zil never gets any action

[16num]

I've tried everything I can think of including synthetic benchmarks but the zil never gets used (not even a little). This is the case even when transferring files form a SSD to a local pool. Sync options haven't appeared to do anything. What else could I be missing? I'm new to ZFS and don't use it professionally so it's most likely an ammeter mistake I'm making.

I know the zil drive below is way overkill but really just interesting in playing with freenas and ZFS at this point.

the current configuration is as follows
2x6TB raidz1
1x intel 750 400GB zil
1x intel 750 400GB l2arc
2x e5-1240v4
64GB 4ch


thanks

 

[m0nkey_]

ZIL? The correct term in SLOG. ZIL is in RAM, SLOG is on a device such as a SSD.

You're probably not touching the SLOG as it's unlikely you're performing synchronous writes.

 

[zambanini]

you can set your dataset to sync always

 

[Stux]

How are you measuring it’s use?

With gstat?

 

[kdragon75]

ZIL is in RAM

Or the main pool (still in RAM too) for sync writes if no SLOG is configured. Thats is if I'm not mistaken.

 

[anmnz]

Or the main pool (still in RAM too) for sync writes if no SLOG is configured. Thats is if I'm not mistaken.

Not sure what you guys mean, because the ZIL is on persistent storage not RAM. Otherwise it wouldn't work...

But yes it's on the pool's normal data storage disks if you don't move it to a SLOG.

 

[kdragon75]

Not sure what you guys mean, because the ZIL is on persistent storage not RAM. Otherwise it wouldn't work...

But yes it's on the pool's normal data storage disks if you don't move it to a SLOG.

My understanding was that its always in RAM but (for sync writes) it's only considered written once is on non-volatile memory i.e. HDD or SSD. Again the "working" copy is always in RAM.

 

[anmnz]

My understanding was that its always in RAM but (for sync writes) it's only considered written once is on non-volatile memory i.e. HDD or SSD. Again the "working" copy is always in RAM.

Async writes get stored in RAM and written to disk as part of the next transaction group.

Sync writes get stored in RAM and written to disk as part of the next transaction group just like async writes, but a copy of the data is also written immediately to disk, without waiting for the next transaction group to be written.

The ZIL is that place where sync writes get written to immediately. Its purpose is just to hold sync write data where the write has been acknowledged but the data has not been written as part of a transaction group. Once the data has been written as part of a transaction group, it's removed from the ZIL. This explains why (a) the ZIL is small, and (b) why you never need to read from the ZIL unless you are recovering from a crash.

But it's certainly on disk. Does that make sense?

 

[pro lamer]

ZIL is on persistent storage not RAM

it's on the pool's normal data storage disks if you don't move it to a SLOG.

Moreover the above is along with the SLOG meaning. SLOG = Separate [zfs] intent LOG (iXsystems university module 101). Opposed to non-separate ZIL being not separated from... the main storage.

Sent from my mobile phone