Lineage you can actually read
The new graph view collapses 200-node pipelines into the five hops that matter.
Lineage graphs are usually accurate and unreadable. A real workspace has 200 views and the graph looks like a bowl of noodles.
Start from the number
The new view starts from a single value on a dashboard and walks back to the raw events that produced it. Everything that did not contribute is folded away.
Hops are grouped by stage: ingest, transform, serve.
Each hop shows its freshness and last change.
One click opens the SQL that produced it.
In testing, engineers found the source of a wrong number in a median of 48 seconds, down from 11 minutes with the full graph.
More from the engine room.
How we cut p99 from 11 ms to 4.2 ms
Three changes to the read path in v4.0: declared sort keys, a per-core allocator and a planner that stopped re-sorting partitions at query time.
Exactly-once without two-phase commit
Idempotent sinks, epoch fencing and a 64-bit sequence per partition. No coordinator, no stalls.
Time travel is just a really good index
Why AS OF queries on 30 days of history cost less than one extra replica.
