Testing and deployment

A plugin that passes unit tests can still stall a cluster. The failure modes that matter — cross-node divergence, replay mismatch, executor double-delivery — only appear when you test for them deliberately.

The testing ladder

Climb it in order. Each rung catches a class of bug the one below cannot.

1. Unit-test codecs and deterministic transitions. Round-trip every encoding. Assert exact bytes, not just equality of decoded objects.

2. Run the state-machine conformance and replay matrix. Replay the same blocks twice on one member and assert an identical root. This catches reproducibility bugs.

3. Test malformed and hostile finalized input through apply(). Truncated bodies, wrong types, deeply nested structures, oversized collections, duplicate ids, and values at every declared bound. Each must produce a deterministic no-op, never an escaping exception.

4. Start an embedded multi-member cluster with appchain-testkit. @AppChainCluster gives you a real multi-member consensus round inside a JUnit 5 test. This is the first rung that can catch cross-node divergence — a HashMap iteration bug passes rungs 1–3 and fails here.

5. Verify root parity, proof keys, restart, catch-up, rollback/reapply, and plugin packaging. Every member must expose the same root at the same height, and a member that joins late must reach it by verified catch-up.

6. For effects, test the failure modes explicitly: crash-before-send, send-before-ack, retry, reconciliation, parking, requeue, and duplicate idempotency. External execution is at-least-once, so a receiver that is not idempotent will be found here or in production.

7. Run a packaged JVM cluster. The real distribution, the real plugins/ directory, the real catalog validation. Packaging bugs — a missing transitive dependency, an embedded host SPI class — only surface here.

Test dependencies

Artifact	Purpose
yano-appchain-core-testkit	JUnit 5 @AppChainCluster embedded clusters.
yano-x-effects-testkit	Effect executor and result-path testing.
yano-x-client	REST, SSE, and client-side proof verification from tests.
yano-x-eutxo-testkit, yano-x-eutxo-zk-testkit	For the eUTxO and ZK ledgers.

Gates in the Yano X build

If you are contributing a plugin to Yano X itself rather than shipping your own, choose gates in proportion to the change:

./gradlew :state-machines:stdlib:test          # focused unit tests
./gradlew verifyArtifactInventory              # module/publication/manifest/bundle changes
./gradlew verifyJvmOnlyBuild                   # build topology changes
./gradlew integrationTest                      # integration domains
./gradlew cryptoTest                           # crypto domains
./gradlew clean build                          # dependency, bundle, isolation, packaging

verifyArtifactInventory checks that every module has exactly one declared artifact identity and that runtime plugins have bundle publications. verifyJvmOnlyBuild rejects accidental native-image tasks.

See Developing Yano X for the full contributor workflow.

Deployment rules

• Give every bundle a stable plugin id and a semantic version. The id is a consensus-visible identity; do not reuse it for different semantics.
• Verify catalog state and health on every member before admitting traffic. A member whose catalog failed to validate should not receive submissions.
• Treat plugin removal or drift as a deployment error, not as something to fall back from automatically. Silent fallback would change semantics.
• Namespace configuration and metrics by plugin and contribution, so an operator can attribute a problem to the right bundle.
• Keep plugin domain APIs read-only unless commands still enter through the authenticated app-chain submission path.

Installing on a cluster

# On every member's distribution:
cp shipment-yano-plugin.jar /opt/yano-x/plugins/
tools/yano-plugins/bin/yano-plugins validate /opt/yano-x/plugins/*.jar

# Before admitting traffic:
./yano.sh appchain doctor shipment-chain --distribution /opt/yano-x
./yano.sh appchain drift  shipment-chain --peer http://node-a:8080

The property is yano.plugins.directory. Restart the nodes; Yano is not rebuilt.

drift is the fastest way to find the one member that is out of step, and it is worth running as a routine check rather than only during an incident.

Operating a plugin

Surface	What it gives you
/status and /ui/app-chain/	Tip, root, agreement, anchor lag, plugin inventory.
Health checks	Per-plugin health contributed through the SPI.
Prometheus metrics	Namespaced by plugin and contribution; effect metrics can be scoped by type.
Effect operations	Inspect, retry, park, requeue, and reconcile stuck effects.
Admin API	Key rotation, snapshots, onboarding.

Distinguish the four states an effect can be in when diagnosing: intent recorded, execution attempted, incorporation completed, and anchoring of the height that carries it. Operators who conflate them chase the wrong problem.

Before a pilot

Beyond the ladder above, plan for:

• application-specific soak and load testing on a public test network;
• key and secret operations, including rotation runbooks for all five secret classes;
• monitoring and SLOs, incident recovery, and restore rehearsal;
• retention, pruning, and oldestProvableHeight policy; and
• governance for any future semantic change.

Tutorial 9 — from demo to pilot turns this into a concrete deployment plan.