Re: kern/54727: writing a large file causes unreasonable system behaviour

["J. Hannken-Illjes" <hannken%eis.cs.tu-bs.de@localhost>]

The following reply was made to PR kern/54727; it has been noted by GNATS.

From: "J. Hannken-Illjes" <hannken%eis.cs.tu-bs.de@localhost>
To: gnats-bugs%netbsd.org@localhost
Cc: David Holland <dholland-bugs%netbsd.org@localhost>
Subject: Re: kern/54727: writing a large file causes unreasonable system
 behaviour
Date: Mon, 9 Dec 2019 12:21:19 +0100

 > On 6. Dec 2019, at 04:45, David Holland <dholland-bugs%netbsd.org@localhost> wrote:
 > 
 > The following reply was made to PR kern/54727; it has been noted by GNATS.
 > 
 > From: David Holland <dholland-bugs%netbsd.org@localhost>
 > To: gnats-bugs%netbsd.org@localhost
 > Cc: 
 > Subject: Re: kern/54727: writing a large file causes unreasonable system
 > behaviour
 > Date: Fri, 6 Dec 2019 03:42:06 +0000
 > 
 > On Sun, Dec 01, 2019 at 07:00:00AM +0000, spz%NetBSD.org@localhost wrote:
 >> 	running "ssh othersystem dobackup > /storage/file" generating
 >>        a large file while otherwise idle, causes the system to hang
 >>        (no response in 5 minutes on serial console, ddb ps shows
 >>        daemons memory starved or missing) or with 
 >>        "sysctl -w kern.maxvnodes=10000", pgdaemon spinning several minutes
 >>        before any new program can start.
 >>        "vmstat 1" shows:
 >> [...]
 > 
 > two notes:
 > 
 > (1) mlelstv mentioned in chat the other day that all file pages were
 > active (rather than inactive) and this seems broken;
 
 This happens for me too.  Looks like:
 
 - we are low on memory.
 - nearly all pages are active and belong to ONE vnode (the large file
   we are currently creating).
 - pagedaemon ends up in uvmpdpol_balancequeue() to increase the number
   of inactive pages.
 - often the one vnode v_interlock is held by another thread
   so uvmpd_trylockowner(p) fails for nearly all active pages.
 
 - the pagedaemon starts busy looping until it finds this vnode
   unlocked and everything proceeds.
 
 > 
 > (2) while chasing something else related today a number of people
 > found that all their vnodes were active (rather than inactive),
 > probably for the same reason, and this also seems broken.
 
 If this is with "pstat -v" there is a small error in the kernel,
 the usecount must be decremented by one.
 
 > also this problem probably should be filed as a release stopper :-(
 
 Not for -9, I had this problem in -8 too.
 
 --
 J. Hannken-Illjes - hannken%eis.cs.tu-bs.de@localhost - TU Braunschweig (Germany)