## Proposed Changes
Remove the remaining Altair `FIXME`s from consensus land.
1. Implement tree hash caching for the participation lists. This required some light type manipulation, including removing the `TreeHash` bound from `CachedTreeHash` which was purely descriptive.
2. Plumb the proposer index through Altair attestation processing, to avoid calculating it for _every_ attestation (potentially 128ms on large networks). This duplicates some work from #2431, but with the aim of getting it in sooner, particularly for the Altair devnets.
3. Removes two FIXMEs related to `superstruct` and cloning, which are unlikely to be particularly detrimental and will be tracked here instead: https://github.com/sigp/superstruct/issues/5
* Adjust beacon node timeouts for validator client HTTP requests (#2352)
Resolves#2313
Provide `BeaconNodeHttpClient` with a dedicated `Timeouts` struct.
This will allow granular adjustment of the timeout duration for different calls made from the VC to the BN. These can either be a constant value, or as a ratio of the slot duration.
Improve timeout performance by using these adjusted timeout duration's only whenever a fallback endpoint is available.
Add a CLI flag called `use-long-timeouts` to revert to the old behavior.
Additionally set the default `BeaconNodeHttpClient` timeouts to the be the slot duration of the network, rather than a constant 12 seconds. This will allow it to adjust to different network specifications.
Co-authored-by: Paul Hauner <paul@paulhauner.com>
* Use read_recursive locks in database (#2417)
Closes#2245
Replace all calls to `RwLock::read` in the `store` crate with `RwLock::read_recursive`.
* Unfortunately we can't run the deadlock detector on CI because it's pinned to an old Rust 1.51.0 nightly which cannot compile Lighthouse (one of our deps uses `ptr::addr_of!` which is too new). A fun side-project at some point might be to update the deadlock detector.
* The reason I think we haven't seen this deadlock (at all?) in practice is that _writes_ to the database's split point are quite infrequent, and a concurrent write is required to trigger the deadlock. The split point is only written when finalization advances, which is once per epoch (every ~6 minutes), and state reads are also quite sporadic. Perhaps we've just been incredibly lucky, or there's something about the timing of state reads vs database migration that protects us.
* I wrote a few small programs to demo the deadlock, and the effectiveness of the `read_recursive` fix: https://github.com/michaelsproul/relock_deadlock_mvp
* [The docs for `read_recursive`](https://docs.rs/lock_api/0.4.2/lock_api/struct.RwLock.html#method.read_recursive) warn of starvation for writers. I think in order for starvation to occur the database would have to be spammed with so many state reads that it's unable to ever clear them all and find time for a write, in which case migration of states to the freezer would cease. If an attack could be performed to trigger this starvation then it would likely trigger a deadlock in the current code, and I think ceasing migration is preferable to deadlocking in this extreme situation. In practice neither should occur due to protection from spammy peers at the network layer. Nevertheless, it would be prudent to run this change on the testnet nodes to check that it doesn't cause accidental starvation.
* Return more detail when invalid data is found in the DB during startup (#2445)
- Resolves#2444
Adds some more detail to the error message returned when the `BeaconChainBuilder` is unable to access or decode block/state objects during startup.
NA
* Use hardware acceleration for SHA256 (#2426)
Modify the SHA256 implementation in `eth2_hashing` so that it switches between `ring` and `sha2` to take advantage of [x86_64 SHA extensions](https://en.wikipedia.org/wiki/Intel_SHA_extensions). The extensions are available on modern Intel and AMD CPUs, and seem to provide a considerable speed-up: on my Ryzen 5950X it dropped state tree hashing times by about 30% from 35ms to 25ms (on Prater).
The extensions became available in the `sha2` crate [last year](https://www.reddit.com/r/rust/comments/hf2vcx/ann_rustcryptos_sha1_and_sha2_now_support/), and are not available in Ring, which uses a [pure Rust implementation of sha2](https://github.com/briansmith/ring/blob/main/src/digest/sha2.rs). Ring is faster on CPUs that lack the extensions so I've implemented a runtime switch to use `sha2` only when the extensions are available. The runtime switching seems to impose a miniscule penalty (see the benchmarks linked below).
* Start a release checklist (#2270)
NA
Add a checklist to the release draft created by CI. I know @michaelsproul was also working on this and I suspect @realbigsean also might have useful input.
NA
* Serious banning
* fmt
Co-authored-by: Mac L <mjladson@pm.me>
Co-authored-by: Paul Hauner <paul@paulhauner.com>
Co-authored-by: Michael Sproul <michael@sigmaprime.io>
## Issue Addressed
- Resolves#2452
## Proposed Changes
I've seen a few people confused by this and I don't think the message is really worth it.
## Additional Info
NA
## Issue Addressed
#635
## Proposed Changes
- Keep attestations that reference a block we have not seen for 30secs before being re processed
- If we do import the block before that time elapses, it is reprocessed in that moment
- The first time it fails, do nothing wrt to gossipsub propagation or peer downscoring. If after being re processed it fails, downscore with a `LowToleranceError` and ignore the message.
## Issue Addressed
NA
## Proposed Changes
Adds a metric to see how many set bits are in the sync aggregate for each beacon block being imported.
## Additional Info
NA
## Issue Addressed
NA
## Proposed Changes
Adds more detail to the log when an attestation is ignored due to a prior one being known. This will help identify which validators are causing the issue.
## Additional Info
NA
## Issue Addressed
- Resolves#2444
## Proposed Changes
Adds some more detail to the error message returned when the `BeaconChainBuilder` is unable to access or decode block/state objects during startup.
## Additional Info
NA
## Issue Addressed
Closes#2245
## Proposed Changes
Replace all calls to `RwLock::read` in the `store` crate with `RwLock::read_recursive`.
## Additional Info
* Unfortunately we can't run the deadlock detector on CI because it's pinned to an old Rust 1.51.0 nightly which cannot compile Lighthouse (one of our deps uses `ptr::addr_of!` which is too new). A fun side-project at some point might be to update the deadlock detector.
* The reason I think we haven't seen this deadlock (at all?) in practice is that _writes_ to the database's split point are quite infrequent, and a concurrent write is required to trigger the deadlock. The split point is only written when finalization advances, which is once per epoch (every ~6 minutes), and state reads are also quite sporadic. Perhaps we've just been incredibly lucky, or there's something about the timing of state reads vs database migration that protects us.
* I wrote a few small programs to demo the deadlock, and the effectiveness of the `read_recursive` fix: https://github.com/michaelsproul/relock_deadlock_mvp
* [The docs for `read_recursive`](https://docs.rs/lock_api/0.4.2/lock_api/struct.RwLock.html#method.read_recursive) warn of starvation for writers. I think in order for starvation to occur the database would have to be spammed with so many state reads that it's unable to ever clear them all and find time for a write, in which case migration of states to the freezer would cease. If an attack could be performed to trigger this starvation then it would likely trigger a deadlock in the current code, and I think ceasing migration is preferable to deadlocking in this extreme situation. In practice neither should occur due to protection from spammy peers at the network layer. Nevertheless, it would be prudent to run this change on the testnet nodes to check that it doesn't cause accidental starvation.
## Issue Addressed
Resolves#2313
## Proposed Changes
Provide `BeaconNodeHttpClient` with a dedicated `Timeouts` struct.
This will allow granular adjustment of the timeout duration for different calls made from the VC to the BN. These can either be a constant value, or as a ratio of the slot duration.
Improve timeout performance by using these adjusted timeout duration's only whenever a fallback endpoint is available.
Add a CLI flag called `use-long-timeouts` to revert to the old behavior.
## Additional Info
Additionally set the default `BeaconNodeHttpClient` timeouts to the be the slot duration of the network, rather than a constant 12 seconds. This will allow it to adjust to different network specifications.
Co-authored-by: Paul Hauner <paul@paulhauner.com>
## Proposed Changes
Implement the consensus changes necessary for the upcoming Altair hard fork.
## Additional Info
This is quite a heavy refactor, with pivotal types like the `BeaconState` and `BeaconBlock` changing from structs to enums. This ripples through the whole codebase with field accesses changing to methods, e.g. `state.slot` => `state.slot()`.
Co-authored-by: realbigsean <seananderson33@gmail.com>
## Issue Addressed
#2377
## Proposed Changes
Implement the same code used for block root lookups (from #2376) to state root lookups in order to improve performance and reduce associated memory spikes (e.g. from certain HTTP API requests).
## Additional Changes
- Tests using `rev_iter_state_roots` and `rev_iter_block_roots` have been refactored to use their `forwards` versions instead.
- The `rev_iter_state_roots` and `rev_iter_block_roots` functions are now unused and have been removed.
- The `state_at_slot` function has been changed to use the `forwards` iterator.
## Additional Info
- Some tests still need to be refactored to use their `forwards_iter` versions. These tests start their iteration from a specific beacon state and thus use the `rev_iter_state_roots_from` and `rev_iter_block_roots_from` functions. If they can be refactored, those functions can also be removed.
This updates some older dependencies to address a few cargo audit warnings.
The majority of warnings come from network dependencies which will be addressed in #2389.
This PR contains some minor dep updates that are not network related.
Co-authored-by: Michael Sproul <michael@sigmaprime.io>
## Issue Addressed
`make lint` failing on rust 1.53.0.
## Proposed Changes
1.53.0 updates
## Additional Info
I haven't figure out why yet, we were now hitting the recursion limit in a few crates. So I had to add `#![recursion_limit = "256"]` in a few places
Co-authored-by: realbigsean <seananderson33@gmail.com>
Co-authored-by: Michael Sproul <michael@sigmaprime.io>
## Proposed Changes
A user on Discord (`@ChewsMacRibs`) reported that the validator monitor was logging `WARN Attested to an incorrect head` for their validator while it was awaiting activation.
This PR modifies the monitor so that it ignores inactive validators, by the logic that they are either awaiting activation, or have already exited. Either way, there's no way for an inactive validator to have their attestations included on chain, so no need for the monitor to report on them.
## Additional Info
To reproduce the bug requires registering validator keys manually with `--validator-monitor-pubkeys`. I don't think the bug will present itself with `--validator-monitor-auto`.
## Issue Addressed
#2293
## Proposed Changes
- Modify the handler for the `eth_chainId` RPC (i.e., `get_chain_id`) to explicitly match against the Geth error string returned for pre-EIP-155 synced Geth nodes
- ~~Add a new helper function, `rpc_error_msg`, to aid in the above point~~
- Refactor `response_result` into `response_result_or_error` and patch reliant RPC handlers accordingly (thanks to @pawanjay176)
## Additional Info
Geth, as of Pangaea Expanse (v1.10.0), returns an explicit error when it is not synced past the EIP-155 block (2675000). Previously, Geth simply returned a chain ID of 0 (which was obviously much easier to handle on Lighthouse's part).
Co-authored-by: Paul Hauner <paul@paulhauner.com>
## Issue Addressed
NA
## Proposed Changes
I am starting to see a lot of slog-async overflows (i.e., too many logs) on Prater whenever we see attestations for an unknown block. Since these logs are identical (except for peer id) and we expose volume/count of these errors via `metrics::GOSSIP_ATTESTATION_ERRORS_PER_TYPE`, I took the following actions to remove them from `DEBUG` logs:
- Push the "Attestation for unknown block" log to trace.
- Add a debug log in `search_for_block`. In effect, this should serve as a de-duped version of the previous, downgraded log.
## Additional Info
TBC
## Issue Addressed
NA
## Proposed Changes
Bump versions.
## Additional Info
This is not exactly the v1.4.0 release described in [Lighthouse Update #36](https://lighthouse.sigmaprime.io/update-36.html).
Whilst it contains:
- Beta Windows support
- A reduction in Eth1 queries
- A reduction in memory footprint
It does not contain:
- Altair
- Doppelganger Protection
- The remote signer
We have decided to release some features early. This is primarily due to the desire to allow users to benefit from the memory saving improvements as soon as possible.
## TODO
- [x] Wait for #2340, #2356 and #2376 to merge and then rebase on `unstable`.
- [x] Ensure discovery issues are fixed (see #2388)
- [x] Ensure https://github.com/sigp/lighthouse/pull/2382 is merged/removed.
- [x] Ensure https://github.com/sigp/lighthouse/pull/2383 is merged/removed.
- [x] Ensure https://github.com/sigp/lighthouse/pull/2384 is merged/removed.
- [ ] Double-check eth1 cache is carried between boots
## Issue Addressed
NA
## Proposed Changes
Reverts #2345 in the interests of getting v1.4.0 out this week. Once we have released that, we can go back to testing this again.
## Additional Info
NA
## Issue Addressed
NA
## Proposed Changes
When observing `jemallocator` heap profiles and Grafana, it became clear that Lighthouse is spending significant RAM/CPU on processing blocks from the RPC. On investigation, it seems that we are loading the parent of the block *before* we check to see if the block is already known. This is a big waste of resources.
This PR adds an additional `check_block_relevancy` call as the first thing we do when we try to process a `SignedBeaconBlock` via the RPC (or other similar methods). Ultimately, `check_block_relevancy` will be called again later in the block processing flow. It's a very light function and I don't think trying to optimize it out is worth the risk of a bad block slipping through.
Also adds a `New RPC block received` info log when we process a new RPC block. This seems like interesting and infrequent info.
## Additional Info
NA
## Issue Addressed
NA
## Proposed Changes
Return a very specific error when at attestation reads shuffling from a frozen `BeaconState`. Previously, this was returning `MissingBeaconState` which indicates a much more serious issue.
## Additional Info
Since `get_inconsistent_state_for_attestation_verification_only` is only called once in `BeaconChain::with_committee_cache`, it is quite easy to reason about the impact of this change.
## Issue Addressed
NA
## Proposed Changes
Whilst investigating #2372, I [learned](https://github.com/sigp/lighthouse/issues/2372#issuecomment-851725049) that the error message returned from some failed Eth1 requests are always `NotReachable`. This makes debugging quite painful.
This PR adds more detail to these errors. For example:
- Bad infura key: `ERRO Failed to update eth1 cache error: Failed to update Eth1 service: "All fallback errored: https://mainnet.infura.io/ => EndpointError(RequestFailed(\"Response HTTP status was not 200 OK: 401 Unauthorized.\"))", retry_millis: 60000, service: eth1_rpc`
- Unreachable server: `ERRO Failed to update eth1 cache error: Failed to update Eth1 service: "All fallback errored: http://127.0.0.1:8545/ => EndpointError(RequestFailed(\"Request failed: reqwest::Error { kind: Request, url: Url { scheme: \\\"http\\\", cannot_be_a_base: false, username: \\\"\\\", password: None, host: Some(Ipv4(127.0.0.1)), port: Some(8545), path: \\\"/\\\", query: None, fragment: None }, source: hyper::Error(Connect, ConnectError(\\\"tcp connect error\\\", Os { code: 111, kind: ConnectionRefused, message: \\\"Connection refused\\\" })) }\"))", retry_millis: 60000, service: eth1_rpc`
- Bad server: `ERRO Failed to update eth1 cache error: Failed to update Eth1 service: "All fallback errored: http://127.0.0.1:8545/ => EndpointError(RequestFailed(\"Response HTTP status was not 200 OK: 501 Not Implemented.\"))", retry_millis: 60000, service: eth1_rpc`
## Additional Info
NA
## Issue Addressed
NA
## Primary Change
When investigating memory usage, I noticed that retrieving a block from an early slot (e.g., slot 900) would cause a sharp increase in the memory footprint (from 400mb to 800mb+) which seemed to be ever-lasting.
After some investigation, I found that the reverse iteration from the head back to that slot was the likely culprit. To counter this, I've switched the `BeaconChain::block_root_at_slot` to use the forwards iterator, instead of the reverse one.
I also noticed that the networking stack is using `BeaconChain::root_at_slot` to check if a peer is relevant (`check_peer_relevance`). Perhaps the steep, seemingly-random-but-consistent increases in memory usage are caused by the use of this function.
Using the forwards iterator with the HTTP API alleviated the sharp increases in memory usage. It also made the response much faster (before it felt like to took 1-2s, now it feels instant).
## Additional Changes
In the process I also noticed that we have two functions for getting block roots:
- `BeaconChain::block_root_at_slot`: returns `None` for a skip slot.
- `BeaconChain::root_at_slot`: returns the previous root for a skip slot.
I unified these two functions into `block_root_at_slot` and added the `WhenSlotSkipped` enum. Now, the caller must be explicit about the skip-slot behaviour when requesting a root.
Additionally, I replaced `vec![]` with `Vec::with_capacity` in `store::chunked_vector::range_query`. I stumbled across this whilst debugging and made this modification to see what effect it would have (not much). It seems like a decent change to keep around, but I'm not concerned either way.
Also, `BeaconChain::get_ancestor_block_root` is unused, so I got rid of it 🗑️.
## Additional Info
I haven't also done the same for state roots here. Whilst it's possible and a good idea, it's more work since the fwds iterators are presently block-roots-specific.
Whilst there's a few places a reverse iteration of state roots could be triggered (e.g., attestation production, HTTP API), they're no where near as common as the `check_peer_relevance` call. As such, I think we should get this PR merged first, then come back for the state root iters. I made an issue here https://github.com/sigp/lighthouse/issues/2377.
## Issue Addressed
#2325
## Proposed Changes
This pull request changes the behavior of the Peer Manager by including a minimum outbound-only peers requirement. The peer manager will continue querying for peers if this outbound-only target number hasn't been met. Additionally, when peers are being removed, an outbound-only peer will not be disconnected if doing so brings us below the minimum.
## Additional Info
Unit test for heartbeat function tests that disconnection behavior is correct. Continual querying for peers if outbound-only hasn't been met is not directly tested, but indirectly through unit testing of the helper function that counts the number of outbound-only peers.
EDIT: Am concerned about the behavior of ```update_peer_scores```. If we have connected to a peer with a score below the disconnection threshold (-20), then its connection status will remain connected, while its score state will change to disconnected.
```rust
let previous_state = info.score_state();
// Update scores
info.score_update();
Self::handle_score_transitions(
previous_state,
peer_id,
info,
&mut to_ban_peers,
&mut to_unban_peers,
&mut self.events,
&self.log,
);
```
```previous_state``` will be set to Disconnected, and then because ```handle_score_transitions``` only changes connection status for a peer if the state changed, the peer remains connected. Then in the heartbeat code, because we only disconnect healthy peers if we have too many peers, these peers don't get disconnected. I'm not sure realistically how often this scenario would occur, but it might be better to adjust the logic to account for scenarios where the score state implies a connection status different from the current connection status.
Co-authored-by: Kevin Lu <kevlu93@gmail.com>
## Issue Addressed
#2282
## Proposed Changes
Reduce the outbound requests made to eth1 endpoints by caching the results from `eth_chainId` and `net_version`.
Further reduce the overall request count by increasing `auto_update_interval_millis` from `7_000` (7 seconds) to `60_000` (1 minute).
This will result in a reduction from ~2000 requests per hour to 360 requests per hour (during normal operation). A reduction of 82%.
## Additional Info
If an endpoint fails, its state is dropped from the cache and the `eth_chainId` and `net_version` calls will be made for that endpoint again during the regular update cycle (once per minute) until it is back online.
Co-authored-by: Paul Hauner <paul@paulhauner.com>
The ordering of adding new peers to the peerdb and deciding when to dial them was not considered in a previous update.
This adds the condition that if a peer is not in the peer-db then it is an acceptable peer to dial.
This makes #2374 obsolete.
## Issue Addressed
NA
## Proposed Changes
Modify the configuration of [GNU malloc](https://www.gnu.org/software/libc/manual/html_node/The-GNU-Allocator.html) to reduce memory footprint.
- Set `M_ARENA_MAX` to 4.
- This reduces memory fragmentation at the cost of contention between threads.
- Set `M_MMAP_THRESHOLD` to 2mb
- This means that any allocation >= 2mb is allocated via an anonymous mmap, instead of on the heap/arena. This reduces memory fragmentation since we don't need to keep growing the heap to find big contiguous slabs of free memory.
- ~~Run `malloc_trim` every 60 seconds.~~
- ~~This shaves unused memory from the top of the heap, preventing the heap from constantly growing.~~
- Removed, see: https://github.com/sigp/lighthouse/pull/2299#issuecomment-825322646
*Note: this only provides memory savings on the Linux (glibc) platform.*
## Additional Info
I'm going to close#2288 in favor of this for the following reasons:
- I've managed to get the memory footprint *smaller* here than with jemalloc.
- This PR seems to be less of a dramatic change than bringing in the jemalloc dep.
- The changes in this PR are strictly runtime changes, so we can create CLI flags which disable them completely. Since this change is wide-reaching and complex, it's nice to have an easy "escape hatch" if there are undesired consequences.
## TODO
- [x] Allow configuration via CLI flags
- [x] Test on Mac
- [x] Test on RasPi.
- [x] Determine if GNU malloc is present?
- I'm not quite sure how to detect for glibc.. This issue suggests we can't really: https://github.com/rust-lang/rust/issues/33244
- [x] Make a clear argument regarding the affect of this on CPU utilization.
- [x] Test with higher `M_ARENA_MAX` values.
- [x] Test with longer trim intervals
- [x] Add some stats about memory savings
- [x] Remove `malloc_trim` calls & code
## Issue Addressed
Windows incompatibility.
## Proposed Changes
On windows, lighthouse needs to default to STDIN as tty doesn't exist. Also Windows uses ACLs for file permissions. So to mirror chmod 600, we will remove every entry in a file's ACL and add only a single SID that is an alias for the file owner.
Beyond that, there were several changes made to different unit tests because windows has slightly different error messages as well as frustrating nuances around killing a process :/
## Additional Info
Tested on my Windows VM and it appears to work, also compiled & tested on Linux with these changes. Permissions look correct on both platforms now. Just waiting for my validator to activate on Prater so I can test running full validator client on windows.
Co-authored-by: ethDreamer <37123614+ethDreamer@users.noreply.github.com>
Co-authored-by: Michael Sproul <micsproul@gmail.com>
## Issue Addressed
The latest version of Rust has new clippy rules & the codebase isn't up to date with them.
## Proposed Changes
Small formatting changes that clippy tells me are functionally equivalent
## Issue Addressed
N/A
## Proposed Changes
Add unit tests for the various CLI flags associated with the beacon node and validator client. These changes require the addition of two new flags: `dump-config` and `immediate-shutdown`.
## Additional Info
Both `dump-config` and `immediate-shutdown` are marked as hidden since they should only be used in testing and other advanced use cases.
**Note:** This requires changing `main.rs` so that the flags can adjust the program behavior as necessary.
Co-authored-by: Paul Hauner <paul@paulhauner.com>
