[Mining Issue] Miner sometimes puts incorrect data to NAS during finalization

[Shekelme]

Recently, 10 bad sectors appeared in one of the deadlines. After one or two WindowPoSts, they have not recovered. Here they are:

Faults:                   10
Faulty Sectors:           [54632 54634 55102 55609 56141 56147 56148 56483 56640 56647]

A little more details (by the way, the command for some reason only displays data about 7 sectors):

admfc@pc3950:~$ lotus-miner proving check 23 | grep bad
23        0          56647   bad (stat /media/filecoin-storage-9/cache/s-t0187709-56647/sc-02-data-tree-r-last-0.dat: no such file or directory)
23        0          55609   bad (stat /media/filecoin-storage-9/cache/s-t0187709-55609/p_aux: no such file or directory)
23        0          56483   bad (stat /media/filecoin-storage-9/cache/s-t0187709-56483/sc-02-data-tree-r-last-3.dat: no such file or directory)
23        0          56141   bad (stat /media/filecoin-storage-9/cache/s-t0187709-56141/t_aux: no such file or directory)
23        0          56640   bad (stat /media/filecoin-storage-9/cache/s-t0187709-56640/p_aux: no such file or directory)
23        0          56147   bad (stat /media/filecoin-storage-9/cache/s-t0187709-56147/sc-02-data-tree-r-last-6.dat: no such file or directory)
23        0          55102   bad (stat /media/filecoin-storage-9/cache/s-t0187709-55102/p_aux: no such file or directory)

Let's look at the first sector 56647: Sure something is wrong! Looks like miner put some "wrong" cache from PC2 worker?

And this is the case with all of these 7 sectors.

Version lotus version 1.10.1+mainnet+git.d920b47e7

[Shekelme]

The issue still persists. lotus-worker version 1.11.3+mainnet+git.a0ddb10de