[OpenAFS-port-darwin] OpenAFS 1.6.2 on OS X 10.8: suspected deadlock

Derrick Brashear shadow@gmail.com
Tue, 9 Apr 2013 08:29:18 -0400


 /afs/your-file-system.com/user/shadow/OpenAFS-1.6.2.1-115-ge8307-MountainLion.dmg

On Tue, Apr 9, 2013 at 8:08 AM, Derrick Brashear <shadow@gmail.com> wrote:
> I'll get you something later today.
>
> Derrick
>
> On Apr 9, 2013, at 6:55 AM, "Duncan S Kincaid" <dsk@MIT.EDU> wrote:
>
>> viel Dank stephan und derrick!
>> if you've an installer you'd like me to test, i'd be delighted.
>>
>> ciao
>> dk
>>
>>
>> On Apr 9, 2013, at 4:02 AM, Stephan Wiesand <Stephan.Wiesand@desy.de>
>> wrote:
>>
>>> 9747 was merged.
>>>
>>> On Apr 8, 2013, at 22:27 , Derrick Brashear <shadow@gmail.com> wrote:
>>>
>>>> Indeed. Sorry. 1.6 version is 9747. Stephan, this is a 1.6.2
>>>> regression and only applies
>>>> to MacOS, so it should be noncontroversial.
>>>>
>>>> On Mon, Apr 8, 2013 at 4:21 PM, Duncan S Kincaid <dsk@mit.edu> wrote:
>>>>> this problem was evident in OpenAFS Client 1.6.1 but resolved in OpenAFS Client 1.6.1a.
>>>>> It seems to have resurfaced under 1.6.2.
>>>>>
>>>>> Issue with 1.6.1 was identified by Derrick and fixed:
>>>>> "the good: you're out of Rx packets.
>>>>> the bad: not sure where they all are yet!"
>>>>> and
>>>>> "never mind. found it. gerrit 7788."
>>>>>
>>>>> with thanks
>>>>> dk
>>>>>
>>>>> ==========================
>>>>> Behaviour: All clients with AFS home directories see 20 minute delay before
>>>>> files accessible. Then suddenly all files available and all proceeds normally.
>>>>>
>>>>> For reference, please find report filed July 14 2012 re 1.6.1 client below:
>>>>> ==========================
>>>>> 1. spinning beach ball for *exactly* 20 minutes, then session returns to normal
>>>>> 2. can confirm user gets tokens and TGT at login (using LoginHook and LaunchAgent scripts)
>>>>> 3. ssh connections as root into the beach balling mac reveal following:
>>>>>      a. no CPU load. All processes 'stuck' or 'sleeping' (apart from 'top', of course)
>>>>>      b. virtually no system.log file entires from time of beach balling. at 20 minute mark lots of 'time out' errors logged.
>>>>>      c. attempts to traverse afs directory tree results in shell locking up completely
>>>>> 4. tcpdumps show very little afs traffic**. mostly pings (rx ack) to servers/databases at rate of 2-5/sec. at the 20 minute mark, a barrage of traffic (rx data). (500-4000/sec)****.
>>>>> 5. local users (those whose home directories reside on computer hard drive) have never seen this problem
>>>>> 6. this problem occurs irrespective of particular user with home directory in AFS, AFS server storing his home directory, AFS cell,  and client computer OS.
>>>>> 7. 'Action At A Distance': when the beach ball spins on Macintosh, mounting the same user's AFS home from another computer (Linux box) takes 2 minutes as opposed to customary 1-2 seconds. subsequent Linux logins are normal.
>>>>>
>>>>>
>>>>> |||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
>>>>> duncan kincaid
>>>>> cron | mit school of architecture and planning
>>>>>
>>>>>
>>>>>
>>>>>
>>>>> _______________________________________________
>>>>> port-darwin mailing list
>>>>> port-darwin@openafs.org
>>>>> https://lists.openafs.org/mailman/listinfo/port-darwin
>>>
>>> --
>>> Stephan Wiesand
>>> DESY - DV -
>>> Platanenallee 6
>>> 15732 Zeuthen, Germany
>>
>>
>> |||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
>> duncan kincaid
>> cron | mit school of architecture and planning
>>
>>
>>
>>



-- 
Derrick