Reproducible jitter

Technical Preview

The Rust SDK is a technical preview. Its API surface may still evolve and changes may not follow semantic versioning. Pin an exact version if you need stability.

Retry backoff, worker startup delay (CAMUNDA_WORKER_STARTUP_JITTER_MAX_SECONDS) and FALCON endpoint selection draw from an injected Random rather than an ambient generator. The default, LiveRandom, is seeded once per process from OS entropy. A SeededRandom replays the same sequence for the same seed, so a test can assert the exact delays the SDK schedules instead of a range:

use camunda_orchestration_sdk::{CamundaClient, CamundaOptions, SeededRandom};
use std::sync::Arc;

// Seeds from CAMUNDA_TEST_SEED when it is set, otherwise from a fresh seed.
let random = Arc::new(SeededRandom::from_env()?);
// Log the source: it names the seed and how to replay this run.
println!("{random}"); // SeededRandom(seed=...; replay with CAMUNDA_TEST_SEED=...)

let client = CamundaClient::new(CamundaOptions::new().with_random(random))?;

The seeded generator is specified across the Camunda SDKs, so one seed produces the same draws in every language. A worker draws its startup delay when it is started (run, start or spawn), not when its task first runs, so workers started in sequence draw in that order however the runtime schedules them.

std::hash::RandomState, the only entropy std exposes, is banned in the runtime by clippy.toml outside the LiveRandom adapter.