Support for single ZIL Log drive?

[markw78]

Under Solaris versions of ZFS, a ZIL log drive can vanish from the system and not take down the pool, but continue to function with degraded performance.

Is there an ETA for that in FreeNAS? Since I'm just getting started with it all, I'd rather not use up that SATA port if at all possible.

Also if I do need a ZIL mirror, can FreeNAS take care of the mirror? Or do I need a RAID controller to do the mirroring for me?

Thanks!
Mark

 

[Brosif_My_Nif]

The most recent release of FreeNAS is currently running v28 which supports removal of a ZIL without loss of data.

 

[cyberjock]

The most recent release of FreeNAS is currently running v28 which supports removal of a ZIL without loss of data.

I just want to clarify this statement so there is no misunderstandings. ZIL, for reliability reasons, should be in a mirror with version prior to 8.2. The current build being worked on (8.3) has the bug fixed. But at this moment 8.3 is only in Release Candidate 1 status. It should be formally released as a final version soon.

 

[paleoN]

The most recent release of FreeNAS is currently running v28 which supports removal of a ZIL without loss of data.

However loss of the ZIL still results in loss of data in the ZIL if you need it. Best practice, in Solaris & FreeBSD, is to still mirror the ZIL, which ZFS can do, if you like your data.

Is there an ETA for that in FreeNAS? Since I'm just getting started with it all, I'd rather not use up that SATA port if at all possible.

Do you even need a SLOG?

 

[markw78]

Do you even need a SLOG?

Yes, it's back end for ESX, I've had terrible performance with ZFS when not using a SLOG. We purchased a 7220 with no SSD drives and it's awful.

In my home built OpenSolaris box I have a 32G SSD partitioned into 2 volumes, one used as log the other as cache. This SSD "disappears" from the OS from time to time with no ill-effect aside from performance degradation, just need to power off and reboot for it to come back. No data issues at all. I believe it's running SNV151.

The goal is to replace this with FreeNAS, awhile back we lost all our data because we had no SMART or notifications for the box under OpenSolaris, lost 1 disk, than lost another on rebuild. FreeNAS appears to take care of all our needs as far as that goes, since we are primarily a Windows shop.

Our cases are small with only 6 SATA ports, so I was hoping to add 2 more ports (via 3ware card) and put a Log and Cache drive on that, then use the 6 on-board ports for the actual data disks. Having to mirror the log disk add's another port and is a bit of a wrench in the plan.

Mark

 

[paleoN]

In my home built OpenSolaris box I have a 32G SSD partitioned into 2 volumes, one used as log the other as cache. This SSD "disappears" from the OS from time to time with no ill-effect aside from performance degradation, just need to power off and reboot for it to come back. No data issues at all.

Assuming the SSD in question obeys cache flush commands I don't see how it would cause data issues as you are able to access it on restart. Of course you won't even need to read it as long as the ZIL writes become fully committed to the pool in their pending txgs.

With your current setup the L2ARC is contending with the SLOG for your SSD's IOPS. This very well may be contributing to it falling off the bus.

The goal is to replace this with FreeNAS, awhile back we lost all our data because we had no SMART or notifications for the box under OpenSolaris, lost 1 disk, than lost another on rebuild.

Use double-parity arrays.

Our cases are small with only 6 SATA ports, so I was hoping to add 2 more ports (via 3ware card) and put a Log and Cache drive on that, then use the 6 on-board ports for the actual data disks. Having to mirror the log disk add's another port and is a bit of a wrench in the plan.

Best practice is to mirror the SLOG despite it being inconvenient for us. I've gotten the impression, historically at least, that FreeBSD is more fragile with SLOGs than Solaris based OSes. If a SLOG faults and then e.g. causes a kernel panic, you've lost the last 10 seconds of writes.