fix: sealing: Finalize snap sectors before submitting proofs

[magik6k]

Related Issues

Might help with the issue raised in https://filecoinproject.slack.com/archives/C038C47NWAE/p1651270743875089

Proposed Changes

Move snapdeals sectors to proving storage before submitting the proof to the chain.

[codecov]

Codecov Report

Merging #8582 (3ab625f) into master (88f8c7d) will decrease coverage by 0.01%.
The diff coverage is n/a.

@@            Coverage Diff             @@
##           master    #8582      +/-   ##
==========================================
- Coverage   40.86%   40.85%   -0.02%     
==========================================
  Files         686      686              
  Lines       75710    75710              
==========================================
- Hits        30939    30929      -10     
+ Misses      39445    39434      -11     
- Partials     5326     5347      +21     

Impacted Files	Coverage Δ
extern/storage-sealing/fsm.go	60.33% <ø> (ø)
chain/events/state/fastapi.go	75.00% <0.00%> (-25.00%)	⬇️
chain/events/message_cache.go	87.50% <0.00%> (-12.50%)	⬇️
chain/events/observer.go	73.79% <0.00%> (-7.59%)	⬇️
chain/events/events_called.go	83.90% <0.00%> (-6.35%)	⬇️
miner/miner.go	57.37% <0.00%> (-3.28%)	⬇️
chain/sync_manager.go	74.53% <0.00%> (-2.18%)	⬇️
extern/storage-sealing/currentdealinfo.go	70.17% <0.00%> (-1.76%)	⬇️
chain/stmgr/searchwait.go	68.58% <0.00%> (-1.29%)	⬇️
chain/sub/incoming.go	60.55% <0.00%> (-0.92%)	⬇️
... and 11 more

[ZenGround0]

I understand the change and like that it moves the replica update flow closer to the commit flow but I don't understand how the current state could cause post failures. Can you describe the mechanism?

[magik6k]

Basically we’d submit the proof while the sector is still in sealing storage, and then start to move it - if the sealing pipeline is backlogged, this can take a while, and we may enter a deadline, and you still can’t prove the sector.

The second, possibly more broken, thing that can happen is that we have moved the sealed file, but haven’t yet moved the cache files, we may still try to prove that sector, and there you’ll see missing p_aux

Both of those should be fixed by finalising early.