Re: Fixes for DDF test case (race conditions in mdmon)

[NeilBrown <neilb@suse.de>]

[-- Attachment #1: Type: text/plain, Size: 1836 bytes --]

On Fri, 25 Oct 2013 12:07:31 +0200 mwilck@arcor.de wrote:

> This is a series of patches to fix the DDF tests case(tests/10ddf-create).
> 
> Patch 3 fixes a problem that was implicitly introduced by commit 
> c1ea5a98 - "mdadm -Db" would now print UUIDs for subarrays, but the
> wrong ones.
> 
> Patch 5-7 are the interesting part. The DDF test case kept failing for
> me right there where a comment already says "# This failed once. 
> The raid5 was resyncing". I tracked this down to race conditions
> between mdmon and the kernel when arrays are stopped - the kernel
> cleans up in sysfs and reading the attributes from there is unreliable.
> Files may have already vanished, read() may fail or even succeed
> with incorrect results.
> 
> The other patches are hopefully self-explaining.
> 
> Regards
> Martin
> 
> Martin Wilck (8):
>   DDF: __write_ddf_structure: Fix wrong reference to ddf->primary
>   DDF: __write_init_super_ddf: just use seq number of active header
>   DDF: brief_detail_super_ddf: print correct UUID for subarrays
>   DDF: add code to debug state changes
>   monitor: don't call pselect() on deleted sysfs files
>   monitor: read_and_act: handle race conditions for resync_start
>   monitor: treat unreadable array_state as clean
>   tests/10ddf-create: omit log output check
> 
>  monitor.c          |   38 ++++++++++++----
>  super-ddf.c        |  124 ++++++++++++++++++++++++++++++++++++++-------------
>  tests/10ddf-create |    6 +--
>  3 files changed, 121 insertions(+), 47 deletions(-)

Thanks for these, and sorry for the delay in getting to them
(though the Date: field was
   Date: Fri, 25 Oct 2013 12:07:31 +0200
 so I've actually processed them 6 months before they will have been sent!!!)

All make sense and have been applied and push out.

NeilBrown

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 828 bytes --]