[2848] in SIPB bug reports
gdbm lossage in BITNET gateway software
daemon@ATHENA.MIT.EDU (mhpower@Athena.MIT.EDU)
Tue May 19 13:07:15 1992
From: mhpower@Athena.MIT.EDU
To: bug-sipb@Athena.MIT.EDU
Date: Tue, 19 May 92 13:06:35 EDT
On (so far rare) occasions, the Zephyr-to-BITNET gateway software
branches into a wrong section of the code. It's not a repeatable
problem. It always seems to occur at places where the code is
conditional on a dbm fetch (it uses gdbm 1.5). An exit/restart fixes
the problem, at least for a while. I thought I'd report this here so
that other people will have heard about it, if a user reports some
anomalous behavior.
Also, maybe someone who reads this has a clue about gdbm. Here's the
situation: there are two system daemons that both open a gdbm data
file read-only when they start up, and both keep it open forever.
There are about 100 entries in the data file. I haven't really tracked
the fetch rate, but I suspect it may be something like 10 per second
at the absolute peaks.
I'm not sure where the problem might be. I should look further at the
bug-gnu-utils discuss archive, but it doesn't seem like there are many
gdbm reports there. Maybe ask mib, too. Also, the distribution on prep
hasn't been updated in over a year. (The local working copy is in
/afs/sipb/project/bitnet/gdbm-1.5/src, which should be world
readable.)
A possible workaround might be to try giving each process its own
private copy of the data file. Another is, of course, to dump gdbm
entirely and rewrite the code to use something else.
Matt
