ZFS Pool Questions

[402blownstroker]

I am getting ready to build a ZFS pool and I have a couple of questions I can not find answers for right off hand.

1) If a disk that is used to either cache or for the logfile fails, can the pool still function?
2) If one of those disks fail can they be swapped out for working versions?

 

[Abit667]

The newer ZFS versions can handle failures of both the L2ARC (cache) and ZIL (log)

L2ARC failure is just read cache. If this disk fails and there are no mirrors, all that happens is you've lost the cache and you'll only have the ARC cache in RAM. On a very busy system that relies heavily on the cache this could be bad, but otherwise not a big deal. You can replace the disk used for L2ARC and add it back to the pool. You will probably experience some stutters in IO as it figures out the cache device is failing.

ZIL failing is a little more serious if it's not mirrored. The contents of the ZIL exist in RAM as the data comes in, and on the ZIL disk before it's flushed to the pool. So, if just the drive fails, and the system stays online (doesn't panic, crash, or do anything else dumb as a result of the drive failing) it will be able to write that data from memory into the pool, remove the ZIL drive and everything will continue on okay.

If the ZIL drive fails, and as a result of the drive failing the machine panics, or ZFS becomes locked up and can't write to the pool you will loose whatever writes were in the ZIL as you will have to reboot. Now, the effects of this could range from not a big deal/no data loss, to data loss depending on how much data was waiting to be written out and what exactly it was doing when it failed.

Personally I'd mirror the ZIL in all use cases where you think a ZIL is necessary. L2ARC doesn't matter as much, unless you have a workload that relies on it and you don't want to take the performance hit when it fails. If this is for a home server/media storage type device you don't need or want an L2ARC and a ZIL is probably pointless as well unless you're doing a ton of sync writes for some reason.

 

[402blownstroker]

So is adding a ZIL to the pool that big of a deal?

 

[danswartz]

you have a zil in the pool. what people mean is an slog (separate log device), but yes, people call it a zil. anyway, depends on what you do. nfs from esxi you need a zil ssd or run sync=disabled if u have a good ups and feel lucky.

 

[zrav]

If the ZIL drive fails, and as a result of the drive failing the machine panics, or ZFS becomes locked up and can't write to the pool you will loose whatever writes were in the ZIL as you will have to reboot.

No, if the SLOG device fails the ZIL is written to the main pool like if it didn't have a separate SLOG device in the first place. There is no impact on data safety, only performance. And all of this is only relevant if you have sync writes.

 

[spazoid]

You can't mirror the L2ARC and there would be no point. If you're worried about losing your entire L2ARC cache at once, use two (or more) smaller SSD's instead. One malfunctioning drive would only result in a fraction of your entire L2ARC being unavailable. As a bonus, two SSD's can typically deliver more IOPS than 1.

 

[danswartz]

No, if the SLOG device fails the ZIL is written to the main pool like if it didn't have a separate SLOG device in the first place. There is no impact on data safety, only performance. And all of this is only relevant if you have sync writes.

if the system crashes and the slog device fails, you can lose some data. not likely true. opensolaris used to be very intolerant of a failed slog device, but nowadays it isn't such a big deal...

 

[zrav]

Indeed. Just clearing up that ZFS does have a fallback