Lotus: splitstore and snapshots preparedness for tipset CIDs

[raulk]

A user has reported that the attached program run against Wallaby is unable to resolve a tipset that the node provided in the first place.

• The tipset CID in this case was bafy2bzacecwnr6sbpqvo6hcezvkqdi5unnd4t4mwa7zqyljtxtfzacy2ojgi6.
• The current epoch on Wallaby is 36100.
• The user was able to successfully resolve tipsets for more recent heights, e.g. 34539.

Theory

The Wallaby JSON-RPC endpoint is running the Lotus splitstore (pending confirmation). Tipset keys are not linked to the state tree or the chain in any manner, nor to one another; therefore the splitstore discards them. That would explain why more recent heights are available, while older ones aren't.

Solution

Maybe add the ability to inform the splitstore about CIDs bound by time, not by linkage?

Repro

	ctx := context.Background()
	api2, _, err := client.NewFullNodeRPCV1(ctx, "http://18.224.22.146:1234/rpc/v1", nil)
	if err != nil {
		panic(err)
	}
	tipSet, err := api2.ChainGetTipSetByHeight(ctx, abi.ChainEpoch(6905), types.EmptyTSK)
	if err != nil {
		panic(err)
	}
	tipSetCid, err := tipSet.Key().Cid()
	if err != nil {
		panic(err)
	}
	fmt.Println(tipSetCid)
	hash, err := api.EthHashFromCid(tipSetCid)
	if err != nil {
		panic(err)
	}
	block, err := api2.EthGetBlockByHash(ctx, hash, true)
	if err != nil {
		panic(err)
	}
	fmt.Println(block)

[raulk]

Node was not running a splitstore, but this will definitely become a problem with it.

[jennijuju]

oh we have an issue for this! so... so not splitstore.. but more of.. we dont think tipset CID is ever put in the block store in the first place lololol

[arajasek]

filecoin-project/lotus#9904 for the non-splitstore part of this issue.

[raulk]

I've renamed this issue to align with what should be covered. I think this is best assigned to @arajasek, so will reassign.

[Stebalien]

Or maybe @vyzo? @arajasek has a lot on his plate right now.

[jennijuju]

i sus the splitstore part of this issue shall be resolved by filecoin-project/lotus#9799

[arajasek]

So there's 3 parts to this:

• making sure we actually preserve CIDs, done in #9904
• making sure splitstore GC these CIDs, done in #9960
• making sure snapshots contain this information, not yet done
  • @jennijuju has a proposal that i like, which is to simply calculate and persist these CIDs on snapshot import -- she will open a lotus issue for that

[jennijuju]

she will open a lotus issue for that

filecoin-project/lotus#9977

[arajasek]

Handing over to Jennifer, who will close when lotus 9977 lands.

[jennijuju]

filecoin-project/lotus#10042 is the last piece. will make sure lotus pr lands so closing this now