beacon rpc sometimes returns LightClientFinalityUpdate with sync_committee length < 2/3

[DragonDev1906]

Describe the bug
When requesting a LightClientFinalityUpdate via the beacon_chain RPC the endpoints sometimes returns an update with only a few entries in the sync committee (as shown in the screenshots). I'm not sure if this is a bug in Nimbus, but I can't think of anything where such a FinalityUpdate would be useful.

According to the consensus specs light clients should:

• check the validity of such updates (see process_light_client_update),
• update store.best_valid_update
• update store.current_max_active_participants if the update has more participants set (which should never get this low),
• update the optimistic header (if the light client wants this updated asap he'd use LightClientOptimisticUpdate and
• apply the update (and thus the finalized header) if the sync committee is large enough (2/3).

Please correct me if I understood something wrong in the consensus specs or if there is another reason why LightClientFinalityUpdate with less than 2/3 of the sync committee would be useful. My current best guess is that it isn't useful at all and probably shouldn't be returned by the /eth/v1/beacon/light_client/finality_update rpc endpoint. I can easily check for the sync committee size to avoid this issue, instead of just checking the finalized blocknumber, which I mainly do to avoid unnecessary communication internally (I don't care about the optimistic updates/blocks/data).

I don't know if this also affects the subscription (I'll likely change it to use the subscription at some point, but for now I'm polling the rpc endpoint).

It could of course also be that my script is wrong and doesn't decode the sync committee correctly, hence the question of whether I understood something wrong when reading the consensus-specs.

To Reproduce
Steps to reproduce the behavior:

1. Platform details (OS, architecture): Arch Linux, x86 (shouldn't be relevant since this is about the API and I'm using the http://unstable.prater.beacon-api.nimbus.team/ node.
2. Branch/commit used: N/A (I don't know which version that node uses currently, I've tested it today).
3. Commands being executed: Custom code that polls this RPC endpoint every 12 seconds, prints the sync committee size of the update

I additionally gives these updates to a light client if the finalized block number changed. Sometimes this then didn't apply the update, which is how I found this.

Screenshots (logs from my application)
Log output of an application polling the RPC endpoint every 12 seconds and outputting the sync committee size:

Log output showing the situation I've found: A finality update with <2/3 sync_committee (the first one I received with a new finalized_header..execution.block_number), together with debug logs of the light client showing that the sync committee was the reason that the update was not applied.

Additional Info
This seems to happen quite regularly, most Epochs give me a FinalityUpdate with less than 2/3 sync committees.

[DragonDev1906]

Here is some more information:
It looks like the FinalityUpdate changes whenever the attested header changes, even if the sync_committee_size is 1. Though I'm still not sure when this would be useful when requesting a LightClientFinalityUpdate.

[etan-status]

• For the /eth/v1/beacon/light_client/finality_update endpoint, part of this is due to some blocks simply not including a good sync aggregate, in this case you have to wait for the next block to see whether it contains better data.
• For the /eth/v1/events push based mechanism, it would indeed be better if a second LightClientFinalityUpdate would be pushed in such cases, once an update with supermajority sync committee participation becomes available (in case that the initial one has low participation):
  • Gossip additional LightClientFinalityUpdate on new supermajority ethereum/consensus-specs#3549
  • send LC finality update on event stream on supermajority #5602

[etan-status]

It could of course also be that my script is wrong and doesn't decode the sync committee correctly, hence the question of whether I understood something wrong when reading the consensus-specs.

You can check the next block (attested_header.beacon.slot + 1) on a block explorer.

For example, https://prater.beaconcha.in/slot/6709376 contains the sync_aggregate for attested_header.beacon.slot = 6709376, confirming that there was indeed only a single signature included in that block (out of 512).

Keep in mind that the Goerli network is soon deprecated, Sepolia / Holesky should have better performance in that regard.

[DragonDev1906]

For the /eth/v1/beacon/light_client/finality_update endpoint, part of this is due to some blocks simply not including a good sync aggregate, in this case you have to wait for the next block to see whether it contains better data.

You're right. I wasn't expecting the specs to say that create_light_client_update should completely ignore the sync committee size when creating an update, but it does make sense if the light client is interested in the optimistic header, as the optimistic header is updated with sync_committee_size>=1. ¹ Also makes sense since the same code is used for create_light_client_finality_update and create_light_client_optimistic_update.

For some reason I expected create_light_client_finality_update to not bother about updates that did not forward the finalized_header, but they can of course still be used to get optimistic header updates. Though unfortunately this means that it can sometimes be impossible to get a finalized_header through the beacon API until the next block is mined.

The spec even says this (apparently I can't read): Full nodes SHOULD provide the LightClientFinalityUpdate with the highest attested_header.beacon.slot

Keep in mind that the Goerli network is soon deprecated, Sepolia / Holesky should have better performance in that regard.

Yeah, that's why I'm currently testing on both Goerli and Sepolia. Higher chance to get unexpected or not quite normal situations that still can happen.

Footnotes

1. https://ethereum.github.io/consensus-specs/specs/altair/light-client/sync-protocol/#process_light_client_update ↩

[etan-status]

With #5602 (unstable branch) it should now be possible for you to simply track the eventstream. When monitoring topics=light_client_finality_update, you will now get an update both when new finality is achieved, and additionally when supermajority is available (if the initial update did not have supermajority participation).

Can also try in browser / curl (but have to wait ~5-10 mins for finality to advance to see something)

• http://unstable.mainnet.beacon-api.nimbus.team/eth/v1/events?topics=light_client_finality_update