Docs Run a node
Run a local cluster
Run twelve validators in three hubs on one machine with acrux net up, send traffic through it, crash a proposer and read the run report.
acrux net up runs a whole New Constellation net on one machine: twelve validators in three hubs, FRA, NYC and TYO, four in each, every validator its own actor with its own key, message state, executor and Votor. They exchange signed messages over loopback QUIC with the delays of their hubs' geography. Acrux Net runs from the same command.
acrux net upWhat runs
| Validators | 12, equal stake, four per hub |
| Proposers | 3 per window, drawn by the schedule |
| Slots | 50 ms ms, 16 per window, 4 per cblock |
| Delays | inside a hub and between hubs as on Parameters |
| Traffic | 50 generated transfers per second, --traffic-rate to change it |
| Gateway | JSON-RPC 127.0.0.1:4930, events ws://127.0.0.1:4931 |
| Metrics | http://127.0.0.1:4989/metrics and /health |
| Transport | QUIC on twelve ports from 127.0.0.1:4950 |
A 60% certificate needs eight of the twelve validators, so every cblock is decided by at least two hubs.
Point the SDK or /net at the gateway and the cluster is a net like any other.
A bounded run
acrux net up --run-seconds 30 --traffic-rate 50The run stops after 30 seconds. Without --run-seconds it runs until Ctrl+C.
Crash a proposer
acrux net up --crash-window 2 --crash-proposer 3 --run-seconds 30Proposer 3 of window 2 is silent from the window's start. The window continues on P = {1, 2}, and each slot's first vote waits for the attestation deadline. The same case runs as the Crash scenario.
Load a program
acrux net up --program CUBR9LNdTwxwp5ARBQj1EopfuH8dr8u4i7DsXcb8HbHN=../chain/target/deploy/demo_pool.soEach --program ADDRESS=PATH loads an SBF program into every validator's SVM before the first slot. Repeat it for more programs.
Stop and the report
Stop with Ctrl+C, or let --run-seconds end the run. The node closes its transport, writes last-run.json into the store directory and prints one summary line:
| Field | Meaning |
|---|---|
report | the path of last-run.json |
finalizedCblocks | cblocks finalized in the run |
skips | cblocks skipped |
slotPacingUs | slot start-to-start pacing: count, p50, p99, maximum, µs |
The report holds, among the rest:
| Field | Meaning |
|---|---|
withinWindowPacingUs, windowHandoverGapUs | pacing inside windows, and the gap at each window boundary |
proposerHandoverGapUs | the handover gap per hub link, keyed like FRA-NYC |
commitSpreadUs | first to last commitment of a slot |
finalizationUs | first slot of a cblock to its finalization |
finalizedCblocks, skippedCblocks | counts |
comparedStateHashHeights, mismatchingStateHashHeights | heights whose state hashes were compared across validators, and how many differed |
maximumExecutionSkewSlots, maximumExecutionLagSlots | how far validators' execution drifted apart and behind |
stateHashes, validatorLastExecuted | each validator's final state hash and last executed slot |
generatedTransactions | transfers the traffic generator sent |
The store
--store names the run's directory, target/runs/net by default, relative to the working directory. Each start is a fresh genesis.
| File | Holds |
|---|---|
genesis.json | the run's configuration, seed, hubs and their members' identities |
net-<start>.redb | slot, cblock and transaction records; reads of evicted slots fall back to it |
last-run.json | the run report, written on stop |
profile.json | phase timings, written on stop |
Give each run its own directory to keep its report: --store target/runs/crash-30s.