Quick wins for a faster PC:
Repair Windows errors before they cause bigger problemsFix Now →Scan for outdated or missing drivers - takes under a minuteDriver Scan →To stop an agent from processing the same event twice, design retries across the whole event path—not just inside the handler. A delivery can be repeated after a timeout, even if the first attempt already changed data or triggered an external action. Classify failures, use bounded backoff with jitter, make each side effect safe to repeat where possible, and define what happens when processing cannot succeed.
Why an agent retry can repeat completed work
An event-driven system typically has a producer that records a change, a transport that routes the event, and a consumer—such as an agent or event handler—that reacts to it. An event is a record of something that happened, not simply a function call waiting to be retried. Google Cloud’s architecture guidance describes events this way.
Trace an event through creation, publication, broker acceptance, delivery, handler execution, side-effect commit, and acknowledgement. A crucial failure point is the gap between committing a side effect and the transport observing the acknowledgement. If the handler completes a database update but times out before its acknowledgement is received, the transport may deliver the event again. The next handler invocation cannot assume the previous one did nothing.
Delivery guarantees describe what the transport does, not whether a business operation happens exactly once. At-least-once delivery permits duplicates; at-most-once delivery can avoid redelivery but may lose work. “Exactly once” is meaningful only when its scope and mechanism are specified. Google Cloud Pub/Sub distinguishes these delivery guarantees, while AWS Durable Execution notes that at-most-once behavior for an individual retry attempt does not ensure a step runs exactly once across an entire workflow.
#1 Best Overall
Choose what to retry, and for how long
Retry transient failures, not every failure. The right classification depends on the transport and downstream service, so check their current error behavior rather than treating these categories as universal rules.
- Often transient: temporary unavailability, throttling, and short-lived connectivity problems.
- Usually needs correction, not repetition: invalid input and authorization or configuration errors.
Use progressively longer waits with jitter, and set limits on both attempts and total elapsed time. Exponential backoff with jitter helps prevent clients that fail together from retrying together and adding load to an already struggling service. AWS guidance recommends backoff for transient errors and cautions that frequent retries can increase contention; its Well-Architected guidance also recommends a maximum retry count and monitoring queue length and backlog. The sources do not establish one universally correct formula or schedule for agent code.
Make the retry budget fit the work’s deadline. Track how old an event is as well as how many times it has been attempted: an event that is no longer useful can become stale work, while retries can consume capacity needed by newer events. Set budgets against the workload’s actual timeout and throughput requirements, and check how the policy behaves under those conditions.
Make repeated processing safe
Idempotency means that repeating an operation does not produce an additional unintended effect. It is a key defense against duplicate delivery, but it must cover each side effect, not just the handler’s database write. A duplicate-suppressed database mutation does not automatically prevent a second payment, email, or external API call.
The Tool Desk
Outbyte PC Repair FREERepair Windows errors before they cause bigger problemsFix Now →Outbyte Driver Updater FREEScan for outdated or missing drivers - takes under a minuteDriver Scan →Google Cloud Eventarc recommends idempotent handlers for at-least-once delivery and summarizes the principle this way: “Idempotency works well with at-least-once delivery, because it makes it safe to retry.” Where possible, persist the event identity and processing state atomically with the business mutation. For an external API, pass a stable idempotency key if that API supports one.
Google Cloud’s Eventarc guidance treats the combination of CloudEvents source and id attributes as a unique event identity; the same combination is considered a duplicate in that guidance. That identity rule is not a universal guarantee that every broker or downstream service deduplicates events for you.
Rank #4
- Used Book in Good Condition
If a side effect cannot safely be repeated, isolate it from automatic replay. One option is to persist intent and result and reconcile an ambiguous outcome before attempting the action again. Another is to use an execution policy that avoids automatic replay where appropriate. Choose identity keys and deduplication windows carefully: an overly broad key or long-lived record can suppress a legitimate new event that happens to share the identity you selected.
Give failed events a terminal path
A retry policy needs a defined outcome for events that cannot be processed. A dead-letter queue or topic can retain them for diagnosis and controlled redrive rather than letting them disappear after retry exhaustion or retention expiry. Make the destination observable, restrict access where appropriate, and decide who or what inspects and recovers its contents.
Do these 3 things before closing this tab:
1Clear out junk files and repair common Windows errors2Fix the driver behind crashes, sound loss and screen glitches3Repair Windows errors before they cause bigger problemsBest Value
Redrive is another delivery attempt, not proof that earlier attempts had no effect. Keep the same idempotency protections in place when replaying an event. Before redrive, inspect the failure reason and determine whether the original attempt could have completed a side effect.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.Provider defaults are examples, not a universal retry recipe
Cloud services make different decisions about retryable errors, time windows, retention, and dead-letter handling. The documented settings below are provider-specific examples, not recommendations for every agent or workload.
| Service | Documented retry and delivery behavior | Exhaustion or retention detail |
|---|---|---|
| Google Eventarc Standard | Documents at-least-once delivery and default retry behavior through Pub/Sub. Its documented Pub/Sub transport backoff bounds are 10 seconds minimum and 600 seconds maximum. | Google documents a 24-hour default message-retention duration. Undelivered events can be discarded when retention expires unless a dead-letter topic is configured. |
| Amazon EventBridge | AWS documents a default retry period of 24 hours and up to 185 attempts, using exponential backoff with jitter. | Events are dropped after retries are exhausted unless a dead-letter queue is configured. |
| Azure Event Grid | Microsoft documents error-dependent decisions to retry, dead-letter, or drop. Its delivery schedule is best effort, includes randomization, and may still result in duplicate delivery. | Some configuration-related errors are not retried, making dead-letter configuration relevant. A comparable numeric attempt limit or retention value is not stated here. |
These values are the services’ documented defaults or behaviors, not independent performance measurements. Provider settings can change, so confirm the current documentation before relying on a particular value.
Check the whole policy, not only the retry counter
Before deploying an agent or handler, check the complete path from delivery through recovery:
- Which delivery guarantee applies, and what exactly is included in any exactly-once claim?
- Which errors retry, immediately dead-letter, or drop?
- What are the attempt limit, elapsed-time limit, and event-retention window?
- Does backoff include jitter, and does the retry budget fit the work’s deadline?
- Can concurrent delivery or ordering affect the handler’s business logic?
- Are database mutations and external API operations protected against duplicate effects?
- Can operators see retry rates, backlog age, exhausted events, and dead-letter contents?
- Does redrive preserve the original event identity and pass through the same idempotency protections?
These checks make retries a reliability policy spanning the handler, transport, and every service where processing creates side effects. A retry loop alone cannot provide that guarantee.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
