Manpages

NAME

makehistory - Initialize or rebuild INN history database

SYNOPSIS

makehistory [-abFIOSx] [-f filename] [-l count] [-L load-average] [-s size] [-T tmpdir]

DESCRIPTION

makehistory rebuilds the history(5) text file, which contains a list of message-IDs of articles already seen by the server. It can also be used to rebuild the overview database. Note that even though the dbz indices for the history file are also rebuilt by makehistory, it is useful to run makedbz(8) after makehistory(8) in order to improve the efficiency of the indices (makehistory does not know how large to make the hash table at first run, unless the size is given by the -s flag).

The default location of the history text file is pathdb/history; to specify an alternate location, use the -f flag.

By default, makehistory will scan the entire spool, using the storage manager, and write a history line for every article. To also generate overview information, use the -O flag.

If a malformed article is found in the news spool, in a way which prevents its integration into the history or overview data, a log line will be output and the malformed article will just be skipped.

OVERVIEW REBUILD

WARNING: If you’re trying to rebuild the overview database, be sure to stop innd(8) and delete or zero out the existing database before you start for the best results. An overview rebuild should not be done while the server is running. Unless the existing overview is deleted, you may end up with problems like out-of-order overview entries, excessively large overview buffers, and the like.

If ovmethod in inn.conf is "ovdb", you must have the ovdb processes running while rebuilding overview. ovdb needs them available while writing overview entries. You can start them by hand separate from the rest of the server by running ovdb_init; see ovdb_init(8) for more details.

Similarly, if ovmethod in inn.conf is "ovsqlite", you must have the ovsqlite-server process running while rebuilding overview. See ovsqlite-server(8) for more details and how to start it by hand.

Rebuilding overview data is as straight-forward as:

1.

Checking that the configuration file of the new overview storage method is present in pathetc and fits your needs (buffindexed.conf, ovdb.conf or ovsqlite.conf). Note that the tradindexed overview storage method does not have a dedicated configuration file.

2.

Making sure that INN is stopped ("rc.news stop" as the news user, or whichever command you’re using).

3.

Setting the new overview storage method in the ovmethod parameter in inn.conf.

4.

Making sure that the directory specified by the pathoverview parameter in inn.conf exists and is empty (or contains freshly created buffindexed buffers, if using that overview storage method). Otherwise, rename the current directory (to backup existing overview data) and re-create pathoverview as the news user.

5.

Starting ovdb_init or ovsqlite-server as the news user if the new overview storage method is respectively ovdb or ovsqlite.

6.

Running "makehistory -O -x -F" and waiting for the command to finish. (You may notice a few logs about articles for which overview data cannot be inserted into the new overview storage method. As long as there aren’t tons of them, it is normal, notably because there is an internal limit in the length of overview data generated by makehistory, contrary to innd. Unfortunately, these rare articles won’t be present in the new overview.)

7.

Stopping ovdb or ovsqlite helper programs if you started them during the previous steps (running "rc.news stop" as the news user will stop them; do not mind the messages related to the fact that the news server was not running).

8.

Starting INN and checking the logs to make sure everything is fine. You will normally notice that the active file is renumbered (rc.news takes care of that when run after an overview rebuild; otherwise, manually run "ctlinnd renumber ''").

OPTIONS

-a

Append to the history file rather than generating a new one. If you append to the main history file, make sure innd(8) is throttled or not running, or you can corrupt the history.

-b

Delete any messages found in the spool that do not have valid Message-ID header fields in them.

-F

Fork a separate process to flush overview data to disk rather than doing it directly. The advantage of this is that it allows makehistory to continue to collect more data from the spool while the first batch of data is being written to the overview database. The disadvantage is that up to twice as much temporary disk space will be used for the generated overview data. This option only makes sense in combination with -O. With buffindexed, the overchan program is invoked to write overview.

-f filename

Rather than writing directly to pathdb/history, instead write to filename, also in pathdb.

-I

Don’t store overview data for articles numbered lower than the lowest article number in active. This is useful if there are for whatever reason old articles on disk that shouldn’t be available to readers or put into the overview database.

-l count

This option specifies how many articles to process before writing the accumulated overview information out to the overview database. The default is "10000". Since overview write performance is faster with sorted data, each "batch" gets sorted. Increasing the batch size with this option may further improve write performance, at the cost of longer sort times. Also, temporary space will be needed to store the overview batches. At a rough estimate, about 300 * count bytes of temporary space will be required (not counting temp files created by sort(1)). See the description of the -T option for how to specify the temporary storage location. This option has no effect with buffindexed, because buffindexed does not need sorted overview and no batching is done.

-L load-average

Temporarily pause activities if the system load average exceeds the specified level load-average. This allows makehistory to run on a system being used for other purposes without monopolizing system resources and thus making the response time for other applications unacceptably slow. Using nice(1) does not help much for that because the problem comes from disk I/O usage, and ionice(1) is not always available or efficient.

-O

Create the overview database as well as the history file. Overview information is only required if the server supports readers; it is not needed for a transit-only server (see enableoverview in inn.conf(5)). If you are using the buffindexed overview storage method, erase all of your overview buffers before running makehistory with -O.

-S

Rather than storing the overview data into the overview database, just write it to standard output in a form suitable for feeding to overchan later if wished. When this option is used, -F, -I, -l, and -T are ignored. This option only makes sense in combination with -O.

-s size

Size the history database for approximately size key-value pairs (i.e. lines in history). Accurately specifying the size is an optimization that will create a more efficient database. (The size should be the estimated eventual number of articles, typically the size of the old history file, in lines.)

By default, makehistory will create a database optimized for handling about 6,000,000 articles (or 500,000 if the slower tagged hash format is used). This size does not limit the number of articles the news server can store in its history file. It will just get slower when that optimal size is exceeded, until the next run of news.daily which will appropriately resize it.

-T tmpdir

If -O is given, makehistory needs a location to write temporary overview data. By default, it uses pathtmp, set in inn.conf, but if this option is given, the provided tmpdir is used instead. This is also used for temporary files created by sort(1) (which is invoked in the process of writing overview information since sorted overview information writes faster). By default, sort usually uses your system temporary directory; see the sort(1) man page on your system to be sure.

-x

If this option is given, makehistory won’t write out history file entries. This is useful mostly for building overview without generating a new history file.

EXAMPLES

Here’s a typical example of rebuilding the entire history and overview database, removing broken articles in the news spool. This uses the default temporary file locations and should be done while innd isn’t running (or is throttled).

makehistory -b -f history.n -O -l 30000 -I

This will rebuild the overview (if using buffindexed, erase the existing overview buffers before running this command) and leave a new history file as "history.n" in pathdb. To preserve all of the history entries from the old history file that correspond to rejected articles or expired articles, follow the above command with:

cd <pathdb>
awk 'NF == 2 { print }' < history >> history.n

(replacing the path with your pathdb, if it isn’t the default). Then look over the new history file for problems and run:

makedbz -s `wc -l < history.n` -f history.n

Then rename all of the files matching "history.n.*" to "history.*", replacing the current history database and indices. After that, it’s safe to unthrottle innd.

For a simpler example:

makehistory -b -f history.n -I -O

will scan the spool, removing broken articles and generating history and overview entries for articles missing from history.

To pre-size the history file for 100,000,000 articles, and generate overview data at the same time, you may directly use the following command:

makehistory -O -s 100000000

You then do not need running makedbz as the history file has already been generated and optimized for the expected number of articles.

To just rebuild overview:

makehistory -O -x -F

FILES

pathdb/history

This is the default output file for makehistory.

pathtmp

Where temporary files are written unless -T is given.

HISTORY

Originally written by Rich $alz <rsalz [AT] uunet.net> for InterNetNews and updated by various other people since.

SEE ALSO

active(5), ctlinnd(8), history(5), inn.conf(5), innd(8), libinn_dbz(3), makedbz(8), ovdb_init(8), overchan(8), ovsqlite-server(8).