Bug report about remove disks while others didn't finish plotting

Support
1

Issue Report

BUG:the PC has six SSDs,named 1,2,3,4,5,6. disk 6 is on plotting while others finished plot. remove disk 5 and start the farmer. It just keeping write to the disk 6 with 10~20M/S speed. didn’t sync piece cache and didn’t start plotting.even shutdown the farmer it would keep writing.
solution: remove disk 6 which didn’t finished plotting. start the farmer and waiting until it finished piece cache sync. add disk 6 and start farmer.
I had reproduced it several times.

Environment

system:7950X+X670E
OS:ubuntu 22.04
farmer:Mar-29

  • Operating System:
  • Pulsar/Advanced CLI/Docker:

Problem

[Paste any errors or relevant logs here]
2

I have never seen anything like this on Linux, but I can imagine how might happen if file system has to allocate something, but it shouldn’t last long. What file system are you using? Simple ext4 is recommended with no RAID of any kind or anything fancy like that.

3

My file system is NTFS. Actually I have more than 100 SSDs. All works well both on windows and ubuntu previous. I will try ext4 later.

4

As soon as changed to ext4 file system. The farmer start receiving data and piece cach sync progressing. Seems latest version didn’t suport NTFS well on ubuntu. But it didn’t affect the plotted disks. Replotting works well on ubuntu with NTFS disks.
image

5

Well, NTFS is your reason. If you let it do its thing, it will finish eventually. Same thing happens in Windows too.
I’m wondering if exfat is any better in this regard given how problematic NTFS is.

Seems latest version didn’t suport NTFS well on ubuntu

It does support NTFS on all OSs, it is just that NTFS is poorly designed for a specific use case that we have here.

6

What surprises me is the write speed so low? It should have intensive write to disk 6 (my wild guess is the OS was trying to re-arrange/sort the big file all over again).

And I really think if you wait longer, you’ll see piece cache sync. It is as expected you didn’t see plotting since the piece cache is not complete (you removed disk 5).

I’d like to have your confirmation that you see the audit read on disk #1 to #4, i.e. farming on those disks. Can you confirm that?

7

And I really think if you wait longer, you’ll see piece cache sync. It is as expected you didn’t see plotting since the piece cache is not complete (you removed disk 5).

I don’t think so. Farmer didn’t download any data.

I’d like to have your confirmation that you see the audit read on disk #1 to #4, i.e. farming on those disks. Can you confirm that?

confirm

I further tested with new PC and new NTFS disks on ubuntu. It didn’t work. After format disks to ext4 it start work.

8

It did download data, then it tried to write it to disk and got stuck because of stupid NTFS doing its things. This is expected on NTFS and apparently happens both on Windows and Linux. There is --plot-cache false that can be used to disable it or you can use a better file system like you already did.