linux-kernel - Re: bcache vs bcachefs

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <20160907155552.4fqndafa7vepattr@kmo-pixel>
Date:   Wed, 7 Sep 2016 07:55:52 -0800
From:   Kent Overstreet <kent.overstreet@...il.com>
To:     "Austin S. Hemmelgarn" <ahferroin7@...il.com>
Cc:     Harald Dunkel <harri@...ics.de>, linux-kernel@...r.kernel.org
Subject: Re: bcache vs bcachefs

On Wed, Sep 07, 2016 at 10:34:15AM -0400, Austin S. Hemmelgarn wrote:
> On 2016-09-06 20:55, Kent Overstreet wrote:
> > On Tue, Sep 06, 2016 at 11:46:28AM +0200, Harald Dunkel wrote:
> > > Hi folks,
> > > 
> > > I am pretty hesitant replacing the rock-solid ext4 by bcachefs on my servers.
> > > Meaning no offense, but surely I would prefer to have ext4 with a thin "SSD
> > > caching layer" over a completely different filesystem, potentially with alot
> > > of teething troubles.
> > > 
> > > Question: Is bcache EOL or can I rely on it for the next 5 to 10 years?
> > 
> > bcache is not EOL - it's still receiving bugfixes.
> > 
> > That said though, there's no reason to expect a long teething period with
> > bcachefs, it's already more reliable than btrfs in single device mode.
> > 
> I'd be curious to see any actual data you have to back that up, especially
> regarding what kernel and userspace were involved with the BTRFS testing.

I'm not a btrfs user, I'm mostly going off feedback from bcachefs users (many or
most of whom are or have been btrfs users too).

On the bcachefs side, here's what we've seen so far:

 - No one has lost data or seen any corruption when using the currently
   supported feature set (multiple devices and caching are not stable just yet).

   I believe the worst bug anyone's hit was 0 length symlinks after unclean
   shutdown, and also 0 length files because we hadn't implemented the
   workarounds for applications that don't fsync yet.

 - I _believe_ no one's seen any crashes/deadlocks/hangs in supported
   configurations - there was a report of a deadlock, but I believe he was using
   tiering and it hasn't reproduced since.

I haven't asked for details about the btrfs issues the bcachefs users have seen
(one guy mentioned his 20 core machine panics running btrfs as the reason he was
considering switching it to bcachefs).

Smaller sample size, unscientific comparisons, etc. etc. but it's certainly
looking more robust so far.

That said, I'm not advocating people rush out to throw bcachefs on their servers
or use it without backups yet, it's still young and needs more widespread
testing.