reid@Glacier.ARPA (Brian Reid) (09/06/85)
We had some problems this weekend when our daily, weekly, and monthly netnews maintenance runs were all trying to run at the same time; one of the side effects was that we then ran for 4 days with rnews turned off and the incoming news accumulating in /usr/spool/news/rnews.$$. This evening I manually ran all of the accumulated news, which means that I got a rare chance to watch half a week's worth of netnews go by with stderr connected to my terminal. There were about 1000 messages piled up, and 31 of them were thrown away before my very eyes, with a message saying "out of sync, skipping xxx", where "xxx" was each line of a discarded message. This means that 1.5 percent of the messages were lost, and in every case, the lost messages were preceded in the batch by a zero-length control message. I remember Chuqui's original posting, then several challenging him saying it wasn't happening, then Chuqui re-asserting that it was happening. One thing that I don't remember is whether or not anybody ever posted a specific fix to unbatch. In other words, do I have to go into unbatch and fix this or has somebody posted the fix only to have the fix gobbled upstream of me by the very bug it was supposed to correct. -- Brian Reid decwrl!glacier!reid Stanford reid@SU-Glacier.ARPA
kre@munnari.OZ (Robert Elz) (09/07/85)
When I first saw Chuq's report of this "bug" I was a little shocked, then I looked at the code, and came to the conclusion that it couldn't happen (which someone else did also & reported to the net). Now Brian has had the same thing happen to him. This gets beyond mere co-incidence, thinks I, lets have a go at forcing this bug to happen. But no luck - I have tried the batching code from 2.10.1, 2.10.2, and 2.10.3 - they all produce identical batches. I have tried unbatching with unbatch from all of those versions, again, all I get is a lot of "duplicate article ignored" messages (because I keep feeding rnews the same old set of articles over & over). I even looked at mcvax's batching code that I know some people use, and couldn't find anything wrong with that either. (I know that at least Chuq's problem wasn't with that one, because he mentioned the #! rnews line, which doesn't exist with that version). So - just what is happening here? Chuq, Brian, which versions of news are you running, what are the id's in your unbatch.c files, and even more important, tell us about your news feed. Where did the bad batches come from? What version of the news code, and what version of batch.c. (Are you possibly both getting news from some site that has broken the batching code locally?) Is the news being just batched, or batched & compressed as well? (ie: is the bug possibly in compress, rather than batch/unbatch). If compress (or anything similar) is being used, which versions at each end? Can you send (mail) me a small piece of a batch that is bad. Say from the #! rnews that precedes the 0 length control article, down to and including the header of the next article. Its important that this problem be located, if only to set our minds at rest that it is something local to the nsc / glacier / ... area. Robert Elz seismo!munnari!kre