Fixes#864 "Sync progress stops at Goerli block 4494913", and equivalent on
other networks.
The block body fetcher in `blockchain_sync.nim` had an incorrect assumption
about how peers respond to `GetBlockBodies`. It was issuing requests for N
block bodies and incorrectly handling replies which contained fewer than N
bodies.
Having received up to 192 headers in a batch, it split the range into smaller
`GetBlockBodies` requests, fetched each reply, then combined replies. The
effect was Nimbus requested batches of 128+64 block bodies, received gaps in
the reply sequence, then aborted.
That meant it repeatedly fetched data, then discarded it, and fetched it again,
dropping good peers in the process.
Aborted and restarted batches occurred with earlier blocks too, but this became
more pronounced until there were no suitable peers at batch 4494913..4495104.
Here's a trace:
```
TRC 2021-09-29 02:40:24.977+01:00 Requesting block headers file=blockchain_sync.nim:224 start=4494913 count=192 peer=<ENODE>
TRC 2021-09-29 02:40:24.977+01:00 >> Sending eth.GetBlockHeaders (0x03) file=protocol_eth65.nim:51 peer=<PEER> startBlock=4494913 max=192
TRC 2021-09-29 02:40:25.005+01:00 << Got reply eth.BlockHeaders (0x04) file=protocol_eth65.nim:51 peer=<PEER> count=192
TRC 2021-09-29 02:40:25.007+01:00 >> Sending eth.GetBlockBodies (0x05) file=protocol_eth65.nim:51 peer=<PEER> count=128
TRC 2021-09-29 02:40:25.209+01:00 << Got reply eth.BlockBodies (0x06) file=protocol_eth65.nim:51 peer=<PEER> count=13
TRC 2021-09-29 02:40:25.210+01:00 >> Sending eth.GetBlockBodies (0x05) file=protocol_eth65.nim:51 peer=<PEER> count=64
TRC 2021-09-29 02:40:25.290+01:00 << Got reply eth.BlockBodies (0x06) file=protocol_eth65.nim:51 peer=<PEER> count=64
WRN 2021-09-29 02:40:25.306+01:00 Bodies len != headers.len file=blockchain_sync.nim:276 bodies=77 headers=192
TRC 2021-09-29 02:40:25.306+01:00 peer disconnected file=blockchain_sync.nim:403 peer=<PEER>
TRC 2021-09-29 02:40:25.306+01:00 Finished obtaining blocks file=blockchain_sync.nim:303 peer=<PEER>
```
In practice, for modern peers, Nimbus received shorter replies than it assumed
depending on the block sizes on the chain. Geth/Erigon has 2MiB `BlockBodies`
response size soft limit. OpenEthereum has 4MiB.
Up to Berlin (EIP-2929), Nimbus's fetcher failed often, but there were still
some peers serving what Nimbus needed.
Just after the start of Berlin, at batch 4494913..4495104 on Goerli, zero peers
responded with full size replies for the whole batch, so Nimbus couldn't
progress past that point. But there was already a problem happening before
that for large blocks, dropping good peers and repeatedly fetching the same
block data.
Signed-off-by: Jamie Lokier <jamie@shareable.org>
Using the same packet tracing format to match `protocol_eth65`.
There aren't many calls, and this makes them clear.
Signed-off-by: Jamie Lokier <jamie@shareable.org>
Move `blockchain_sync.nim` from `nim-eth` to `nimbus-eth1`.
This lets `blockchain_sync` use the `eth/65` protocol to synchronise with more
modern peers than before.
Practically, the effect is the sync process runs more quickly and reliably than
before. It finds usable peers, and they are up to date.
Note, this is mostly old code, and it mostly performs "classic sync", the
original Ethereum method. Here's a summary of this code:
- It decides on a blockchain canonical head by sampling a few peers.
- Starting from block 0 (genesis), it downloads each block header and
block, mostly in order.
- After it downloads each block, it executes the EVM transactions in that block
and updates state trie from that, before going to the next block.
- This way the database state is updated by EVM executions in block order,
and new state is persisted to the trie database after each block.
Even though it mentions Geth "fast sync" (comments near end of file), and has
some elements, it isn't really. The most obvious missing part is this code
_doesn't download a state trie_, it calculates all state from block 0.
Geth "fast sync" has several parts:
1. Find an agreed common chain among several peers to treat as probably secure,
and a sufficiently long suffix to provide "statistical economic consensus"
when it is validated.
2. Perform a subset of PoW calculations, skipping forward over a segment to
verify some of the PoWs according to a pattern in the relevant paper.
3. Download the state trie from the block at the start of that last segment.
4. Execute only the blocks/transactions in that last segment, using the
downloaded state trie, to fill out the later states and properly validate the
blocks in the last segment.
Some other issues with `blockchain_sync` code:
- If it ever reaches the head of the chain, it doesn't follow new blocks with
increasing block numbers, at least not rapidly.
- If the chain undergoes a reorg, this code won't fetch a block number it has
already fetched, so it can't accept the reorg. It will end up conflicted
with peers. This hasn't mattered because the development focus has been on
the bulk of the catching up process, not the real-time head and reorgs.
- So it probably doesn't work correctly when it gets close to the head due to
many small reorgs, though it might for subtle reasons.
- Some of the network message handling isn't sufficiently robust, and it
discards some replies that have valid data according to specification.
- On rare occasions the initial query mapping block hash to number can
fail (because the peer's state changes).
- It makes some assumptions about the state of peers based on their responses
which may not be valid (I'm not convinced they are). The method for working
out "trusted" peers that agree a common chain prefix is clever. It compares
peers by asking each peer if it has the header matching another peer's
canonical head block by hash. But it's not clear that merely knowing about a
block constitutes agreement about the canonical chain. (If it did, query by
block number would give the same answer more authoritatively.)
Nonetheless, being able to run this sync process on `eth/65` is useful.
<# interactive rebase in progress; onto 66532e8a
Signed-off-by: Jamie Lokier <jamie@shareable.org>
Jamie Lokier