Subject: Re: that VM/fs cache coherency problem
To: None <email@example.com>
From: Niklas Hallqvist <firstname.lastname@example.org>
Date: 08/01/1995 09:02:42
>>>>> "Allen" == Allen Briggs <email@example.com> writes:
>> tail -f on logfiles is coredumping. (The problem seems to be
>> related to tail's tendency to read the file via mmap().)
Allen> Is this over NFS? I do this all the time on my mac68k box with
Allen> all local filesystems and have never seen tail -f dump or fail
Allen> at all.
Allen> The way I usually build the world is to nohup and background
Allen> the make, then tail -f the log file sometimes from multiple
Allen> terminals at once...
Actually, yesterday I had tail consistently fail for me exactly like
in "der mouse"'s case, on a UFS partition. I cannot remember off-hand
if I've seen it before. Perhaps the existing heuristics (or whatever
they are) for cache-coherency (or lack of) is usually good, like in
99% of the cases. This was on a 950226 -current kernel, but I don't
think anything has changed wrt VM/buffer cache, since then. Hmm,
perhaps it all depends on the paging activity, it ought to have been
low for me yesterday..., so old pages would've been kept longer...