Jimmy Chen

In the test the 95th percentile was used (instead of 50th), and showing very different results: > I used `histogram_quantile(0.95, rate(beacon_data_column_sidecar_gossip_verification_seconds_bucket{network=~'$network', execution_client=~'$execution', consensus_client=~'$consensus'}[$__rate_interval]))` so that we could see the data...

Thanks @pawanjay176!

@pawanjay176 I was thinking if the following issue may be related - multiple concurrent writes holding up the fork choice write lock https://github.com/sigp/lighthouse/issues/6439

I've done some more testing and I'm seeing numbers between 10ms and 20ms (`time.busy`), so the metrics were actually not very accurate due to bad prometheus buckets: ``` 2025-07-06T14:59:16.779858Z INFO...

Resolved in #7921

Nice, thanks @eserilev

Completed in #7582

That'd be great, thanks @gitToki ☺

I did a bit of testing and got a breakdown here: https://hackmd.io/5qRLqpJaSFy7DOHSuyJWxw?view#gossip_data_column The finding suggested: 1. Gossip verification time metric on Grafana was a bit inflated because of how the...

The same is also happening with `compute_cells`, which is a really fast operation - benchmark is 12ms for 20 blobs on my Mac M1 Pro. However metrics shows that it's...