[Bugs] [Bug 1717824] Fencing: Added the tcmu-runner ALUA feature support but after one of node is rebooted the glfs_file_lock() get stucked

bugzilla at redhat.com bugzilla at redhat.com
Tue Jul 23 07:48:58 UTC 2019


https://bugzilla.redhat.com/show_bug.cgi?id=1717824



--- Comment #16 from Xiubo Li <xiubli at redhat.com> ---
(In reply to Susant Kumar Palai from comment #15)
> (In reply to Xiubo Li from comment #12)
> > @Susant,
> > 
> > There 2 issues are found:
> > 
> > 
> > 1, From my test the glfs_file_lock() sometimes it will takes around 43
> > seconds, is it normal ? And why ?
> > 
> >  90456 2019-07-23 10:08:57.444 31411 [INFO] tcmu_glfs_lock:901 glfs/block0:
> > lxb--------------glfs_file_lock start
> >  ...
> >  319959 2019-07-23 10:09:40.183 31411 [INFO] tcmu_glfs_lock:905 glfs/block0:
> > lxb--------------glfs_file_lock end
> 
> Checked the time taken for file_lock and it completes immediately for me.
> 
>     ret = glfs_fsetxattr(fd1, GF_ENFORCE_MANDATORY_LOCK, "set", 8, 0);
>     if (ret < 0) {
>         LOG_ERR("glfs_fsetxattr", errno);
>         ret = -1;
>         goto out;
>     }
> 
>     time(&before);
>     /* take a write mandatory lock */
>     ret = glfs_file_lock(fd1, F_SETLKW, &lock, GLFS_LK_MANDATORY);
>     if (ret) {
>         LOG_ERR("glfs_file_lock", errno);
>         goto out;
>     }
>     time(&after);
>     diff = (unsigned long )after - before;
>     fprintf(fp, "time %lu %lu %lu\n", diff, before, after);
> 
> time 0 1563867824 1563867824
> 
> Can you attach the brick log here when you run the test next time?
> 
> > 
> > 2, After the lock is broke and all the FIOs callback will return -1, the
> > -EPERM, not the -EBUSY as we discussed before. Is there any change about the
> > return value ? I am only checking the -EBUSY and -ENOTCONN then only after
> > that the lock state in local tcmu node will be changed. Or the local state
> > in tcmu-runner service will always in LOCKED state, but it actually already
> > lost the lock and should be in UNLOCKED state, so all the IOs will fail.
> 
> Please attach the brick log after enabling trace logging. <gluster v set
> <vol> brick-log-level TRACE>
> > 
> > 
> > 

Sure, I will do that after my current work handy, possibly late today or
tomorrow morning.
Thanks
BRs

-- 
You are receiving this mail because:
You are on the CC list for the bug.


More information about the Bugs mailing list