ZFS test plan

[sphinx99]

Before committing to it, I want to load up my ZFS pool with a bunch of data then put it through some worst case scenarios, in order to test,
- that it works properly
- that my hardware works properly with it
- that the user (ME) works properly in dealing with problems

This is pretty standard RAIDZ pools of SATA disks. I happen to be doing this under ESXi if it matters, using Solaris 11.

I could use advice on what to test. What I have done or plan to do:
- offline disk while copying data to a pool, then online it again, then do a scrub, all while copying
- while writing as much data as possible to a pool, simply pull one of my SATA drive data cables.
- power it all down, swap SATA cables (device names change?) then power it back up and see what happens. I tend to mix up cables when re-organizing/re-wiring the internals....
- I sure would like to test / see how ZFS handles corruption and recovering from it, but I am not sure how to test. A part of me thought, use Winhex to mess up some random bytes near the start of one disk, then see what happens. any suggestions here?

- Since I'm provisioning these disks as whole-disk VMDKs to ZFS, I probably out to drop a datastore or something while Solaris is live to see what happens.

 

[Suprnaut]

Your test is not going to be a true test of ZFS since you are using VMDK disks instead of raw devices. I am interested in seeing your results since a friend of mine is thinking of using ZFS off of vmdk disks because his hardware doesn't support vt-d for passthrough. Can you also post hardware specs/IO benchmarks?

 

[sphinx99]

This is a kick-the-tires environment. P55 motherboard with i5 750 processor, and I have about a dozen Samsung 2TB F4 drives to play with. Right now using ICH10, while waiting for a LSI card and a M1015 card to show up. Also have some SSDs and WD greens and blacks to test, all SATA. Have not done much benchmarking, but a dd if=hfhddhhd of=/dev/null has four 2tb drives doing about 240mb/sec. Not bad since i don't know where on the platters that file is. Can't do much serious testing until i get more RAM for more VMs on the box, or get around to LACP for more bandwidth.

I would prefer to use RDMs but the option was grayed out. I did some digging and it seems ESXi can be picky about RDMs from the GUI using local SATA disk? Have to investigate further. Don't know if I have VT-d on this platform or if it is turned on. TBH I was planning to turn this hardware into a PC for my girlfriend, and get something with ECC. Meanwhile, I wanted to get comfy with the common ZFS failure modes.

 

[TeeJayHoward]

- power it all down, swap SATA cables (device names change?) then power it back up and see what happens. I tend to mix up cables when re-organizing/re-wiring the internals....

I've done this. With the drives in the wrong spot on my controller, Solaris still went "Oh, hey, there's a pool for you to import!". Worked just fine. Actually, I think I got lazy and the disks are still swapped...