Skip site navigation (1)Skip section navigation (2)
Date:      Thu, 5 Sep 2013 11:16:42 -0400
From:      Zaphod Beeblebrox <zbeeble@gmail.com>
To:        =?ISO-8859-2?Q?Edward_Tomasz_Napiera=B3a?= <trasz@freebsd.org>
Cc:        FreeBSD Stable <freebsd-stable@freebsd.org>, Ian Lepore <ian@freebsd.org>
Subject:   Re: gmirror crash writing to disk? Or is it su+j crash?
Message-ID:  <CACpH0MfrO-0m1PN9w3LpTUO6cN4OkZGuZ8HmhijjeG12cxZR-g@mail.gmail.com>
In-Reply-To: <CACpH0Mc5GEg4EJfMzpg3c0H7BQti%2BV6Gu_E9q7uZCVwEo-OvOw@mail.gmail.com>
References:  <CACpH0MdU5NMi=GvYbG5W-F-UkH=s89d8Docvz8_KS0%2BfqNH1cQ@mail.gmail.com> <370A25C8-7747-4B96-A506-EB92FD0F77CF@FreeBSD.org> <1377895898.1111.341.camel@revolution.hippie.lan> <CACpH0MeOG9v422AdpGpbos6mTLFu69PNXmP3ZFTV66Vc=H9Eew@mail.gmail.com> <CACpH0Mc%2Begqb7V9=bJ73VpzLvc5K0cTRN6fQZOkBwPD_TetRDg@mail.gmail.com> <A8B0CAB6-6646-4AA0-A03C-412C2438716D@freebsd.org> <CACpH0MdQPGB-psLj-xyALtrid6P2ByXH9sZsDxuhT1owdtg4zw@mail.gmail.com> <CACpH0Mc5GEg4EJfMzpg3c0H7BQti%2BV6Gu_E9q7uZCVwEo-OvOw@mail.gmail.com>

next in thread | previous in thread | raw e-mail | index | archive | help
Replying to myself again, I again doubled the bio_transient_maxcnt:
original value 160, failed doubling 360, new value 720; and the machine was
able to successfully "for i in jot 10; do make -j4 buildkernel; done" ...

But doesn't this mean that we still have a resource exhaustion to worry
about?  Isn't this just another race waiting for the the right set of
conditions?


On Tue, Sep 3, 2013 at 11:06 AM, Zaphod Beeblebrox <zbeeble@gmail.com>wrote=
:

> Since there weren't any more ideas here, I tried turning off
> hyper-threading.  This is an old pentium-D type CPU --- that is: one core
> with HT.  I'm wondering if the HT nature is helping this resource
> exhaustion, so I turned off HT (basically making this a single-threaded
> CPU) and it seems to have made the problem go away.
>
> That is not to say that the problem is fixed: it simply means that
> replication may be tied to multiple CPUs and/or the allocation of resourc=
es
> by an HT CPU core.
>
>
> On Mon, Sep 2, 2013 at 3:53 AM, Zaphod Beeblebrox <zbeeble@gmail.com>wrot=
e:
>
>> The first one (kern.geom.transient_map_retries) causes the system to
>> wedge.
>>
>> The second one (default is 180, I doubled to 360) causes the system to
>> crash but not dump.
>>
>> So... neither fixes the problem.
>>
>>
>> On Sat, Aug 31, 2013 at 5:27 AM, Edward Tomasz Napiera=B3a <
>> trasz@freebsd.org> wrote:
>>
>>> Wiadomo=B6=E6 napisana przez Zaphod Beeblebrox <zbeeble@gmail.com> w dn=
iu
>>> 31 sie 2013, o godz. 00:49:
>>> > Because someone said that there would be no logging of unerlying ATA
>>> errors without verbose, I rebooted with verbose and tried the same make=
 -j4
>>> again... and here is the relatively similar core.txt.5
>>> >
>>> >
>>> https://uk.eicat.ca/owncloud/public.php?service=3Dfiles&t=3Dd99648ef587=
6b91c5957148445e60c87
>>> >
>>> > Looking at it, gmirror is dropping the same error and the underlying
>>> hardware is not causing the error...
>>>
>>> Let me quote Konstantin:
>>>
>>> > It is either an exhaustion of the transient map, or a deadlock.
>>> > For the first, setting kern.geom.transient_map_retries to 0 could hel=
p.
>>> > For the second, the count of the transient buffers must be increased,
>>> > by kern.bio_transient_maxcnt loader tunable.
>>>
>>> Could you try both and tell which one of them fixed the problem?  Thank=
s!
>>>
>>>
>>
>



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?CACpH0MfrO-0m1PN9w3LpTUO6cN4OkZGuZ8HmhijjeG12cxZR-g>