Re: Segmentation faults in ceph-osd

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

 



On 21/05/2013, at 21.00, Samuel Just <sam.just@xxxxxxxxxxx> wrote:
> How large are the xattrs?
> -Sam

Sam,

They are quite small, as far as I can see from 10-50 chars each.

However, as you just explained on irc, our current unbounded 1:1 files vs. objects aproach, which in rare cases leads to ~25GB large objects, is not a good idea.. Could those large objects be what triggers the large leveldb allocation (and eventuelly the segfault)? 


-- 
Anders

> On Tue, May 21, 2013 at 10:13 AM, Emil Renner Berthing <ceph@xxxxxxxx> wrote:
>> On 21 May 2013 19:05, Samuel Just <sam.just@xxxxxxxxxxx> wrote:
>>> Do you use xattrs at all?
>> 
>> Yes, on each object we set between 2 to 4 attributes at write time
>> which are then left unchanged.
>> /Emil
>> 
>>> -Sam
>>> 
>>> On Tue, May 21, 2013 at 9:34 AM, Anders Saaby <anders@xxxxxxxxx> wrote:
>>>> On 21/05/2013, at 18.19, Gregory Farnum <greg@xxxxxxxxxxx> wrote:
>>>>> On Tue, May 21, 2013 at 9:01 AM, Emil Renner Berthing <ceph@xxxxxxxx> wrote:
>>>>>> On 21 May 2013 17:55, Gregory Farnum <greg@xxxxxxxxxxx> wrote:
>>>>>>> On Tue, May 21, 2013 at 8:44 AM, Emil Renner Berthing <ceph@xxxxxxxx> wrote:
>>>>>>>> Hi Greg,
>>>>>>>> 
>>>>>>>> Here are some more stats on our servers:
>>>>>>>> - each server has 64GB ram,
>>>>>>>> - there are 12 OSDs pr. server,
>>>>>>>> - each OSD uses around 1.5 GB of memory,
>>>>>>>> - we have 18432 PGs,
>>>>>>>> - around 5 to 10 MB writes/s is written to each OSD and almost no reads (yet).
>>>>>>> 
>>>>>>> What interface are you writing with? How many OSD servers are there?
>>>>>> 
>>>>>> We're using librados and there are 132 OSDs so far.
>>>>> 
>>>>> Okay, so the allocation is happening in the depths of LevelDB — maybe
>>>>> the issue is there somewhere. Are you doing anything weird with omap,
>>>>> snapshots, or xattrs?
>>>> 
>>>> I can help; No, we are not using omap, snaps or weird stuff with xattrs.
>>>> 
>>>> We are storing objects of sizes from few KB to GBs. Also, we have a quirk in the application design right now, which means that we store an object, write it again under a new name and delete the original object. - if that is of any potential value to finding this.
>>>> 
>>>> 
>>>> --
>>>> Anders

--
To unsubscribe from this list: send the line "unsubscribe ceph-devel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html




[Index of Archives]     [CEPH Users]     [Ceph Large]     [Information on CEPH]     [Linux BTRFS]     [Linux USB Devel]     [Video for Linux]     [Linux Audio Users]     [Yosemite News]     [Linux Kernel]     [Linux SCSI]
  Powered by Linux