Wear levels on SSD's increasing daily.

[300cpilot]

Anyone else seeing this amount of wear on ssd's?

I have been running this node for about 2 years, with the drives that are in it. I was watching to see if they would withstand being used this way, because they are desktop drives, so I expected limits, but something has changed. Any way about a year ago we noted the disk health and they were not showing anything. I had not looked at it again until yesterday and the boot drive is now at 37% wearout and the other three drives were at 7% wearout. Today however one drive has increased to 9%. A 2% increase over night.

The os is on the latest patches/updates VE 8.3.2.

 

[Kingneutron]

You should probably proactively replace the 240GB before it gets to ~80%, but what do you expect with consumer-level SSD? Proxmox docs explicitly say they are not recommended. Even for homelab use it's chancy. Plus you have all 3 the same make/model, so likely they're all wearing out around the same time.

If you are doing zfs-on-zfs in-guest, you probably have crazy write amplification.

You don't give any details on how your VMs and backing storage are configured, so it's basically impossible to do anything but theorize here. For all I know, you have your VMs hosted on a 3-disk raidz1, and that's definitely not ideal.

https://howtoaskfor.help/

 

[300cpilot]

I think I was not clear, I am looking to see if anyone else noticed anything similar. If they have then its a warning, if not my drives did not last as long as I thought they would, my fault. Three years isn't bad, they were moved from a different server 2 years ago. I thought I was clear in the description that I knew the drives were not recommended, sorry. New drives have been ordered and should be here in a couple of days.

This is a small setup, single Dell 610 server 192 gigs ram, 12 cores, hba adapter, single drive for os (bought 6/2020), 3-2tb drives(bought 6/2022) in a single ZFS raidz1-0. I use PBS to backup on separate ma, has 2 CT's and 5 VM's on a separate machine. The screen shot shows how the drives are configured.

What is running on it.
A postfix DB server that records environmental variables from a temp, humid, Radon, Co2 and Co sensors, every 10 minutes.
Second postfix server, it is only used when I am developing, which has not been for about 6 months.
A git server, which is rarely written to.
Home assistant VM
Webserver for local Graphana server, displays environmental data on a remote raspberry pi. It will read the data from DB once an hour.
Opnsense firewall, runs a single user's office
PiHole DNS

 

[mr44er]

I think I was not clear, I am looking to see if anyone else noticed anything similar.

No problem and yes, this is normal. I saw on fresh Samsung 0% changing over night to 1%. The change from 1% to 2% took longer.
The wearout-% are not live, so a 2% jump is not unusual and this comes from SMART and it is a prediction, not an accurate state to trust fully.

Three years isn't bad

Yes, indeed. You are lucky with it.

 

[andrew-transparent]

Yes, I see the same thing in my home lab on consumer SSDs when using Ceph/ZFS. Those are busy file systems (CoW) compared to EXT4/XFS. I have a few SSDs running non-important VMs 5+ years old at 90%+ and they're still going. I plan on running them into the ground to see how far they can go. I'm getting 'spurious read errors' in Ceph lately, so maybe that's an early warning sign?

 

[mr44er]

I'm getting 'spurious read errors' in Ceph lately, so maybe that's an early warning sign?

Yes, either the disk gives wrong data back (not that data that was written onto it) or it takes too much time for error correction the disk itself has running. (retry re-read until it fits or forever or give up after ~10-20th attempt, depends on firmware)

 

[_gabriel]

more over HBA like LSI H310 doesn't support TRIM on consumer SSD.
https://forums.servethehome.com/ind...-support-without-drat-rzat-requirement.26450/