Asynchronous Task & Retry Simulator

Explore worker saturation, retries, backoff, idempotency, and dead-letter handling with reproducible simulations.

Runs in your browser

First time here?Quick start & guided walkthrough

See how worker capacity, failures, retries, and duplicate deliveries affect a waiting queue.

Pick a scenarioStart with the baseline, worker saturation, retry storm, or backoff example. Keep the random seed fixed when comparing changes.
Change capacity or failure behaviorAdjust workers or incoming tasks, then inspect retry limit, timeout, backoff, and jitter. Change one factor at a time so the comparison remains understandable.
Run and inspect the timelineRun the seeded simulation. Check completed logical jobs, attempts, waiting work, and dead letters together. The observation window can end with unfinished work; Stop cancels a long run.

PresetArrival rate (tasks/s)WorkersAttempt duration (ms)Failure probability (0–1)Retry limit after first attemptBase retry delay (ms)Attempt timeout (ms)Observation window (s)Random seedDuplicate delivery probabilityMaximum waiting queue

Exponential backoff Full jitter Deduplicate logical jobs Retain exhausted attempts in dead-letter count

Load an example or enter your own input to begin.

How to use this tool

Choose a preset, set the arrival rate, worker count, attempt duration, failure probability, retry policy, and observation window, then run the seeded simulation.

Keep the same seed and parameters to reproduce a run. Compare retry storms with backoff and jitter. Change duplicate delivery probability and deduplication to examine their effect. Stop execution or reset without losing control of the page.

How it works

A seeded discrete-event model schedules Poisson arrivals, attempt completions, and retries. Each worker handles one attempt at a time. Attempts use a fixed duration; configured timeouts end an attempt in the model.

Retries use a bounded delay, optional exponential growth, and optional full jitter. Queue admission has a configured limit. Exhausted attempts contribute to either the dead-letter or discarded count.

Completion statistics are observations from this simulated run, not production measurements. Work still waiting, running, or scheduled to retry at the end is reported separately.

Scope and limitations

Failure events are independent, the worker count is fixed, and broker delivery/network timing is simplified.
Deduplication illustrates policy; it does not implement production transactional idempotency or exactly-once delivery.
At most 5,000 expected arrivals, 100,000 processed events, 64 workers, and eight retries. The display trace is limited to 500 entries.

Technical references

Discuss the engineering behind this tool

Your input stays in this tab. These tools do not upload, store, or send it to an AI service.