NETZlive in production

Backs Act 2: how we made it real.

Copied documentation

netzlive.cluster and the pages under it are copies of the NETZlive docs site’s systeme/cluster/ section (~/Work/NETZlive/docs), taken 2026-10-09. That site is authoritative. Pages directly in this section cover what the talk needs that those pages don’t: the history, the incidents, the reasoning.

Questions to answer

• Service registration via OTP start phases. Where it happens, and what “ready” means.
• The race condition — a service advertised before its dependencies loaded. Date, symptoms, fix. Reconstruct from the commit or PR, not from memory.
• defdelegate_api. What the macro expands to: node lookup, RPC, error wrapping.
• Versioned contracts. How a service declares its version and how a client requires one.
• The silent contract break — scenario_id → scenario_ref. Is this a real incident or a composite? The talk must not present a composite as a single event.
• NodeError vs RPCError, and where retry config lives.
• Two-tier messaging. The RPC-vs-Kafka rule, and why Phoenix.PubSub stays local.

Where the evidence lives

• netzlive.cluster/lib/ — the wrapper library.
• redispatch/ and the other services, for real call sites.
• Git history and incident notes for the war stories.