description: Notes from the regular proof of stake [Eth2] implementers call
# PoS Implementers’ Call #72 - 2021-09-09
[Quick contemporaneous notes by Ben Edgington; fka "Eth2 Implementers' Call"]
Prater fork was generally successful. A couple of bumps were swiftly ironed out. Still seeing lower than expected sync aggregate participation on both Pyrmont and Prater. We need to look into this more closely to find the root cause.
In view of this, and Prysm availability, suggest we plan to do releases mid/late-September and upgrade in October. But we also do not wish to block work on The Merge with Altair stuff.
Test coverage on Altair is still being improved. Release today, and another one coming at the end of the week.
[Terence] Main need is for more enhanced testing. Suggest some "chaos monkey" scenarios around timing of attestations and blocks. Could skew the clocks of a subset of nodes on Pyrmont.
There are remaining [scenario tests](https://github.com/eth2-clients/eth2-networks/issues/59) to run on Pyrmont.
When should we set the fork epoch? In order to do releases in the last week of Sept, we must select the epoch at the next call at the latest.
## Client updates
Started implementing Merge code. Reduced memory cost of state storage by half. Have shared design for light client server with teams. Seeking feedback.
Shipped beta for Altair. Now testing the code for mainnet. Will support multi-core BLS verification as an opt-in. Will ship installers for Linux, Windows, Mac. Started work on light client. Now preparing for The Merge, both Nimbus Eth2 and Eth1. Working on mobile UI for monitoring and managing beacon nodes.
Some optimisations. Focus remains on multiple runtimes for the Altair fork. Getting closer to a working version.
Preparing next major release with weak subjectivity sync and remote signer support. Analysing attestation packing behaviour on mainnet and have discussed with all teams. Talked to a staking pool about their late block proposals, and added better handling of incoming late blocks to LH. Now implementing latest Merge spec.
Fixed Altair upgrade issue on Prater - it was an interesting edge case that prevented proposals for the first epoch. In fixing that we removed the spec `get_domain()` function entirely, which makes handling forks very explicit. Put in a bunch of fixes for including redundant attestations. Looking into attestation caches and duplicate attestation handling. Trying to do all deduplication at the gossip level rather than at the application level. There seem to be lots of old attestations hanging around on the network. A bunch more optimisation work, especially number of calls between validator client and beacon node.
Mostly merging the Altair upgrade code into the main branch. Intense code review of the consensus code and caching. Preparing for v2 release. Beta.3 spec tests revealed a consensus bug which is now fixed. Reviewing cache implementations for mainnet, since the size of the network is increasing. Fixed some attestation packing/duplication issues. A couple of new hires :tada:
[AdrianS] has [a tool](https://github.com/ajsutton/validator-monitor) for detecting blocks with redundant attestations. Most blocks do: one or two, but sometimes more significant.
## Merge discussion
Later today will share doc with targets for Merge devnets in October.
Engine API discussion continued in a separate meeting today.
Dmitry from TX/RX is working on test vectors for the Merge spec. Goal is for them to be usable by clients, with stubbed execution engine.
There is [a PR](https://github.com/ethereum/consensus-specs/pull/2581) to use the RANDAO mix for the previous slot rather than current for the execution layer. Danny has swung more in favour of this than he was initially. Partially as it would play more nicely with executable shards in future. Please review.
## Research updates
## Spec discussion and AOB
[Mikhail] How best to proceed with the engine API discussion? [Danny] Suggest putting the doc in a repo with a PR per set of methods. This would allow wider engagement and better tracking. HackMD doc is limited in how it can be commented on, and tracking updates.
* * *
# Chat highlights
From danny to Everyone: 03:01 PM
From danny to Everyone: 03:06 PM
From Adrian Sutton to Everyone: 03:24 PM
: https://github.com/ajsutton/validator-monitor for anyone interested in the redundant attestation detector. I’m thinking there’s probably a few bad behaviours it could detect over time.
From Mikhail Kalinin to Everyone: 03:26 PM