ZFS freeze/livelock
Willem Jan Withagen
wjw at digiware.nl
Thu Oct 14 09:32:01 UTC 2010
On 2010-10-14 11:12, Borja Marcos wrote:
>
> On Oct 13, 2010, at 6:16 PM, Willem Jan Withagen wrote:
>
>> On 2010-10-13 13:08, Borja Marcos wrote:
>
>> Well I think what I did more or less fits your desciption.
>>
>> But thusfar it did not happen.
>> And I'm (very slowly) redoing some of these steps, with all debugging settings in the kernel.
>
> Sometimes it isn't easy to reproduce, but I found a way. Whenever a new version of FreeBSD comes out I run the following test.
>
> Start two machines. A and B
> Update them (via make buildworld, etc)
> set up an automatic replication for a dataset from A to B in 30 second or 1 minute intervals
> The chosen dataset is the one which contains /usr/src and /usr/obj
> run a loop of make buildworld&& make clean on A
> And on B I run a couple of tasks or so that simple keep a loop copying the contents of the destination dataset (example, pool/srcobj) to a different place, for instance using "tar", so that I keep a heavy I/O activity.
>
> So far I can reproduce the phenomenon in less than 20 minutes.
'mmmm, I'n not really pounding my B-system that hard....
But you haven't found a way to see what lock is the actual cullpit?
Let alone, try and determine how all contenders actually got there?
Perhaps you should file a PR with the above means of reproducing it,
just for history sake. Or perhaps you already did?
But my B-system at the moment is already 3 days underway to receive a
225G volume. :( It makes progress but really really very slow.
--WjW
More information about the freebsd-fs
mailing list