Re: June 2025 stabilization week

  Hi,

TLDR: advisory freeze thawed.  Some details below.

On Tue, Jun 24, 2025 at 10:40:33AM -0700, Gleb Smirnoff wrote:
T> Status update for Tuesday:
T> 
T> - Personal desktop with nVidia graphics - all OK.
T> - Home router/server - all OK.
T> - Laptop with Intel graphics - drm61 kmod is broken due to disappeared
T>   vm_page_lock_KBI().

The drm kmods just need to be built manually from ports with this update,
since FreeBSD-kmod repo doesn't yet have them built.

T> At Netflix the A/B testing against previous stabweek didn't discover any
T> stability issues, however there is quite a lot of performance churn (nothing
T> critical though!) that we are now interpreting.

We observed the following three differencies between May stabweek and today:

1) Amount of wired memory increased significantly.  Our average across many
   different boxes is +9 Gb.  We have hypothesis that this is related to
   the ZFS merge and we are testing that now.

2) SSD transaction time slightly increased.

3) Number of sendfile() calls that hit a busy page increased.  Number of
   previously cached pages encountered by sendfile() increased.  Number of
   overall sendfile calls decreased for the same amount of output.  Right
   now we speculatively attribute 3) as a side effect of 2).

T> There regression with devfs rules, reported by Michael in this thread.

This doesn't seem to be reproducible always, so considering it a less
serios issue. Investigation continues.

Sounds like a bunch of problems, but none are critical, so declaring
advisory freeze to be over.

-- 
Gleb Smirnoff