Re: RHEL6.6 dm-cache and dm-era

[Heinz Mauelshagen <heinzm@xxxxxxxxxx>]

On 09/29/2014 04:31 AM, Romu Hu wrote:

On 2014/9/26 18:33, Heinz Mauelshagen wrote:

Romu,

did you initialize the metadata device before trying to create the era target?
Zeroes in the first KBs will do.

Thanks Heinz, it solved the problem.  Is it the proper way to initialize the metadata device of an era target or just a workaround?

It is the proper way to address it when using dmsetup.

MInd you dmsetup is a low-level tool which is very handy for testing/evatuation
like in your use case.

Ultimately dm-era will be supported by lvm and eventually other higher level tools,
which will be in charge of doing metadata device initialization.

Heinz

Thanks
Romu

Yes, check for NULL in era_destroy is needed to  handle the error path you hit properly.

Heinz

On 09/26/2014 05:16 AM, Romu Hu wrote:

On 2014/9/25 22:57, Heinz Mauelshagen wrote:

Romu,

"dmsetup create MyEra --table "0 20971520 /dev/mapper/mpathbp1 /dev/mapper/mpathap1 8"

should read

dmsetup create MyEra --table "0 20971520 era /dev/mapper/mpathbp1 /dev/mapper/mpathap1 8"

Sorry my bad.

I ran 'dmsetup create MyEra --table "0 20971520 era /dev/mapper/mpathbp1 /dev/mapper/mpathap1 8"' but kernel (2.6.32-494.el6.i686) oopsed:

Sep 25 18:34:11 localhost kernel: device-mapper: era: sb_check failed: magic 0: wanted 2126579579
Sep 25 18:34:11 localhost kernel: device-mapper: block manager: superblock validator check failed for block 0
Sep 25 18:34:11 localhost kernel: device-mapper: era: couldn't read_lock superblock
Sep 25 18:34:11 localhost kernel: BUG: unable to handle kernel NULL pointer dereference at 00000008
Sep 25 18:34:11 localhost kernel: IP: [<f7e2b3c7>] era_destroy+0x7/0x60 [dm_era]Sep 25 18:34:11 localhost kernel: *pdpt = 0000000032457001 *pde = 00000003fb5fa067
Sep 25 18:34:11 localhost kernel: Oops: 0000 [#1] SMP
Sep 25 18:34:11 localhost kernel: last sysfs file: /sys/module/dm_persistent_data/initstate

Perhaps this should be cherry-picked?
https://git.kernel.org/cgit/linux/kernel/git/torvalds/linux.git/commit/?id=989f26f5ad308f40a95f280bf9cd75e558d4f18d

The log says sb_check failed, anything wrong with the dmsetup command parameters?

Thanks
Romu

On 09/25/2014 04:02 PM, Romu wrote:

2014-09-25 18:19 GMT+08:00 Heinz Mauelshagen <heinzm@xxxxxxxxxx>:

On 09/25/2014 08:10 AM, Romu Hu wrote:

On 2014/9/24 12:10, Brassow Jonathan wrote:

On Sep 22, 2014, at 7:35 AM, Romu wrote:

I tried the following command to set up my era target but the command immediately panics the system and the system reboots.

# dmsetup create MyEra --table "0 41941903 era /dev/mapper/VG-CacheDataLV_cmeta /dev/mapper/VG-OriginLV 4096"

The metadata dev and the origin dev are all part of a LVM cache LV.  VG-CacheDataLV_cmeta is the cache metadata LV on the smaller and faster device, VG-OriginLV is the origin LV on the faster and slower device, 41941903 is the total sector number of the device of OriginLV (the LV takes 100% space of the device), 4096 is the block size of OriginLV, I have run 'mkfs.ext4 /dev/mapper/VG-OriginLV' before running the dmsetup command.

Below is the message in /var/log/messages after running the dmsetup comnmand:

kernel: device-mapper: era: sb_check failed: magic 1623043: wanted 2126579579

kernel: device-mapper: block manager: superblock validator check failed for block 0

kernel: device-mapper: era: couldn't read_lock superblock

Any idea?

Sorry, I haven't used dm-era yet.  However, it does appear that you are perhaps specifying the wrong devices when creating the era device?  Looks like you might be allowing the era target and the cache target to use the same metadata area at the same time - causing them to corrupt each other's metadata area?

 brassow

I think the dmsetup command to set up an era target should be something like this:

# dmsetup create MyEra --table "0 sector_number era metadeta_dev origin_dev block_size"

My questions:

1) How to calculate sector_number?  According to https://www.kernel.org/doc/Documentation/device-mapper/era.txt, I guess it should be (4 * nr_blocks) bytes + buffers, but what is nr_blocks and what is buffers?

No calculation: it's the size of your era target in sectors. Typically "blockdev --getsz origin_dev"

2) Any documentation for dmsetup tables?

Look for examples in the kernel source trees Documentaion/device.mapper.

table line syntax is: "start_sector length_in_sectors <target> <target_arguments{0,N}>"

3) Both my metadata_dev and origin_dev are 10G partitions, is this all right?

Metadata device size is plenty but that depends on how many eras with how many updates you want to have.

4) My origin_dev is a ext4 filesystem with a block size of 4096, so the block_size in the command line should also be 4096?

You may use 4096 = _8_ sectors, you used 4096 sectors = 2MB below.
It's the granularity the era target housekeeps blocks for.

I tried the following command:

# dmsetup create MyEra --table "0 160000 era /dev/mapper/mpathap1 /dev/mapper/mpathbp1 4096"

May not be  same physical device for data and metadata!
If it is, thta'd explain your oops.

# dmsetup create MyEra --table "0 $(blockdev --getsz /dev/mapper/mpathbp) era whatever_disctinct_metadata_device /dev/mapper/mpathbp1 8"

Would use 8 sector block size (which is tiny!) with disctinct metadata and data devices (presumabyl your ext4 sits on /dev/mapper/mpathbp1)

Heinz

Heinz, thank you for your help!

My origin dev is 10G so total sector number is 20971520.  /dev/mapper/mpathbp1 (/dev/dm-6) is the metadata dev, /dev/mapper/mpathap1 (/dev/dm-7) is the origin dev, they are on different LUNs.

I tried the following commands:

# dmsetup create MyEra --table "0 20971520 /dev/mapper/mpathbp1 /dev/mapper/mpathap1 8"

# Target type name /dev/mapper/mpathbp1 is too long.

# Command failed

# dmsetup create MyEra --table "0 20971520 /dev/dm-6 /dev/dm-7 8"

# device-mapper: reload ioctl on MyEra failed: Invalid argument

# Command failed

And when executing the second command I see the following in /var/log/messages:

Sep 25 06:37:48 localhost kernel: device-mapper: table: 253:8: /dev/dm-6: unknown target type

Sep 25 06:37:48 localhost kernel: device-mapper: ioctl: error adding target to table

Sep 25 06:37:48 localhost multipathd: dm-8: remove map (uevent)

Sep 25 06:37:48 localhost multipathd: dm-8: devmap not registered, can't remove

Sep 25 06:37:48 localhost multipathd: dm-8: remove map (uevent)

Sep 25 06:37:48 localhost multipathd: dm-8: devmap not registered, can't remove

Then I run 'mkfs.ext4 /dev/dm-6' and tried the second command again but got the same result.

Any idea?

Thanks

Romu

--
dm-devel mailing list
dm-devel@xxxxxxxxxx
https://www.redhat.com/mailman/listinfo/dm-devel

--
dm-devel mailing list
dm-devel@xxxxxxxxxx
https://www.redhat.com/mailman/listinfo/dm-devel

--
dm-devel mailing list
dm-devel@xxxxxxxxxx
https://www.redhat.com/mailman/listinfo/dm-devel

-- 
===============================================================
Heinz Mauelshagen                               +49 2626 141200
Consulting Development Engineer             FAX +49 2626 924446
Red Hat GmbH
Am Sonnenhang 11
56242 Marienrachdorf
Germany                                       heinzm@xxxxxxxxxx
===============================================================

--
dm-devel mailing list
dm-devel@xxxxxxxxxx
https://www.redhat.com/mailman/listinfo/dm-devel

--
dm-devel mailing list
dm-devel@xxxxxxxxxx
https://www.redhat.com/mailman/listinfo/dm-devel

--
dm-devel mailing list
dm-devel@xxxxxxxxxx
https://www.redhat.com/mailman/listinfo/dm-devel