[Cluster-devel] problem about dlm posix file lock (sorry for missing subjuct)

[Cluster-devel] problem about dlm posix file lock (sorry for missing subjuct)

[David Teigland]

On Tue, Oct 13, 2015 at 04:30:53AM -0600, Zhen Ren wrote:
> It expects alarm timeout to send SIGALRM, and wake up the sleep process,
> as "man fcntl" says: "If a  signal  is  caught  while waiting,  then
> the call is interrupted and (after the signal handler has returned)
> returns immediately (with return value -1 and errno set to EINTR".

Hi, thanks for investigating this.  Have you verified that SIGALRM
interrupts the process on a local fs?  Is it any signal?  Is this causing
problems for a real program?

> dlm_posix_lock -> wait_event_killable
> And wait_event_killable will put process into "TASK_KILLABLE" state
> which's like "UNINTERRUPTABLE" but can be waked up by fatal signals. I
> did some tests, SIGTERM can did it, but SIGALRM cannot.

Perhaps we should replace that wait_event_killable with
wait_event_interruptible.  Could you recompile with that change and see if
it does what you want?

> Did this go against posix file lock semanteme? Any hints would be very
> appreciated!  I can provide any infos as I can if needed;-)

Posix locks on a cluster file system are always going to behave somewhat
differently than a local fs, but this may be a case where the difference
can be easily avoided with the change above.  Changing the behavior after
so long is a slightly concerning, though.

Dave

[Cluster-devel] problem about dlm posix file lock (sorry for missing subject)

[Eric Ren]

Hi Divad,

Please see comments in lines;-)

David Teigland <teigland@redhat.com> write?
> 
>> On Tue, Oct 13, 2015 at 04:30:53AM -0600, Zhen Ren wrote:
>> It expects alarm timeout to send SIGALRM, and wake up the sleep process,
>> as "man fcntl" says: "If a  signal  is  caught  while waiting,  then
>> the call is interrupted and (after the signal handler has returned)
>> returns immediately (with return value -1 and errno set to EINTR".
> 
> Hi, thanks for investigating this.  Have you verified that SIGALRM
> interrupts the process on a local fs?  Is it any signal?  Is this causing
> problems for a real program?

Yeah,I tried on ext4. Two processes fcntl the same file with F_SETLKW in F_WRLCK mode. The second blocked process can be interrupted by SIGALRM.

Also, for ocfs2,  on local node did the same as ext4, SIGALRM works! And I noticed that the status of the blocked process on multiple nodes was "Dl",but was "D+" on local ocfs2. Don't know if it matters. It's some odd.

> 
>> dlm_posix_lock -> wait_event_killable
>> And wait_event_killable will put process into "TASK_KILLABLE" state
>> which's like "UNINTERRUPTABLE" but can be waked up by fatal signals. I
>> did some tests, SIGTERM can did it, but SIGALRM cannot.
> 
> Perhaps we should replace that wait_event_killable with
> wait_event_interruptible.  Could you recompile with that change and see if
> it does what you want?

OK,I'll try;-)

> 
>> Did this go against posix file lock semanteme? Any hints would be very
>> appreciated!  I can provide any infos as I can if needed;-)
> 
> Posix locks on a cluster file system are always going to behave somewhat
> differently than a local fs, but this may be a case where the difference
> can be easily avoided with the change above.  Changing the behavior after
> so long is a slightly concerning, though.

Yeah, more test after change, ensure no unexpected effects.

thanks,
Eric

> 
> Dave
>