Util for miners to extend all v1 sectors #5924
Conversation
arajasek
requested review from
Kubuxu,
magik6k and
whyrusleeping
as code owners
arajasek
force-pushed
the
asr/extend-v1-util
branch
from
67e7b44
to
ced47ca
Compare
cmd/lotus-storage-miner/sectors.go
Outdated
| if cctx.Bool("v1-sectors") { | ||
| extensions := map[miner.SectorLocation]map[abi.ChainEpoch][]uint64{} | ||
| // are given durations within a week? | ||
| closeEnough := func(a, b abi.ChainEpoch) bool { |
cmd/lotus-storage-miner/sectors.go
Outdated
| } | ||
|
|
||
| for _, si := range sis { | ||
| if si.SealProof < abi.RegisteredSealProof_StackedDrg2KiBV1_1 { |
There was a problem hiding this comment.
maybe invert and use a continue?
cmd/lotus-storage-miner/sectors.go
Outdated
| for _, si := range sis { | ||
| if si.SealProof < abi.RegisteredSealProof_StackedDrg2KiBV1_1 { | ||
|
|
||
| ml := builtin3.SealProofPoliciesV11[si.SealProof].SectorMaxLifetime |
There was a problem hiding this comment.
Let's move this into chain/actors/policy and make it a function over the version.
cmd/lotus-storage-miner/sectors.go
Outdated
| } | ||
|
|
||
| newExp := ml - (miner3.WPoStProvingPeriod * 2) + si.Activation | ||
| p, err := api.StateSectorPartition(ctx, maddr, si.SectorNumber, types.EmptyTSK) |
There was a problem hiding this comment.
This is really inefficient (searches all partitions/deadlines). OK for a first pass, but this could be very slow for some miners.
On the other hand, may be not... let's wait and see.
There was a problem hiding this comment.
Yeah, I tried it for the largest miners, and it is...not fast. Easy optimization is to only load the actor state once, but that is not gonna be the critical path. Alternative is to iterate over deadlines, and pre-compute one very large map of sectors -> locations, and then hit that up as needed.
This is a "needs to be run once" command, though, so eh?
cmd/lotus-storage-miner/sectors.go
Outdated
| for l, exts := range extensions { | ||
| for newExp, numbers := range exts { | ||
| scount += len(numbers) | ||
| if scount > miner.AddressedSectorsMax || len(p.Extensions) == miner3.DeclarationsMax { |
cmd/lotus-storage-miner/sectors.go
Outdated
| } | ||
| p := &miner3.ExtendSectorExpirationParams{} |
There was a problem hiding this comment.
why by reference? We're dereferencing everywhere.
| } | ||
|
|
||
| p := &miner3.ExtendSectorExpirationParams{} | ||
| for l, numbers := range sectors { |
There was a problem hiding this comment.
Can we dedup this with the above loop? It's doing the exact same thing (except now we aren't sending multiple messages).
There was a problem hiding this comment.
Eh == we aren't sending multiple messages here so it'll fail in some cases (although I guess we can say that that's pretty unlikely).
arajasek
force-pushed
the
asr/extend-v1-util
branch
from
ced47ca
to
2855ec2
Compare
arajasek
force-pushed
the
asr/extend-v1-util
branch
from
2855ec2
to
72c410c
Compare
| } | ||
|
|
||
| p := &miner3.ExtendSectorExpirationParams{} | ||
| for l, numbers := range sectors { |
There was a problem hiding this comment.
Eh == we aren't sending multiple messages here so it'll fail in some cases (although I guess we can say that that's pretty unlikely).
Run
lotus-miner sectors extend --v1-sectors.The
toleranceflag can be passed to indicate what durations aren't "worth" extending. It defaults to one week, which means that sectors whose current lifetime's are within one week of the maximum possible lifetime will not be extended.The
expiration-cutoffflag can be passed to skip sectors whose expiration is past a certain point from the current head. It defaults to infinity (no cutoff), but if, say,28800was specified, then only sectors expiring in the next 10 days would be extended (2880 epochs in 1 day).Tested manually on butterfly (kinda)
Should be tested on a large mainnet miner