dave@bvax.UUCP (Dave Wallace) (05/06/87)
I recently joined the 'net but have a problem. I am running System V, Release 2.2 on a VAX 730( No, that's NOT the problem). When I get my 'feed', uuxqt runs, forking a sh, who forks a sh, who forks a compress and rnews,( plus the compress forks an unbatch). I can get up to 6 of these running at once, which causes the 730 to slow down just a bit (-: . I currently get my feed from one site and don't send to anyone else ( until I get this problem resolved. ) It takes the 730 most of a working day to process 6 or 7 batch files in /usr/spool/uucp! It seems like these processes are fighing them selves. Any body have any suggestions? I appologize for the length of this posting, but I have included a sample ps to show what is going on. F S UID PID PPID C PRI NI ADDR SZ WCHAN TTY TIME COMD 3 S 0 0 0 0 0 20 322 4 8003bfb0 ? 0:01 swapper 1 S 0 1 0 0 39 20 4cb 93 7ffff800 ? 3:10 init 3 S 0 0 0 0 0 20 4d2 4 8000456c ? 3:01 swapper 1 S 0 44 1 0 28 20 59d 56 8003c864 console 0:02 getty 1 S 0 9609 1 0 28 20 f05 56 8003a834 ttyp12 0:02 getty 1 S 0 33 1 0 26 20 72d 94 8002a802 ? 0:54 cron 1 S 0 35 1 0 39 20 6e6 12 7ffff800 console 0:35 update 1 S 0 37 1 0 26 20 6d1 32 8001e828 console 0:00 errdemon 1 S 71 40 1 0 26 20 67d 90 8002a142 ? 0:03 lpsched 1 S 100 10040 1 0 30 20 c40 97 8002d300 ttyp03 0:11 sh 1 S 9 10119 1 0 30 20 f45 117 8002d350 ttyp12 0:02 uuxqt 1 S 0 48 1 0 28 20 86c 56 8003a5cc ttyp05 0:02 getty 1 S 0 49 1 0 28 20 8a2 56 8003a624 ttyp06 0:02 getty 1 S 0 50 1 0 28 20 8b8 56 8003a67c ttyp07 0:02 getty 1 S 9 10343 10119 0 30 20 d20 90 8002d490 ttyp12 0:00 sh 1 S 0 52 1 0 28 20 8e6 56 8003a7dc ttyp11 0:02 getty 1 S 9 10344 10343 0 30 20 b83 90 8002d530 ttyp12 0:01 sh 1 S 0 54 1 0 28 20 90e 56 8003a88c ttyp13 0:02 getty 1 S 0 55 1 0 28 20 92d 56 8003a8e4 ttyp14 0:02 getty 1 S 9 10368 1 0 30 20 c3e 117 8002d620 ttyp12 0:01 uuxqt 1 S 9 10369 10368 0 30 20 c74 90 8002d670 ttyp12 0:00 sh 1 S 101 10438 1 0 28 20 e00 94 8003a574 ttyp04 0:09 sh 1 S 9 10384 1 0 30 20 b60 117 8002d710 ttyp12 0:01 uuxqt 1 S 9 10385 10384 0 30 20 ac2 90 8002d760 ttyp12 0:00 sh 1 R 9 10579 10387 10 63 20 9e6 203 ttyp12 0:46 rnews 1 S 9 10370 10369 0 30 20 860 90 8002d800 ttyp12 0:01 sh 1 S 107 4224 1 0 30 20 9eb 139 8002d850 ttyp02 0:15 csh 1 S 9 10371 10370 0 30 20 c60 56 8002d8a0 ttyp12 0:07 unbatch 1 S 9 10346 10344 0 30 20 a5e 56 8002d8f0 ttyp12 0:07 unbatch 1 S 9 10386 10385 0 30 20 5df 90 8002d940 ttyp12 0:01 sh 1 S 9 10372 10371 0 26 20 d0c 866 8002aec4 ttyp12 0:18 compress 1 S 9 10387 10386 0 30 20 c72 56 8002d9e0 ttyp12 0:07 unbatch 1 S 9 10388 10387 0 26 20 9c4 866 8002b884 ttyp12 0:16 compress 1 S 9 10347 10346 0 26 20 80b 866 8002c484 ttyp12 0:17 compress 1 R 5 10580 10281 4 64 22 b74 203 ? 0:28 rnews 1 R 5 10581 10529 4 64 22 bd6 203 ? 0:20 rnews 1 R 9 10583 10371 29 74 20 e13 203 ttyp12 0:21 rnews 1 S 3 10525 10522 0 39 26 eda 95 7ffff800 ? 0:02 sadc 1 S 107 10263 4224 3 28 20 ce5 316 8003a4c4 ttyp02 1:10 vi 1 S 5 10279 10278 0 30 22 fcc 90 8002dc60 ? 0:00 sh 1 S 5 10280 10279 0 30 22 505 90 8002dcb0 ? 0:01 sh 1 S 5 10527 10526 0 30 22 7e0 90 8002dd00 ? 0:00 sh 1 S 3 10522 1 0 30 26 f37 89 8002dd50 ? 0:01 sh 1 S 5 10278 1 0 30 22 b62 116 8002dda0 ? 0:01 uuxqt 1 S 5 10281 10280 0 30 22 bab 56 8002ddf0 ? 0:11 unbatch 1 S 9 10586 10346 13 20 20 fb4 203 8006ddc8 ttyp12 0:11 rnews 1 S 5 10528 10527 0 30 22 bba 90 8002de90 ? 0:01 sh 1 S 5 10526 1 0 30 22 dbf 116 8002dee0 ? 0:01 uuxqt 1 S 5 10529 10528 0 30 22 bc6 56 8002df30 ? 0:02 unbatch 1 S 5 10530 10529 0 26 22 4ff 866 8002a324 ? 0:05 compress 1 R 100 10588 10040 15 67 20 ad3 102 ttyp03 0:03 ps 1 S 103 4257 1 0 28 20 fd2 97 8003a784 ttyp10 0:12 sh Thankyou --------------------------------------------------------------------- Dave Wallace Bell Canada ( on week days ) Toronto, Ontario, Canada mnetor!genet!clunk!bvax!dave or utzoo!psddevl!hqtd!dave or (416) 599-1588 ____________________________________________________________________
jgd@pollux.UUCP (Dr. James George Dunham) (05/19/87)
In article <116@bvax.UUCP> dave@bvax.UUCP (Dave Wallace) writes: > >I recently joined the 'net but have a problem. I am running System V, >Release 2.2 on a VAX 730( No, that's NOT the problem). When I get my >'feed', uuxqt runs, forking a sh, who forks a sh, who forks a >compress and rnews,( plus the compress forks an unbatch). I can >get up to 6 of these running at once, which causes the 730 to slow down >just a bit (-: . I currently get my feed from one site and don't send >to anyone else ( until I get this problem resolved. ) It takes the 730 >most of a working day to process 6 or 7 batch files in /usr/spool/uucp! It >seems like these processes are fighing them selves. Any body have >any suggestions? We reduced the priority of our news jobs so they would not interfere with other users doing work on our system and have a similiar problem. It seems that the lock on uuxqt is a timed lock, and so if you have a uuxqt job running for a long time and someone else calls, they can start a new uuxqt job if the time on the lock is expired. We have seen several uuxqt jobs get started when we have long jobs running in the background. On our system the problem get worse since it appears there is a limit to the number of processes one user can run. When the limit is reached, then everything comes to a halt. I feel that the correct solution is to make the lock on uuxqt a pid lock rather than a timed lock. This way, only one copy of uuxqt can run at a time. If uuxqt is killed for some reason, then the pid will be wrong and a new one is allowed to start up. A quick fix is to lengthen the time of the lock which is X_LOCKTIME in uucp.h and when multiple jobs are found, kill all but the oldest one. I am hoping to get the lock changed to a pid lock this summer.
grr@cbmvax.cbm.UUCP (George Robbins) (05/22/87)
In article <311@pollux.UUCP> jgd@pollux.UUCP (Dr. James George Dunham) writes: > In article <116@bvax.UUCP> dave@bvax.UUCP (Dave Wallace) writes: > > > >I can > >get up to 6 of these running at once, which causes the 730 to slow down > >just a bit (-: . I currently get my feed from one site and don't send > >to anyone else ( until I get this problem resolved. ) It takes the 730 > >most of a working day to process 6 or 7 batch files in /usr/spool/uucp! It > >seems like these processes are fighing them selves. Any body have > >any suggestions? > > We reduced the priority of our news jobs so they would not interfere > with other users doing work on our system and have a similiar problem. One good solution is to create an rnews script that gets executed via uucp that simply copies the files into a temporary directory with a unique name. The you can have something running from crontab that each hour starts up and processes the queued news. Of course this script can test, either via ps or with a lock file, whether one is still running and abort... This kind of approach can be extended to also keep the incoming stream around for a couple of days to make for easy recovery. Another area to look at is how much time is being sucked up by compress. Compress used to get real hoggy if you used full 16-bit compression. While this may have been improved, you may find that changing your feed to 12-bit compression reduces the cpu/memory load at only a slight increase in your telephone costs. -- George Robbins - now working for, uucp: {ihnp4|seismo|rutgers}!cbmvax!grr but no way officially representing arpa: cbmvax!grr@seismo.css.GOV Commodore, Engineering Department fone: 215-431-9255 (only by moonlite)
rick@seismo.CSS.GOV (Rick Adams) (05/22/87)
One good solution is to create an rnews script that gets executed via uucp that simply copies the files into a temporary directory with a unique name. The you can have something running from crontab that each hour starts up and processes the queued news. Of course this script can test, either via ps or with a lock file, whether one is still running and abort... If you define SPOOLNEWS when you compile rnews, you will get this behaviour automatically. (Just remember to run rnews -U from crontab or you will wonder where your news is going!) ---rick
avolio@decuac.dec.com (Frederick M. Avolio) (05/22/87)
In article <116@bvax.UUCP> dave@bvax.UUCP (Dave Wallace) writes: > >I recently joined the 'net but have a problem. I am running System V, >Release 2.2 on a VAX 730( No, that's NOT the problem). ... >> ... takes the 730 >>most of a working day to process 6 or 7 batch files in /usr/spool/uucp! It >>seems like these processes are fighing them selves. Any body have >>any suggestions? 2.11 news allows you to accept news inm batched for and spirit it away for later processing. Why not set things up to unbatch news only during the night between certain hours (see inews regarding "inews -U"). Ask you news feed not to call you during that time. Fred
sl@van-bc.UUCP (Stuart Lynne) (05/23/87)
In article <311@pollux.UUCP> jgd@pollux.UUCP (Dr. James George Dunham) writes: >In article <116@bvax.UUCP> dave@bvax.UUCP (Dave Wallace) writes: >> >>I recently joined the 'net but have a problem. I am running System V, >>Release 2.2 on a VAX 730( No, that's NOT the problem). When I get my >>'feed', uuxqt runs, forking a sh, who forks a sh, who forks a >>compress and rnews,( plus the compress forks an unbatch). I can >>get up to 6 of these running at once, which causes the 730 to slow down >>just a bit (-: . I currently get my feed from one site and don't send >>to anyone else ( until I get this problem resolved. ) It takes the 730 >>most of a working day to process 6 or 7 batch files in /usr/spool/uucp! It >>seems like these processes are fighing them selves. Any body have >>any suggestions? Whats probably happening is the mulitple copies of compress fighting for table space. > > We reduced the priority of our news jobs so they would not interfere >with other users doing work on our system and have a similiar problem. >It seems that the lock on uuxqt is a timed lock, and so if you have a uuxqt >job running for a long time and someone else calls, they can start a new >uuxqt job if the time on the lock is expired. We have seen several uuxqt >jobs get started when we have long jobs running in the background. On our >system the problem get worse since it appears there is a limit to the number >of processes one user can run. When the limit is reached, then everything >comes to a halt. > I feel that the correct solution is to make the lock on uuxqt a pid >lock rather than a timed lock. This way, only one copy of uuxqt can run >at a time. If uuxqt is killed for some reason, then the pid will be wrong >and a new one is allowed to start up. A quick fix is to lengthen the time >of the lock which is X_LOCKTIME in uucp.h and when multiple jobs are found, >kill all but the oldest one. I am hoping to get the lock changed to a pid >lock this summer. The best (?) way around this problem seems to be to define SPOOLNEWS and re make your news system. rnews/inews will then simply copy the incoming files to /usr/spool/news/.rnews. Then once(twice) per hour run a script from crontab to run rnews -U. With a couple of tests you can figure out if rnews is already unbatching and exit if it is. Defining NICENESS does seem to be a big win for small systems. You stop noticing when news is being unbatched. The other problem with running rnews from uuxqt is that if it does lock up correctly so that no second copy of rnews starts up, your mail gets delayed until the news is unbatched. Which on some small systems can be hours. Spooling the news takes only a few seconds and then the system is ready to forward incoming mail. In general you should try avoid running jobs which take a long time to execute directly from uuxqt. Instead try and use uuxqt to spool the information and initiate the job as a separate task allowing uuxqt to exit or run the job from a cron entry. --------- Cut here for sample news hourly script ----------- # hourly script for news # # check for various lock files # initiate rnews -U to unspool news # ACTIVELOCK="active.lock" ACTIVEBATCH="active.unbatch" RNEWSLOCK="/usr/spool/news/.rnews.lock" cd /usr/local/lib/news if test -f $ACTIVELOCK then echo newshourly: $ACTIVELOCK exists exit 1 fi if test -f $ACTIVEBATCH then echo newshourly: $ACTIVEBATCH exists exit 1 fi if test -f $RNEWSLOCK then echo newshourly: $RNEWSLOCK exists exit 1 fi ln active $ACTIVEBATCH /usr/local/bin/rnews -U rm $ACTIVEBATCH # end of news hourly -- Stuart Lynne ihnp4!alberta!ubc-vision!van-bc!sl Vancouver,BC,604-937-7532
dave@lsuc.UUCP (05/25/87)
In article <43878@beno.seismo.CSS.GOV> rick@seismo.CSS.GOV (Rick Adams) writes: > > One good solution is to create an rnews script that gets > executed via uucp that simply copies the files into a temporary > directory with a unique name... > >If you define SPOOLNEWS when you compile rnews, you will get this >behaviour automatically. Almost, but not quite. With SPOOLNEWS (which requires news 2.11, incidentally), unbatching and installing of news will be postponed, but uncompression will be done right away, if your feeds are sending you files which are totally compressed (2.10.*-style). If you're on a system with limited memory, you may not want compress running during the day either. We use a "cunbatch" which is basically just a "cat". David Sherman The Law Society of Upper Canada -- { seismo!mnetor cbosgd!utgpu watmath decvax!utcsri ihnp4!utzoo } !lsuc!dave
scotty@l5comp.UUCP (06/10/87)
First, some further questions about uuxqt and the news. I get my news by calling in for it to a remote site. The remote site batches it and sets it up to be processed by cunbatch on my system. I have spooling turned on in my news stuff. I also have nice set. Uuxqt still runs for a good hour after the actual uucp operation finishes. I've also noticed that articles get posted onto my local system while uuxqt is running. I suspect that after the first file is decompressed news starts running and posting begins. This has to slow down further de-compressing. Which also of course lengthens the time that uuxqt is left active. Do I have some switch set wrong in my system? Second, there be time slot hoppers in the world. I have an assigned dial in time, but every so often someone jumps it. Last week my system didn't get through till 05:00 (time slot was at 02:00) one night and then didn't finish uucp till 06:30. However, I have an expire-daily that runs at 06:00. The system was still running the expire AND the rnews -U at 13:00. I assume this was a result of expire running at the same time as news was being munched down, BUT I don't know enough about the news system to say for sure. I had to kill -9 the rnews -U at 13:20 to shut everything down. The pacct files had run up to 2.6 megs so I figured something was fatally wrong. :) As a side issue, is it 'safe' to kill -9 an active expire? Was the expire-daily the problem? And if so what should I do to prevent this mess from happening again? Move expire-daily till MUCH further past the 02:00 news feed slot? Third, I'm sure everyone here has seen the 'rebel approved' posting to comp.sources.misc. Is there anything I can do to the news software on my system to prevent someone posting such a message using my system? I know trust is a great protection against alot of stuff, and I DO trust my regular users, but that posting seems to have been made from a 'guest' account (if that wasn't faked as well.) I'd rather build in protection than yank the guest account on my system. And to whomever made that posting: You should have signed it rather than skulking around like a criminal with a mask on. Your efforts are more likely to focus attention on locking the system up rather than on unmoderating newsgroups. When I read your posting there was an answering pang of rebellion in my heart, then I looked to see whom to converse with and my heart froze as cold as the ice in Siberia. Thanks in advance for any help, Scott Turner -- UUCP-stick: stride!l5comp!scotty | If you want to injure my goldfish just make UUCP-auto: scotty@l5comp.UUCP | sure I don't run up a vet bill. GEnie: JST | "The bombs drop in 5 minutes" R. Reagan Disclaimer? I own L5 Computing. Isn't that enough?
news@jpusa1.UUCP (06/13/87)
Summary:
Expires:
In article <287@l5comp.UUCP> scotty@l5comp.UUCP (Scott Turner) writes:
-First, some further questions about uuxqt and the news.
-I get my news by calling in for it to a remote site. The remote site batches it
-and sets it up to be processed by cunbatch on my system.
-I have spooling turned on in my news stuff. I also have nice set.
-Uuxqt still runs for a good hour after the actual uucp operation finishes.
You should turn on spooling to cut down on uucp time. What's going on is uuxqt
is firing up news with the associated compress/batch stuff and the machine is
bogging down. Much better to let uucp squirel away the batches to be processed
later. I use batching and if the uucp port is active, I don't start the unbatch.
This is handled in a script run hourly by cron.
-
-The system was still running the expire AND the rnews -U at 13:00.
News 2.11 handles the expire/rnews interlocking so this shouldn't be a problem.
At the end of expire, an 'rnews -U' is started up so it's normal to see this.
-As a side issue, is it 'safe' to kill -9 an active expire?
You'll probably trash your history file if you do. Use an 'expire -r' to fix it.
-Was the expire-daily the problem? And if so what should I do to prevent
-this mess from happening again? Move expire-daily till MUCH further past
-the 02:00 news feed slot?
Separating the uuxqt from news via spooling is what's needed. I also don't
start up expire if news or the uucp port are active. The expire.sh script
checks for this and sleeps 5 min till things have quieted down. This is more
to keep the load down than rnews/expire paranoia as the interlocking seems to
work now.
-Third, I'm sure everyone here has seen the 'rebel approved' posting to
-comp.sources.misc. Is there anything I can do to the news software on my
-system to prevent someone posting such a message using my system? I know
-trust is a great protection against alot of stuff, and I DO trust my
-regular users, but that posting seems to have been made from a 'guest'
-account (if that wasn't faked as well.) I'd rather build in protection than
-yank the guest account on my system.
You can hack Pnews but anyone who knows how can post an article even if you have
no news connections or news software.
Stu Heiss {gargoyle,ihnp4}!jpusa1!stu
jerry@oliveb.UUCP (Jerry F Aguirre) (06/15/87)
Even with "spooling" turned on you may still have lengthy overhead at uuxqt time. If the transmitting site is sending you news in "cunbatch" format rather than "#!cunbatch" format then the "uncompress", which is a large part of the overhead, will happen at uuxqt time. The rnews invoked will save the article for later processing but the uncompress overhead will slow things down. If the timeout on the uuxqt lock file expires then you will get another "cunbatch" will happen with another uncompress running in parallel. Older versions of news (2.10) execute the "cunbatch" command instead of "rnews". Newer versions (2.11) will use this format if the "-o" option of "sendbatch" is used. Frequently this is used when the new site is feeding a site requiring the old version. The preferable sollution is for the site feeding you to switch to the new format. (Remove the "-o" option to "sendbatch".). If they are queueing a single batch for multiple sites and some of them are old versions then this may be a problem. If your feed is itself running an old version then urge them to upgrade. The second alternative is to replace "cunbatch" with something that just saves the "cunbatch" for later processing. Several shell scripts have been posted in the past. I have some C programs that I use here but they are somewhat BSD dependant. The uncompression, unbatching, and rnews processing is done only when the load is below a specified level. As far as the rnews/expire timing problem goes, wouldn't it make more sense to run expire BEFORE polling for new news? If you ran expire at midnight, and then polled at 2AM then you would have less chance of an overlap. After all, you control the expire, you can only make a try for the uucp. This also makes more sense if you think of the expire clearing out disk space for the news that is about to arrive.