Ted Unangst <t...@tedunangst.com> wrote: > Todd C. Miller wrote: > > On Mon, 07 Jan 2019 22:55:54 +0000, Christian Weisgerber wrote: > > > > > Oooh, that short period can be the time between disk flushes (30s?) > > > with softupdates. I think that's one of the scenarios where you > > > can run out of disk space if you have softupdates enabled. > > > > Yes, I think I've had that happen in the (now rather distant) past. > > Has anybody seen this recently? Many years ago I had a diff to fix it. Lost > the diff, but I remember it's not super complicated. I figured the problem > went away (due to less tight partitioning) and haven't seen it myself. If > people are still having to workaround it, I can cook something up.
Wait. Stop. Some architectures triggered this "delayed release of softdep buffers" worse. In order of danger: macppc, hppa, sgi, ... To me, it seems to be related to "running servicing functions". I don't think this needs a workaround. I despise the word. The proper bug needs to be found, and fixed. Not some workaround installed which seems to fix the issue on machines people don't have, and golly gee it still remains or surfaces in some other way.. and then OH WE KNOW THE REAL PROBLEM, here is the real fix, and years later "someone forgot their broken workaround still remains in the tree causing other effects". Don't "workaround it". Workaround it by fixing it. I think macppc and hppa are teaching a lesson. deep rm -rf on hppa is super slow. On macppc, softdep is very slow at releasing buffers. Where are the MD mistakes to cause this?
