Resonate Recursive Fan-Out Pattern — Rust
SDK note. The Rust SDK is in active development (v0.6.0, on crates.io). This pattern uses
ctx.run().spawn()andctx.rpc().spawn()— both synchronous as of 0.5.0+. Verify against current SDK source before shipping.
Overview
Recursive fan-out spawns multiple child invocations in parallel, awaits them individually, and (optionally) recurses deeper. The Rust expression uses .spawn() on a Context execution builder to get a DurableFuture, collects those futures into a Vec, then awaits each in turn.
For the language-agnostic mental model, see resonate-recursive-fan-out-pattern-typescript.
When to use
- Batch processing where items are independent
- Web crawling / tree traversal with dynamic depth
- Map-reduce shaped workflows
- Any fan-out where each leaf is a discrete, retryable unit
Parallel fan-out in the same process
use resonate::prelude::*;
#[resonate::function]
async fn enrich_batch(ctx: &Context, order_ids: Vec<String>) -> Result<Vec<String>> {
let mut futures = Vec::with_capacity(order_ids.len());
// fire off N children in parallel
for id in order_ids {
let fut = ctx.run(enrich_one, id).spawn()?;
futures.push(fut);
}
// await all; order preserved
let mut results = Vec::with_capacity(futures.len());
for f in futures {
results.push(f.await?);
}
Ok(results)
}
#[resonate::function]
async fn enrich_one(order_id: String) -> Result<String> {
Ok(format!("enriched-{}", order_id))
}
Each enrich_one is a durable promise. If the parent crashes mid-loop, the children are stored server-side; on parent replay, f.await? lookups hit the stored results.
Parallel fan-out across workers
#[resonate::function]
async fn parallel_enrich(ctx: &Context, ids: Vec<String>) -> Result<Vec<String>> {
let mut futures = Vec::with_capacity(ids.len());
for id in ids {
let fut = ctx
.rpc::<String>("enrich_one", id)
.target("poll://any@enrichment-workers")
.spawn()?;
futures.push(fut);
}
let mut results = Vec::with_capacity(futures.len());
for f in futures {
results.push(f.await?);
}
Ok(results)
}
The Resonate server handles fair dispatch across any process polling enrichment-workers.
Recursive fan-out
A durable function calling itself for tree traversal:
#[derive(serde::Serialize, serde::Deserialize)]
struct CrawlInput {
url: String,
depth: u32,
}
#[derive(serde::Serialize, serde::Deserialize)]
struct CrawlOutput {
url: String,
depth: u32,
links: Vec<String>,
children: Vec<CrawlOutput>,
}
#[resonate::function]
async fn crawl(ctx: &Context, input: CrawlInput) -> Result<CrawlOutput> {
let links = ctx.run(fetch_links, input.url.clone()).await?;
if input.depth == 0 {
return Ok(CrawlOutput {
url: input.url,
depth: 0,
links,
children: vec![],
});
}
// fan out to each link, one depth deeper
let mut futures = Vec::with_capacity(links.len());
for link in &links {
let sub = CrawlInput { url: link.clone(), depth: input.depth - 1 };
let fut = ctx.rpc::<CrawlOutput>("crawl", sub).spawn()?;
futures.push(fut);
}
let mut children = Vec::with_capacity(futures.len());
for f in futures {
children.push(f.await?);
}
Ok(CrawlOutput { url: input.url, depth: input.depth, links, children })
}
#[resonate::function]
async fn fetch_links(url: String) -> Result<Vec<String>> {
// ...
Ok(vec![])
}
Bounded parallelism
Process in chunks of K to cap in-flight invocations:
#[resonate::function]
async fn process_bounded(
ctx: &Context,
(items, concurrency): (Vec<String>, usize),
) -> Result<Vec<String>> {
let mut results = Vec::with_capacity(items.len());
for chunk in items.chunks(concurrency) {
let mut futures = Vec::with_capacity(chunk.len());
for item in chunk {
let fut = ctx.run(process_one, item.clone()).spawn()?;
futures.push(fut);
}
for f in futures {
results.push(f.await?);
}
}
Ok(results)
}
#[resonate::function]
async fn process_one(item: String) -> Result<String> {
Ok(format!("processed-{}", item))
}
Each chunk of up to concurrency runs in parallel; the next chunk only starts after the previous fully settles.
Partial-failure tolerance
By default, a ? on .await? bubbles the first child error up. To tolerate individual failures:
#[resonate::function]
async fn enrich_tolerant(ctx: &Context, ids: Vec<String>) -> Result<Vec<std::result::Result<String, String>>> {
let mut futures = Vec::with_capacity(ids.len());
for id in ids {
let fut = ctx.run(enrich_one, id).spawn()?;
futures.push(fut);
}
let mut out = Vec::with_capacity(futures.len());
for f in futures {
match f.await {
Ok(v) => out.push(Ok(v)),
Err(e) => out.push(Err(e.to_string())),
}
}
Ok(out)
}
(Note the return type uses std::result::Result to disambiguate from Resonate's aliased Result.)
Distinct Rust idioms
.spawn()?— synchronous: returnsResult<DurableFuture<T>>immediately (no await needed); unwrap with?, then await theDurableFuturelater to get the actualTVec::with_capacity(n)— idiomatic capacity pre-allocation for known-size collections.chunks(n)slice method — bounded parallelism cleaner than iterator fiddling- Turbofish
ctx.rpc::<T>(...)— needed when the return type isn't inferable from context .map_err(|e| e.to_string())for error-type narrowing in tolerant-fan-out casesCloneon input types — borrowed slices (&[String]) can't crossctx.runboundaries; own the data
Avoid
- Unbounded recursion without a depth counter — the promise tree grows; Rust's default stack is fine but the server-side promise graph isn't free
- Holding borrowed references across
.await?—.spawn()and.await?mean owned data only
Related skills
resonate-basic-durable-world-usage-rust—ctx.run,ctx.rpc,.spawn(), builder optionsresonate-saga-pattern-rust— fan-out inside or alongside sagasdurable-execution— foundational replay semanticsresonate-recursive-fan-out-pattern-typescript/-python— sibling SDKs