[3481] in SIPB bug reports
No subject found in mail header
daemon@ATHENA.MIT.EDU (Jonathan I. Kamens)
Mon Feb 1 15:46:14 1993
Date: Mon, 1 Feb 93 15:46:12 -0500
From: "Jonathan I. Kamens" <jik@Aktis.COM>
To: usenet@Athena.MIT.EDU, bug-sipb@Athena.MIT.EDU
Cc: ellis@Athena.MIT.EDU, bjaspan@Athena.MIT.EDU
It turns out that Barry and Steve were right -- something *was*
something done incorrectly during the transition over to
senator-bedfellow that caused xrn to run out of memory.
In particular, I noticed today that my xrn window claimed that there
were over 11,000 old articles (i.e., articles that I'd already read)
in alt.config.
I figured that this must be bogus, so I logged into senator-bedfellow
and did a little bit of digging around.
It appears that at some point during the transition, senator-bedfellow
was putting new articles into the newsgroups but using its original
article numbering, rather than the article numbering that was
transferred over from bloom-picayune.
I can speculate on how this might have occurred, but it's only
speculation. Perhaps people forgot to throttle the server while doing
the transition. Or, perhaps they forgot to tell the server to reread
the active file after copying over the new active file from
bloom-picayune, which means that until someone did a `ctlinnd renumber
""' command (which happens automatically during the nightly expire),
the server numbered articles using senator-bedfellow's original
numbering rather than bloom-picayune's.
In any case, I have fixed this by doing two things:
1) Running a perl script over the active file and news spool to find
any newsgroups with a gap of more than 1,000 articles between two
articles in the newsgroup. For all newsgroups for which such a gap
was found, all articles before the gap were deleted (i.e., given a
".#" prefix). If there were multiple gaps, the latest one was used.
The perl script I used is in /mit/jik/tmp/check-gaps.pl
2) Running `ctlinnd renumber ""' to renumber the active file so that
the deletions I did would be noticed.
Note that I did the first few newsgroups by hand rather than using the
script, so there are some newsgroups where old articles with gaps
smaller than 1,000 were still removed. Note, further, that I
accidentally used rm rather than delete on about five articles (total)
in two or three different newsgroups, but I really don't feel
compelled to worry about them :-).
I deleted the files, rather than rm'ing them, so that if someone
decides that my solution to the problem was the wrong one, it can be
backed out (by undeleting the files and again renumbering the active
file). Another solution I could have used is to throttle the server,
rename all of the incorrectly numbered articles with new, correct
numbers, renumber the active file, and then untrottle the server.
However, this would have required a history file rebuild, because the
renamed files could no longer be found in order to expire them, and I
didn't want to bother with a history file rebuild.
jik
