reid@Glacier.ARPA (Brian Reid) (08/11/85)
John Quarterman kindly posted "uuhosts" recently, and like most of the SA's I know, I installed it. Since then I have experienced two severe total lockups of the entire news/mail system, both of which were traceable to uuhosts; furthermore, by killing uuhosts I was able to unlock things. I think that the basic problem is that when you put a line like this in your sys file: maps:mod.map.uucp:B:uuhosts -x that the "uuhosts" program is run in a fork hierarchy approximately like this (I haven't checked to see if inews forks uuhosts directly or if some other agent is involved; the principle is the same regardless): uucico -- uuxqt -- rnews -- inews -- uuhosts The kicker is that uhosts can take a couple of hours to run on a heavily loaded system. In particular, for the recent distribution of eur.*, uuhosts took about 4 wall clock hours to run. This means that while uuhosts was running for those 4 hours, inews was waiting for it to finish. While inews was waiting for it to finish, the news system was locked up. While uuxqt was waiting for rnews to wait for inews to finish, the LCK.XQT file was set and no other uucp traffic was moving. For some reason that I still can't quite figure out, this uucp-world jamup managed to propagate over to the sendmail--/usr/spool/mqueue world, and lock THAT up for 4 hours. After 4 hours of being locked up, some process abandoned ship somewhere, and I had to manually restart sendmail to get the mail going again (I have seen this failure in sendmail -q1h before; running sendmail -q from crontab once per hour works better). What I recommend that people do is to copy Rick Adams' "turning off rnews" hack, and to capture the maps by calling a shell script that saves its standard input in /usr/spool/news/uucpmaps.$$, where $$ is the pid, and then have some crontab-driven thing that comes around later and runs uuhosts -x over all of those files. I just installed this on Glacier, and it's too soon to report success, but it has to work better than running it directly from the sys file. -- Brian Reid decwrl!glacier!reid Stanford reid@SU-Glacier.ARPA
jsq@im4u.UUCP (08/12/85)
Interesting. We haven't seen that problem here, but we run uuhosts
from a sys file on a 780 with fast disks.
The idea of having the sys file just collect the articles
for later extraction by uuhosts run from cron sounds good to me.
That would also allow avoiding indexing twice for two articles which
came in close together. I will put something together with some
other changes I've made and make another distribution soon.
--
John Quarterman, UUCP: {ihnp4,seismo,harvard,gatech}!ut-sally!jsq
ARPA Internet and CSNET: jsq@ut-sally.ARPA, soon to be jsq@sally.UTEXAS.EDU