[sheepdog] [PATCH v2 0/3] sheep: writeback cache semantics in backend store

Liu Yuan namei.unix at gmail.com
Mon Sep 3 05:43:01 CEST 2012


On 09/03/2012 01:45 AM, Hitoshi Mitake wrote:
> On Sun, Sep 2, 2012 at 2:58 PM, Liu Yuan <namei.unix at gmail.com> wrote:
>> On 09/02/2012 10:34 AM, Hitoshi Mitake wrote:
>>> This patch series implements writeback cache semantics in backend
>>> store of sheep. Current backend store farm and plain_store calls
>>> open() with O_DSYNC, so every object write causes slow disk
>>> access. This patch series try to reduce this overhead with flushing
>>> disk cache with the invocation of SD_OP_FLUSH_VDI instead of every
>>> object write can improve performance when a number of VMs is not so
>>> large.
>>>
>>> Hitoshi Mitake (3):
>>>   sheep: change gateway_forward_request() for forwarding requests to
>>>     all other nodes
>>>   sheep: add new operation flush() to store_driver
>>>   sheep: add SD_OP_SYNC_VDI and SD_OP_FLUSH_PEER for writeback cache
>>>     semantics
>>
>> This patch set can't pass test scripts.
>>
>> You can run tests as following (see HOWTO for more info):
>>
>> cd tests
>> sudo ./check -corosync
>>
> 
> Thanks for your review and sorry for my bad patch series.
> I'll fix the problem later.
> 

Well, your patch set code looks great to me. With your patch set, I just
met random segfault running 26. But I am not sure if your patch set is
root cause, because I enabled tracer meanwhile. I'll dig more today.

Thanks,
Yuan



More information about the sheepdog mailing list