Soenneker.Redis.WorkQueue 4.0.2

Soenneker.Redis.WorkQueue

A durable, partition-aware Redis work queue with scheduling, leases, retries, deduplication, and round-robin fairness.

Installation

dotnet add package Soenneker.Redis.WorkQueue

Registration

services.AddRedisWorkQueueAsSingleton<SendMessageWork>("send-message", options =>
{
    options.MaxConcurrentItemsPerPartition = 2;
    options.ClaimLeaseDuration = TimeSpan.FromMinutes(1);
    options.MaximumAttempts = 10;
    options.MaintenanceInterval = TimeSpan.FromSeconds(1);
});

Registration starts a background maintenance service by default. Every process may register it; a distributed Redis semaphore ensures that only one process maintains a logical queue at a time. Set EnableBackgroundMaintenance to false only when another component calls RunMaintenance periodically.

The partition key is the fairness and concurrency boundary. In Leadping, for example, use the business ID. Ready partitions are dispatched round-robin, and Soenneker.Redis.Semaphores prevents more than the configured number of claims for one partition across every process.

Enqueue and process

await queue.Enqueue(new RedisWorkQueueItem<SendMessageWork>
{
    Id = message.Id,
    PartitionKey = message.BusinessId,
    Value = new SendMessageWork(message.Id),
    AvailableAt = DateTimeOffset.UtcNow // set a future value to schedule it
}, cancellationToken);

await using RedisWorkQueueClaim<SendMessageWork>? claim =
    await queue.TryClaim(workerId, cancellationToken);

if (claim is null)
    return;

try
{
    await Process(claim.Item.Value, claim.OwnershipLostToken);
    await queue.Complete(claim, cancellationToken);
}
catch (Exception exception)
{
    await queue.Retry(claim, new RedisWorkQueueFailure
    {
        Reason = exception.GetType().Name,
        Details = exception.Message
    }, TimeSpan.FromSeconds(10), cancellationToken);
    throw;
}

The queue provides at-least-once delivery. Claims automatically renew both their work-item lease and their partition semaphore permit. Background maintenance promotes scheduled items, recovers expired claims, repairs ready partitions, and expires completed-item deduplication markers.

Retry moves work to the dead-letter store when it reaches MaximumAttempts. Use GetDeadLetter and RequeueDeadLetter to inspect and recover valid items. Missing or unreadable payloads are dead-lettered automatically so they cannot block a partition. Use Abandon when a healthy worker needs to relinquish a claim immediately without treating it as a processing failure.

The implementation uses typed Soenneker.Redis.Util operations and Redis transactions with optimistic conditions. It does not use Redis scripts.