Whole buncha d-u-m-b questions

Heath Kehoe heath.kehoe at intermec.com
Thu Jul 6 18:22:21 UTC 2000


>
>Now, the most bothersome thing:  ovdb seems to be leaving around
>all its old logfiles.  Who is supposed to clean up after it, and
>when?

The "dbprocs" script is supposed to periodically remove old log
files.  It runs "db_archive -a" which ouputs a list of log files
that are no longer needed; then removes those log files.


>I created the overview database which resulted in a few logs
>being left around, but as soon as I started innd, it looks like
>there was some process that replayed them and deleted all but
>the most recent one.  Since then, none of the files has been
>deleted and I had to wipe out the first bunch by hand so as
>not to run out of space.
>
>I see that rc.news starts something which eventually is supposed
>to rm the file mentioned in tmp/dba.whatever, but I have a couple
>old files around -- would that prevent things from cleaning up?

Make sure that the db_checkpoint and db_archive commands are in
news' PATH.  Log in or su to your news user, then:

  . /path/to/innshellvars
  db_archive -a

If you get 'not found' for db_archive, then the dbprocs script
can't run it, either (see below).


>
>Now, is this to be expected?  I created 32 overview files, for a
>full-feed reader machine, and most of them are roughly the same
>size, except for one which is significantly larger:
>-rw-rw-r--   1 news     235888640 Jul  3 23:25 /news/overview/ov00019
>-rw-rw-r--   1 news     251199488 Jul  3 23:25 /news/overview/ov00020
>-rw-rw-r--   1 news     255762432 Jul  3 23:25 /news/overview/ov00021
>-rw-rw-r--   1 news     685096960 Jul  3 23:25 /news/overview/ov00022
>-rw-rw-r--   1 news     214360064 Jul  3 23:25 /news/overview/ov00023
>
>There are a couple that just top the 300MB mark, but only this
>one above is more than twice the size of any other at nearly
>700MB (it's been that way for a while).

This is normal.  The overview data is divided amongst the ovNNNNN
files by newsgroup.  So a newsgroup with an unusually large number
of articles will cause its corresponding ovNNNNN file to be larger
than the others.

It so happens that control.cancel goes into ov00022 (when
numdbfiles == 32) which probably explains why that one is so big.


>
>This is with 2.7.7 on Slowaris 2.7.  I haven't noticed any reader
>problems at all, so apart from all this, it seems to be fine.
>
>
>Now that I freed up a few gigs of space before stopping innd,
>since I restarted it, I'm getting this message every few minutes:
>db_archive: log_get:  /news/overview/log.0000001529: No such file or directory
>Probably since I deleted it.
>
>If there is not some bug resulting in the old log.* files sticking
>around when they should be deleted, then when is it safe to delete
>the logfiles?  And/or how?
>

If you delete a log before all of the transactions in it have been
checkpointed, you'll get those errors.  Be sure a "db_checkpoint"
process is running (it's normally started by "dbprocs").

Use the command "db_archive -a" to print the log files that can be
safely removed.  Again, dbprocs should do this for you.

Make sure you have the db_checkpoint, db_archive, and db_deadlock
commands which are a part of the BerkeleyDB distribution.  When
you install BerkeleyDB from the source distribution, those commands
get put into /usr/local/BerkeleyDB/bin.  Also, make sure those
commands are in your news user's PATH.


>As usual, a pointer to a description, or a timeframe where I can
>find discussion, of the Berzerk overview and how to run it with
>INN would be appreciated.
>

I have a web page with ovdb info: http://www.avalon.net/~hakehoe/inn/
Also there are man pages for ovdb and dbprocs in the INN distribution.

-heath




More information about the inn-workers mailing list