## Issue Addressed
#4146
## Proposed Changes
Removes the `ExecutionOptimisticForkVersionedResponse` type and the associated Beacon API endpoint which is now deprecated. Also removes the test associated with the endpoint.
## Issue Addressed
N/A
## Proposed Changes
Adds a flag for disabling peer scoring. This is useful for local testing and testing small networks for new features.
## Issue Addressed
Attempt to fix#3812
## Proposed Changes
Move web3signer binary download script out of build script to avoid downloading unless necessary. If this works, it should also reduce the build time for all jobs that runs compilation.
> This is currently a WIP and all features are subject to alteration or removal at any time.
## Overview
The successor to #2873.
Contains the backbone of `beacon.watch` including syncing code, the initial API, and several core database tables.
See `watch/README.md` for more information, requirements and usage.
It is possible that when we go to ban a peer, there is already an unbanned message in the queue. It could lead to the case that we ban and immediately unban a peer leaving us in a state where a should-be banned peer is unbanned.
If this banned peer connects to us in this faulty state, we currently do not attempt to re-ban it. This PR does correct this also, so if we do see this error, it will now self-correct (although we shouldn't see the error in the first place).
I have also incremented the severity of not supporting protocols as I see peers ultimately get banned in a few steps and it seems to make sense to just ban them outright, rather than have them linger.
## Issue Addressed
Addresses #4117
## Proposed Changes
See https://github.com/ethereum/keymanager-APIs/pull/58 for proposed API specification.
## TODO
- [x] ~~Add submission to BN~~
- removed, see discussion in [keymanager API](https://github.com/ethereum/keymanager-APIs/pull/58)
- [x] ~~Add flag to allow voluntary exit via the API~~
- no longer needed now the VC doesn't submit exit directly
- [x] ~~Additional verification / checks, e.g. if validator on same network as BN~~
- to be done on client side
- [x] ~~Potentially wait for the message to propagate and return some exit information in the response~~
- not required
- [x] Update http tests
- [x] ~~Update lighthouse book~~
- not required if this endpoint makes it to the standard keymanager API
Co-authored-by: Paul Hauner <paul@paulhauner.com>
Co-authored-by: Jimmy Chen <jimmy@sigmaprime.io>
## Issue Addressed
#3212
## Proposed Changes
- Introduce a new `rate_limiting_backfill_queue` - any new inbound backfill work events gets immediately sent to this FIFO queue **without any processing**
- Spawn a `backfill_scheduler` routine that pops a backfill event from the FIFO queue at specified intervals (currently halfway through a slot, or at 6s after slot start for 12s slots) and sends the event to `BeaconProcessor` via a `scheduled_backfill_work_tx` channel
- This channel gets polled last in the `InboundEvents`, and work event received is wrapped in a `InboundEvent::ScheduledBackfillWork` enum variant, which gets processed immediately or queued by the `BeaconProcessor` (existing logic applies from here)
Diagram comparing backfill processing with / without rate-limiting:
https://github.com/sigp/lighthouse/issues/3212#issuecomment-1386249922
See this comment for @paulhauner's explanation and solution: https://github.com/sigp/lighthouse/issues/3212#issuecomment-1384674956
## Additional Info
I've compared this branch (with backfill processing rate limited to to 1 and 3 batches per slot) against the latest stable version. The CPU usage during backfill sync is reduced by ~5% - 20%, more details on this page:
https://hackmd.io/@jimmygchen/SJuVpJL3j
The above testing is done on Goerli (as I don't currently have hardware for Mainnet), I'm guessing the differences are likely to be bigger on mainnet due to block size.
### TODOs
- [x] Experiment with processing multiple batches per slot. (need to think about how to do this for different slot durations)
- [x] Add option to disable rate-limiting, enabed by default.
- [x] (No longer required now we're reusing the reprocessing queue) Complete the `backfill_scheduler` task when backfill sync is completed or not required
## Issue Addressed
Update the database-migrations to include v4.0.1 for database version v16:
## Proposed Changes
Update the table by adding a row
## Additional Info
Please provide any additional information. For example, future considerations
or information useful for reviewers.
## Issue Addressed
#4127. PR to test port conflicts in CI tests .
## Proposed Changes
See issue for more details, potential solution could be adding a cache bound by time to the `unused_port` function.
## Issue Addressed
#3708
## Proposed Changes
- Add `is_finalized_block` method to `BeaconChain` in `beacon_node/beacon_chain/src/beacon_chain.rs`.
- Add `is_finalized_state` method to `BeaconChain` in `beacon_node/beacon_chain/src/beacon_chain.rs`.
- Add `fork_and_execution_optimistic_and_finalized` in `beacon_node/http_api/src/state_id.rs`.
- Add `ExecutionOptimisticFinalizedForkVersionedResponse` type in `consensus/types/src/fork_versioned_response.rs`.
- Add `execution_optimistic_finalized_fork_versioned_response`function in `beacon_node/http_api/src/version.rs`.
- Add `ExecutionOptimisticFinalizedResponse` type in `common/eth2/src/types.rs`.
- Add `add_execution_optimistic_finalized` method in `common/eth2/src/types.rs`.
- Update API response methods to include finalized.
- Remove `execution_optimistic_fork_versioned_response`
Co-authored-by: Michael Sproul <michael@sigmaprime.io>
## Title
Optimise `update_validators` by decrypting key cache only when necessary
## Issue Addressed
Resolves [#3968: Slow performance of validator client PATCH API with hundreds of keys](https://github.com/sigp/lighthouse/issues/3968)
## Proposed Changes
1. Add a check to determine if there is at least one local definition before decrypting the key cache.
2. Assign an empty `KeyCache` when all definitions are of the `Web3Signer` type.
3. Perform cache-related operations (e.g., saving the modified key cache) only if there are local definitions.
## Additional Info
This PR addresses the excessive CPU usage and slow performance experienced when using the `PATCH lighthouse/validators/{pubkey}` request with a large number of keys. The issue was caused by the key cache using cryptography to decipher and cipher the cache entities every time the request was made. This operation called `scrypt`, which was very slow and required a lot of memory when there were many concurrent requests.
These changes have no impact on the overall functionality but can lead to significant performance improvements when working with remote signers. Importantly, the key cache is never used when there are only `Web3Signer` definitions, avoiding the expensive operation of decrypting the key cache in such cases.
Co-authored-by: Maksim Shcherbo <max.shcherbo@consensys.net>
## Issue Addressed
Which issue # does this PR address?
https://github.com/sigp/lighthouse/issues/3669
## Proposed Changes
Please list or describe the changes introduced by this PR.
- A new API to fetch fork choice data, as specified [here](https://github.com/ethereum/beacon-APIs/pull/232)
- A new integration test to test the new API
## Additional Info
Please provide any additional information. For example, future considerations
or information useful for reviewers.
- `extra_data` field specified in the beacon-API spec is not implemented, please let me know if I should instead.
Co-authored-by: Michael Sproul <micsproul@gmail.com>
## Issue Addressed
The minimum supported Rust version has been set to 1.66 as of Lighthouse v4.0.0. This PR updates Rust to 1.66 in lcli Dockerfile.
Co-authored-by: Jimmy Chen <jimmy@sigmaprime.io>
## Proposed Changes
To prevent breakages from `cargo update`, this updates the `arbitrary` crate to a new commit from my fork. Unfortunately we still need to use my fork (even though my `bound` change was merged) because of this issue: https://github.com/rust-lang/rust-clippy/issues/10185.
In a couple of Rust versions it should be resolved upstream.
## Issue Addressed
NA
## Proposed Changes
- Bump versions.
- Bump openssl version to resolve various `cargo audit` notices.
## Additional Info
- Requires further testing
## Issue Addressed
NA
## Proposed Changes
- Implements https://github.com/ethereum/consensus-specs/pull/3290/
- Bumps `ef-tests` to [v1.3.0-rc.4](https://github.com/ethereum/consensus-spec-tests/releases/tag/v1.3.0-rc.4).
The `CountRealizedFull` concept has been removed and the `--count-unrealized-full` and `--count-unrealized` BN flags now do nothing but log a `WARN` when used.
## Database Migration Debt
This PR removes the `best_justified_checkpoint` from fork choice. This field is persisted on-disk and the correct way to go about this would be to make a DB migration to remove the field. However, in this PR I've simply stubbed out the value with a junk value. I've taken this approach because if we're going to do a DB migration I'd love to remove the `Option`s around the justified and finalized checkpoints on `ProtoNode` whilst we're at it. Those options were added in #2822 which was included in Lighthouse v2.1.0. The options were only put there to handle the migration and they've been set to `Some` ever since v2.1.0. There's no reason to keep them as options anymore.
I started adding the DB migration to this branch but I started to feel like I was bloating this rather critical PR with nice-to-haves. I've kept the partially-complete migration [over in my repo](https://github.com/paulhauner/lighthouse/tree/fc-pr-18-migration) so we can pick it up after this PR is merged.
## Issue Addressed
NA
## Proposed Changes
Sets the mainnet Capella fork epoch as per https://github.com/ethereum/consensus-specs/pull/3300
## Additional Info
I expect the `ef_tests` to fail until we get a compatible consensus spec tests release.
## Issue Addressed
NA
## Proposed Changes
Replaces #4058 to attempt to reduce `ERRO Failed to send scheduled attestation` spam and provide more information for diagnosis. With this PR we achieve:
- When dequeuing attestations after a block is received, send only one log which reports `n` failures (rather than `n` logs reporting `n` failures).
- Make a distinction in logs between two separate attestation dequeuing events.
- Add more information to both log events to help assist with troubleshooting.
## Additional Info
NA
This PR enables the user to adjust the shuffling cache size.
This is useful for some HTTP API requests which require re-computing old shufflings. This PR currently optimizes the
beacon/states/{state_id}/committees HTTP API by first checking the cache before re-building shuffling.
If the shuffling is set to a non-default value, then the HTTP API request will also fill the cache when as it constructs new shufflings.
If the CLI flag is not present or the value is set to the default of 16 the default behaviour is observed.
Co-authored-by: Michael Sproul <michael@sigmaprime.io>
Currently Lighthouse will remain uncontactable if users port forward a port that is not the same as the one they are listening on.
For example, if Lighthouse runs with port 9000 TCP/UDP locally but a router is configured to pass 9010 externally to the lighthouse node on 9000, other nodes on the network will not be able to reach the lighthouse node.
This occurs because Lighthouse does not update its ENR TCP port on external socket discovery. The intention was always that users should use `--enr-tcp-port` to customise this, but this is non-intuitive.
The difficulty arises because we have no discovery mechanism to find our external TCP port. If we discovery a new external UDP port, we must guess what our external TCP port might be. This PR assumes the external TCP port is the same as the external UDP port (which may not be the case) and thus updates the TCP port along with the UDP port if the `--enr-tcp-port` flag is not set.
Along with this PR, will be added documentation to the Lighthouse book so users can correctly understand and configure their ENR to maximize Lighthouse's connectivity.
This relies on https://github.com/sigp/discv5/pull/166 and we should wait for a new release in discv5 before adding this PR.
If a node is also a bootnode it can try to add itself to its own local routing table which will emit an error.
The error is entirely harmless but we would prefer to avoid emitting the error.
This PR does not attempt to add a boot node ENR if that ENR corresponds to our local peer-id/node-id.
## Issue Addressed
Resolves#4061
## Proposed Changes
Adds a message to tell users to check their EE.
## Additional Info
I really struggled to come up with something succinct and complete, so I'm totally open to feedback.
## Issue Addressed
NA
## Proposed Changes
When producing a block from a builder, there are two points where we could consider the block "broadcast":
1. When the blinded block is published to the builder.
2. When the un-blinded block is published to the P2P network (this is always *after* the previous step).
Our logging for late block broadcasts was using (2) for builder-blocks, which was creating a lot of false-positive logs. This is because the builder publishes the block on the P2P network themselves before returning it to us and we perform (2). For clarity, the logs were false-positives because we claim that the block was published late by us when it was actually published earlier by the builder.
This PR changes our logging behavior so we do our logging at (1) instead. It also updates our metrics for block broadcast to distinguish between local and builder blocks. I believe the metrics change will be natively compatible with existing Grafana dashboards.
## Additional Info
One could argue that the builder *should* return the block to us faster, however that's not the case. I think it's more important that we don't desensitize users with false-positives.
## Issue Addressed
Closes#3814, replaces #3818.
## Proposed Changes
* Add a WARN log for the case where we are attempting to sync chain segments but can't process them because they're building on an invalid parent. The most common case where we see this is when the execution node database is corrupt, causing sync to stall mysteriously (because we're currently logging the failure only at debug level).
* Additionally I've bumped up the logging for invalid execution payloads to `WARN`. This may result in some duplicate logs as we log errors from the `beacon_chain` and then again from the beacon processor. Invalid payloads and corrupt DBs _should_ be rare enough that this doesn't produce overwhelming log volume.
## Issue Addressed
Added note in lighthouse book to instruct users to use a min lighthouse requirement to run Siren Ui.
Which issue # does this PR address?
## Proposed Changes
Please list or describe the changes introduced by this PR.
## Additional Info
Please provide any additional information. For example, future considerations
or information useful for reviewers.
There is a race condition which occurs when multiple discovery queries return at almost the exact same time and they independently contain a useful peer we would like to connect to.
The condition can occur that we can add the same peer to the dial queue, before we get a chance to process the queue.
This ends up displaying an error to the user:
```
ERRO Dialing an already dialing peer
```
Although this error is harmless it's not ideal.
There are two solutions to resolving this:
1. As we decide to dial the peer, we change the state in the peer-db to dialing (before we add it to the queue) which would prevent other requests from adding to the queue.
2. We prevent duplicates in the dial queue
This PR has opted for 2. because 1. will complicate the code in that we are changing states in non-intuitive places. Although this technically adds a very slight performance cost, its probably a cleaner solution as we can keep the state-changing logic in one place.
## Issue Addressed
#3938
## Proposed Changes
- `network::Processor` is deleted and all it's logic is moved to `network::Router`.
- The `network::Router` module is moved to a single file.
- The following functions are deleted: `on_disconnect` `send_status` `on_status_response` `on_blocks_by_root_request` `on_lightclient_bootstrap` `on_blocks_by_range_request` `on_block_gossip` `on_unaggregated_attestation_gossip` `on_aggregated_attestation_gossip` `on_voluntary_exit_gossip` `on_proposer_slashing_gossip` `on_attester_slashing_gossip` `on_sync_committee_signature_gossip` `on_sync_committee_contribution_gossip` `on_light_client_finality_update_gossip` `on_light_client_optimistic_update_gossip`. This deletions are possible because the updated `Router` allows the underlying methods to be called directly.
## Issue Addressed
- Add a complete match for `Protocol` here.
- The incomplete match was causing us not to append context bytes to the light client protocols
- This is the relevant part of the spec and it looks like context bytes are defined https://github.com/ethereum/consensus-specs/blob/dev/specs/altair/light-client/p2p-interface.md#getlightclientbootstrap
Disclaimer: I have no idea if people are using it but it shouldn't have been working so not sure why it wasn't caught
Co-authored-by: realbigsean <seananderson33@gmail.com>
## Issue Addressed
In #4027 I forgot to add the `parent_block_number` to the payload attributes SSE.
## Proposed Changes
Compute the parent block number while computing the pre-payload attributes. Pass it on to the SSE stream.
## Additional Info
Not essential for v3.5.1 as I suspect most builders don't need the `parent_block_root`. I would like to use it for my dummy no-op builder however.
## Issue Addressed
Add support for ipv6 and dual stack in lighthouse.
## Proposed Changes
From an user perspective, now setting an ipv6 address, optionally configuring the ports should feel exactly the same as using an ipv4 address. If listening over both ipv4 and ipv6 then the user needs to:
- use the `--listen-address` two times (ipv4 and ipv6 addresses)
- `--port6` becomes then required
- `--discovery-port6` can now be used to additionally configure the ipv6 udp port
### Rough list of code changes
- Discovery:
- Table filter and ip mode set to match the listening config.
- Ipv6 address, tcp port and udp port set in the ENR builder
- Reported addresses now check which tcp port to give to libp2p
- LH Network Service:
- Can listen over Ipv6, Ipv4, or both. This uses two sockets. Using mapped addresses is disabled from libp2p and it's the most compatible option.
- NetworkGlobals:
- No longer stores udp port since was not used at all. Instead, stores the Ipv4 and Ipv6 TCP ports.
- NetworkConfig:
- Update names to make it clear that previous udp and tcp ports in ENR were Ipv4
- Add fields to configure Ipv6 udp and tcp ports in the ENR
- Include advertised enr Ipv6 address.
- Add type to model Listening address that's either Ipv4, Ipv6 or both. A listening address includes the ip, udp port and tcp port.
- UPnP:
- Kept only for ipv4
- Cli flags:
- `--listen-addresses` now can take up to two values
- `--port` will apply to ipv4 or ipv6 if only one listening address is given. If two listening addresses are given it will apply only to Ipv4.
- `--port6` New flag required when listening over ipv4 and ipv6 that applies exclusively to Ipv6.
- `--discovery-port` will now apply to ipv4 and ipv6 if only one listening address is given.
- `--discovery-port6` New flag to configure the individual udp port of ipv6 if listening over both ipv4 and ipv6.
- `--enr-udp-port` Updated docs to specify that it only applies to ipv4. This is an old behaviour.
- `--enr-udp6-port` Added to configure the enr udp6 field.
- `--enr-tcp-port` Updated docs to specify that it only applies to ipv4. This is an old behaviour.
- `--enr-tcp6-port` Added to configure the enr tcp6 field.
- `--enr-addresses` now can take two values.
- `--enr-match` updated behaviour.
- Common:
- rename `unused_port` functions to specify that they are over ipv4.
- add functions to get unused ports over ipv6.
- Testing binaries
- Updated code to reflect network config changes and unused_port changes.
## Additional Info
TODOs:
- use two sockets in discovery. I'll get back to this and it's on https://github.com/sigp/discv5/pull/160
- lcli allow listening over two sockets in generate_bootnodes_enr
- add at least one smoke flag for ipv6 (I have tested this and works for me)
- update the book
## Proposed Changes
The current `/lighthouse/nat` implementation checks for _zero_ address updated messages, when it should check for a _non-zero_ number. This was spotted while debugging an issue on Discord where a user's ports weren't forwarded but `/lighthouse/nat` was still returning `true`.
## Issue Addressed
#3435
## Proposed Changes
Fire a warning with the path of JWT to be created when the path given by --execution-jwt is not found
Currently, the same error is logged if the jwt is found but doesn't match the execution client's jwt, and if no jwt was found at the given path. This makes it very hard to tell if you accidentally typed the wrong path, as a new jwt is created silently that won't match the execution client's jwt. So instead, it will now fire a warning stating that a jwt is being generated at the given path.
## Additional Info
In the future, it may be smarter to handle this case by adding an InvalidJWTPath member to the Error enum in lib.rs or auth.rs
that can be handled during upcheck()
This is my first PR and first project with rust. so thanks to anyone who looks at this for their patience and help!
Co-authored-by: Sebastian Richel <47844429+sebastianrich18@users.noreply.github.com>
## Proposed Changes
Two tiny updates to satisfy Clippy 1.68
Plus refactoring of the `http_api` into less complex types so the compiler can chew and digest them more easily.
Co-authored-by: Michael Sproul <michael@sigmaprime.io>
## Issue Addressed
#4040
## Proposed Changes
- Add the `always_prefer_builder_payload` field to `Config` in `beacon_node/client/src/config.rs`.
- Add that same field to `Inner` in `beacon_node/execution_layer/src/lib.rs`
- Modify the logic for picking the payload in `beacon_node/execution_layer/src/lib.rs`
- Add the `always-prefer-builder-payload` flag to the beacon node CLI
- Test the new flags in `lighthouse/tests/beacon_node.rs`
Co-authored-by: Paul Hauner <paul@paulhauner.com>
## Issue Addressed
NA
## Proposed Changes
In #4024 we added metrics to expose the latency measurements from a VC to each BN. Whilst playing with these new metrics on our infra I realised it would be great to have a single metric to make sure that the primary BN for each VC has a reasonable latency. With the current "metrics for all BNs" it's hard to tell which is the primary.
## Additional Info
NA
## Issue Addressed
NA
## Proposed Changes
- Adds a `WARN` statement for Capella, just like the previous forks.
- Adds a hint message to all those WARNs to suggest the user update the BN or VC.
## Additional Info
NA
## Proposed Changes
Fix the cargo audit failure caused by [RUSTSEC-2023-0018](https://rustsec.org/advisories/RUSTSEC-2023-0018) which we were exposed to via `tempfile`.
## Additional Info
I've held back the libp2p crate for now because it seemed to introduce another duplicate dependency on libp2p-core, for a total of 3 copies. Maybe that's fine, but we can sort it out later.
## Issue Addressed
Closes#3963 (hopefully)
## Proposed Changes
Compute attestation selection proofs gradually each slot rather than in a single `join_all` at the start of each epoch. On a machine with 5k validators this replaces 5k tasks signing 5k proofs with 1 task that signs 5k/32 ~= 160 proofs each slot.
Based on testing with Goerli validators this seems to reduce the average time to produce a signature by preventing Tokio and the OS from falling over each other trying to run hundreds of threads. My testing so far has been with local keystores, which run on a dynamic pool of up to 512 OS threads because they use [`spawn_blocking`](https://docs.rs/tokio/1.11.0/tokio/task/fn.spawn_blocking.html) (and we haven't changed the default).
An earlier version of this PR hyper-optimised the time-per-signature metric to the detriment of the entire system's performance (see the reverted commits). The current PR is conservative in that it avoids touching the attestation service at all. I think there's more optimising to do here, but we can come back for that in a future PR rather than expanding the scope of this one.
The new algorithm for attestation selection proofs is:
- We sign a small batch of selection proofs each slot, for slots up to 8 slots in the future. On average we'll sign one slot's worth of proofs per slot, with an 8 slot lookahead.
- The batch is signed halfway through the slot when there is unlikely to be contention for signature production (blocks are <4s, attestations are ~4-6 seconds, aggregates are 8s+).
## Performance Data
_See first comment for updated graphs_.
Graph of median signing times before this PR:
Graph of update attesters metric (includes selection proof signing) before this PR:
Median signing time after this PR (prototype from 12:00, updated version from 13:30):
99th percentile on signing times (bounded attestation signing from 16:55, now removed):
Attester map update timing after this PR:
Selection proof signings per second change:
## Link to late blocks
I believe this is related to the slow block signings because logs from Stakely in #3963 show these two logs almost 5 seconds apart:
> Feb 23 18:56:23.978 INFO Received unsigned block, slot: 5862880, service: block, module: validator_client::block_service:393
> Feb 23 18:56:28.552 INFO Publishing signed block, slot: 5862880, service: block, module: validator_client::block_service:416
The only thing that happens between those two logs is the signing of the block:
0fb58a680d/validator_client/src/block_service.rs (L410-L414)
Helpfully, Stakely noticed this issue without any Lighthouse BNs in the mix, which pointed to a clear issue in the VC.
## TODO
- [x] Further testing on testnet infrastructure.
- [x] Make the attestation signing parallelism configurable.
## Issue Addressed
Closes#3896Closes#3998Closes#3700
## Proposed Changes
- Optimise the calculation of withdrawals for payload attributes by avoiding state clones, avoiding unnecessary state advances and reading from the snapshot cache if possible.
- Use the execution layer's payload attributes cache to avoid re-calculating payload attributes. I actually implemented a new LRU cache just for withdrawals but it had the exact same key and most of the same data as the existing payload attributes cache, so I deleted it.
- Add a new SSE event that fires when payloadAttributes are calculated. This is useful for block builders, a la https://github.com/ethereum/beacon-APIs/issues/244.
- Add a new CLI flag `--always-prepare-payload` which forces payload attributes to be sent with every fcU regardless of connected proposers. This is intended for use by builders/relays.
For maximum effect, the flags I've been using to run Lighthouse in "payload builder mode" are:
```
--always-prepare-payload \
--prepare-payload-lookahead 12000 \
--suggested-fee-recipient 0x0000000000000000000000000000000000000000
```
The fee recipient is required so Lighthouse has something to pack in the payload attributes (it can be ignored by the builder). The lookahead causes fcU to be sent at the start of every slot rather than at 8s. As usual, fcU will also be sent after each change of head block. I think this combination is sufficient for builders to build on all viable heads. Often there will be two fcU (and two payload attributes) sent for the same slot: one sent at the start of the slot with the head from `n - 1` as the parent, and one sent after the block arrives with `n` as the parent.
Example usage of the new event stream:
```bash
curl -N "http://localhost:5052/eth/v1/events?topics=payload_attributes"
```
## Additional Info
- [x] Tests added by updating the proposer re-org tests. This has the benefit of testing the proposer re-org code paths with withdrawals too, confirming that the new changes don't interact poorly.
- [ ] Benchmarking with `blockdreamer` on devnet-7 showed promising results but I'm yet to do a comparison to `unstable`.
Co-authored-by: Michael Sproul <micsproul@gmail.com>
