AAC driver. No kernel error messages for failed raid5?

[Chris St Denis]

I recently setup a new FreeBSD 7.0 AMD64 server with an IBM ServeRAID-8k
configured with a raid5 for data and raid1 for OS.

I pulled out one of the raid5 drives to test the functionality and
noticed that FreeBSD didn't seem to notice the disk failure at all. I
was expecting kernel messages about it, but got nothing.


This is a bit concerning because I don't look at the server physically
every day and want to know ASAP if a drive fails.

I tried setting options AAC_DEBUG=1 but that produces a lot of messages
even during normal use so I'd rather not leave that on.


Is this a bug in the driver, or simply missing functionality?

[Ed Maste]

Post by Chris St Denis
I pulled out one of the raid5 drives to test the functionality and
noticed that FreeBSD didn't seem to notice the disk failure at all. I
was expecting kernel messages about it, but got nothing.

This is missing functionality in the aac(4) driver. For now about
the best you can do is regularly poll the status using Adaptec's CLI
tool "arcconf."

-Ed

[Chris St Denis]

Post by Ed Maste

Post by Chris St Denis
I pulled out one of the raid5 drives to test the functionality and
noticed that FreeBSD didn't seem to notice the disk failure at all. I
was expecting kernel messages about it, but got nothing.

This is missing functionality in the aac(4) driver. For now about
the best you can do is regularly poll the status using Adaptec's CLI
tool "arcconf."
-ED

The tool you need to look into is: sysutils/aaccli . I think arctool
is only for arcmsr-devices.
unfortunately aaccli doesn't provide the possibility to use it with
parameters, so you probably
need to use it with an expect-script.
On the adaptec 2410SA I activated the "alarm" feature in the
controller bios, which helps me,
because its a home server but this will not help you if your server is
sitting somewhere else.
Because of this limitation (and a few other things with the
controller) I'm certainly looking for
an other solution. In my case a ZFS-based software RAID will suffice,
but this might not be
desirable for you.
Take care & good luck,
Robert

...

arcconf from ports works fine. It hangs on exit but does die off
eventually and doesn't do any harm sitting in background for a while
when run from cron, and from the commandline I can just ^c it.

Here is what I used in cron for anyone who is interested:
/usr/local/sbin/arcconf GETCONFIG 1 LD | egrep '(name|Status)'
It gives results like this which work well. Could probably be
incorporated into the daily run output, but I don't know exactly how off
hand.

Logical device name : Boot mirror
Status of logical device : Optimal
Logical device name : Data raid5
Status of logical device : Optimal


According to arcconf my card doesn't have an audible alarm :(

[Robert Jesacher]

Post by Ed Maste

Post by Chris St Denis
I pulled out one of the raid5 drives to test the functionality and
noticed that FreeBSD didn't seem to notice the disk failure at all. I
was expecting kernel messages about it, but got nothing.

This is missing functionality in the aac(4) driver. For now about
the best you can do is regularly poll the status using Adaptec's CLI
tool "arcconf."
-ED

The tool you need to look into is: sysutils/aaccli . I think arctool
is only for arcmsr-devices.
unfortunately aaccli doesn't provide the possibility to use it with
parameters, so you probably
need to use it with an expect-script.

On the adaptec 2410SA I activated the "alarm" feature in the
controller bios, which helps me,
because its a home server but this will not help you if your server is
sitting somewhere else.

Because of this limitation (and a few other things with the
controller) I'm certainly looking for
an other solution. In my case a ZFS-based software RAID will suffice,
but this might not be
desirable for you.

Take care & good luck,
Robert

[Dan Nelson]

Post by Ed Maste

Post by Chris St Denis
I pulled out one of the raid5 drives to test the functionality and
noticed that FreeBSD didn't seem to notice the disk failure at all.
I was expecting kernel messages about it, but got nothing.

This is missing functionality in the aac(4) driver. For now about
the best you can do is regularly poll the status using Adaptec's CLI
tool "arcconf."
-ED

The tool you need to look into is: sysutils/aaccli . I think arctool
is only for arcmsr-devices. unfortunately aaccli doesn't provide the
possibility to use it with parameters, so you probably need to use it
with an expect-script.

...

aaccli is most definitely scriptable:

$ aaccli "open /readonly aac0 : enclosure show status : disk show smart : container list /full"

[Chris St Denis]

Doesn't seem to work with my IBM ServeRAID 8k

CLI > open /readonly aac0
Executing: open /readonly=TRUE "aac0"
Command Error: <The current AFAAPI.DLL is too old to work with the
current controller software.>

Seems a little odd it's referencing a dll (which doesn't exist on the
system)

Post by Dan Nelson

Post by Ed Maste

Post by Chris St Denis
I pulled out one of the raid5 drives to test the functionality and
noticed that FreeBSD didn't seem to notice the disk failure at all.
I was expecting kernel messages about it, but got nothing.

This is missing functionality in the aac(4) driver. For now about
the best you can do is regularly poll the status using Adaptec's CLI
tool "arcconf."
-ED

The tool you need to look into is: sysutils/aaccli . I think arctool
is only for arcmsr-devices. unfortunately aaccli doesn't provide the
possibility to use it with parameters, so you probably need to use it
with an expect-script.

$ aaccli "open /readonly aac0 : enclosure show status : disk show smart : container list /full"

...

[Ed Maste]

Post by Chris St Denis
Doesn't seem to work with my IBM ServeRAID 8k
CLI > open /readonly aac0
Executing: open /readonly=TRUE "aac0"
Command Error: <The current AFAAPI.DLL is too old to work with the
current controller software.>

You can avoid this issue by building a kernel with the latest aac(4)
driver from RELENG_6/RELENG_7/HEAD, but you're probably better off
just using arcconf instead of aaccli.

-Ed