=head1 NAME

his - routines for managing INN history

=head1 SYNOPSIS

B<#include E<lt>inn/history.hE<gt>>

B<struct history;>

B<struct histstats {>
B<    int hitpos;>
B<    int hitneg;>
B<    int misses;>
B<    int dne;>
B<};>

B<#define HIS_RDONLY ...>
B<#define HIS_RDWR ...>
B<#define HIS_CREAT ...>
B<#define HIS_ONDISK ...>
B<#define HIS_INCORE ...>
B<#define HIS_MMAP ...>

B<enum {>
B<    HISCTLG_PATH,>
B<    HISCTLS_PATH,>
B<    HISCTLS_SYNCCOUNT,>
B<    HISCTLS_NPAIRS,>
B<    HISCTLS_IGNOREOLD,>
B<    HISCTLS_STATINTERVAL>
B<};>

B<struct history *HISopen(const char *>I<path>B<, const char *>I<method>B<, int >I<flags>B<);>

B<bool HISclose(struct history *>I<history>B<);>

B<bool HISsync(struct history *>I<history>B<);>

B<void HISsetcache(struct history *>I<history>B<, size_t >I<size>B<);>

B<bool HISlookup(struct history *>I<history>B<, const char *>I<key>B<, time_t *>I<arrived>B<, time_t *>I<posted>B<, time_t *>I<expires>B<, TOKEN *>I<token>B<);>

B<bool HIScheck(struct history *>I<history>B<, const char *>I<key>B<);>

B<bool HISwrite(struct history *>I<history>B<, const char *>I<key>B<, time_t >I<arrived>B<, time_t >I<posted>B<, time_t >I<expires>B<, const TOKEN *>I<token>B<);>

B<bool HISremember(struct history *>I<history>B<, const char *>I<key>B<, time_t >I<arrived>B<);>

B<bool HISreplace(struct history *>I<history>B<, const char *>I<key>B<, time_t >I<arrived>B<, time_t >I<posted>B<, time_t >I<expires>B<, const TOKEN *>I<token>B<);>

B<bool HISexpire(struct history *>I<history>B<, const char *>I<path>B<, const char *>I<reason>B<, bool >I<writing>B<, void *>I<cookie>B<, time_t >I<threshold>B<, bool (*>I<exists>B<)(void *cookie, time_t arrived, time_t posted, time_t expires, const TOKEN *token));>

B<bool HISwalk(struct history *>I<history>B<, const char *>I<reason>B<, void *>I<cookie>B<, bool (*>I<callback>B<)(void *cookie, time_t arrived, time_t posted, time_t expires, const TOKEN *token));>

B<struct histstats HISstats(struct history *>I<history>B<);>

B<const char *HISerror(struct history *>I<history>B<);>

B<bool HISctl(struct history *>I<history>B<, int >I<request>B<, void *>I<val>B<);>

=head1 DESCRIPTION

These functions provide provide access to the INN history
database. They maintain key/value pairs in an opaque database whilst
providing for expiry of outdated information.

The history structure is an opaque handle returned from HISopen.

The B<HISopen> function opens the history file designated by I<path>
using the mode I<flags> using the specified I<method>. I<flags> may be
B<HIS_RDONLY> to indicate that read-only access to the history
database is desired, or B<HIS_RDWR> for read/write access. History
methods are defined at build time; the history method currently
available is "hisv6". On success a newly initialised history handle is
returned, or B<NULL> on failure.

B<HIS_ONDISK>, B<HIS_INCORE> and B<HIS_MMAP> may be logically ORed
into I<flags> to provide a hint to the underlying history manager as
to how it should handle its data files; B<HIS_ONDISK> indicates that
the caller would like as much of the data to be kept on disk (and out
of memory), B<HIS_INCORE> indicates that the data files should be kept
in main memory where possible and B<HIS_MMAP> that the files should be
mmap()ed into the processes address space. B<HIS_INCORE> is typically
used where a mass rebuild of the history database is being performed;
the underlying history manager may assume that the caller will call
B<HISsync>() to sync the data files to disk.

The B<HIS_CREAT> flag indicates that the history database should be
initialised as new; if any options which affect creation of the
database need to be set an anonymous history handle should be created
by calling B<HISopen> with I<path> set to B<NULL>, any options set
using B<HISctl>, then the database opened by calling B<HISctl> with
B<HISCTLS_PATH>.

The B<HISclose> function closes the handle I<history> and deallocates
any resources associated with it. It returns B<false> on failure or
B<true> on success.

The B<HISsync> function synchronises any outstanding transactions
associated with I<history> to disk.

B<HISsetcache> associates a cache used for speeding up HIScheck with
I<history>. The cache will occupy approximately I<size> bytes.

B<HISlookup> retrieves a token from I<history> based on the passed
I<key> (normally the Message-ID). If no entry with an associated token
can be found, B<HISlookup> will return B<false>. If a token is found
I<arrived>, I<expires>, and I<posted> are filled in with the message
arrival, expiry, and posting times respectively (or zero, if the time
component is not available), in addition to I<token> being set to the
retrieved token and a function return value of B<true>. Any of
arrived, expires, posted, or token may be B<NULL> in which case that
component is not returned to the caller, without affecting the return
value.

B<HIScheck> checks the database I<history> for I<key> (normally the
Message-ID); if I<key> has previously been set via B<HISwrite>,
B<HIScheck> returns B<true>, else B<false>.

B<HISwrite> writes a new entry to the database I<history> associated
with I<key>. I<arrived>, I<posted>, and I<expired> specify the arrival,
posting, and expiry time respectively; I<posted> and I<expired> may be
specifed as <= 0 in which case that component shall be treated as
absent in the database. I<token> is associated with the specified
I<key>. B<HISwrite> returns B<true> on success, or B<false> on
failure. The behaviour when I<key> is not unique with respect to the
existing entries in I<history> is unspecified.

B<HISremember> writes a new entry to the database I<history>
associated with I<key>, merely remembering that this I<key> has been
seen, together with its arrival time I<arrived>. B<HISremember>
returns B<true> on success, or B<false> on failure. The behaviour when
I<key> is not unique with respect to the existing entries in
I<history> is unspecified.

B<HISreplace> replaces an existing entry in the database I<history>,
associated with I<key>. I<arrived>, I<posted>, I<expired> specify the
arrival, posting and expiry time respectively; I<posted> and
I<expired> may be specifed as <= 0 in which case that component shall
be treated as absent in the database. I<token> is associated with the
specified I<key>; if B<NULL> then the history database merely
remembers that this I<key> has been seen, together with its arrival
time. B<HISreplace> returns B<true> on success, or B<false> on
failure.

B<HISexpire> expires the history database associated with I<history>,
creating a new, replacement, database in the same location if I<path>
is B<NULL>, or in I<path> if not B<NULL>; if I<path> is not B<NULL>
then the replacement of the old history database with the new one is
assumed to be performed out of band by the caller. The I<writing> flag
is normally passed as B<true>, if you wish to inhibit writing of the
new database (and so merely see the callbacks), I<writing> may be set
B<false>.

If the underlying history mechanism needs to pause the server, the
I<reason> string is used as the argument to the `ctlinnd pause'
command, and as such the server should be reserved by the caller prior
to calling B<HISexpire>; if the caller wishes to inhibit pausing of
the server, passing B<NULL> will achieve this. If I<reason> is not
B<NULL>, then on successful return from B<HISexpire> the server will
be left paused and the caller should unpause it.

The history database is scanned and entries with an associated storage
token are passed to the discrimination function I<exists>.

If I<exists>() returns B<false> it indicates that stored entity
associated with token is no longer available (or no longer required),
and therefore the associated history entry may be expired once it
meets the I<threshold> constraint. If I<exists>() returns B<true> the
entry is kept as-is in the newly expired history database.

The I<exists> function is passed the arrival, posting and expiry
times, in addition to the token associated with the entry. Note that
posting and/or expiry may be zero, but that the token will never be
B<NULL> (such entries are handled solely via the threshold
mechanism). The storage token passed to the discrimination function
may updated if required (for example, as might be needed by a
hierachical storage management implementation).

Entries in the database with an arrival time less than I<threshold>
with no token associated with them are deleted from the database.

The parameter I<cookie> is passed to the discrimination function, and
may be used for any purpose required by the caller.

If the discrimination function attempts to access the underlying
database (for read or write) during the callback, the behaviour is
unspecified.

B<HISwalk> provides an iteration function for the specified I<history>
database. For every entry in the history database, I<callback> is
invoked, passing the I<cookie>, arrival, posting, and expiry times, in
addition to the token associated with the entry. If the I<callback>()
returns B<false> the iteration is aborted and B<HISwalk> returns
B<false> to the caller.

To process the entire database in the presence of a running server,
I<reason> may be passed; if this argument is not B<NULL>, it is used
as an an argument to the `ctlinnd (reserve|pause|go)' commands. If
I<reason> is B<NULL> and the server is running, the behaviour of
B<HISwalk> is undefined.

If the callback function attempts to access the underlying database
during the callback, the behaviour is unspecified.

B<HISstats> returns statistics on the history cache mechanism; given a
handle I<history>, the return value is a I<struct histstats>
detailing:

=over 4

=item C<hitpos>

The number of times an item was found directly in the cache and known
to exist in the underlying history manager.

=item C<hitneg>

The number of times an item was found directly in the cache and known
not to exist in the underlying history manager.

=item C<misses>

The number of times an item was not found directly in the cache, but
on retrieval from the underlying history manager was found to exist.

=item C<dne>

The number of times an item was not found directly in the cache, but
on retrieval from the underlying history manager was found not to exist.

=back

Note that the history cache is only checked by B<HIScheck> and only
affected by B<HIScheck>, B<HISwrite>, B<HISremember> and
B<HISreplace>. Following a call to B<HISstats> the history statistics
associated with I<history> are cleared.

B<HISerror> returns a string describing the most recent error
associated with I<history>; the format and content of these strings is
history manager dependent. Note that on setting an error, the history
API will call the B<warn> function from libinn(3).

B<HISctl> provides a control interface to the underlying history
manager. The I<request> argument determines the type of the request
and the meaning of the I<val> argument. The values for I<request> are:

=over 4

=item C<HISCTLG_PATH> (const char **)

Get the base file path which the history handle represents. I<val>
should be a pointer to a location of type B<const char *>.  The
result must not later be passed to free(3).


=item C<HISCTLS_PATH> (const char *)

Set the base file path which this history handle should use; typically
this is used after an anonymous handle has been created using
B<HISopen(NULL, ...)>. I<val> should be a value of type B<const char
*> and will be copied before being stored internally.

=item C<HISCTLS_SYNCCOUNT> (size_t *)

Set an upper bound on how many history operations may be pending in
core before being synced to permanent storage; B<0> indicates
unlimited. I<val> should be a pointer to a value of type B<size_t> and
will not be modified by the call.

=item C<HISCTLS_NPAIRS> (size_t *)

Set a hint to the to the underlying history manager as to how many
entries there are expected to be in the history database; B<0>
indicates that an automatic or default sizing should be made. I<val>
should be a pointer to a value of type B<size_t> and will not be
modified by the call.

=item C<HISCTLS_IGNOREOLD> (bool *)

Instruct the underlying history manager to ignore existing database
when creating new ones; typically this option may be set to B<true> if
the administrator believes that the existing history database is
corrupt and that ignoring it may help. I<val> should be a pointer to a
value of type B<bool> and will not be modified by the call.

=item C<HISCTLS_STATINTERVAL> (time_t *)

For the history v6 and tagged hash managers, set the interval, in
seconds, between stat(2)s of the history files checking for replaced
files (as happens during expire); this option is typically used by
nnrpd(8) like applications. I<val> should be a pointer to a value of
type B<time_t> and will not be modified by the call.

=head1 HISTORY

Written by Alex Kiernan <alexk@demon.net> for InterNetNews 2.4.0.

$Id: libinnhist.pod 5909 2002-12-03 05:17:18Z vinocur $