Re: [ceph-commit] Ceph Zfs

Re: [ceph-commit] Ceph Zfs

[Sage Weil]

[moved to ceph-devel]

On Thu, 25 Oct 2012, Raghunandhan wrote:
> Hi All,
> 
> I have been working around ceph quite a long and trying to stitch zfs with
> ceph. I was able to do it to certain extent as follows:
> 1. zpool creation
> 2. set dedup
> 3. create a mountable volume of zfs (zfs create)
> 4. format the volume with ext4 and enabling xattr
> 5. mkcephfs on the volume
> 
> This actually works and dedup is perfect. But i need to avoid multiple layers
> on the storage since the performance is very slow and the kernel timeout
> occurs often for a 8GB RAM. I want to test the performance between btrfs and
> zfs. I want to avoid the above multiple layering on storage and make the ceph
> cluster aware of zfs. Let me know if anyone has workaround this.

I'm not familiar enough with zfs to know what 'mountable volume' means.. 
is that a block device/lun that you're putting ext4 on?  Probably the best 
results will come from creating a zfs *file system* (using the ZPL or 
whatever it is) and running ceph-osd on top of that.

There is at least one open bug from someone having problems there, but 
we'd very much like to sort out the problem.

sage

Re: [ceph-commit] Ceph Zfs

[Raghunandhan]

Hi Sage,

Thanks for replying back, Once a zpool is created if i mount it on 
/var/lib/ceph/osd/ceph-0 the cephfs doesnt recognize it as a superblock 
and hence it fails, Im trying to build this on our cloud storage since 
btrfs has not been stable nor they have come up with online dedup i have 
no other choice for now to work with zfs ceph which makes sense.

So what i exactly did was created a zpool store
1 Then used the same store and made a block device from it using zfs 
create
2 Once the zfs create was successful i was able to format with ext4 
using xattr
3 On top of it was the ceph

Following this process doesnt make sense because of multiple layer on 
the storage and the ceph consumes a lot of RAM and cpu cycles which ends 
up in kernel hung task. It would be great if there is a way i could 
directly use the zfs pool with ceph and make it work.

---
Regards,
Raghunandhan.G
IIHT Cloud Solutions Pvt. Ltd.
#15, 4th Floor, 'A' Wing, Sri Lakshmi Complex,
St. Marks Road, Bangalore - 560 001, India

On 25-10-2012 22:06, Sage Weil wrote:
> [moved to ceph-devel]
>
> On Thu, 25 Oct 2012, Raghunandhan wrote:
>> Hi All,
>>
>> I have been working around ceph quite a long and trying to stitch 
>> zfs with
>> ceph. I was able to do it to certain extent as follows:
>> 1. zpool creation
>> 2. set dedup
>> 3. create a mountable volume of zfs (zfs create)
>> 4. format the volume with ext4 and enabling xattr
>> 5. mkcephfs on the volume
>>
>> This actually works and dedup is perfect. But i need to avoid 
>> multiple layers
>> on the storage since the performance is very slow and the kernel 
>> timeout
>> occurs often for a 8GB RAM. I want to test the performance between 
>> btrfs and
>> zfs. I want to avoid the above multiple layering on storage and make 
>> the ceph
>> cluster aware of zfs. Let me know if anyone has workaround this.
>
> I'm not familiar enough with zfs to know what 'mountable volume' 
> means..
> is that a block device/lun that you're putting ext4 on?  Probably the 
> best
> results will come from creating a zfs *file system* (using the ZPL or
> whatever it is) and running ceph-osd on top of that.
>
> There is at least one open bug from someone having problems there, 
> but
> we'd very much like to sort out the problem.
>
> sage

Re: [ceph-commit] Ceph Zfs

[Dan Mick]

On 10/25/2012 09:46 PM, Raghunandhan wrote:
> Hi Sage,
>
> Thanks for replying back, Once a zpool is created if i mount it on
> /var/lib/ceph/osd/ceph-0 the cephfs doesnt recognize it as a superblock
> and hence it fails,

I assume you mean "once a zfs is created"?  One can't mount zpools, can one?

> Im trying to build this on our cloud storage since
> btrfs has not been stable nor they have come up with online dedup i have
> no other choice for now to work with zfs ceph which makes sense.
>
> So what i exactly did was created a zpool store
> 1 Then used the same store and made a block device from it using zfs create
> 2 Once the zfs create was successful i was able to format with ext4
> using xattr
> 3 On top of it was the ceph
>
> Following this process doesnt make sense because of multiple layer on
> the storage and the ceph consumes a lot of RAM and cpu cycles which ends
> up in kernel hung task. It would be great if there is a way i could
> directly use the zfs pool with ceph and make it work.

Have you actually tried making a zfs filesystem in the zpool, and using 
that as backing store for the osd?

>
> ---
> Regards,
> Raghunandhan.G
> IIHT Cloud Solutions Pvt. Ltd.
> #15, 4th Floor, 'A' Wing, Sri Lakshmi Complex,
> St. Marks Road, Bangalore - 560 001, India
>
> On 25-10-2012 22:06, Sage Weil wrote:
>> [moved to ceph-devel]
>>
>> On Thu, 25 Oct 2012, Raghunandhan wrote:
>>> Hi All,
>>>
>>> I have been working around ceph quite a long and trying to stitch zfs
>>> with
>>> ceph. I was able to do it to certain extent as follows:
>>> 1. zpool creation
>>> 2. set dedup
>>> 3. create a mountable volume of zfs (zfs create)
>>> 4. format the volume with ext4 and enabling xattr
>>> 5. mkcephfs on the volume
>>>
>>> This actually works and dedup is perfect. But i need to avoid
>>> multiple layers
>>> on the storage since the performance is very slow and the kernel timeout
>>> occurs often for a 8GB RAM. I want to test the performance between
>>> btrfs and
>>> zfs. I want to avoid the above multiple layering on storage and make
>>> the ceph
>>> cluster aware of zfs. Let me know if anyone has workaround this.
>>
>> I'm not familiar enough with zfs to know what 'mountable volume' means..
>> is that a block device/lun that you're putting ext4 on?  Probably the
>> best
>> results will come from creating a zfs *file system* (using the ZPL or
>> whatever it is) and running ceph-osd on top of that.
>>
>> There is at least one open bug from someone having problems there, but
>> we'd very much like to sort out the problem.
>>
>> sage
>
> --
> To unsubscribe from this list: send the line "unsubscribe ceph-devel" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [ceph-commit] Ceph Zfs

[Raghunandhan]

Hi Dan,

Yes once a zpool is created there is a way we can use the zpool and 
make a partition out of it using "zfs create -V". The newly created 
partition will be available on fdisk. Later the same partition can be 
formatted with ext4 and used with ceph-osd.

I have also tried using a zfs filesystem in the zpool and mapped it 
with osd. When i run mkcephfs i get "error creating empty object store 
/osd.0: (22) invalid argument

== osd.0 ===
2012-10-27 10:40:33.939961 7f6e6165d780 -1 filestore(/osd.0) mkjournal 
error creating journal on /osd.0/journal: (22) Invalid argument
2012-10-27 10:40:33.939981 7f6e6165d780 -1 OSD::mkfs: FileStore::mkfs 
failed with error -22
2012-10-27 10:40:33.940036 7f6e6165d780 -1  ** ERROR: error creating 
empty object store in /osd.0: (22) Invalid argument
failed: '/sbin/mkcephfs -d /tmp/mkcephfs.3zqOx7Btvl --init-daemon 
osd.0'

---
Regards,
Raghunandhan.G
IIHT Cloud Solutions Pvt. Ltd.
#15, 4th Floor, 'A' Wing, Sri Lakshmi Complex,
St. Marks Road, Bangalore - 560 001, India

On 27-10-2012 02:08, Dan Mick wrote:
> On 10/25/2012 09:46 PM, Raghunandhan wrote:
>> Hi Sage,
>>
>> Thanks for replying back, Once a zpool is created if i mount it on
>> /var/lib/ceph/osd/ceph-0 the cephfs doesnt recognize it as a 
>> superblock
>> and hence it fails,
>
> I assume you mean "once a zfs is created"?  One can't mount zpools, 
> can one?
>
>> Im trying to build this on our cloud storage since
>> btrfs has not been stable nor they have come up with online dedup i 
>> have
>> no other choice for now to work with zfs ceph which makes sense.
>>
>> So what i exactly did was created a zpool store
>> 1 Then used the same store and made a block device from it using zfs 
>> create
>> 2 Once the zfs create was successful i was able to format with ext4
>> using xattr
>> 3 On top of it was the ceph
>>
>> Following this process doesnt make sense because of multiple layer 
>> on
>> the storage and the ceph consumes a lot of RAM and cpu cycles which 
>> ends
>> up in kernel hung task. It would be great if there is a way i could
>> directly use the zfs pool with ceph and make it work.
>
> Have you actually tried making a zfs filesystem in the zpool, and
> using that as backing store for the osd?
>
>>
>> ---
>> Regards,
>> Raghunandhan.G
>> IIHT Cloud Solutions Pvt. Ltd.
>> #15, 4th Floor, 'A' Wing, Sri Lakshmi Complex,
>> St. Marks Road, Bangalore - 560 001, India
>>
>> On 25-10-2012 22:06, Sage Weil wrote:
>>> [moved to ceph-devel]
>>>
>>> On Thu, 25 Oct 2012, Raghunandhan wrote:
>>>> Hi All,
>>>>
>>>> I have been working around ceph quite a long and trying to stitch 
>>>> zfs
>>>> with
>>>> ceph. I was able to do it to certain extent as follows:
>>>> 1. zpool creation
>>>> 2. set dedup
>>>> 3. create a mountable volume of zfs (zfs create)
>>>> 4. format the volume with ext4 and enabling xattr
>>>> 5. mkcephfs on the volume
>>>>
>>>> This actually works and dedup is perfect. But i need to avoid
>>>> multiple layers
>>>> on the storage since the performance is very slow and the kernel 
>>>> timeout
>>>> occurs often for a 8GB RAM. I want to test the performance between
>>>> btrfs and
>>>> zfs. I want to avoid the above multiple layering on storage and 
>>>> make
>>>> the ceph
>>>> cluster aware of zfs. Let me know if anyone has workaround this.
>>>
>>> I'm not familiar enough with zfs to know what 'mountable volume' 
>>> means..
>>> is that a block device/lun that you're putting ext4 on?  Probably 
>>> the
>>> best
>>> results will come from creating a zfs *file system* (using the ZPL 
>>> or
>>> whatever it is) and running ceph-osd on top of that.
>>>
>>> There is at least one open bug from someone having problems there, 
>>> but
>>> we'd very much like to sort out the problem.
>>>
>>> sage
>>
>> --
>> To unsubscribe from this list: send the line "unsubscribe 
>> ceph-devel" in
>> the body of a message to majordomo@vger.kernel.org
>> More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [ceph-commit] Ceph Zfs

[Sage Weil]

On Sat, 27 Oct 2012, Raghunandhan wrote:
> Hi Dan,
> 
> Yes once a zpool is created there is a way we can use the zpool and make a
> partition out of it using "zfs create -V". The newly created partition will be
> available on fdisk. Later the same partition can be formatted with ext4 and
> used with ceph-osd.
> 
> I have also tried using a zfs filesystem in the zpool and mapped it with osd.
> When i run mkcephfs i get "error creating empty object store /osd.0: (22)
> invalid argument
> 
> == osd.0 ===
> 2012-10-27 10:40:33.939961 7f6e6165d780 -1 filestore(/osd.0) mkjournal error
> creating journal on /osd.0/journal: (22) Invalid argument
> 2012-10-27 10:40:33.939981 7f6e6165d780 -1 OSD::mkfs: FileStore::mkfs failed
> with error -22
> 2012-10-27 10:40:33.940036 7f6e6165d780 -1  ** ERROR: error creating empty
> object store in /osd.0: (22) Invalid argument
> failed: '/sbin/mkcephfs -d /tmp/mkcephfs.3zqOx7Btvl --init-daemon osd.0'

Can you generate a log with 'debug filestore = 20' of this happening so we 
can see exactly which operation is failing with -EINVAL?  There is 
probably some ioctl or syscall that is going awry.

Thanks!
sage


> 
> ---
> Regards,
> Raghunandhan.G
> IIHT Cloud Solutions Pvt. Ltd.
> #15, 4th Floor, 'A' Wing, Sri Lakshmi Complex,
> St. Marks Road, Bangalore - 560 001, India
> 
> On 27-10-2012 02:08, Dan Mick wrote:
> > On 10/25/2012 09:46 PM, Raghunandhan wrote:
> > > Hi Sage,
> > > 
> > > Thanks for replying back, Once a zpool is created if i mount it on
> > > /var/lib/ceph/osd/ceph-0 the cephfs doesnt recognize it as a superblock
> > > and hence it fails,
> > 
> > I assume you mean "once a zfs is created"?  One can't mount zpools, can one?
> > 
> > > Im trying to build this on our cloud storage since
> > > btrfs has not been stable nor they have come up with online dedup i have
> > > no other choice for now to work with zfs ceph which makes sense.
> > > 
> > > So what i exactly did was created a zpool store
> > > 1 Then used the same store and made a block device from it using zfs
> > > create
> > > 2 Once the zfs create was successful i was able to format with ext4
> > > using xattr
> > > 3 On top of it was the ceph
> > > 
> > > Following this process doesnt make sense because of multiple layer on
> > > the storage and the ceph consumes a lot of RAM and cpu cycles which ends
> > > up in kernel hung task. It would be great if there is a way i could
> > > directly use the zfs pool with ceph and make it work.
> > 
> > Have you actually tried making a zfs filesystem in the zpool, and
> > using that as backing store for the osd?
> > 
> > > 
> > > ---
> > > Regards,
> > > Raghunandhan.G
> > > IIHT Cloud Solutions Pvt. Ltd.
> > > #15, 4th Floor, 'A' Wing, Sri Lakshmi Complex,
> > > St. Marks Road, Bangalore - 560 001, India
> > > 
> > > On 25-10-2012 22:06, Sage Weil wrote:
> > > > [moved to ceph-devel]
> > > > 
> > > > On Thu, 25 Oct 2012, Raghunandhan wrote:
> > > > > Hi All,
> > > > > 
> > > > > I have been working around ceph quite a long and trying to stitch zfs
> > > > > with
> > > > > ceph. I was able to do it to certain extent as follows:
> > > > > 1. zpool creation
> > > > > 2. set dedup
> > > > > 3. create a mountable volume of zfs (zfs create)
> > > > > 4. format the volume with ext4 and enabling xattr
> > > > > 5. mkcephfs on the volume
> > > > > 
> > > > > This actually works and dedup is perfect. But i need to avoid
> > > > > multiple layers
> > > > > on the storage since the performance is very slow and the kernel
> > > > > timeout
> > > > > occurs often for a 8GB RAM. I want to test the performance between
> > > > > btrfs and
> > > > > zfs. I want to avoid the above multiple layering on storage and make
> > > > > the ceph
> > > > > cluster aware of zfs. Let me know if anyone has workaround this.
> > > > 
> > > > I'm not familiar enough with zfs to know what 'mountable volume' means..
> > > > is that a block device/lun that you're putting ext4 on?  Probably the
> > > > best
> > > > results will come from creating a zfs *file system* (using the ZPL or
> > > > whatever it is) and running ceph-osd on top of that.
> > > > 
> > > > There is at least one open bug from someone having problems there, but
> > > > we'd very much like to sort out the problem.
> > > > 
> > > > sage
> > > 
> > > --
> > > To unsubscribe from this list: send the line "unsubscribe ceph-devel" in
> > > the body of a message to majordomo@vger.kernel.org
> > > More majordomo info at  http://vger.kernel.org/majordomo-info.html
> 
> 

Re: [ceph-commit] Ceph Zfs

[Raghunandhan]

On 27-10-2012 23:45, Sage Weil wrote:
> On Sat, 27 Oct 2012, Raghunandhan wrote:
>> Hi Dan,
>>
>> Yes once a zpool is created there is a way we can use the zpool and 
>> make a
>> partition out of it using "zfs create -V". The newly created 
>> partition will be
>> available on fdisk. Later the same partition can be formatted with 
>> ext4 and
>> used with ceph-osd.
>>
>> I have also tried using a zfs filesystem in the zpool and mapped it 
>> with osd.
>> When i run mkcephfs i get "error creating empty object store /osd.0: 
>> (22)
>> invalid argument
>>
>> == osd.0 ===
>> 2012-10-27 10:40:33.939961 7f6e6165d780 -1 filestore(/osd.0) 
>> mkjournal error
>> creating journal on /osd.0/journal: (22) Invalid argument
>> 2012-10-27 10:40:33.939981 7f6e6165d780 -1 OSD::mkfs: 
>> FileStore::mkfs failed
>> with error -22
>> 2012-10-27 10:40:33.940036 7f6e6165d780 -1  ** ERROR: error creating 
>> empty
>> object store in /osd.0: (22) Invalid argument
>> failed: '/sbin/mkcephfs -d /tmp/mkcephfs.3zqOx7Btvl --init-daemon 
>> osd.0'
>
> Can you generate a log with 'debug filestore = 20' of this happening 
> so we
> can see exactly which operation is failing with -EINVAL?  There is
> probably some ioctl or syscall that is going awry.
>
> Thanks!
> sage

Above issue was rectified with journal dio=false in ceph.conf

ceph status when used with zfs filesystem OSD dies on one node but its 
still up on other node.

# ceph -s
    health HEALTH_WARN 407 pgs degraded; 169 pgs down; 169 pgs peering; 
15 pgs recovering; 323 pgs stuck unclean; recovery 38/42 degraded 
(90.476%); 19/21 unfound (90.476%); 1/2 in osds are down
    monmap e1: 2 mons at {a=11.0.0.2:6789/0,b=11.0.0.3:6789/0}, election 
epoch 4, quorum 0,1 a,b
    osdmap e7: 2 osds: 1 up, 2 in
     pgmap v10: 576 pgs: 15 active+recovering+degraded, 169 
down+peering, 392 active+degraded; 8059 bytes data, 1004 MB used, 85683 
MB / 86687 MB avail; 38/42 degraded (90.476%); 19/21 unfound (90.476%)
    mdsmap e5: 1/1/1 up {0=a=up:active}, 1 up:standby

Log file generated when 2 osd's where up and later it went down.

2012-10-27 11:14:07.152741 mon.0 11.0.0.2:6789/0 27 : [INF] osdmap e5: 
2 osds: 2 up, 2 in
2012-10-27 11:14:07.192719 mon.0 11.0.0.2:6789/0 28 : [INF] pgmap v6: 
576 pgs: 576 creating; 0 bytes data, 0 KB used, 0 KB / 0 KB avail
2012-10-27 11:14:12.007671 mon.0 11.0.0.2:6789/0 29 : [INF] pgmap v7: 
576 pgs: 272 creating, 43 active, 253 active+clean, 8 active+recovering; 
1243 bytes data, 1003 MB used, 85684 MB / 86687 MB avail; 9/18 degraded 
(50.000%)
2012-10-27 11:14:32.014302 mon.0 11.0.0.2:6789/0 30 : [DBG] osd.0 
11.0.0.2:6801/24250 reported failed by osd.1 11.0.0.3:6801/8443
2012-10-27 11:14:37.033547 mon.0 11.0.0.2:6789/0 31 : [DBG] osd.0 
11.0.0.2:6801/24250 reported failed by osd.1 11.0.0.3:6801/8443
2012-10-27 11:14:42.060678 mon.0 11.0.0.2:6789/0 32 : [DBG] osd.0 
11.0.0.2:6801/24250 reported failed by osd.1 11.0.0.3:6801/8443
2012-10-27 11:14:42.060827 mon.0 11.0.0.2:6789/0 33 : [INF] osd.0 
11.0.0.2:6801/24250 failed (3 reports from 1 peers after 30.046376 >= 
grace 20.000000)
2012-10-27 11:14:42.157536 mon.0 11.0.0.2:6789/0 34 : [INF] osdmap e6: 
2 osds: 1 up, 2 in

osd.0 dies after a while:

2012-10-27 11:19:46.751562 mon.0 11.0.0.2:6789/0 40 : [INF] osd.0 out 
(down for 304.604259)
2012-10-27 11:19:46.785574 mon.0 11.0.0.2:6789/0 41 : [INF] osdmap e8: 
2 osds: 1 up, 1 in
2012-10-27 11:19:46.811588 mon.0 11.0.0.2:6789/0 42 : [INF] pgmap v12: 
576 pgs: 15 active+recovering+degraded, 169 down+peering, 392 
active+degraded; 8059 bytes data, 1004 MB used, 85683 MB / 86687 MB 
avail; 38/42 degraded (90.476%); 19/21 unfound (90.476%)
2012-10-27 11:19:49.591172 mon.0 11.0.0.2:6789/0 43 : [INF] pgmap v13: 
576 pgs: 15 active+recovering+degraded, 169 down+peering, 392 
active+degraded; 8059 bytes data, 1004 MB used, 85683 MB / 86687 MB 
avail; 38/42 degraded (90.476%); 19/21 unfound (90.476%)
2012-10-27 11:20:04.671337 mon.0 11.0.0.2:6789/0 44 : [INF] pgmap v14: 
576 pgs: 15 active+recovering+degraded, 169 down+peering, 392 
active+degraded; 8059 bytes data, 1004 MB used, 85683 MB / 86687 MB 
avail; 38/42 degraded (90.476%); 19/21 unfound (90.476%)

status of osd.1 as of now:
2012-10-28 10:48:54.022338 osd.1 11.0.0.3:6801/8443 396978 : [WRN] slow 
request 84884.436995 seconds old, received at 2012-10-27 
11:14:09.585282: osd_op(mds.0.1:28 200.00000001 [write 131~671] 
1.6e5f474 RETRY) v4 currently delayed
2012-10-28 10:48:54.022343 osd.1 11.0.0.3:6801/8443 396979 : [WRN] slow 
request 84851.874118 seconds old, received at 2012-10-27 
11:14:42.148159: osd_op(mds.0.1:29 200.00000000 [writefull 0~84] 
1.844f3494 RETRY) v4 currently delayed
2012-10-28 10:48:54.022346 osd.1 11.0.0.3:6801/8443 396980 : [WRN] slow 
request 81939.241084 seconds old, received at 2012-10-27 
12:03:14.781193: osd_op(mds.0.1:30 200.00000001 [write 802~183] 
1.6e5f474) v4 currently delayed
2012-10-28 10:48:54.022350 osd.1 11.0.0.3:6801/8443 396981 : [WRN] slow 
request 81939.240915 seconds old, received at 2012-10-27 
12:03:14.781362: osd_op(mds.0.1:31 200.00000000 [writefull 0~84] 
1.844f3494) v4 currently delayed

---
Regards,
Raghunandhan.G

>>
>> ---
>> Regards,
>> Raghunandhan.G
>> IIHT Cloud Solutions Pvt. Ltd.
>> #15, 4th Floor, 'A' Wing, Sri Lakshmi Complex,
>> St. Marks Road, Bangalore - 560 001, India
>>
>> On 27-10-2012 02:08, Dan Mick wrote:
>> > On 10/25/2012 09:46 PM, Raghunandhan wrote:
>> > > Hi Sage,
>> > >
>> > > Thanks for replying back, Once a zpool is created if i mount it 
>> on
>> > > /var/lib/ceph/osd/ceph-0 the cephfs doesnt recognize it as a 
>> superblock
>> > > and hence it fails,
>> >
>> > I assume you mean "once a zfs is created"?  One can't mount 
>> zpools, can one?
>> >
>> > > Im trying to build this on our cloud storage since
>> > > btrfs has not been stable nor they have come up with online 
>> dedup i have
>> > > no other choice for now to work with zfs ceph which makes sense.
>> > >
>> > > So what i exactly did was created a zpool store
>> > > 1 Then used the same store and made a block device from it using 
>> zfs
>> > > create
>> > > 2 Once the zfs create was successful i was able to format with 
>> ext4
>> > > using xattr
>> > > 3 On top of it was the ceph
>> > >
>> > > Following this process doesnt make sense because of multiple 
>> layer on
>> > > the storage and the ceph consumes a lot of RAM and cpu cycles 
>> which ends
>> > > up in kernel hung task. It would be great if there is a way i 
>> could
>> > > directly use the zfs pool with ceph and make it work.
>> >
>> > Have you actually tried making a zfs filesystem in the zpool, and
>> > using that as backing store for the osd?
>> >
>> > >
>> > > ---
>> > > Regards,
>> > > Raghunandhan.G
>> > > IIHT Cloud Solutions Pvt. Ltd.
>> > > #15, 4th Floor, 'A' Wing, Sri Lakshmi Complex,
>> > > St. Marks Road, Bangalore - 560 001, India
>> > >
>> > > On 25-10-2012 22:06, Sage Weil wrote:
>> > > > [moved to ceph-devel]
>> > > >
>> > > > On Thu, 25 Oct 2012, Raghunandhan wrote:
>> > > > > Hi All,
>> > > > >
>> > > > > I have been working around ceph quite a long and trying to 
>> stitch zfs
>> > > > > with
>> > > > > ceph. I was able to do it to certain extent as follows:
>> > > > > 1. zpool creation
>> > > > > 2. set dedup
>> > > > > 3. create a mountable volume of zfs (zfs create)
>> > > > > 4. format the volume with ext4 and enabling xattr
>> > > > > 5. mkcephfs on the volume
>> > > > >
>> > > > > This actually works and dedup is perfect. But i need to 
>> avoid
>> > > > > multiple layers
>> > > > > on the storage since the performance is very slow and the 
>> kernel
>> > > > > timeout
>> > > > > occurs often for a 8GB RAM. I want to test the performance 
>> between
>> > > > > btrfs and
>> > > > > zfs. I want to avoid the above multiple layering on storage 
>> and make
>> > > > > the ceph
>> > > > > cluster aware of zfs. Let me know if anyone has workaround 
>> this.
>> > > >
>> > > > I'm not familiar enough with zfs to know what 'mountable 
>> volume' means..
>> > > > is that a block device/lun that you're putting ext4 on?  
>> Probably the
>> > > > best
>> > > > results will come from creating a zfs *file system* (using the 
>> ZPL or
>> > > > whatever it is) and running ceph-osd on top of that.
>> > > >
>> > > > There is at least one open bug from someone having problems 
>> there, but
>> > > > we'd very much like to sort out the problem.
>> > > >
>> > > > sage
>> > >
>> > > --
>> > > To unsubscribe from this list: send the line "unsubscribe 
>> ceph-devel" in
>> > > the body of a message to majordomo@vger.kernel.org
>> > > More majordomo info at  
>> http://vger.kernel.org/majordomo-info.html
>>
>>

Re: [ceph-commit] Ceph Zfs

[Raghunandhan]

ceph status when used with zfs filesystem osd dies.

# ceph -s
    health HEALTH_WARN 407 pgs degraded; 169 pgs down; 169 pgs peering; 
15 pgs recovering; 323 pgs stuck unclean; recovery 38/42 degraded 
(90.476%); 19/21 unfound (90.476%); 1/2 in osds are down
    monmap e1: 2 mons at {a=11.0.0.2:6789/0,b=11.0.0.3:6789/0}, election 
epoch 4, quorum 0,1 a,b
    osdmap e7: 2 osds: 1 up, 2 in
     pgmap v10: 576 pgs: 15 active+recovering+degraded, 169 
down+peering, 392 active+degraded; 8059 bytes data, 1004 MB used, 85683 
MB / 86687 MB avail; 38/42 degraded (90.476%); 19/21 unfound (90.476%)
    mdsmap e5: 1/1/1 up {0=a=up:active}, 1 up:standby

Log file generated when 2 osd's where up and later it went down.

2012-10-27 11:14:07.152741 mon.0 11.0.0.2:6789/0 27 : [INF] osdmap e5: 
2 osds: 2 up, 2 in
2012-10-27 11:14:07.192719 mon.0 11.0.0.2:6789/0 28 : [INF] pgmap v6: 
576 pgs: 576 creating; 0 bytes data, 0 KB used, 0 KB / 0 KB avail
2012-10-27 11:14:12.007671 mon.0 11.0.0.2:6789/0 29 : [INF] pgmap v7: 
576 pgs: 272 creating, 43 active, 253 active+clean, 8 active+recovering; 
1243 bytes data, 1003 MB used, 85684 MB / 86687 MB avail; 9/18 degraded 
(50.000%)
2012-10-27 11:14:32.014302 mon.0 11.0.0.2:6789/0 30 : [DBG] osd.0 
11.0.0.2:6801/24250 reported failed by osd.1 11.0.0.3:6801/8443
2012-10-27 11:14:37.033547 mon.0 11.0.0.2:6789/0 31 : [DBG] osd.0 
11.0.0.2:6801/24250 reported failed by osd.1 11.0.0.3:6801/8443
2012-10-27 11:14:42.060678 mon.0 11.0.0.2:6789/0 32 : [DBG] osd.0 
11.0.0.2:6801/24250 reported failed by osd.1 11.0.0.3:6801/8443
2012-10-27 11:14:42.060827 mon.0 11.0.0.2:6789/0 33 : [INF] osd.0 
11.0.0.2:6801/24250 failed (3 reports from 1 peers after 30.046376 >= 
grace 20.000000)
2012-10-27 11:14:42.157536 mon.0 11.0.0.2:6789/0 34 : [INF] osdmap e6: 
2 osds: 1 up, 2 in

2012-10-27 11:19:46.751562 mon.0 11.0.0.2:6789/0 40 : [INF] osd.0 out 
(down for 304.604259)
2012-10-27 11:19:46.785574 mon.0 11.0.0.2:6789/0 41 : [INF] osdmap e8: 
2 osds: 1 up, 1 in
2012-10-27 11:19:46.811588 mon.0 11.0.0.2:6789/0 42 : [INF] pgmap v12: 
576 pgs: 15 active+recovering+degraded, 169 down+peering, 392 
active+degraded; 8059 bytes data, 1004 MB used, 85683 MB / 86687 MB 
avail; 38/42 degraded (90.476%); 19/21 unfound (90.476%)
2012-10-27 11:19:49.591172 mon.0 11.0.0.2:6789/0 43 : [INF] pgmap v13: 
576 pgs: 15 active+recovering+degraded, 169 down+peering, 392 
active+degraded; 8059 bytes data, 1004 MB used, 85683 MB / 86687 MB 
avail; 38/42 degraded (90.476%); 19/21 unfound (90.476%)
2012-10-27 11:20:04.671337 mon.0 11.0.0.2:6789/0 44 : [INF] pgmap v14: 
576 pgs: 15 active+recovering+degraded, 169 down+peering, 392 
active+degraded; 8059 bytes data, 1004 MB used, 85683 MB / 86687 MB 
avail; 38/42 degraded (90.476%); 19/21 unfound (90.476%)


---
Regards,
Raghunandhan.G


On 27-10-2012 02:08, Dan Mick wrote:
> On 10/25/2012 09:46 PM, Raghunandhan wrote:
>> Hi Sage,
>>
>> Thanks for replying back, Once a zpool is created if i mount it on
>> /var/lib/ceph/osd/ceph-0 the cephfs doesnt recognize it as a 
>> superblock
>> and hence it fails,
>
> I assume you mean "once a zfs is created"?  One can't mount zpools, 
> can one?
>
>> Im trying to build this on our cloud storage since
>> btrfs has not been stable nor they have come up with online dedup i 
>> have
>> no other choice for now to work with zfs ceph which makes sense.
>>
>> So what i exactly did was created a zpool store
>> 1 Then used the same store and made a block device from it using zfs 
>> create
>> 2 Once the zfs create was successful i was able to format with ext4
>> using xattr
>> 3 On top of it was the ceph
>>
>> Following this process doesnt make sense because of multiple layer 
>> on
>> the storage and the ceph consumes a lot of RAM and cpu cycles which 
>> ends
>> up in kernel hung task. It would be great if there is a way i could
>> directly use the zfs pool with ceph and make it work.
>
> Have you actually tried making a zfs filesystem in the zpool, and
> using that as backing store for the osd?
>
>>
>> ---
>> Regards,
>> Raghunandhan.G
>> IIHT Cloud Solutions Pvt. Ltd.
>> #15, 4th Floor, 'A' Wing, Sri Lakshmi Complex,
>> St. Marks Road, Bangalore - 560 001, India
>>
>> On 25-10-2012 22:06, Sage Weil wrote:
>>> [moved to ceph-devel]
>>>
>>> On Thu, 25 Oct 2012, Raghunandhan wrote:
>>>> Hi All,
>>>>
>>>> I have been working around ceph quite a long and trying to stitch 
>>>> zfs
>>>> with
>>>> ceph. I was able to do it to certain extent as follows:
>>>> 1. zpool creation
>>>> 2. set dedup
>>>> 3. create a mountable volume of zfs (zfs create)
>>>> 4. format the volume with ext4 and enabling xattr
>>>> 5. mkcephfs on the volume
>>>>
>>>> This actually works and dedup is perfect. But i need to avoid
>>>> multiple layers
>>>> on the storage since the performance is very slow and the kernel 
>>>> timeout
>>>> occurs often for a 8GB RAM. I want to test the performance between
>>>> btrfs and
>>>> zfs. I want to avoid the above multiple layering on storage and 
>>>> make
>>>> the ceph
>>>> cluster aware of zfs. Let me know if anyone has workaround this.
>>>
>>> I'm not familiar enough with zfs to know what 'mountable volume' 
>>> means..
>>> is that a block device/lun that you're putting ext4 on?  Probably 
>>> the
>>> best
>>> results will come from creating a zfs *file system* (using the ZPL 
>>> or
>>> whatever it is) and running ceph-osd on top of that.
>>>
>>> There is at least one open bug from someone having problems there, 
>>> but
>>> we'd very much like to sort out the problem.
>>>
>>> sage
>>
>> --
>> To unsubscribe from this list: send the line "unsubscribe 
>> ceph-devel" in
>> the body of a message to majordomo@vger.kernel.org
>> More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [ceph-commit] Ceph Zfs

[Raghunandhan]

Also the open bug which is pending i have tried with it. Ceph-osd 
starts up with zfs volume after the ceph service is up in sometime the 
osd's stop working. I have been working around with releases from 
ceph-0.30 till the latest 0.54 to check with zfs compatibility.

Kindly let me know if this can happen in any way it would become a 
breakthrough in our storage design until btrfs becomes stable.

---
Regards,
Raghunandhan.G
IIHT Cloud Solutions Pvt. Ltd.
#15, 4th Floor, 'A' Wing, Sri Lakshmi Complex,
St. Marks Road, Bangalore - 560 001, India

On 25-10-2012 22:06, Sage Weil wrote:
> [moved to ceph-devel]
>
> On Thu, 25 Oct 2012, Raghunandhan wrote:
>> Hi All,
>>
>> I have been working around ceph quite a long and trying to stitch 
>> zfs with
>> ceph. I was able to do it to certain extent as follows:
>> 1. zpool creation
>> 2. set dedup
>> 3. create a mountable volume of zfs (zfs create)
>> 4. format the volume with ext4 and enabling xattr
>> 5. mkcephfs on the volume
>>
>> This actually works and dedup is perfect. But i need to avoid 
>> multiple layers
>> on the storage since the performance is very slow and the kernel 
>> timeout
>> occurs often for a 8GB RAM. I want to test the performance between 
>> btrfs and
>> zfs. I want to avoid the above multiple layering on storage and make 
>> the ceph
>> cluster aware of zfs. Let me know if anyone has workaround this.
>
> I'm not familiar enough with zfs to know what 'mountable volume' 
> means..
> is that a block device/lun that you're putting ext4 on?  Probably the 
> best
> results will come from creating a zfs *file system* (using the ZPL or
> whatever it is) and running ceph-osd on top of that.
>
> There is at least one open bug from someone having problems there, 
> but
> we'd very much like to sort out the problem.
>
> sage