gmirror and a flaky member

[Andriy Gapon]

On 06/01/2017 17:09, Miroslav Lachman wrote:
> Andriy Gapon wrote on 2017/01/06 11:12:
>> On 06/01/2017 11:54, Andriy Gapon wrote:
>>>
>>> Can a geom mirror handle a member that gets disconnected and then reappears
>>> again?
>>>
>>> What I am seeing right now is that the mirror does not pick up the member when
>>> it reappears.   I have to add it back manually.
> 
> It is intentional to mark disappeared device as broken.
> If you want to remove working device from gmirror, you must use gmirror remove
> command (or gmirror deactivate).
> 
>> To add more substance, here is what gets logged when the disk disappears:
>>
>> GEOM_MIRROR: Request failed (error=6). ada0p2[READ(offset=2517700608,
>> length=4096)]
>> GEOM_MIRROR: Device swap: provider ada0p2 disconnected.
>>
>> And here's what gets logged when the disk reappears:
>> GEOM_MIRROR: Component ada0p2 (device swap) broken, skipping.
>> GEOM_MIRROR: Cannot add disk ada0p2 to swap (error=22).
> 
> Was the disk removed by user or was it by some bad event?

The latter.  I suspect some problem with an SSD's controller or firmware.
Basically, the disk disappeared and then re-appeared half a minute later.

> >>> Even worse, the commands I have
>>> to execute are:
>>> $ gmirror forget ...
>>> $ gmirror insert ...
>>>
>>> This does not appear to be a graceful way of reactivating the member.
> 
> You can re-activate only member which was correctly deactivated. Not one yanked
> out without any "graceful" command.

I see.

> And as gmirror doesn't work as ZFS mirror (cannot do resilver) the re-added
> device is always fully rebuilt.

Indeed.

But it would be nice if gmirror was able to handle my situation automatically:
- when a disk disappears, mark it broken (missing to be more precise)
- when it re-appears, rebuild it and add back to the mirror

It seems that the main problem at the moment is that gmirror doesn't distinguish
between a disk getting many errors (e.g. disk going bad) and a disk disappearing
(maybe permanently, maybe temporarily).  But I could be mistaken about this.


-- 
Andriy Gapon