kdump/kexec vmcore size doubles each time

kdump/kexec vmcore size doubles each time

[joe shmoe]

Hi,
I am trying to configure kexec/kdump on 2.6.35-14. 


$> kexec --version
kexec-tools-testing 20080330 released 30th March 2007


I did manage to configure kdump to use raw device (/dev/sda7) for saving the dump from within the initrd.

When the production kernel boots, it copies core from /dev/sda7 to /var/crash/ path.
I see that the size of data copied over from /dev/sda7 to /var/crash almost doubles with every crash (370MB to 620MB to 1.2 GB to 2.1GB). 


The size returns back to normal, should there be a "normal" reboot cycle anytime. After that every subsequent crash again doubles the size. This is very baffling. 

On a related note, I wanted to understand if there are any known issues with having swap device as a dump location. Does kdump start 
early enough to ensure that vmcore can be safely copied off swap 
/dev/sda7 and THEN mkswap/swapon commands set up the swap space?


Thanks much.

Re: kdump/kexec vmcore size doubles each time

[Daniel Kiper]

On Thu, Jul 19, 2012 at 03:08:44PM -0700, joe shmoe wrote:
> Hi,
> I am trying to configure kexec/kdump on 2.6.35-14.
>
>
> $> kexec --version
> kexec-tools-testing 20080330 released 30th March 2007

[...]

It is very ancient version. Please use at least Ver. 2.0.3
(http://kernel.org/pub/linux/utils/kernel/kexec/) or current
development version (git://git.kernel.org/pub/scm/utils/kernel/kexec/kexec-tools.git).

Daniel

Re: kdump/kexec vmcore size doubles each time

[joe shmoe]

Hi,
Thank you for the reply. I ended up opting for writing to filesystem rather than raw_device and this seems to have helped me sidestep the problem for now. Unfortunately, I can't quite make a jump to later sources.

I do have another question, I see KDUMP_KEEP_OLD_DUMPS mentioned to automatically clean up prior dumps (for ex: http://dedoimedo.com/computers/kdump.html and other SUSE docs).
However, I cannot find any logic in sources to do this (and sure enough I don't see older cores in my /var/crash/ path clean up as new cores are dumped).

Could someone please comment if:
(i) KDUMP_KEEP_OLD_DUMPS is unsupported or not. If it IS supported, could someone please help with the relevant source code so I could debug why it isn't working? Yes, I did download kexec-tools git but cannot find anything relevant.

(ii) If there is NO explicit source code to achieve this -- then do folks achieve such a cleanup through KDUMP_PRE and KDUMP_POST scripts? If yes, is there a standard-ish script out there to achieve this?

Thanks much.

----- Original Message -----
From: Daniel Kiper <dkiper@net-space.pl>
To: joe shmoe <joeshmoeypeter@yahoo.com>
Cc: "linux-kernel@vger.kernel.org" <linux-kernel@vger.kernel.org>; daniel.kiper@oracle.com
Sent: Friday, July 20, 2012 2:36 AM
Subject: Re: kdump/kexec vmcore size doubles each time

On Thu, Jul 19, 2012 at 03:08:44PM -0700, joe shmoe wrote:
> Hi,
> I am trying to configure kexec/kdump on 2.6.35-14.
>
>
> $> kexec --version
> kexec-tools-testing 20080330 released 30th March 2007

[...]

It is very ancient version. Please use at least Ver. 2.0.3
(http://kernel.org/pub/linux/utils/kernel/kexec/) or current
development version (git://git.kernel.org/pub/scm/utils/kernel/kexec/kexec-tools.git).

Daniel

Re: kdump/kexec vmcore size doubles each time

[Daniel Kiper]

On Mon, Jul 23, 2012 at 04:48:19PM -0700, joe shmoe wrote:
> Hi,
> Thank you for the reply.

You are welcome.

> I ended up opting for writing to filesystem rather than
> raw_device and this seems to have helped me sidestep the
> problem for now.

Good move.

> Unfortunately, I can't quite make a jump to later sources.
>
> I do have another question, I see KDUMP_KEEP_OLD_DUMPS mentioned
> to automatically clean up prior dumps (for ex: http://dedoimedo.com/computers/kdump.html
> and other SUSE docs).
> However, I cannot find any logic in sources to do this (and sure
> enough I don't see older cores in my /var/crash/ path clean up
> as new cores are dumped).
>
> Could someone please comment if:
> (i) KDUMP_KEEP_OLD_DUMPS is unsupported or not. If it IS supported,
> could someone please help with the relevant source code so I could
> debug why it isn't working? Yes, I did download kexec-tools git but
> cannot find anything relevant.
>
> (ii) If there is NO explicit source code to achieve this -- then
> do folks achieve such a cleanup through KDUMP_PRE and KDUMP_POST scripts?
> If yes, is there a standard-ish script out there to achieve this?

Sorry but I could not help because I do not use SUSE.
But maybe guys from SUSE could point out something.

Daniel