ZFS Zil Log txg_sync causing high IO every 5 seconds ? Any solution

bitblue.lab

Member
Oct 7, 2015
75
0
6
Is there any solution for Zil Log I/O...I know that every 5 seconds it flushes the data to disk but that is causing high I/O jump every 5 seconds even in small loads of Hard Drives ?

My pool of zfs is in mirror of 2x2TB 7200RPM Drives.
 
That is not ZIL I/O. It is normal I/O. If you use a slog then the writes to it will happen when they occur. That applies to writes without a slog, if they are sync writes. The 5 seconds flush is normal. If you see high I/O and throughput, maybe you need more drives. If you see high I/O and low throughput then either your drives have issues (I have two of these with no SMART errors but sluggish write speed) or your pool is too full and/or very fragmented.
 
Every ~5 seconds ZFS do write flush.

If your
sync=standard : sync writes are written 2 times (first to LOG, second as normal write every ~5 seconds), async write are written only once (every ~5 seconds)
sync=always : sync writes and async writes are written 2 times (first to LOG, second as normal write every ~5 seconds).
sync=disabled : sync writes and async writes are written only once (every ~5 seconds)

Of course ZFS can flush it write cache between 5 second period but the biggest flush is every ~5 seconds.

Maybe you can run every second 'sync' command to flush it, but I dont think it can solve something.
 
I think after kernel update etc is more stabilised now...I havent changed anythink in config yet...only enabled lzr compression and is going well.
 
Yes, the recent kernel and also zfs-on-linux updates yield to better stability. I was looking forward to this. My stability problems are still there, but at a much higher stress level as before. I'd like I would not see the problems after all :-/
 
Today I had a lot of problems with high io load and high cpu usage from z_wr_iss and now I see in updates a new kernel and zfs intramfs so I hope its going to be better after upgrade and reboot to use new updates and kernel, I will keep you guys updated here.