## Description
Currently lighthouse load-balances across peers a single finalized chain. The chain is selected via the most peers. Once synced to the latest finalized epoch Lighthouse creates chains amongst its peers and syncs them all in parallel amongst each peer (grouped by their current head block).
This is typically fast and relatively efficient under normal operations. However if the chain has not finalized in a long time, the head chains can grow quite long. Peer's head chains will update every slot as new blocks are added to the head. Syncing all head chains in parallel is a bottleneck and highly inefficient in block duplication leads to RPC timeouts when attempting to handle all new heads chains at once.
This PR limits the parallelism of head syncing chains to 2. We now sync at most two head chains at a time. This allows for the possiblity of sync progressing alongside a peer being slow and holding up one chain via RPC timeouts.
The changes are somewhat simple but should solve two issues:
- When quickly changing between chains once and a second time back again, batchIds would collide and cause havoc.
- If we got an out of range response from a peer, sync would remain in syncing but without advancing
Changes:
- remove the batch id. Identify each batch (inside a chain) by its starting epoch. Target epochs for downloading and processing now advance by EPOCHS_PER_BATCH
- for the same reason, move the "to_be_downloaded_id" to be an epoch
- remove a sneaky line that dropped an out of range batch without downloading it
- bonus: put the chain_id in the log given to the chain. This is why explicitly logging the chain_id is removed
## Proposed Changes
To mitigate the impact of minority forks on RAM and disk usage, this change rejects blocks whose parent lies more than 320 slots (10 epochs, ~1 hour) in the past. The behaviour is configurable via `lighthouse bn --max-skip-slots N`, and can be turned off entirely using `--max-skip-slots none`.
Co-authored-by: Paul Hauner <paul@paulhauner.com>
## Issue Addressed
NA
## Proposed Changes
Moves beacon block processing over to the newly-added `GossipProcessor`. This moves the task off the core executor onto the blocking one.
## Additional Info
- With this PR, gossip blocks are being ignored during sync.
## Issue Addressed
NA
## Proposed Changes
- Adds a new function to allow getting a state with a bad state root history for attestation verification. This reduces unnecessary tree hashing during attestation processing, which accounted for 23% of memory allocations (by bytes) in a recent `heaptrack` observation.
- Don't clone caches on intermediate epoch-boundary states during block processing.
- Reject blocks that are known to fork choice earlier during gossip processing, instead of waiting until after state has been loaded (this only happens in edge-case).
- Avoid multiple re-allocations by creating a "forced" exact size iterator.
## Additional Info
NA
## Issue Addressed
There is currently an issue with yamux when connecting to prysm peers. The source of the issue is currently unknown.
This PR removes yamux support to force mplex negotation. We can add back yamux support once we have isolated and corrected the issue.
## Issue Addressed
Resolves#1489
## Proposed Changes
- Change starting metadata seq num to 0 according to the [spec](https://github.com/ethereum/eth2.0-specs/blob/dev/specs/phase0/p2p-interface.md#metadata).
- Remove metadata field from `NetworkGlobals`
- Persist metadata to disk on every update
- Load metadata seq number from disk on restart
- Persist enr to disk on update to ensure enr sequence number increments are persisted as well.
## Additional info
Since we modified starting metadata seq num to 0 from 1, we might still see `Invalid Sequence number provided` like in #1489 from prysm nodes if they have our metadata cached.
## Issue Addressed
#1384
Only catch, as currently implemented, when dialing the multiaddr nodes, there is no way to ask the peer manager if they are already connected or dialing
## Discovery v5 update
In this update we remove the openssl dependency in favour of rust-crypto.
The update also removes a series of unnecessary async functions which may improve some of the issues we have been experiencing.
## Issue Addressed
- Resolves#1451
## Proposed Changes
- Restricts the `contains_block` and `contains_block` so they only indicate a block is present if it descends from the finalized root. This helps to ensure that fork choice never points to a block that has been pruned from the database.
- Resolves#1451
- Before importing a block, double-check that its parent is known and a descendant of the finalized root.
- Split a big, monolithic block verification test into smaller tests.
## Additional Notes
I suspect there would be a craftier way to do the `is_descendant_of_finalized` check, but we're a bit tight on time now and we can optimize later if it starts showing in benches.
## TODO
- [x] Tests
## Issue Addressed
N/A
## Proposed Changes
Introduces the `GossipProcessor`, a multi-threaded (multi-tasked?), non-blocking processor for some messages from the network which require verification and import into the `BeaconChain`.
Initial testing indicates that this massively improves system stability by (a) moving block tasks from the normal executor (b) spreading out attestation load.
## Additional Info
TBC
These started to appear when I `^C` `curl -N http://localhost:5052/beacon/fork/stream`: `Aug 12 13:00:01.539 ERRO Couldn't stream piece hyper::Error(ChannelClosed), service: http`
Something must have changed in hyper since SSE has been implemented because I'm sure I haven't seen those errors before.
This PR properly detects a closed SSE stream and cleans up.
## Issue Addressed
NA
## Proposed Changes
Adds support for using the [`cross`](https://github.com/rust-embedded/cross) project to produce cross-compiled binaries using Docker images.
Provides quite clean and simple cross-compiles cause all the complexity is hidden in Dockerfiles. It does require you to be in the `docker` group though.
## Details
- Adds shortcut commands to `Makefile`
- Ensures `reqwest` and `discv5` use vendored openssl libs (i.e., static not shared).
- Switches to a [commit](284f705964) of blst that has a renamed C function to avoid a collision with openssl (upstream issue: https://github.com/supranational/blst/issues/21).
- Updates `ring` to the latest satisfiable version, since an earlier version was causing issues with `cross`.
- Off-topic, but adds extra message about Windows support as suggested by Discord user.
## Additional Info
- ~~Blocked on #1495~~
- There are no tests in CI for this yet for a few reasons:
- I'm hesitant to add more long-running tasks.
- Short-term bitrot should be avoided since we'll use it each release.
- In the long term I think it would be good to automate binary creation on a release.
- I observed the binaries increase in size from 50mb to 52mb after these changes.
limit simultaneous outgoing connections attempts to a reasonable top as an extra layer of protection
also shift the keep alive logic of the rpc handler to avoid needing to update it by hand. I think In rare cases this could make shutting down a connection a bit faster.
## Issue Addressed
#1419
## Proposed Changes
Creates a `--graffiti` cli flag in the validator client. If the flag is set, it overrides graffiti in the beacon node.
## Additional Info
## Issue Addressed
#1483
## Proposed Changes
Upgrades the log to a critical if a listener fails. We are able to listen on many interfaces so a single instance is not critical. We should however gracefully shutdown the client if we have no listeners, although the client can still function solely on outgoing connections.
For now a critical is raised and I leave #1494 for more sophisticated handling of this.
This also updates discv5 to handle errors of binding to a UDP socket such that lighthouse is now able to handle them.
## Issue Addressed
Some nodes not following head, high CPU usage and HTTP API delays
## Proposed Changes
Patches gossipsub. Gossipsub was using an `lru_time_cache` to check for duplicates. This contained an `O(N)` lookup for every gossipsub message to update the time cache. This was causing high cpu usage and blocking network threads.
This PR introduces a custom cache without `O(N)` inserts.
This also adds built in safety mechanisms to prevent gossipsub from excessively retrying connections upon failure. A maximum limit is set after which we disconnect from the node from too many failed substream connections.
## Issue Addressed
Peers that connected after the peer limit may remain connected in some circumstances.
This ensures peers not in the peer manager's list get disconnected. Further logging is also added to track this behaviour.
## Issue Addressed
NA
## Proposed Changes
- When producing a block, go and ensure every attestation in the naive aggregation pool is included in the operation pool. This should help us increase the number of useful attestations in a block.
- Lift the `RwLock`s inside `NaiveAggregationPool` up into a single high-level lock. There were race conditions in the existing setup and it was hard to reason about.
## Additional Info
NA
## Issue Addressed
#1028
A bit late, but I think if `BlockError` had a kind (the current `BlockError` minus everything on the variants that comes directly from the block) and the original block, more clones could be removed
## Issue Addressed
NA
## Proposed Changes
- Moves the git-based versioning we were doing into the `lighthouse_version` crate in `common`.
- Removes the `beacon_node/version` crate, replacing it with `lighthouse_version`.
- Bumps the version to `v0.2.0`.
## Additional Info
There are now two types of version string:
1. `const VERSION: &str = Lighthouse/v0.2.0-1419501f2+`
1. `version_with_platform() = Lighthouse/v0.2.0-1419501f2+/x86_64-linux`
(1) is handy cause it's a `const` and shorter. (2) has platform info so it's more useful. Note that the plus-sign (`+`) indicates the the git commit is dirty (it used to be `(modified)` but I had to shorten it to fit into graffiti).
These version strings are now included on:
- `lighthouse --version`
- `lcli --version`
- `curl localhost:5052/node/version`
- p2p messages when we communicate our version
You can update the version by changing this constant (version is not related to a `Cargo.toml`):
b9ad7102d5/common/lighthouse_version/src/lib.rs (L4-L15)
## Issue Addressed
Sync was breaking occasionally. The root cause appears to be identify crashing as events we being sent to the protocol after nodes were banned. Have not been able to reproduce sync issues since this update.
## Proposed Changes
Only send messages to sub-behaviour protocols if the peer manager thinks the peer is connected. All other messages are dropped.
## Issue Addressed
The most recent gossipsub update had an issue where some privacy settings lead to not sending a sequence number with the message. Although Lighthouse treats these as valid (based on current configuration) other clients may not.
This corrects gossipsub to send sequence numbers where expected and based on the configuration settings.
## Issue Addressed
#1056
## Proposed Changes
- Add a rate limiter to the RPC behaviour. This also means the rate limiting occurs just before the door to the application level, so the number of connections a peer opens does not affect this (this would happen in the future if put on the handler)
- The algorithm used is the leaky bucket as a meter / token bucket implemented the GCRA way
- Each protocol has its own limit. Due to the way the algorithm works, the "small" protocols have a hard limit, while bbrange and bbroot allow [burstiness](https://www.wikiwand.com/en/Burstiness). This is so that a peer can't request hundreds of individual requests expecting only one block in a short period of time, it also allows a peer to send two half size requests instead of one with max if they want to without getting limited, and.. it also allows a peer to request a batch of the maximum size and then send _appropriately spaced_ requests of really small sizes. From what I've seen in sync this is plausible when reaching the target slot.
## Additional Info
Needs to be heavily tested
## Issue Addressed
NA
## Proposed Changes
Allows for multiple "hardcoded" testnets.
## Additional Info
This PR is incomplete.
## TODO
- [x] Add flag to CLI, integrate with rest of Lighthouse.
Co-authored-by: Pawan Dhananjay <pawandhananjay@gmail.com>
Co-authored-by: Michael Sproul <michael@sigmaprime.io>
## Issue Addressed
Recurring sync loop and invalid batch downloading
## Proposed Changes
Shifts the batches to include the first slot of each epoch. This ensures the finalized is always downloaded once a chain has completed syncing.
Also add in logic to prevent re-dialing disconnected peers. Non-performant peers get disconnected during sync, this prevents re-connection to these during sync.
## Additional Info
N/A
## Issue Addressed
N/A
## Proposed Changes
This provides a number of corrections and improvements to gossipsub. Specifically
- Enables options for greater privacy around the message author
- Provides greater flexibility on message validation
- Prevents unvalidated messages from being gossiped
- Shifts the duplicate cache to a time-based cache inside gossipsub
- Updates the message-id to handle bytes
- Bug fixes related to mesh maintenance and topic subscription. This should improve our attestation inclusion rate.
## Issue Addressed
#1388 partially (eth2_libp2p & network)
## Proposed Changes
TLDR at the end
- *Complex types* are 3 on the handlers/Behaviours but the types are `Poll<ComplexType>` where `ComplexType` comes from the traits of libp2p. Those, I don't thing are worth an alias. A couple more were from using tokio combinators and were removed writing things the async way and using [`BoxFuture`](https://docs.rs/futures/0.3.5/futures/future/type.BoxFuture.html)
- The *cognitive complexity*.. I tried to address those before (they come from the poll functions too) and tbh they are cognitively simpler to understand the way they are now. Moving separate parts to functions doesn't add much since that code is not repeated and they all do early returns. If moved those returns would now need to be wrapped in an Option, probably, and checked to be returned again. I would leave them like that but that's just preference.
- *Too many arguments*: They are not easily put together in a wrapping struct since the parameters don't relate semantically (Ex: fn new with a log, a reference to the chain, a peer, etc) but some may differ.
- *Needless returns* were indeed needless
## Additional Info
TLDR: removed needless return, used BoxFuture and async, left the rest untouched since those lgtm
