Mapping 2,000 services without asking anyone to draw a diagram
Traces, network flows and deploy config, combined into one living graph with a confidence score on every edge and a history you can rewind.
Every platform team we meet has an architecture diagram. Almost none of them are current. Services get added on Tuesday, split on Thursday and renamed the following sprint. Forge needs an accurate map to reason about incidents, so we decided early that nobody should have to draw one.
Three sources, one graph
Forge builds the map from what your systems already emit:
- Traces show which services call which, and how often.
- Network flows fill in the calls that are not instrumented, such as legacy jobs and database connections.
- Configuration from your deploy manifests adds owners, environments and declared dependencies.
Each edge carries a confidence score. An edge seen in traces every minute is near certain. An edge that only appears in a config file, but never in traffic, is marked as declared and unconfirmed. That difference is often where incidents hide.
Keeping it fresh
The map updates continuously, and Forge records every change with a timestamp. During an incident you can ask what the system looked like an hour ago, before the deploy, and compare it with now. On our largest customer, that history covers 2,140 services and 9,800 edges, and a full rebuild takes under four minutes.
What maps are good for
Grouping alerts, ranking incidents and drafting safe fixes all depend on knowing what talks to what. But the most common thing customers do with the map is simpler: they open it during onboarding and find services nobody knew still existed.