On Tue, Feb 12, 2013 at 06:17:04AM +0000, Norman Cheung wrote:
> I am not sure if this forum is the same as xfs@xxxxxxxxxxx, if so, my apology
> for double posting. I appreciate in sight or work around on this.
>
> Every 3 - 4 days, my application will hang in D state at file close. And
> shortly after that flush (from a different partition) is locked in D state
> also.
>
> My application runs continuously, 5 threads are writing data at a rate of
> 1.5M/sec to 5 different XFS partitions. Each of these partitions is a 2 disk
> RAID 0. In addition, I have other threads consuming 100% CPU at all time, and
> most of these threads are tied to its own CPU.
>
> There are 5 data writing threads are also set to run in specific CPU (one CPU
>
> per thread), with priority set to high (-2). The data writing pattern is:
> each disk writing thread will write a file 1.5 Gig. Then the thread will
> pause for about 3 minutes. Hence we have 5 files of 1.5Gig each after one
> processing cycle. And we keep 5 sets and delete the older ones.
>
> After about 300 - 800 cycle, one or two of these disk writing threads will go
> into D state. And within a second flush of another partition will show up in
> D state. then after 15 minutes of no activities, the parent task will lower
> the priority of all threads (to noraml 20) and abort the threads. In all
> cases, lowering the priority will get threads out of D states. I have also
> tried running the disk writing threads with normal priority (20). Same
> hangs. Also the fclose of all 5 files to 5 different partitions happens
> around the same time.
>
> Thanks in advance,
>
> Norman
>
>
> Below is the sysrq for the 2 offending threads.
>
> 1. the disk writing thread hung in fclose
>
>
> Tigris_IMC.exe D 0000000000000000 0 4197 4100 0x00000000
> ffff881f3db921c0 0000000000000086 0000000000000000 ffff881f42eb8b80
> ffff880861419fd8 ffff880861419fd8 ffff880861419fd8 ffff881f3db921c0
> 0000000000080000 0000000000000000 00000000000401e0 00000000061805c1 Call
> Trace:
> [<ffffffff810d89ed>] ? zone_statistics+0x9d/0xa0 [<ffffffffa0402682>] ?
> xfs_iomap_write_delay+0x172/0x2b0 [xfs] [<ffffffff813c7e35>] ?
> rwsem_down_failed_common+0xc5/0x150
> [<ffffffff811f32a3>] ? call_rwsem_down_write_failed+0x13/0x20
> [<ffffffff813c74ec>] ? down_write+0x1c/0x1d [<ffffffffa03fba8e>] ?
> xfs_ilock+0x7e/0xa0 [xfs] [<ffffffffa041b64b>] ? __xfs_get_blocks+0x1db/0x3d0
> [xfs] [<ffffffff81103340>] ? kmem_cache_alloc+0x100/0x130
> [<ffffffff8113fa2e>] ? alloc_page_buffers+0x6e/0xe0 [<ffffffff81141cdf>] ?
> __block_write_begin+0x1cf/0x4d0 [<ffffffffa041b850>] ?
> xfs_get_blocks_direct+0x10/0x10 [xfs] [<ffffffffa041b850>] ?
> xfs_get_blocks_direct+0x10/0x10 [xfs] [<ffffffff8114226b>] ?
> block_write_begin+0x4b/0xa0 [<ffffffffa041b8fb>] ?
> xfs_vm_write_begin+0x3b/0x70 [xfs] [<ffffffff810c0258>] ?
> generic_file_buffered_write+0xf8/0x250
> [<ffffffffa04207b5>] ? xfs_file_buffered_aio_write+0xc5/0x130 [xfs]
> [<ffffffffa042099c>] ? xfs_file_aio_write+0x17c/0x2a0 [xfs]
> [<ffffffff81115b28>] ? do_sync_write+0xb8/0xf0 [<ffffffff8119daa4>] ?
> security_file_permission+0x24/0xc0
> [<ffffffff8111630a>] ? vfs_write+0xaa/0x190 [<ffffffff81116657>] ?
> sys_write+0x47/0x90 [<ffffffff813ce412>] ? system_call_fastpath+0x16/0x1b
Can you please post non-mangled traces? these have all the lines run
together and then wrapped by you mailer....
Cheers,
Dave.
--
Dave Chinner
david@xxxxxxxxxxxxx
|