Re: MD: "sync_action" issues: pausing resync/recovery automatically restarts.

[Benjamin ESTRABAUD <be@mpstor.com>]

Neil Brown wrote:
> On Thu, 11 Feb 2010 12:02:56 +0000
> Benjamin ESTRABAUD <be@mpstor.com> wrote:
>
>   
>> Hi everybody,
>>
>> I am getting a weird issue when I am writing values to 
>> "/sys/block/mdX/md/sync_action".
>> For instance, I would like to pause a resync or/and a recovery when they 
>> are happening.
>> I create a RAID 5 as follow:
>>
>> mdadm --create -vvv --force --run --metadata=1.2 /dev/md/d0 --level=5 
>> --size=9429760 --chunk=64 --name=1056856 -n5 --bitmap=internal 
>> --bitmap-chunk=4096 --layout=ls /dev/sde2 /dev/sdb2 /dev/sdc2 /dev/sdf2 
>> /dev/sdd2
>>
>> The RAID is resyncing:
>>
>> # cat /proc/mdstat
>> Personalities : [linear] [raid0] [raid1] [raid10] [raid6] [raid5] [raid4]
>> md_d0 : active raid5 sdd2[4] sdf2[3] sdc2[2] sdb2[1] sde2[0]
>>       37719040 blocks super 1.2 level 5, 64k chunk, algorithm 2 [5/5] 
>> [UUUUU]
>>       [====>................]  resync = 22.2% (2101824/9429760) 
>> finish=2.6min speed=46186K/sec
>>       bitmap: 1/1 pages [64KB], 4096KB chunk
>>
>> unused devices: <none>
>>
>> I then decide to pause its resync:
>>
>> # echo idle > /sys/block/md_d0/md/sync_action
>>
>> The RAID resync should have paused by now, let's check the sys properties:
>>
>> # cat /sys/block/md_d0/md/sync_action
>> resync
>>
>> The resync seems to have not stopped/restarted, let's check dmesg:
>>
>> [157287.049715] raid5: raid level 5 set md_d0 active with 5 out of 5 
>> devices, algorithm 2
>> [157287.057601] RAID5 conf printout:
>> [157287.060909]  --- rd:5 wd:5
>> [157287.063700]  disk 0, o:1, dev:sde2
>> [157287.067182]  disk 1, o:1, dev:sdb2
>> [157287.070664]  disk 2, o:1, dev:sdc2
>> [157287.074147]  disk 3, o:1, dev:sdf2
>> [157287.077628]  disk 4, o:1, dev:sdd2
>> [157287.086813] md_d0: bitmap initialized from disk: read 1/1 pages, set 
>> 2303 bits
>> [157287.094134] created bitmap (1 pages) for device md_d0
>> [157287.113475] md: resync of RAID array md_d0
>> [157287.117650] md: minimum _guaranteed_  speed: 1000 KB/sec/disk.
>> [157287.123555] md: using maximum available idle IO bandwidth (but not 
>> more than 200000 KB/sec) for resync.
>> [157287.133011] md: using 2048k window, over a total of 9429760 blocks.
>> [157345.158535] md: md_do_sync() got signal ... exiting
>> [157345.166057] md: checkpointing resync of md_d0.
>> [157345.179819] md: resync of RAID array md_d0
>> [157345.183993] md: minimum _guaranteed_  speed: 1000 KB/sec/disk.
>> [157345.189899] md: using maximum available idle IO bandwidth (but not 
>> more than 200000 KB/sec) for resync.
>> [157345.199353] md: using 2048k window, over a total of 9429760 blocks.
>>
>> The resync seem to stop at some stage since:
>>
>> [157345.158535] md: md_do_sync() got signal ... exiting
>>
>> But it seems to be restarting right after this:
>>
>> [157345.179819] md: resync of RAID array md_d0
>>
>> I read in the md.txt documentation that pausing a resync could sometimes 
>> not work if a n event or trigger was triggering it to automatically 
>> restart. However, I don't think I have any trigger that would cause it 
>> to restart.
>> it then builds perfectly fine.
>>
>> I now want to check if the same issue occurs while recovering, after 
>> all, I especially want to be able to pause a recovery, while I don't 
>> really need to pause/restart resyncs.
>>
>> Let's say I pull a disk from the bay, fail it and remove it as follow:
>>
>> # mdadm --fail /dev/md/d0 /dev/sde2
>> mdadm: set /dev/sde2 faulty in /dev/md/d0
>>
>> # mdadm --remove /dev/md/d0 /dev/sde2
>> mdadm: hot removed /dev/sde2
>>
>> Now let's add a spare:
>>
>> # /opt/soma/bin/mdadm/mdadm --add /dev/md/d0 /dev/sda2  
>> raid manager: added /dev/sda2
>>
>> The RAID is now recovering:
>>
>> # cat /proc/mdstat
>> Personalities : [linear] [raid0] [raid1] [raid10] [raid6] [raid5] [raid4]
>> md_d0 : active raid5 sda2[5] sdd2[4] sdf2[3] sdc2[2] sdb2[1]
>>       37719040 blocks super 1.2 level 5, 64k chunk, algorithm 2 [5/4] 
>> [_UUUU]
>>       [>....................]  recovery =  1.7% (169792/9429760) 
>> finish=0.9min speed=169792K/sec
>>       bitmap: 0/1 pages [0KB], 4096KB chunk
>>
>> unused devices: <none>
>>
>> # cat /sys/block/md_d0/md/sync_action
>> recover
>>
>> Let's try and stop this recovery:
>>
>> # echo idle > /sys/block/md_d0/md/sync_action
>>
>> [157641.618291]  disk 3, o:1, dev:sdf2
>> [157641.621774]  disk 4, o:1, dev:sdd2
>> [157641.632057] md: recovery of RAID array md_d0
>> [157641.636413] md: minimum _guaranteed_  speed: 1000 KB/sec/disk.
>> [157641.642314] md: using maximum available idle IO bandwidth (but not 
>> more than 200000 KB/sec) for recovery.
>> [157641.651940] md: using 2048k window, over a total of 9429760 blocks.
>> [157657.120722] md: md_do_sync() got signal ... exiting
>> [157657.267055] RAID5 conf printout:
>> [157657.270381]  --- rd:5 wd:4
>> [157657.273171]  disk 0, o:1, dev:sda2
>> [157657.276650]  disk 1, o:1, dev:sdb2
>> [157657.280129]  disk 2, o:1, dev:sdc2
>> [157657.283605]  disk 3, o:1, dev:sdf2
>> [157657.287087]  disk 4, o:1, dev:sdd2
>> [157657.290568] RAID5 conf printout:
>> [157657.293876]  --- rd:5 wd:4
>> [157657.296660]  disk 0, o:1, dev:sda2
>> [157657.300139]  disk 1, o:1, dev:sdb2
>> [157657.303615]  disk 2, o:1, dev:sdc2
>> [157657.307096]  disk 3, o:1, dev:sdf2
>> [157657.310579]  disk 4, o:1, dev:sdd2
>> [157657.320835] md: recovery of RAID array md_d0
>> [157657.325194] md: minimum _guaranteed_  speed: 1000 KB/sec/disk.
>> [157657.331091] md: using maximum available idle IO bandwidth (but not 
>> more than 200000 KB/sec) for recovery.
>> [157657.340713] md: using 2048k window, over a total of 9429760 blocks.
>> [157657.347047] md: resuming recovery of md_d0 from checkpoint.
>>
>> I am getting the same issue, the recovery stops, but restarts 200 
>> milliseconds later.
>>     
>
> So clearly the resync is pausing - for 200milliseconds....
>
> 'idle' is only really useful to top a 'check' or 'repair'.
> A 'sync' or 'recovery' md really wants to do, so whenever it seems to be
> needed it, it does it.
>
> What you want is "frozen" which is only available since 2.6.31.
>
>   
Hi Neil, and thanks a lot for your reply.

I understand what you mean by this.

2.6.31 would have the perfect feature for me, but unfortunately I cannot 
change to this Kernel.
>> This clearly indicates that some sort of trigger is automatically 
>> restarting the resync and recovery, but I have no clue as of what could 
>> it be.
>>
>> Would anyone here had a similar experience with trying to stop resyncs? 
>> Is there a "magic" variable that would enable or disable automatic 
>> restart of resync/recoveries?
>>
>> Would anyone know of a standard event or trigger that would cause a 
>> resync or recovery to automatically restart?
>>
>> Thank you very much in advance for your help.
>>
>> My Kernel version is:
>>
>> 2.6.26.3
>>
>>     
>
> So with that kernel, you cannot freeze a recovery.
>
> Why do you want to?
>
>   
I would like to minimize IO penalities when rebuilding (I know of the 
sync_min and sync_max but even rebuilding at a very low speed makes the 
whole IOs run much slower. Therefore, "pausing" the resync is a perfect 
solution while rebuilding. It can then be restarted when the file copy 
is done for instance.
> A possible option is the mark the array read-only 
>    "mdadm  --read-only /dev/mdXX".
>
>   
This is a good solution for me, the array is not mounted in my case as 
it is being used as raw storage.

Thanks a lot for this suggestion!
> This doesn't work if the array is mounted, but does stop any recovery from
> happening.
>
> NeilBrown
>
>
>   
Ben.