>I raised this with you in February, I wonder if there has been any
>progress.
>
>We run an Intranet news server, with 400 users accessing 50,000 messages
>in 100 groups (200Mb of news spool). News never expires. We use the FTS
>facility to provide a search engine. Hardware is a Sun Ultra 1 with
>128Mb RAM and 256Mb spool file, running Solaris 2.6.
>
>Our problem is that memory usage grows constantly, until all virtual
>memory is exhausted after 10-20 days. A restart of DNEWS is necessary to
>clear the problem.
>
>Interestingly, memory use does not steadily increase; there are 'jumps'
>of 10-20Mb followed by a day or so of steady utilisation. An extract
>from dnews.log around the time of a 'jump' is shown below.
>
Hi, sorry Dave somehow this never came to my attention, or dropped
out of my memory cache which seems to have an intermittent fault :-)
Anyway, it looks like a memory leak in the fts module which wouldn't
surprise me and shouldn't be hard to find. Could you send
me the output of
tellnews showmem
on the system after it's been up for a wee while. Or ideally also right after
restarting it so I can see the differences in memory tables and I'll get
back to
you shortly on it, if you don't hear for a week send me a reminder.
ChrisP.
>29 09:47:22 :warn: feed: Starting SMTP-feed to {web001.hmgcc}
>29 09:48:29 :warn: feed: Starting SMTP-feed to {web001.hmgcc}
>29 09:49:05 51:warn: xov: Xover file out of order (3) ?
>hmgcc.general:6354 6355 1323752
>29 09:51:46 16:warn: Sorting xover for (hmgcc.discuss.xxx) as it seems
>to have gaps 2/285/388 r=381 len=449
>29 09:51:48 :warn: Idle Expire xover group hmgcc.discuss.xxx
>29 09:51:49 :warn: xidle: Sort completed
>29 09:52:46 51:warn: Sorting xover for (hmgcc.discuss.yyy) as it seems
>to have gaps 2/1094/1381 r=1373 len=317
>29 09:52:48 :warn: Idle Expire xover group hmgcc.discuss.yyy
>29 09:52:54 :warn: xidle: Sort completed
>29 09:55:22 :error: Time to switch to the new .fts files
>
>The last line is interesting (but perhaps unrelated) - it occurs every
>30 minutes or so.
>
>Output of 'tellnews status' is below.
>
># tellnews status
>200 news.hmgcc DNEWS Version 5.4g5, S0, posting OK
>
>Feed[0]: me *
>Feed[1]: ftsfeed
>*,!hmgcc.qqq.*,!hmgcc.www.eee.support,!hmgcc.www.ttt.*,!
>Feed[2]: web001.hmgcc SMTP Posts=0 0.00/s Rej=0 Req=0 Dup=0 R=100:0:0
>Tot=0.00/
>s 0.00k/s
>Live cache misses 0/0
>
>{connection list snipped}
>
>Disk space 500MB - history*2=5MB,xover=12MB,dbi.idx*2=1MB = 481MB for
>piles
> Pile 0, 0.0/0.0% 0MB, days=0 none-old-format
> Pile 1, 39.4/95.0% 190MB, days=10000 *
>
>Expire started at None, start 0% end 0%
> Processed 0 removed 0 deleting 0/0/0 speed 0 pass 0
> Pile hist_mins 0, xover_mins 0, resort_mins 0, fillin 0, copy 0
>
>Nothing is being sucked at present.
>DB used 189.50MB, Slack 9.52MB 5.03%, DBI Misses 0/0, blocks 0=0k h=98
>History: disk 2557k memory 407k/840k, Str 1763k/2066k/11450k Malloc
>20952k/20991
>k/18397k
>Used 197MB=39%, Free work 704Mb, hist 704Mb, spool 704Mb, log 27Mb
>xbuild: q=0 sorted=17, fillin=5156 copy=17324
>Groups Xover 96 12MB, Cached 119 of 123, GID 125, Uptime 0 day(s) 2
>hour(s), XO
>ver Cache 12.53%
>
>.
>281 Completed
>205 closing connection - goodbye!
>
>
>We also run DNews as an Internet news server under Linux, handling 100Mb
>per day. There are no memory problems with this setup.
>
>TIA
>
>Dave Pickles
