Re: [6.2 regression][bisected]discard storm on idle since v6.1-rc8-59-g63a7cb130718 discard=async

[Chris Mason <clm@meta.com>]

On 4/5/23 2:18 AM, Christoph Hellwig wrote:
> On Tue, Apr 04, 2023 at 02:15:38PM -0400, Chris Mason wrote:
>> It seems like a good time to talk through a variations of discard usage
>> in fb data centers.  We run a pretty wide variety of hardware from
>> consumer grade ssds to enterprise ssds, and we've run these on
>> ext4/btrfs/xfs.
> 
> Remember what you guys call consumer grade is the top of the shelve
> consumer grade SSDs, or in fact low-end enterprise ones that are based
> on the latest and greatest top shelve SSDs with firmware specifically
> tweaked for you.  That is a whole differnet level from random crappy
> no-name SSD from 5 years ago.
> 

This is definitely true, our providers are reliable and support the
products well.  But, you also have to factor in the write cycles, drive
lifetimes, and just the constant threat of statistics at scale.  Drives
might come in off something close to the top shelf, but datacenters are
pretty hard on storage.

Going back to the fedora thread, we've definitely had ssds with
reliability issues caused by trims, but we haven't had any I can think
of where mount -o discard was fundamentally less reliable than periodic
full device trimming.

-chris