Re: librbd - threads grow with each Image object

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



Hmm, I checked again with librbd codes. Each image will create two
threads one for finisher, another is SafeTimer. So your case is
reasonable.

@jason @josh We may reduce this via lookup_or_create_singleton_object too?

On Sat, Nov 21, 2015 at 3:51 PM, Allen Liao <aliao.svsgames@xxxxxxxxx> wrote:
> I'm on 0.94.5.
>
> No, rbd cache is not enabled.  Even if each Image creates only one extra
> thread, if I have tens of thousands of Image objects open, there will be
> tens of thousands of threads in my process.  Practically speaking, am I not
> allowed to cache Image objects?
>
> On Fri, Nov 20, 2015 at 8:24 PM, Haomai Wang <haomaiwang@xxxxxxxxx> wrote:
>>
>> What's your ceph version?
>>
>> Do you enable rbd cache? By default, each Image should only have one
>> extra thread(maybe we also should obsolete this?).
>>
>> On Sat, Nov 21, 2015 at 9:26 AM, Allen Liao <aliao.svsgames@xxxxxxxxx>
>> wrote:
>> > I am developing a python application (using rbd.py) that requires
>> > querying
>> > information about tens of thousands of rbd images.  I have noticed that
>> > the
>> > number of threads in my process grow linearly with each Image object
>> > that is
>> > created.
>> >
>> > After creating about 800 Image objects (that all share a single ioctx),
>> > my
>> > process already has more than 2000 threads.  I get the thread count
>> > using
>> > `ps huH p <pid> | wc -l`.
>> >
>> > If I call close() on each Image object after operating on it then the
>> > threads are cleaned up.  However, I want to cache these objects and
>> > reuse
>> > them since it is expensive to create tens of thousands of these objects
>> > all
>> > the time.
>> >
>> > Is it correct for librbd to create 4-5 threads for each Image object
>> > created?
>> >
>> > For example, I'm doing something similar to:
>> > -------------
>> >
>> > import rbd
>> > import rados
>> >
>> > cluster = rados.Rados(conffile='my_ceph.conf')
>> > cluster.connect()
>> > ioctx = cluster.open_ioctx('mypool')
>> >
>> > # With each object, new threads are created until close() is called
>> > image0 = rbd.Image(ioctx, 'myimage0')
>> > image1 = rbd.Image(ioctx, 'myimage1')
>> > ...
>> > image9000 = rbd.Image(ioctx, 'myimage9000')
>> >
>> >
>> > _______________________________________________
>> > ceph-users mailing list
>> > ceph-users@xxxxxxxxxxxxxx
>> > http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com
>> >
>>
>>
>>
>> --
>> Best Regards,
>>
>> Wheat
>
>



-- 
Best Regards,

Wheat
_______________________________________________
ceph-users mailing list
ceph-users@xxxxxxxxxxxxxx
http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com



[Index of Archives]     [Information on CEPH]     [Linux Filesystem Development]     [Ceph Development]     [Ceph Large]     [Linux USB Development]     [Video for Linux]     [Linux Audio Users]     [Yosemite News]     [Linux Kernel]     [Linux SCSI]     [xfs]


  Powered by Linux