Request for testing: ata(4) MFC

[Jeremy Chadwick]

On Thu, Oct 16, 2008 at 01:15:46PM +0400, Andrey V. Elsukov wrote:
>> - Can we please see about adding the FreeNAS project's ata timeout
>>   sysctls?  I see lots of delays/sleeps in numerous pieces of code that
>>   pertain to soft or hard resets of AHCI controllers, and I often worry
>>   about the implications of hard-coded timeouts.
>>
>>   http://freenas.svn.sourceforge.net/viewvc/freenas/trunk/build/kernel-patches/ata/files/patch-ata.diff?view=markup
>
> As i remember changes in FreeNAS don't change timeouts for resets.

You're quite right -- they define the allowed time between the moment
the OS sends a command to the device/disk to the time the device/disk
sends back an OK/response.

The point is that all these timeouts (well, not every single one, but
the obvious ones) should be sysctl adjustables.

I worry that too strict a timeout could cause the device/controller to
be inappropriately reset during times when a disk may be doing something
internally (such was the case with old IBM disks and their ADM feature;
true, the feature is no longer implemented, but you get the point), and
I also worry that too loose of a timeout might cause the system to lock
up for long durations of time when a controller reset actually is
needed.  The default values we have are fine, but letting people tune
them based on their system would be ideal.

The FreeNAS guys have already provided evidence that tuning such
variables is beneficial for some users, depending upon the type of disk
and bus they're using.

> In any case, these changes can not be in 7.1-RELEASE. My patch
> targeted to move changes from CURRENT to 7.1. But it seems there
> are too few testers and patch can not be commited.

This doesn't bode well.  :-(  I think the problem is:

1) Not a lot of people have "dev boxes" they can try this stuff out on,
(Personally, I do have such boxes, but I don't have physical access to
them (I would have to go to our datacenter) to try failure scenarios;
"general" usability testing I can do of course),

2) Very few (if any at all) are willing to put the patch on a production
machine, as the risks outweigh the benefits for most,

3) Lack of eyeballs -- I have no idea how many FreeBSD users are
subscribed to -stable, and of those, how many use ATA or even care about
the patch ("my stuff works, why would I want to try this?")

Are we hoping that the patch will be included in 7.2?

-- 
| Jeremy Chadwick                                jdc at parodius.com |
| Parodius Networking                       http://www.parodius.com/ |
| UNIX Systems Administrator                  Mountain View, CA, USA |
| Making life hard for others since 1977.              PGP: 4BD6C0CB |