Use single state tree at startup

[twoeths]

Problem description

Right now if user provides a checkpointState or checkpointSyncUrl we'll deserialize states to 2 separate trees: one with the last state stored in db and the other one downloaded from checkpointState or checkpointSyncUrl. This takes a lot of memory and time of the node unnecessarily

Solution description

Once we load state with the last db value, use loadState() util to load the other state from bytes. This ensures we use same state tree across application

Additional context

this will also gives us more confident on bringing n-historical state flags to production because we use loadState() api at every startup

[twoeths]

tested with finalized state slot 1885504 of holesky

const chainForkConfig = createChainForkConfig(holeskyChainConfig);

const now = Date.now();
const wsState = getStateTypeFromBytes(chainForkConfig, stateBytes).deserializeToViewDU(stateBytes);
console.log("@@@ loaded state", wsState.slot, Date.now() - now, "ms");
const root = wsState.hashTreeRoot();
console.log("@@@ state root", toRootHex(root), Date.now() - now, "ms");

this gives

@@@ loaded state 1885504 2201 ms
@@@ state root 0x72e5a7e32dceb275f4268b28180d95fa40e3767c31040f7471f3a402de8edbfa 45998 ms

the initial full hashTreeRoot() time is almost 44s, if we use the db state to load downloaded ws state, it would save us at least 40s due to a lot of nodes are now shared between 2 merkle trees

note that initially we always have to compute state root

lodestar/packages/cli/src/cmds/beacon/initBeaconState.ts

Line 53 in 6d01593

anchorCheckpoint = getCheckpointFromState(store);

export function getCheckpointFromState(state: BeaconStateAllForks): Checkpoint {
  return {
    // the correct checkpoint is based on state's slot, its latestBlockHeader's slot's epoch can be
    // behind the state
    epoch: computeCheckpointEpochAtStateSlot(state.slot),
    root: getLatestBlockRoot(state),
  };
}

export function getLatestBlockRoot(state: BeaconStateAllForks): Root {
  const header = ssz.phase0.BeaconBlockHeader.clone(state.latestBlockHeader);
  if (ssz.Root.equals(header.stateRoot, ZERO_HASH)) {
    header.stateRoot = state.hashTreeRoot();
  }
  return ssz.phase0.BeaconBlockHeader.hashTreeRoot(header);
}

[twoeths]

Aug-26 06:40:22.054[]                 ^[[32minfo^[[39m: Lodestar network=holesky, version=v1.21.0/te/shared_buffer_sync_pubkeys_2/93e2c66, commit=93e2c66c4d93b5a145bcaf183a0f28982508046b
Aug-26 06:40:22.199[]                 ^[[32minfo^[[39m: Connected to LevelDB database path=/home/devops/beacon/chain-db
Aug-26 06:40:46.360[]                 ^[[33mwarn^[[39m: Forced syncing from checkpoint even though db state is within weak subjectivity period
Aug-26 06:40:46.361[]                 ^[[33mwarn^[[39m: Please consider removing --forceCheckpointSync flag unless absolutely necessary
Aug-26 06:40:46.361[]                 ^[[32minfo^[[39m: Fetching checkpoint state checkpointSyncUrl=https://beaconstate-holesky.chainsafe.io
Aug-26 06:40:51.258[]                 ^[[32minfo^[[39m: Download completed stateId=finalized
Aug-26 06:41:15.124[]                 ^[[32minfo^[[39m: Initializing beacon from a valid checkpoint state slot=2395936, epoch=74873, stateRoot=0x0e528419032ef08a1f740ecb3ffb6e62949610dc95242f64d097c8769be816b8, isWithinWeakSubjectivityPeriod=true
Aug-26 06:41:17.650[chain]            ^[[32minfo^[[39m: Historical state worker started
Aug-26 06:41:17.661[eth1]             ^[[32minfo^[[39m: Eth1 provider urls=http://127.0.0.1:8545
Aug-26 06:41:17.664[execution]        ^[[32minfo^[[39m: Execution client urls=http://127.0.0.1:8551
Aug-26 06:41:48.002[chain]         ^[[36mverbose^[[39m: Clock slot slot=2396009

on a common node, when start up we block for 3 times:

• hashTreeRoot() db state, ~24s
• hashTreeRoot() downloaded state, ~24s
• populate pubkey cache, ~30s

the goal is to get rid of the 2nd blocking time