You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Label a beacon block published through the Beacon API as local, not gossip #846
Problem.publish_beacon_block hands the block a validator client published to the chain actor as BlockSource::Gossip (crates/net/p2p/src/gossipsub/handler.rs:999-1004), though it never crossed the wire.
Its telemetry rows (#765) are therefore filed under source = 'gossip' with near-zero arrival latency, mixed into the arrival distribution every cross-client comparison reads. docs/telemetry.md ("Known gaps") lists this, with the workaround of filtering on the missing decode_end row. The same path already labels the block's custody columns local (handler.rs:985-992), so blocks and columns disagree.
Why not in #765.BlockSource is not a telemetry type. It is the chain actor's view of how a block arrived, and it feeds more than the rows: import metrics and timeliness bookkeeping, for instance. A new variant is a change to ethlambda-network-api and every match on it, to be checked for each reader rather than slipped into a telemetry PR.
Fix.
Add a BlockSource::Local variant, sent by publish_beacon_block.
Give it the label local in import_timing::source_label.
Check every other reader of BlockSource and keep its current behaviour for a local block where that behaviour is right. Fork choice's timeliness treats the node's own block as timely, as gossip does now.
Remove the known-gaps entry from docs/telemetry.md.
A test: a block handed over by publish_beacon_block writes its rows under local.
Activity