From: "Darrick J. Wong" <darrick.wong@oracle.com>
To: Markus Stockhausen <stockhausen@collogia.de>
Cc: "'linux-xfs@vger.kernel.org'" <linux-xfs@vger.kernel.org>
Subject: Re: XFS hang - 4.4.73 longterm
Date: Mon, 10 Jul 2017 09:51:41 -0700 [thread overview]
Message-ID: <20170710165141.GK4103@magnolia> (raw)
In-Reply-To: <12EF8D94C6F8734FB2FF37B9FBEDD173010E0284F8@EXCHANGE.collogia.de>
On Sun, Jul 09, 2017 at 05:01:25PM +0000, Markus Stockhausen wrote:
> > -----Ursprüngliche Nachricht-----
> > Von: Darrick J. Wong [mailto:darrick.wong@oracle.com]
> > Gesendet: Freitag, 7. Juli 2017 05:47
> > An: Markus Stockhausen
> > Cc: 'linux-xfs@vger.kernel.org'
> > Betreff: Re: XFS hang - 4.4.73 longterm
> >
> > > > ...
> > > > This is the known deadlock in SEEK_HOLE/SEEK_DATA; I have patches queued to fix it in 4.13, as soon as the dust settles and I send the pull req.
> > >
> > > Short, precise, frightening.
> > >
> > > Can you advise what will the best option to avoid that error.
> > > First things that come to my mind would be:
> > >
> > > - get back to original 3.10 stable kernel from CentOS Distro
> > > - lower NFS mount version
> > > - Maybe remove some single patch that introduced the error?
> >
> > Unfortunately, SEEK_HOLE (and the deadlocky code) go all the way back to the beginning of SEEK_HOLE support, so I don't know that
> > there's a good workaround for this. There's not really a way to shut off support for it, either, that doesn't involve gross code surgery.
>
> Not good news. Finally I disabled SEEK support in nfsd by partially
> reverting 3.18 patch 24bab491220faa446d94 "NFSD: Implement SEEK".
> While before NFS/XFS crashed with active VM snapshots and heavy disk
> writes after minutes it now runs for 2 days of intensive testing
> without errors.
>
> At least now I have full discard support in side VMs (via hole
> punching of disk files).
I'd advise you to set an extent size hint to reduce fragmentation, the
same as we tell everyone else. :)
I'd also advise you not to punch too many holes in the VM files because
that increases extent map overhead. There's also an unfortunate design
problem in the in-memory extent map -- all the extents have to fit in a
single chunk of memory -- and punching a lot of small holes when memory
is tight can lead to allocation failures.
(Even after we fix that problem I'd still advise against punching a lot
of small holes purely due to the overhead that creates.)
> I'm not familiar with SEEK_HOLE/SEEK_data feature. Do you see negative
> impact if disabling these? In contrast to falling back to NFS 4
> without seek + discard support.
Well, whatever program was using SEEK_{HOLE,DATA} to skip sparse holes
in the files will likely revert to a much slower "read the whole file
looking for block aligned zeroes" algorithm... assuming that was why it
was doing that in the first place.
FWIW I just sent off the 4.13 merge request with the SEEK_HOLE fixes,
though they depend on iomap, which (iirc) wasn't added to the vfs until
4.8 or so.
--D
>
> Markus
> ****************************************************************************
> Diese E-Mail enth??lt vertrauliche und/oder rechtlich gesch??tzte
> Informationen. Wenn Sie nicht der richtige Adressat sind oder diese E-Mail
> irrt??mlich erhalten haben, informieren Sie bitte sofort den Absender und
> vernichten Sie diese Mail. Das unerlaubte Kopieren sowie die unbefugte
> Weitergabe dieser Mail ist nicht gestattet.
>
> ??ber das Internet versandte E-Mails k??nnen unter fremden Namen erstellt oder
> manipuliert werden. Deshalb ist diese als E-Mail verschickte Nachricht keine
> rechtsverbindliche Willenserkl??rung.
>
> Collogia
> Unternehmensberatung AG
> Ubierring 11
> D-50678 K??ln
>
> Vorstand:
> Kadir Akin
> Dr. Michael H??hnerbach
>
> Vorsitzender des Aufsichtsrates:
> Hans Kristian Langva
>
> Registergericht: Amtsgericht K??ln
> Registernummer: HRB 52 497
>
> This e-mail may contain confidential and/or privileged information. If you
> are not the intended recipient (or have received this e-mail in error)
> please notify the sender immediately and destroy this e-mail. Any
> unauthorized copying, disclosure or distribution of the material in this
> e-mail is strictly forbidden.
>
> e-mails sent over the internet may have been written under a wrong name or
> been manipulated. That is why this message sent as an e-mail is not a
> legally binding declaration of intention.
>
> Collogia
> Unternehmensberatung AG
> Ubierring 11
> D-50678 K??ln
>
> executive board:
> Kadir Akin
> Dr. Michael H??hnerbach
>
> President of the supervisory board:
> Hans Kristian Langva
>
> Registry office: district court Cologne
> Register number: HRB 52 497
>
> ****************************************************************************
next prev parent reply other threads:[~2017-07-10 16:51 UTC|newest]
Thread overview: 7+ messages / expand[flat|nested] mbox.gz Atom feed top
2017-07-05 19:19 XFS hang - 4.4.73 longterm Markus Stockhausen
2017-07-06 0:24 ` Darrick J. Wong
2017-07-06 4:45 ` AW: " Markus Stockhausen
2017-07-07 3:47 ` Darrick J. Wong
2017-07-09 17:01 ` AW: " Markus Stockhausen
2017-07-10 16:51 ` Darrick J. Wong [this message]
2017-07-10 18:10 ` Markus Stockhausen
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20170710165141.GK4103@magnolia \
--to=darrick.wong@oracle.com \
--cc=linux-xfs@vger.kernel.org \
--cc=stockhausen@collogia.de \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).