Skip to main content

Transaction remediation

Dekaf exposes two distinct force-abort operations:

  • ForceTerminateTransactionAsync targets one transactional ID through its transaction coordinator. It fences the active producer by requesting a new producer epoch, which terminates any ongoing transaction. This is the operational recovery API for a stuck transactional producer or two-phase-commit participant.
  • AbortTransactionAsync writes an abort marker for one topic partition. It requires producer and coordinator metadata obtained from DescribeProducersAsync and is intended for partition-scoped repair.
await using var admin = Kafka.CreateAdminClient()
    .WithBootstrapServers("localhost:9092")
    .Build();

var result = await admin.ForceTerminateTransactionAsync(
    "payments-processor",
    new ForceTerminateTransactionOptions { TimeoutMs = 30_000 });

if (result.ErrorCode != ErrorCode.None)
{
    Console.WriteLine($"Termination failed: {result.ErrorCode}");
    Console.WriteLine($"Retriable: {result.IsRetriable}");
}

The timeout is sent to the transaction coordinator for its producer-epoch update. When omitted, it inherits AdminClientOptions.RequestTimeoutMs. The cancellation token independently controls the caller's wait and all coordinator/network operations.

Dekaf's built-in and in-memory admin clients expose this operation through ITransactionRemediationAdminClient. The IAdminClient extension method preserves the same call syntax without adding a binary-breaking member to the base interface. Custom admin-client implementations must implement the capability interface to support force termination.

ForceTerminateTransactionAsync resolves a Transaction coordinator and negotiates InitProducerId; it does not route through a topic leader or the KRaft controller. A broker that does not advertise InitProducerId produces BrokerVersionException. Controller-only bootstrap endpoints reject this coordinator operation with UnsupportedEndpointType.

Successful termination returns the newly allocated producer ID and epoch. A non-retriable broker response is preserved in ErrorCode; IsRetriable applies Dekaf's Kafka error classification. Retriable coordinator and transport failures use the admin client's configured retry policy and throw a KafkaException if retries are exhausted.

Force termination fences the current producer. Any later operation from that producer should be treated as failed, and applications must create or initialize a new producer before continuing.