
everything else. However, it is as often as not most convenient to keep in
tradspool special hierarchies like local hierarchies and hierarchies that
should never expire or through the spool of which you need to go manually.
Written by Katsuhiro Kondou <kondou@nec.co.jp> for InterNetNews. Rewritten into POD by Julien Elie.
cycbuff.conf(5), expire.ctl(5), expireover(8), inn.conf(5), innd(8), libinn_uwildmat(3).
| < readers.conf | Russ Allbery > Software > INN > INN CURRENT Documentation | subscriptions > |