[Ocfs2-devel] dlm stress test hangs OCFS2

[Sunil Mushran]

You will have to trace thru process_blocked_lock() to make sense of this.

Coly Li wrote:
> Sunil Mushran Wrote:
>   
>> So the thread is removing the lockres from the list (thus making
>> the count 0), and then calling ocfs2_process_blocked_lock() which
>> is adding that lockres back in the list (thus 1).
>>
>> Trace ocfs2_process_blocked_lock() to see as to why it is putting it
>> back on the list.
>>     
>>> 2) when the blocking happens, the number sequence of
>>> osb->blocked_lock_count is
>>> always like this,
>>> ocfs2_downconvert_thread_do_work:3725: osb->blocked_lock_count: 0
>>> ocfs2_downconvert_thread_do_work:3725: osb->blocked_lock_count: 1
>>> ocfs2_downconvert_thread_do_work:3725: osb->blocked_lock_count: 0
>>> ocfs2_downconvert_thread_do_work:3725: osb->blocked_lock_count: 1
>>> ocfs2_downconvert_thread_do_work:3725: osb->blocked_lock_count: 0
>>> ocfs2_downconvert_thread_do_work:3725: osb->blocked_lock_count: 1
>>> ocfs2_downconvert_thread_do_work:3725: osb->blocked_lock_count: 0
>>> ocfs2_downconvert_thread_do_work:3725: osb->blocked_lock_count: 1
>>> ocfs2_downconvert_thread_do_work:3725: osb->blocked_lock_count: 0
>>> ocfs2_downconvert_thread_do_work:3725: osb->blocked_lock_count: 1
>>> ocfs2_downconvert_thread_do_work:3725: osb->blocked_lock_count: 0
>>> ocfs2_downconvert_thread_do_work:3725: osb->blocked_lock_count: 1
>>> ocfs2_downconvert_thread_do_work:3725: osb->blocked_lock_count: 0
>>> ocfs2_downconvert_thread_do_work:3725: osb->blocked_lock_count: 1
>>> ocfs2_downconvert_thread_do_work:3725: osb->blocked_lock_count: 0
>>> ocfs2_downconvert_thread_do_work:3725: osb->blocked_lock_count: 1
>>> ocfs2_downconvert_thread_do_work:3725: osb->blocked_lock_count: 0
>>> ocfs2_downconvert_thread_do_work:3725: osb->blocked_lock_count: 1
>>> (all are 0-1-0-1-0-1-... in a regular sequence)
>>>
>>>       
>
> I also observed even when no file is created or unlinked, I did nothing. From
> dmesg output, the above information still infinite running, on both nodes.
>
>
>
>