Re: 3.8.3 Bitrot signature process

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

 



Hi Amudhan,

I don't think it's the limitation with read data from the brick.
To limit the usage of CPU, throttling is done using token bucket
algorithm. The log message showed is related to it. But even then
I think it should not take 12 minutes for check-sum calculation unless
there is an fd open (might be internal). Could you please cross verify
if there are any fd opened on that file by looking into /proc? I will
also test it out in the mean time and get back to you.

Thanks and Regards,
Kotresh H R

----- Original Message -----
> From: "Amudhan P" <amudhan83@xxxxxxxxx>
> To: "Kotresh Hiremath Ravishankar" <khiremat@xxxxxxxxxx>
> Cc: "Gluster Users" <gluster-users@xxxxxxxxxxx>
> Sent: Tuesday, September 20, 2016 3:19:28 PM
> Subject: Re:  3.8.3 Bitrot signature process
> 
> Hi Kotresh,
> 
> Please correct me if i am wrong, Once a file write completes and as soon as
> closes fds, bitrot waits for 120 seconds and starts hashing and update
> signature for the file in brick.
> 
> But, what i am feeling that bitrot takes too much of time to complete
> hashing.
> 
> below is test result i would like to share.
> 
> 
> writing data in below path using dd :
> 
> /mnt/gluster/data/G (mount point)
> -rw-r--r-- 1 root root  10M Sep 20 12:19 test53-bs10M-c1.nul
> -rw-r--r-- 1 root root 100M Sep 20 12:19 test54-bs10M-c10.nul
> 
> No any other write or read process is going on.
> 
> 
> Checking file data in one of the brick.
> 
> -rw-r--r-- 2 root root 2.5M Sep 20 12:23 test53-bs10M-c1.nul
> -rw-r--r-- 2 root root  25M Sep 20 12:23 test54-bs10M-c10.nul
> 
> file's stat and getfattr info from brick, after write process completed.
> 
> gfstst-node5:/media/disk2/brick2/data/G$ stat test53-bs10M-c1.nul
>   File: ‘test53-bs10M-c1.nul’
>   Size: 2621440         Blocks: 5120       IO Block: 4096   regular file
> Device: 821h/2081d      Inode: 536874168   Links: 2
> Access: (0644/-rw-r--r--)  Uid: (    0/    root)   Gid: (    0/    root)
> Access: 2016-09-20 12:23:28.798886647 +0530
> Modify: 2016-09-20 12:23:28.994886646 +0530
> Change: 2016-09-20 12:23:28.998886646 +0530
>  Birth: -
> 
> gfstst-node5:/media/disk2/brick2/data/G$ stat test54-bs10M-c10.nul
>   File: ‘test54-bs10M-c10.nul’
>   Size: 26214400        Blocks: 51200      IO Block: 4096   regular file
> Device: 821h/2081d      Inode: 536874169   Links: 2
> Access: (0644/-rw-r--r--)  Uid: (    0/    root)   Gid: (    0/    root)
> Access: 2016-09-20 12:23:42.902886624 +0530
> Modify: 2016-09-20 12:23:44.378886622 +0530
> Change: 2016-09-20 12:23:44.378886622 +0530
>  Birth: -
> 
> gfstst-node5:/media/disk2/brick2/data/G$ sudo getfattr -m. -e hex -d
> test53-bs10M-c1.nul
> # file: test53-bs10M-c1.nul
> trusted.bit-rot.version=0x020000000000000057daa7b50002e5b4
> trusted.ec.config=0x0000080501000200
> trusted.ec.size=0x0000000000a00000
> trusted.ec.version=0x00000000000000500000000000000050
> trusted.gfid=0xe2416bd1aae4403c88f44286273bbe99
> 
> gfstst-node5:/media/disk2/brick2/data/G$ sudo getfattr -m. -e hex -d
> test54-bs10M-c10.nul
> # file: test54-bs10M-c10.nul
> trusted.bit-rot.version=0x020000000000000057daa7b50002e5b4
> trusted.ec.config=0x0000080501000200
> trusted.ec.size=0x0000000006400000
> trusted.ec.version=0x00000000000003200000000000000320
> trusted.gfid=0x54e018dd8c5a4bd79e0317729d8a57c5
> 
> 
> 
> file's stat and getfattr info from brick, after bitrot signature updated.
> 
> gfstst-node5:/media/disk2/brick2/data/G$ stat test53-bs10M-c1.nul
>   File: ‘test53-bs10M-c1.nul’
>   Size: 2621440         Blocks: 5120       IO Block: 4096   regular file
> Device: 821h/2081d      Inode: 536874168   Links: 2
> Access: (0644/-rw-r--r--)  Uid: (    0/    root)   Gid: (    0/    root)
> Access: 2016-09-20 12:25:31.494886450 +0530
> Modify: 2016-09-20 12:23:28.994886646 +0530
> Change: 2016-09-20 12:27:00.994886307 +0530
>  Birth: -
> 
> 
> gfstst-node5:/media/disk2/brick2/data/G$ sudo getfattr -m. -e hex -d
> test53-bs10M-c1.nul
> # file: test53-bs10M-c1.nul
> trusted.bit-rot.signature=0x0102000000000000006de7493c5c90f643357c268fbaaf461c1567e0334e4948023ce17268403aa37a
> trusted.bit-rot.version=0x020000000000000057daa7b50002e5b4
> trusted.ec.config=0x0000080501000200
> trusted.ec.size=0x0000000000a00000
> trusted.ec.version=0x00000000000000500000000000000050
> trusted.gfid=0xe2416bd1aae4403c88f44286273bbe99
> 
> 
> gfstst-node5:/media/disk2/brick2/data/G$ stat test54-bs10M-c10.nul
>   File: ‘test54-bs10M-c10.nul’
>   Size: 26214400        Blocks: 51200      IO Block: 4096   regular file
> Device: 821h/2081d      Inode: 536874169   Links: 2
> Access: (0644/-rw-r--r--)  Uid: (    0/    root)   Gid: (    0/    root)
> Access: 2016-09-20 12:25:47.510886425 +0530
> Modify: 2016-09-20 12:23:44.378886622 +0530
> Change: 2016-09-20 12:38:05.954885243 +0530
>  Birth: -
> 
> 
> gfstst-node5:/media/disk2/brick2/data/G$ sudo getfattr -m. -e hex -d
> test54-bs10M-c10.nul
> # file: test54-bs10M-c10.nul
> trusted.bit-rot.signature=0x010200000000000000394c345f0b0c63ee652627a62eed069244d35c4d5134e4f07d4eabb51afda47e
> trusted.bit-rot.version=0x020000000000000057daa7b50002e5b4
> trusted.ec.config=0x0000080501000200
> trusted.ec.size=0x0000000006400000
> trusted.ec.version=0x00000000000003200000000000000320
> trusted.gfid=0x54e018dd8c5a4bd79e0317729d8a57c5
> 
> 
> (Actual time taken for reading file from brick for md5sum)
> 
> gfstst-node5:/media/disk2/brick2/data/G$ time md5sum test53-bs10M-c1.nul
> 8354dcaa18a1ecb52d0895bf00888c44  test53-bs10M-c1.nul
> 
> real    0m0.045s
> user    0m0.007s
> sys     0m0.003s
> 
> gfstst-node5:/media/disk2/brick2/data/G$ time md5sum test54-bs10M-c10.nul
> bed3c0a4a1407f584989b4009e9ce33f  test54-bs10M-c10.nul
> 
> real    0m0.166s
> user    0m0.062s
> sys     0m0.011s
> 
> As you can see that 'test54-bs10M-c10.nul' file took around 12 minutes to
> update bitort signature (pls refer stat output for the file).
> 
> what would be the cause for such a slow read?. Any limitation in read data
> from brick?
> 
> Also, i am seeing this line bitd.log, what does this mean?
> [bit-rot.c:1784:br_rate_limit_signer] 0-glsvol1-bit-rot-0: [Rate Limit
> Info] "tokens/sec (rate): 131072, maxlimit: 524288
> 
> 
> Thanks
> Amudhan P
> 
> 
> 
> On Mon, Sep 19, 2016 at 1:00 PM, Kotresh Hiremath Ravishankar <
> khiremat@xxxxxxxxxx> wrote:
> 
> > Hi Amudhan,
> >
> > Thanks for testing out the bitrot feature and sorry for the delayed
> > response.
> > Please find the answers inline.
> >
> > Thanks and Regards,
> > Kotresh H R
> >
> > ----- Original Message -----
> > > From: "Amudhan P" <amudhan83@xxxxxxxxx>
> > > To: "Gluster Users" <gluster-users@xxxxxxxxxxx>
> > > Sent: Friday, September 16, 2016 4:14:10 PM
> > > Subject: Re:  3.8.3 Bitrot signature process
> > >
> > > Hi,
> > >
> > > Can anyone reply to this mail.
> > >
> > > On Tue, Sep 13, 2016 at 12:49 PM, Amudhan P < amudhan83@xxxxxxxxx >
> > wrote:
> > >
> > >
> > >
> > > Hi,
> > >
> > > I am testing bitrot feature in Gluster 3.8.3 with disperse EC volume 4+1.
> > >
> > > When i write single small file (< 10MB) after 2 seconds i can see bitrot
> > > signature in bricks for the file, but when i write multiple files with
> > > different size ( > 10MB) it takes long time (> 24hrs) to see bitrot
> > > signature in all the files.
> >
> >    The default timeout for signing to happen is 120 seconds. So the
> > signing will happen
> >   120 secs after the last fd gets closed on that file. So if the file is
> > being written
> >   continuously, it will not be signed until 120 secs after it's last fd is
> > closed.
> > >
> > > My questions are.
> > > 1. I have enabled scrub schedule as hourly and throttle as normal, does
> > this
> > > make any impact in delaying bitrot signature?
> >       No.
> > > 2. other than "bitd.log" where else i can watch current status of bitrot,
> > > like number of files added for signature and file status?
> >      Signature will happen after 120 sec of last fd closure, as said above.
> >      There is not status command which tracks the signature of the files.
> >      But there is bitrot status command which tracks the number of files
> >      scrubbed.
> >
> >      #gluster vol bitrot <volname> scrub status
> >
> >
> > > 3. where i can confirm that all the files in the brick are bitrot signed?
> >
> >      As said, signing information of all the files is not tracked.
> >
> > > 4. is there any file read size limit in bitrot?
> >
> >      I didn't get. Could you please elaborate this ?
> >
> > > 5. options for tuning bitrot for faster signing of files?
> >
> >      Bitrot feature is mainly to detect silent corruption (bitflips) of
> > files due to long
> >      term storage. Hence the default is 120 sec of last fd closure, the
> > signing happens.
> >      But there is a tune able which can change the default 120 sec but
> > that's only for
> >      testing purposes and we don't recommend it.
> >
> >       gluster vol get master features.expiry-time
> >
> >      For testing purposes, you can change this default and test.
> > >
> > > Thanks
> > > Amudhan
> > >
> > >
> > > _______________________________________________
> > > Gluster-users mailing list
> > > Gluster-users@xxxxxxxxxxx
> > > http://www.gluster.org/mailman/listinfo/gluster-users
> >
> 
_______________________________________________
Gluster-users mailing list
Gluster-users@xxxxxxxxxxx
http://www.gluster.org/mailman/listinfo/gluster-users




[Index of Archives]     [Gluster Development]     [Linux Filesytems Development]     [Linux ARM Kernel]     [Linux ARM]     [Linux Omap]     [Fedora ARM]     [IETF Annouce]     [Bugtraq]     [Linux OMAP]     [Linux MIPS]     [eCos]     [Asterisk Internet PBX]     [Linux API]

  Powered by Linux