| name | resonate-basic-debugging-java |
| description | Debug and troubleshoot Resonate applications using the Java SDK. Use when investigating stuck or never-resuming workflows, duplicated side effects after replay, untyped-handle decode surprises (Integer vs Long), CLI positional-argument arity mismatches, the detached by-name-only constraint, Java 21 / virtual-thread requirements, rejected-promise error handling, or r.stop() silently killing a live worker. Verified against the resonatehq-examples/*-java repos and develop/java.mdx (docs PR |
| license | Apache-2.0 |
Resonate Basic Debugging — Java
Prerelease note. resonate-sdk-java is published on Maven Central — pin io.resonatehq:resonate-sdk-java:0.1.1. The API mirrors the Python SDK and may change before a stable 1.0. Requires Java 21+ — the SDK uses virtual threads, a feature generally available in Java 21. Every code block here is compile-verified against 0.1.1 with a Java 21 toolchain, and cross-checked against the resonatehq-examples/*-java repos and develop/java.mdx (docs PR #230).
Overview
Java's type system catches many bugs at compile time — typed method references and a typed ResonateHandle close off whole classes of error the dynamically-typed SDKs hit. The traps that remain are mostly at the durability boundary: untyped-handle decoding, CLI argument binding, replay double-fires, and the by-name-only constraints. This skill is a symptom-first guide to those failure modes.
For the language-agnostic replay and recovery mental model, read durable-execution first.
Triage flow
- Does it even build / start? Confirm a Java 21+ toolchain — the SDK uses virtual threads and will not run on Java 8/11/17.
- Is the worker connected? Confirm the builder's
url(...) (or RESONATE_URL) points at a running server, and r was built without throwing.
- Is the function registered? By-name dispatch (
rpc, the CLI) needs r.register(Owner::fn) on the executing group.
- Is the promise stuck? Run
resonate promise get <id> to check state (pending / resolved / rejected / timedout).
- Is the workflow replaying but producing duplicates? An un-checkpointed side effect is re-running above a durable boundary.
- Is a CLI-invoked function arity-mismatching? Each
--arg binds to one positional parameter.
- Is the worker up but not picking up work?
r.stop() may have been called on a live worker.
Stuck / never-resuming workflows
Latent promise never settled
Symptom: future.await() blocks indefinitely; resonate promise get <id> shows state pending.
Causes:
- The external actor never settled the promise. Resolve it via the CLI (
resonate promise resolve <id> --data '"approved"') or r.promises.resolve(id, new Value(null, "approved")).
- The decoded type does not match what the awaiter expects, so
await throws a decode error rather than returning.
Unlike the Go SDK (which has no promises sub-client and requires manual base64 encoding of the JSON payload), the Java SDK ships r.promises — resolve directly:
import io.resonatehq.resonate.Types.Value;
r.promises.resolve("approval-1", new Value(null, "approved")).join();
See resonate-human-in-the-loop-pattern-java for the full mechanics.
ctx.run leaf blocks indefinitely
Symptom: the workflow task lease expires and the server reassigns the task; the workflow appears to restart rather than resume; attempts never complete.
Cause: the runtime drains every ctx.run-spawned future before it can suspend or settle the parent task. A function that does external I/O, waits on a lock, or sleeps for a long time inside ctx.run holds the lease open until the TTL expires (default 60s).
Fix: move long-running or external-blocking work into ctx.rpc (remote dispatch — the workflow suspends cleanly) or ctx.promise (latent promise settled by an external actor). Reserve ctx.run for in-process computation that returns promptly.
Duplicated side effects (replay)
Symptom: emails, charges, log entries, or DB writes happen more than once per logical invocation.
Cause: the entire workflow body re-runs from the top on every resume. Durable child promises short-circuit work that already settled, but code that runs before reaching a durable boundary (ctx.sleep, ctx.rpc, ctx.promise) executes again on each replay pass.
public static String myWorkflow(Context ctx, String id) {
System.out.println("charging card for order " + id);
return ctx.run(MyWorkflow::chargeCard, id).await();
}
public static String myWorkflow(Context ctx, String id) {
return ctx.run(MyWorkflow::chargeCard, id).await();
}
Rule: any observable side effect (network call, write, notification) belongs inside its own ctx.run or ctx.rpc so the durable promise records the result and short-circuits on replay.
Decode and error handling
Untyped handle: Integer vs Long
Symptom: ClassCastException reading a numeric result from an untyped handle, e.g. (Long) handle.result() throws on a small number.
Cause: the by-name r.run / r.rpc forms and r.get return ResonateHandle<Object>. Jackson decodes a small JSON integer as Integer and a large one as Long. A direct cast to one or the other is fragile.
Fix: read through Number (this is exactly what example-recursive-factorial-java does, because 13! overflows int):
long result = ((Number) handle.result()).longValue();
A method-reference invocation returns a typed ResonateHandle<R> and avoids the issue entirely — prefer it where the function is registered locally.
Generic collection result from a by-name invocation
Symptom: a function that returns List<MyRecord> (or Map<String, MyRecord>) comes back as List<LinkedHashMap> when read from a by-name call — a ClassCastException or surprising field access on the elements.
Cause: the by-name forms (ctx.rpc("name", ...), r.rpc(id, "name", ...), r.get(id)) decode against Object (Resonate.java), so Jackson reconstructs only generic JSON shapes — LinkedHashMap per element — and the record type parameter is lost at the boundary.
Fix: use the method-reference form, which decodes against the function's full generic return type and gives a typed future:
ResonateFuture<List<Delivery>> f = ctx.run(Orders::deliverAll, batch);
List<Delivery> delivered = f.await();
If you must read it from an untyped by-name result, reshape it with a Jackson ObjectMapper.convertValue(result, new TypeReference<List<Delivery>>() {}) — note that pulls Jackson onto your compile classpath (the SDK only depends on it transitively at runtime), so add com.fasterxml.jackson.core:jackson-databind to your build to do this.
CLI invocation arity mismatch
Symptom: invoking from the CLI fails to bind arguments, or the function receives the wrong values / an arity error.
Cause: resonate invoke <id> --func f --arg 5 --arg 1 passes the args as a positional list, and the Java SDK binds them one per declared parameter. A function declaring a single int[] (or List) parameter arity-mismatches two --arg values.
Fix: declare one parameter per --arg:
public static String countdown(Context ctx, int count, int delaySeconds) { ... }
example-quickstart-java is the canonical reference. (This is the opposite of Go, where the same invoke binds the whole list to a single []int.)
Instance method reference registered as a durable function
Symptom: an Owner::fn reference compiles, but execution fails or behaves unexpectedly when the referenced method is an instance method (this::fn, someObject::fn).
Cause: durable functions must be public static. The SDK recovers the method behind a reference by reflection and invokes it without an object instance, so an instance method — which needs a receiver — has nowhere to run from on the worker.
Fix: make every registered function and every ctx.run / ctx.rpc target a public static method. Hold any per-instance state as a dependency (r.withDependency / ctx.getDependency) instead of closing over this.
Rejected promise re-throws the application error
Symptom: handle.result() or future.await() throws even though no local exception was raised at the call site.
Cause: the promise was rejected — either by the registered function throwing, or by an external resonate promise reject <id>. The error is re-thrown by its real type where reconstructable across the durability boundary, otherwise as an ApplicationError carrying the message.
import io.resonatehq.resonate.Errors.ApplicationError;
try {
String result = handle.result();
} catch (ApplicationError ae) {
System.err.println("workflow rejected: " + ae.getMessage());
}
ctx.detached with a method reference does not compile
Symptom: ctx.detached(Owner::fn, args) fails to compile.
Cause: ctx.detached is by-name String only — there is no method-reference overload (verified Context.java:600), unlike ctx.run / ctx.rpc.
Fix: pass the registered name as a String, and ensure the target is registered on whichever group executes it:
String auditId = ctx.detached("writeAuditLog", orderId).id();
Setup footguns
Wrong Java version
Symptom: UnsupportedClassVersionError, or virtual-thread APIs missing at runtime.
Cause: the SDK requires Java 21 (virtual threads). Java 8/11/17 will not work.
Fix: set the toolchain to 21:
java { toolchain { languageVersion = JavaLanguageVersion.of(21) } }
r.stop() on a long-running worker
Symptom: the worker process is running and healthy-looking, but stops picking up new tasks.
Cause: r.stop() closes the server connection, stops the heartbeat loop, and cancels the subscription-refresh loop. In-flight leased tasks have their TTL expire and the server reassigns them. The process keeps running, but the dispatch pipeline is dead.
Rule: call r.stop() only in one-shot binaries, demos, and CI tasks that exit after their work finishes. A long-running worker should stay up — block on a latch and let SIGINT / SIGTERM end the process:
import java.util.concurrent.CountDownLatch;
r.register(Quickstart::countdown);
new CountDownLatch(1).await();
Inspection tools
resonate dev # local dev server
resonate promise get <id> # single promise state + value
resonate promise search 'order:*' # prefix search across promises
resonate promise resolve <id> --data '"approved"' # settle a pending latent promise
resonate tree <id> # call graph for an invocation
See the resonate-cli skill for the full command surface. The CLI is SDK-agnostic; the same commands work against any worker language.
Durable sleep tolerance: a 24h ctx.sleep firing in 23–25h is within the server's timer tolerance window, not a bug.
Avoid
- Branching on
System.currentTimeMillis(), Math.random(), or UUID.randomUUID() directly inside a workflow body — non-deterministic values change between replay passes and cause divergent execution. Move them into a leaf so the result is checkpointed.
- Casting a numeric result from an untyped
ResonateHandle<Object> straight to Long (or Integer) — read through Number.
- Declaring a single array/
List parameter for a CLI-invoked function — --arg values bind positionally, one per parameter.
- Passing more than five application arguments to a durable function —
Fn.F0–Fn.F5 caps it at five beyond the Context. Bundle extras into a record.
- Passing non-serializable types (open streams, thread handles) as workflow args — they encode into the durable promise via JSON.
Related skills
resonate-basic-durable-world-usage-java — Context APIs (ctx.run, ctx.rpc, ctx.sleep, ctx.promise, ctx.detached), the replay model
resonate-basic-ephemeral-world-usage-java — the builder, register, typed vs untyped handles, stop semantics
resonate-human-in-the-loop-pattern-java — latent-promise resolution via r.promises.resolve
resonate-cli — full CLI command surface for promise inspection and settlement
resonate-defaults — default TTL, retry policy, and timeout values across all SDKs
durable-execution — foundational replay and recovery model
resonate-basic-debugging-python — the closest sibling; the Java API mirrors Python