Clear state simply

[snadrus]

I propose that old miners with zero power lose their sector state tree on the next network upgrade.

Old Miners will be those with IDs created before 6 months ago. A simple less-than check should work.

Reasoning:
All this information is at the top of the state tree and should be fast and straightforward. This reduces the state tree easily. It doesn't include complexity nor the timing risks of compaction of existing miners.

The miners remain valid for future use.

Why:
Provers are incentivized to prefer this fork as it will significantly reduce their state storage obligation by removing terminated sectors. The migration time will be consistent for everyone and is expected to be very short. Therefore no network expenditure will be necessary.

This will result in smaller snapshots. It will dispel concerns temporarily about state size.

Looking Forward:
FOC PDP has its own funded cleanup mechanism. Compacting existing miners remains viable if further reductions become necessary from that part of the tree.

[LesnyRumcajs]

Would it make sense to do it on all future network upgrades, i.e., introduce this cleanup as a permanent step in the network upgrade process?

[snadrus]

do it on all future network upgrades

Pausing the chain to fiddle with global state is a process that should not be taken lightly. Also, after this clear, GiB will become a mix of Sectors, PDP, and other state. It's unlikely to have the same impact (in percentage terms) again.

So it'll be available to enact again cheaply, but only with careful testing first (as this round should also get).

[ThatGuyLLC]

would you like to discuss this on next core devs?

[snadrus]

Yes

[snadrus]

2 additional considerations:

1. We should check that they haven't posted WindowPoST for the past 2 days (as there's a delay to getting the first power).
2. There is a non-zero chance that some miner will still be affected (just ramping up, missed PoSTs), but it's very small.

[ZenGround0]

A quick AI assisted measurement of state suggests we can cut state costs by half if we do this simple state clearing procedure:

├───────────────────────────────────────────────┼───────────────────────────────────────────────────────────────────┤
  │ 1. Sectors not live in any deadline           │ 636,415,512                                                       │
  ├───────────────────────────────────────────────┼───────────────────────────────────────────────────────────────────┤
  │ 2. Their sector-info bytes                    │ 65,047,019,932 (60.58 GiB) — +1.079 GiB pro-rata AMT node         │
  │                                               │ overhead = 61.66 GiB                                              │
  ├───────────────────────────────────────────────┼───────────────────────────────────────────────────────────────────┤
  │ 3. Sectors held by zero-power miners (749,264 │ 623,009,546                                                       │
  │  of them)                                     │                                                                   │
  ├───────────────────────────────────────────────┼───────────────────────────────────────────────────────────────────┤
  │ 4. Their full sectors-AMT graph               │ 64,786,873,573 (60.34 GiB)                                        │
  └───────────────────────────────────────────────┴───────────────────────────────────────────────────────────────────┘

I have not validated the result myself so take it with some uncertainty.