[zfs-discuss] Does ZFS On Linux need TLER?

Richard Yao ryao at gentoo.org
Wed Feb 17 09:35:00 EST 2016



> On Feb 16, 2016, at 12:26 AM, Ryan How via zfs-discuss <zfs-discuss at list.zfsonlinux.org> wrote:
> 
> Hey Everyone,
> 
> I know this has been brought up a few times, but I haven't come to any definitive conclusions.
> 
> I've got some backup pools that are made up of a bit of a mixture of leftover drives. As they fail, I have been replacing them with some HGST enterprise drives.
> 
> On one of them I was just going to replace them with some WD blacks (I can't get any HGST drives for a few weeks as they are out of stock and I live in the middle of the sahara [apparently]).
> 
> Anyway, there is currently some piece of junk seagates in there. Every time a drive has issues it locks up the bus, causing several drives to go down at once and I can't figure out which drive is the issue - the smart readings are all over the place. The server also sometimes doesn't boot when one of these drives is "playing up".
> 
> Back to topic. Would that be in any way related to lack of TLER support?. Do I need TLER in this pool? Or is it just a "nice to have to keep the pool from temporarily hanging on a faulty drive" scenario.
> 
> 
> For TL;DR - Does ZFS work ok with no TLER on drives? - I don't mind about temporary hangs for a minute, or will it cause indefinite hanging, like I am finding with these seagates, until I physically remove the drive.

If a read IO hangs, whatever is waiting on the read IO will hang. I had taken a stab at trying to fail such IOs in the past, but I did not get it right and quite frankly, if we issue reads and we do not hear back, bad things are bound to happen regardless of any code intended to compensate for it.

In the case of no TLER, we should hear back eventually, but any time period that should constitute an acceptable latency would have long since passed. Indefinite hangs should not happen though, although if multiple read IOS that are issued in serial require error correction, then the time period in which things really blocked could appear indefinitely long.
> 
> 
> Thanks, Ryan
> 
> 
> 
> 
> 
> -- 
> Ryan How | BitWorks
> 0439 906 130 | ryan at bitworks.com.au | http://bitworks.com.au
> 
> _______________________________________________
> zfs-discuss mailing list
> zfs-discuss at list.zfsonlinux.org
> http://list.zfsonlinux.org/cgi-bin/mailman/listinfo/zfs-discuss
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://list.zfsonlinux.org/pipermail/zfs-discuss/attachments/20160217/ca641793/attachment-0003.html>


More information about the zfs-discuss mailing list