[Bug 290207] [ZFS] lowering "vfs.zfs.arc.max" to a low value causes kernel threads of "arc_evict" to use 91% CPU and disks to wait. System gets unresponsive...

From: <bugzilla-noreply_at_freebsd.org>
Date: Sun, 02 Nov 2025 17:42:53 UTC
https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=290207

--- Comment #29 from Mark Millard <marklmi26-fbsd@yahoo.com> ---
(In reply to Nils Beyer from comment #25)

One of the problems with only presenting the sorted
by size*used is that there are lines with duplicate
labels and "vm pgcache:" happens to be an example
of such (from your before text):

vm pgcache:            4096,      0,   15841,     . . .,   63364
vm pgcache:            4096,      0,   41224,     . . .,  164896

There really is insufficient context for indication
of which vm pgcache each is. The detailed order of
the output matters.

It turns out that they are back-to-back lines
(a live system here):

# vmstat -z | grep -C3 '^vm pgcache:'
buffer arena-61:      61440,      0,       0,       0,       0,   0,   0,   0
buffer arena-65:      65536,      0,       0,       0,       0,   0,   0,   0
buf free cache:         432,      0,       0,    2818, 2880972,   0,   0,   0
vm pgcache:            4096,      0,   52616,    9322,15441768,   0,   0,   0
vm pgcache:            4096,      0, 5590381,   14523,13978355,   0,   0,   0
UMA Kegs:               384,      0,     162,       1,     162,   0,   0,   0
UMA Zones:             4608,      0,     188,       2,     188,   0,   0,   0
UMA Slabs 0:             80,      0, 3986534, 1631910, 7345505,   0,   0,   0

So not sorting but greping for the 2 lines is
sufficient to identify which is which:

# vmstat -z | grep ^vm\ pgcache:
vm pgcache:            4096,      0,   52616,    9322,15442016,   0,   0,   0
vm pgcache:            4096,      0, 5590377,   14527,13978594,   0,   0,   0

(It may be that the used count is a good hint of itself.
But I cannot tell if there are guarantees about such
relative sizes, not knowing the distinction between the
2 lines.)


zfs_btree_leaf_cache does not seem to have this issue:
there is only one such line . . .

# vmstat -z | grep ^zfs_btree_leaf_cache:
zfs_btree_leaf_cache:  4096,      0,     727,    2015, 1588501,   0,   0,   0

-- 
You are receiving this mail because:
You are the assignee for the bug.