In Praise of ZFS on Linux's ZED 'ZFS Event Daemon'

That's neat, certainly, but I'm struggling to think what to actually use this for? The article mentions a couple cases, of which "taking action if devices fail" seems the more concrete; AFAIK that would make it probably straightforward to replace a drive with a spare? Anybody want to share any other concrete uses?

... thank you, that was the example that made me realize I can/should go use this at work because it will solve a problem we have:) And a good point in general.

Mail on certain events (scrub start/finish, warnings, errors, failures), automatically replace failed drives, and all sorts of other things. ZED, for example, can be used to create systemd mounts at boot.

Drive controller noticing a failure vs ZFS throwing a disk are very distinct things. You want alerting at both levels.

I think you are referring to one of two HPE components: Either the iLO, or the iLO combined with a Smart Array controller.

(Dell has equivalent products: For this discussion, iDRAC is the equivalent of iLO, and PERC is the equivalent to Smart Array.)

With those products, the RAID controller (Smart Array or PERC) will be connected to internal and/or external drives, will handle RAID in hardware (ideally with a battery backup write cache), and (through the iLO or iDRAC) generate alerts when a drive fails (or is close to failure).

In the context of ZFS, you don't have that. Your drive controller is either on-motherboard, or a PCIe card like (for example) the (Broadcom) LSI SAS 9300-8e. Those cards do have a RAID option (MegaRAID), but they are often used without.

The rest of the ZFS storage setup is pretty similar to the setup you are used to: Internal drives will have a SAS expander (if needed) on the motherboard, or will use a SAS expander card (for example, an HPE part #870549-B21 SAS expander card). External drives will be in a JBOD that has one or two expanders, and which is connected back to the server using SAS cables. One ZFS difference is that if you have many JBODs, instead of daisy-chaining arrays, you might choose to use a SAS switch (for example, an A54812-SW-01).

With all that I described with ZFS, I haven't mentioned how RAID is handled: ZFS handles RAID in software. RAID Z-1 is equivalent to RAID 5, Z-2 to 6, and Z-3 to 7. RAID also supports RAID 1 (two-drive mirrors), as well as a RAID 10 equivalent (striping across mirrored pairs of drives).

Since RAID is handled in software, and with the physical equipment I described, it is left to the OS to handle almost all monitoring an alerting. The one exception is that JBODs and rack-mount SAS switches (like the Astek) often have an Ethernet connection for monitoring and (basic) hardware control. But even that can often be handled within the OS, using SCSI Enclosure Services (SES, where the enclosure/switch itself is a device the OS can see and query).