Subject: livelock with FFS/Soft Updates
To: <>
From: Ethan Solomita <>
List: current-users
Date: 03/30/2000 15:12:14
	I thought I'd let people know about a Soft Updates bug where the fix is
in the pipeline, but not done yet. The symptoms are the same as a prior
bug which was fixed -- the system is continually pushing data to the
disk, but never finishes and it will appear to be permanently locked.

	Hopefully the code will be in soon, but it's on a queue of several
changes that are going in. In case anyone is interested, it happens when
BQ_AGE contains only buffers with dependencies on buffers in BQ_LRU. The
OS keeps trying to push out BQ_AGE buffers before BQ_LRU, but they keep
getting put back on the dirty lists because of their dependencies on
buffers in BQ_AGE.

	My understanding is that this bug is unusual, but reliable. ie. most
people won't see it, but if you do see it then you'll see it often. I
haven't seen anyone report it here yet, but if someone has it you can
send me email and I'll send you a temporary fix until things get into
the tree.
	-- Ethan