way too high reconstruction speed - bug?

way too high reconstruction speed - bug?

[Tomasz Chmielewski]

I noticed that I have a wrong reconstruction speed when looking at 
/proc/mdstat:

# cat /proc/mdstat
Personalities : [raid0] [raid1] [raid10]
md11 : active raid10 sda2[0] sdd2[3] sdc2[2] sdb2[1]
       779264640 blocks super 1.0 64K chunks 2 near-copies [4/4] [UUUU]
       [===>.................]  resync = 16.4% (127955392/779264640) 
finish=4.1min speed=2588238K/sec


2588238K/sec is about 100 more to what I used to see before.


Should I start to worry?

I'm using kernel 2.6.16.27 on ARM CPU.

[42949428.580000] md: bind<sdb2>
[42949428.580000] md: bind<sdc2>
[42949428.590000] md: bind<sdd2>
[42949428.590000] md: bind<sda2>
[42949428.590000] md: md11: raid array is not clean -- starting 
background reconstruction
[42949428.620000] raid10: raid set md11 active with 4 out of 4 devices
[42949428.630000] md: syncing RAID array md11
[42949428.630000] md: minimum _guaranteed_ reconstruction speed: 1000 
KB/sec/disc.
[42949428.640000] md: using maximum available idle IO bandwidth (but not 
more than 200000 KB/sec) for reconstruction.
[42949428.650000] md: using 128k window, over a total of 779264640 blocks.
[42949428.650000] md: resuming recovery of md11 from checkpoint.



-- 
Tomasz Chmielewski
http://wpkg.org

Re: way too high reconstruction speed - bug?

[Paul Clements]

Tomasz Chmielewski wrote:
> [42949428.590000] md: md11: raid array is not clean -- starting 
> background reconstruction
> [42949428.620000] raid10: raid set md11 active with 4 out of 4 devices
> [42949428.630000] md: syncing RAID array md11
> [42949428.630000] md: minimum _guaranteed_ reconstruction speed: 1000 
> KB/sec/disc.
> [42949428.640000] md: using maximum available idle IO bandwidth (but not 
> more than 200000 KB/sec) for reconstruction.
> [42949428.650000] md: using 128k window, over a total of 779264640 blocks.
> [42949428.650000] md: resuming recovery of md11 from checkpoint.

The recovery was resumed from a previous checkpoint. I guess you 
rebooted or stopped the array during recovery?

At any rate, md doesn't calculate resync speed correctly for 
checkpointed or bitmapped resyncs (it assumes that the skipped portion 
was actually resynced).

Neil fairly recently sent in patches to fix this, but I think they won't 
be in mainline until 2.6.18 or so.

--
Paul

Re: way too high reconstruction speed - bug?

[Tomasz Chmielewski]

Paul Clements wrote:
> Tomasz Chmielewski wrote:
>> [42949428.590000] md: md11: raid array is not clean -- starting 
>> background reconstruction
>> [42949428.620000] raid10: raid set md11 active with 4 out of 4 devices
>> [42949428.630000] md: syncing RAID array md11
>> [42949428.630000] md: minimum _guaranteed_ reconstruction speed: 1000 
>> KB/sec/disc.
>> [42949428.640000] md: using maximum available idle IO bandwidth (but 
>> not more than 200000 KB/sec) for reconstruction.
>> [42949428.650000] md: using 128k window, over a total of 779264640 
>> blocks.
>> [42949428.650000] md: resuming recovery of md11 from checkpoint.
> 
> The recovery was resumed from a previous checkpoint. I guess you 
> rebooted or stopped the array during recovery?
> 
> At any rate, md doesn't calculate resync speed correctly for 
> checkpointed or bitmapped resyncs (it assumes that the skipped portion 
> was actually resynced).
> 
> Neil fairly recently sent in patches to fix this, but I think they won't 
> be in mainline until 2.6.18 or so.

All right, thanks for explanation.

After yet another reboot I had something like speed=7588238K/sec, I can
only wonder when hard drives will have this speed :)


-- 
Tomasz Chmielewski
http://wpkg.org