Hi, Toby
On 19 May 2021, at 15:24, Toby Darling <toby(a)mrc-lmb.cam.ac.uk> wrote:
In the last couple of weeks we've been getting BlueFS spillover warnings on multiple
(>10) osds, eg
BLUEFS_SPILLOVER BlueFS spillover detected on 1 OSD(s)
osd.327 spilled over 58 MiB metadata from 'db' device (30 GiB used of 66 GiB)
to slow device
I know this can be corrected with a 'ceph tell osd.$osd compact' or ignored with
"bluestore_warn_on_bluefs_spillover=false", but my concern is that these
warnings have only recently started.
Could this be a sign of something nasty heading our way that I'm not aware of? Is
there a performance penalty by just ignoring, rather than compacting?
Many thanks for any pointers.
It's just enough to upgrade to Nautilus at least 14.2.19, where Igor developed new
bluestore levels policy (bluestore_volume_selection_policy) in value
'use_some_extra' - any BlueFS spillover should be mitigated!
Cheers,
k