Well hello again, I have just learned that the host that recently had both nvme drives fail upon drive replacement, now has new problems: the filesystem report permanent data errors affecting the database of both, Matrix server and Telegram bridge.
I have just rented a new machine and am about to restore the database snapshot of the 26. of july, just in case. All the troubleshooting the recent days was very exhausting, however, i will try to do or at least prepare this within the upcoming hours.
Update
After a rescan the errors have gone away, however the drives logged errors too. It’s now the question as to whether the data integrety should be trusted.
Status august 1st
Well … good question… optimizations have been made last night, the restore was successful and … we are back to debugging outgoing federation :(
The new hardware also will be a bit more powerful… and yes, i have not forgotten that i wanted to update that database. It’s just that i was busy debugging federation problems.
References
- federation issues after restore: https://github.com/matrix-org/synapse/issues/16025
- why we had to restore initially: https://text.tchncs.de/tchncs/about-the-matrix-incident-on-july-26-2023
Dang the old host was deleted from the monitoring – however looking on at least one smart thing from my emails, there were no errors logged before the drives gave up on life during replacement. They just had a ton read/written and the used counter at 255% (even tho rw and age were not equal, its weird and one reason why i wanted to have at least one replaced in the first place). This is the one that had more:
SMART/Health Information (NVMe Log 0x02) Critical Warning: 0x04 Temperature: 53 Celsius Available Spare: 98% Available Spare Threshold: 10% Percentage Used: 255% Data Units Read: 7,636,639,249 [3.90 PB] Data Units Written: 2,980,551,083 [1.52 PB] Host Read Commands: 87,676,174,127 Host Write Commands: 28,741,297,023 Controller Busy Time: 705,842 Power Cycles: 7 Power On Hours: 17,437 Unsafe Shutdowns: 1 Media and Data Integrity Errors: 0 Error Information Log Entries: 0 Warning Comp. Temperature Time: 0 Critical Comp. Temperature Time: 0 Temperature Sensor 1: 53 Celsius Temperature Sensor 2: 64 Celsius Error Information (NVMe Log 0x01, 16 of 64 entries) No Errors LoggedThe new ones now, where the zpool errors happened look like this
SMART/Health Information (NVMe Log 0x02) Critical Warning: 0x00 Temperature: 24 Celsius Available Spare: 100% Available Spare Threshold: 5% Percentage Used: 3% Data Units Read: 122,135,021 [62.5 TB] Data Units Written: 31,620,076 [16.1 TB] Host Read Commands: 1,014,224,069 Host Write Commands: 231,627,064 Controller Busy Time: 3,909 Power Cycles: 2 Power On Hours: 117 Unsafe Shutdowns: 0 Media and Data Integrity Errors: 0 Error Information Log Entries: 4 Warning Comp. Temperature Time: 0 Critical Comp. Temperature Time: 0 Temperature Sensor 1: 24 Celsius Error Information (NVMe Log 0x01, 16 of 256 entries) Num ErrCount SQId CmdId Status PELoc LBA NSID VS 0 4 0 0x0000 0x8004 0x000 0 0 -SMART/Health Information (NVMe Log 0x02) Critical Warning: 0x00 Temperature: 24 Celsius Available Spare: 100% Available Spare Threshold: 5% Percentage Used: 2% Data Units Read: 153,193,333 [78.4 TB] Data Units Written: 29,787,075 [15.2 TB] Host Read Commands: 1,262,977,843 Host Write Commands: 230,135,280 Controller Busy Time: 4,804 Power Cycles: 11 Power On Hours: 119 Unsafe Shutdowns: 5 Media and Data Integrity Errors: 0 Error Information Log Entries: 14 Warning Comp. Temperature Time: 0 Critical Comp. Temperature Time: 0 Temperature Sensor 1: 24 Celsius Error Information (NVMe Log 0x01, 16 of 256 entries) Num ErrCount SQId CmdId Status PELoc LBA NSID VS 0 14 0 0x100d 0x8004 0x000 0 0 -You‘re not telling me telling me you‘re reading 62 TB in 117 hours, right? Right? xD the old ones were even petabytes.
Those numbers are just insane. I have worked with AI training and storage. I have never seen such numbers.
Well, I suppose that nvme was very much eol. Now I understand the behavior. This many operations in such a short time will put serious strain on your system. No wonder parts can give up. Are you using a raid config? Sorry if you already mentioned it.
i am not sure about those numbers on the new ones … it was one db restore and a few hrs of uptime … a scrub… , then i rsynced some stuff over and since then the thing is in idle 🤷
sample of the current active system … i think at time of arrival it was 2+tb written or something
SMART/Health Information (NVMe Log 0x02) Critical Warning: 0x00 Temperature: 37 Celsius Available Spare: 100% Available Spare Threshold: 10% Percentage Used: 0% Data Units Read: 88,116,921 [45.1 TB] Data Units Written: 43,968,235 [22.5 TB] Host Read Commands: 689,015,212 Host Write Commands: 409,762,513 Controller Busy Time: 1,477 Power Cycles: 4 Power On Hours: 248 Unsafe Shutdowns: 0 Media and Data Integrity Errors: 0 Error Information Log Entries: 0 Warning Comp. Temperature Time: 0 Critical Comp. Temperature Time: 0 Temperature Sensor 1: 37 Celsius Temperature Sensor 2: 46 Celsius Error Information (NVMe Log 0x01, 16 of 64 entries) No Errors LoggedI might now understand what happened to your nvme (just a guess):
SSDs have „spare“ sectors, not available to you until the old ones are used up. Then the new ones get cycled in.
The other info said: no spare available, usage 250%
I have read about this I think. If the spare sectors run out and the drive starts to get smaller and smaller, the system will fill it up to its old capacity and overwrite data, thus corrupting itself.
That what happens to phony ssds that get sold as tb drives but are 250 gig usb drives instead. As long as you only fill 250, you will not recognize something is wrong. Once you go above, you start losing data.
Not totally sure it works that way in ssds but I‘m somewhat sure this 250% usage is an indicator of a run down ssd.
And I still think it is pure negligence of hetzner to not have swapped them out then they were due.
Didn’t they run in raid 1 or something? Usually, if a drive fails, the second one should hold.
I am a bit confused now… the spare was 98% as to read in my snippet above … where does it say “no spare available”? I think it is on me to request a swap, and thats what i did as also the one with slightly less wear reported 255% used – which afaik is an aprox. lifetime left estimation based on rw cycles (not sure about all factors).
The one the hoster left in for me to play with, said no:
[Wed Jul 26 19:19:10 2023] nvme nvme1: I/O 9 QID 0 timeout, disable controller [Wed Jul 26 19:19:10 2023] nvme nvme1: Device shutdown incomplete; abort shutdown [Wed Jul 26 19:19:10 2023] nvme nvme1: Removing after probe failure status: -4Tried multiple kernelflags n stuff but couldn’t get past that error. Would have been interesting to have the hoster ship the thing to me (and maybe that would have been a long enough cooldown to have the thing working again), but i assume that would have been expensive from helsinki.
My bad. I must have misread. Sorry.
Yes, shipping it to you would have probably been a good idea. Does it cost a lot less to use the helsinki location? Otherwise Falkenstein would be a pretty good alternative I guess.