Back to Pulsar

PIP-439: Adding Transaction Support to Pulsar Functions Through Managed Transaction Wrapping

pip/pip-439.md

4.2.118.4 KB
Original Source

PIP-439: Adding Transaction Support to Pulsar Functions Through Managed Transaction Wrapping

Background knowledge

Apache Pulsar transactions enable atomic operations across multiple topics, allowing producers to send messages and consumers to acknowledge messages as a single unit of work. This provides the foundation for exactly-once processing semantics in streaming applications.

Transaction Architecture

Pulsar's transaction system consists of four key components:

  1. Transaction Coordinator (TC): A broker module that manages transaction lifecycles, allocates transaction IDs, and orchestrates the commit/abort process.

  2. Transaction Log: A persistent topic storing transaction metadata and state changes, enabling recovery after failures.

  3. Transaction Buffer: Temporarily stores messages produced within transactions, making them visible to consumers only after commit.

  4. Pending Acknowledge State: Tracks message acknowledgments within transactions, preventing conflicts between competing transactions.

Transaction Lifecycle

Transactions follow a defined lifecycle:

  1. OPEN: Client obtains a transaction ID from the Transaction Coordinator.
  2. PRODUCING/ACKNOWLEDGING: Client registers topic partitions/subscriptions with the TC, then produces/acknowledges messages within the transaction.
  3. COMMITTING/ABORTING: Client requests to end the transaction, TC begins two-phase commit.
  4. COMMITTED/ABORTED: After processing all partitions, TC finalizes the transaction state.
  5. TIMED_OUT: Transactions exceeding their timeout are automatically aborted.

Transaction Guarantees

Pulsar transactions provide:

  • Atomic writes across multiple topics
  • Conditional acknowledgment to prevent duplicate processing by "zombie" instances
  • Visibility control ensuring consumers only see committed transaction messages
  • Support for exactly-once processing in consume-transform-produce patterns

Pulsar Functions

Pulsar Functions is a lightweight compute framework integrated with Apache Pulsar that enables stream processing without managing infrastructure. Key characteristics include:

  • Simple Programming Model: Functions receive messages, process them, and optionally produce output
  • Processing Patterns: Supports both synchronous and asynchronous message processing
  • Context Object: Provides access to message metadata, output production, and state storage
  • Integration: Natively integrated with Pulsar's pub-sub messaging system
  • Deployment: Managed by Pulsar with automatic scaling and fault tolerance

Functions operate on a per-message basis, making them ideal for implementing stream processing with exactly-once semantics when combined with transactions.

Motivation

Currently, Pulsar Functions cannot publish to multiple topics transactionally, which is a significant limitation for use cases requiring atomic multi-topic publishing. For instance, if a function processes an input message and needs to publish related updates to several output topics, there's no guarantee that all operations will succeed atomically.

This limitation prevents building robust stream processing applications that require exactly-once semantics across multiple input and output topics. Without transaction support in Functions, developers must implement their own error handling and retry mechanisms, which can be complex and error-prone.

Adding transaction support to Pulsar Functions would finally ensure message processing atomicity.

Goals

In Scope

  1. Enable automatic transaction support for Pulsar Functions through configuration
  2. Allow Functions to publish messages to multiple topics within a single transaction
  3. Support transactional acknowledgment of input messages
  4. Ensure transactions are committed only if message processing completes successfully
  5. Provide transaction timeout configuration for Functions
  6. Add transaction support for async functions
  7. Handling multiple transactions in batches to improve performance, added in a later phase of implementation

Out of Scope

  1. Exposing explicit transaction management APIs in the Functions interface
  2. Supporting multi-function transactions (transactions spanning multiple function invocations)
  3. Adding transaction support to Pulsar IO connectors
  4. Changes to the Function interface itself

High Level Design

The proposed solution introduces managed transaction wrapping for Pulsar Functions through configuration settings. When enabled, each function execution will be automatically wrapped in a transaction without requiring code changes to the function implementation.

The general flow will be:

  1. Function is configured with transactionMode: MANAGED
  2. When a message arrives, the function runtime creates a new transaction
  3. The function processes the message with an enhanced Context that uses the transaction
  4. Any output messages are published using the transaction
  5. Input message acknowledgment is performed within the transaction
  6. If the function completes successfully, the transaction is committed
  7. If the function throws an exception, the transaction is aborted

This approach provides transaction support in a way that is transparent to function implementers, requiring only configuration changes rather than code changes.

Detailed Design

Design & Implementation Details

Configuration Classes

We will update the FunctionConfig to include transaction-related settings through a new TransactionConfig class:

java
public enum TransactionMode {
  OFF,
  MANAGED
}

public class TransactionConfig {
  private TransactionMode transactionMode = TransactionMode.OFF;
  private Long transactionTimeoutMs = 60000L;
  private Integer transactionBatchingMaxEntries = 1;
  private Long transactionBatchingQuietPeriodMs = 100L;

  // Getters and setters...
}

public class FunctionConfig {
  // Existing fields...

  private TransactionConfig transaction = new TransactionConfig();

  // Getter and setter ...
}
java
We also need to update the protobuf definition for FunctionDetails to include these fields:

message TransactionSpec {
  enum TransactionMode {
      OFF = 0;
      MANAGED = 1;
  }
  TransactionMode transactionMode = 1;
  int64 transactionTimeoutMs = 2;
  int64 transactionBatchingMaxEntries = 3;
  int64 transactionBatchingQuietPeriodMs = 4;
}

message FunctionDetails {
  // Other existing fields...
  TransactionSpec transaction = 24;
}

Modifications to ContextImpl

java
class ContextImpl implements Context, SinkContext, SourceContext, AutoCloseable {
    // Existing fields...

    // Finds the proper transaction to tie to current function execution (sync/async)
    private Transaction getManagedTransaction() {

        // implementation...
    }

    // Existing methods...

    public void setCurrentTransaction(Transaction transaction) {
        this.currentTransaction = transaction;
    }

    @Override
    public <T> TypedMessageBuilder<T> newOutputMessage(String topicName, Schema<T> schema)
          throws PulsarClientException {
      MessageBuilderImpl<T> messageBuilder = new MessageBuilderImpl<>();
      TypedMessageBuilder<T> typedMessageBuilder;
      Producer<T> producer = getProducer(topicName, schema);
      Transaction managedTransaction = getManagedTransaction(); 
    
      if (currentTransaction != null) {
          if (schema != null) {
              // Uses the new API that supports both schema and transaction
              typedMessageBuilder = producer.newMessage(schema, managedTransaction);
          } else {
              typedMessageBuilder = producer.newMessage(managedTransaction);
          }
      } else if (schema != null) {
          typedMessageBuilder = producer.newMessage(schema);
      } else {
          typedMessageBuilder = producer.newMessage();
      }
    
      messageBuilder.setUnderlyingBuilder(typedMessageBuilder);
      return messageBuilder;
    }
}

Asynchronous Functions Support

It's important to note that Pulsar Functions supports asynchronous processing, where functions can return CompletableFuture objects. This proposal ensures that transaction support works seamlessly with both synchronous and asynchronous functions.

For asynchronous functions:

  1. The transaction is created at the beginning of message processing, just like for synchronous functions
  2. When the function returns a CompletableFuture, the transaction is maintained until the future completes
  • Any Context-related operations inside of the returned 'CompletableFuture' objects are tied to the correct transaction
  1. When the future completes successfully, the transaction is committed
  2. If the future completes exceptionally, the transaction is aborted

Batch Processing of Transactions

To optimize performance and reduce the overhead on the Transaction Coordinator, this proposal introduces transaction batching. Transaction batching allows multiple incoming messages to be processed within the same transaction, reducing the total number of transactions created.

Transaction Batching Concept

Transaction batching is distinct from Pulsar's message batching. While message batching combines multiple messages into a single "batch message" for efficient network transfer, transaction batching processes multiple incoming messages (or batch messages) within the scope of a single transaction.

Key benefits of transaction batching include:

  1. Reduced Load on Transaction Coordinator: Fewer transactions means less coordination overhead
  2. Improved Throughput: Higher message processing capacity with lower per-message overhead
  3. Optimized Resource Usage: Better utilization of transaction resources
  4. Consistent Performance at Scale: Maintains performance characteristics under high load

Transaction Batching Parameters

Transaction batching is controlled by two main parameters:

  1. transactionBatchingMaxEntries: The maximum number of entries (incoming messages or batch messages) to process within a single transaction before committing it and starting a new one.
  • An "entry" refers to an incoming batch message which could itself contain multiple individual messages
  • Default: 1 (recommended minimum to ensure batch index acknowledgment state doesn't span transactions)
  • Setting this to 0 disables transaction batching!
  • Setting this to higher values increases throughput but may impact latency
  1. transactionBatchingQuietPeriodMs: The maximum amount of time to wait for additional messages before committing a transaction if transactionBatchingMaxEntries is not reached.
  • Default: 1ms (matching the Pulsar Client producer's default batchingMaxPublishDelay value)
  • This parameter handles scenarios where message flow isn't continuous
  • When the quiet period elapses without new messages arriving, the current transaction is committed

Public-facing Changes

Configuration

Transaction support will be configured using the new TransactionConfig class:

java
TransactionConfig txnConfig = new TransactionConfig();
txnConfig.setTransactionMode(TransactionMode.MANAGED);
txnConfig.setTransactionTimeoutMs(30000L); // 30 seconds
txnConfig.setTransactionBatchingMaxEntries(10); // Process up to 10 entries per transaction
txnConfig.setTransactionBatchingQuietPeriodMs(1L); // Commit after 1ms of inactivity

FunctionConfig functionConfig = new FunctionConfig();
functionConfig.setTransaction(txnConfig);
// Other configuration...

The Functions worker configuration will include a setting to en-/disable pulsar_function_txn_latency:

yaml
# Transaction metrics configuration
transactionMetrics:
  # Enable/disable specific transaction metrics individually
  txnLatency: false          # Transaction duration histograms (high cardinality)
    # ... disable any future metric deemed as too expensive to enable per default

CLI

The Pulsar Admin CLI will be updated to support these new configuration options:

bash
$ pulsar-admin functions create \
  --auto-transactions-enabled true \
  --transaction-timeout-ms 30000 \
  --transaction-batching-max-entries 10 \
  --transaction-batching-quiet-period-ms 1 \
  ... other options ...

Similarly, the CLI will support updating these options with the update command.

Metrics

The following new metrics will be added to track transaction usage in functions:

  1. pulsar_function_txn_created_total: Counter tracking the total number of transactions created by functions
  • Labels: tenant, namespace, name (function name), instance_id, cluster
  • Unit: Count
  1. pulsar_function_txn_committed_total: Counter tracking successfully committed transactions
  • Labels: tenant, namespace, name (function name), instance_id, cluster
  • Unit: Count
  1. pulsar_function_txn_aborted_total: Counter tracking aborted transactions
  • Labels: tenant, namespace, name (function name), instance_id, cluster
  • Unit: Count
  1. pulsar_function_txn_timeout_total: Counter tracking transactions that timed out
  • Labels: tenant, namespace, name (function name), instance_id, cluster
  • Unit: Count
  1. pulsar_function_txn_latency: Histogram of transaction duration from creation to commit/abort (configurable in functions worker)
  • Labels: tenant, namespace, name (function name), instance_id, cluster
  • Unit: Milliseconds
  • turned off by default to avoid adding too much volume and will be configurable in the functions worker
  1. pulsar_function_txn_batch_size: Histogram of transaction batch sizes (number of entries processed per transaction)
  • Labels: tenant, namespace, name (function name), instance_id, cluster
  • Unit: Count
  1. pulsar_function_txn_batch_commit_reason: Counter tracking transaction batch commits by reason
  • Labels: tenant, namespace, name (function name), instance_id, cluster, reason ("max_entries", "quiet_period", "function_close")
  • Unit: Count
  1. pulsar_function_txn_entries_per_second: Gauge tracking the rate of entries processed in transactions
  • Labels: tenant, namespace, name (function name), instance_id, cluster
  • Unit: Entries/second

Monitoring

To monitor the transaction functionality in Pulsar Functions, users should:

  1. Monitor the transaction metrics mentioned above to track transaction usage and success/failure rates.
  2. Set up alerts for high transaction abort rates or timeouts, which may indicate issues with function processing or transaction configuration.
  3. Monitor function processing latency metrics to ensure that using transactions doesn't introduce unacceptable overhead.
  4. Set up alerts for transaction timeout occurrences, which might indicate that the configured timeout is too low for the function's processing time.
  5. Configure appropriate logging levels to capture transaction-related errors for debugging.

Example alert thresholds:

  • Transaction abort rate > 5% over 5 minutes
  • Transaction timeout rate > 1% over 5 minutes
  • Transaction duration approaching timeout value (e.g., > 80% of configured timeout)

Security Considerations

The proposed transaction support doesn't introduce new security concerns as it builds on top of existing Pulsar transaction mechanisms. All security aspects of transactions, including authentication and authorization, are inherited from the Pulsar client's transaction implementation.

Functions will only be able to create transactions and operate on topics they already have permission to access. No additional permissions are required beyond what's already needed for the function to operate.

Backward & Forward Compatibility

Upgrade

The proposed changes are backward compatible with existing Pulsar Functions:

  1. The new transaction mode defaults to OFF so existing functions will continue to operate without transaction support.
  2. Existing functions can enable transaction support by simply updating their configuration.
  3. The Function interface remains unchanged, so existing function implementations will work with transaction support when enabled.

To enable transaction support for existing functions:

  1. Ensure transactions are enabled at the broker level.
  2. Update the function configuration to set transactionMode: MANAGED and optionally configure transactionTimeoutMs.
  3. Restart or update the function to apply the new configuration.

Downgrade / Rollback

When downgrading to a version that doesn't support the transaction features, several steps need to be taken to ensure compatibility:

  1. Configuration Reset:
  • Update all function configurations to set transactionMode: OFF before downgrading
  • If the downgrade target doesn't recognize the transaction configuration fields, these fields should be explicitly removed from the function configuration to prevent errors
  1. Function Updates:
  • After downgrading Pulsar, any functions that were using managed transactions must be updated or redeployed
  • This ensures that the runtime configurations are compatible with the older version

Pulsar Upgrade & Downgrade/Rollback Considerations

  1. Transactions created by functions in one cluster will not span to other clusters; they are local to the cluster where the function executes.
  2. Functions running in different clusters should each be configured for transaction support independently.
  3. During rolling upgrades, ensure that transaction-enabled functions are only deployed to clusters that support transactions.
  4. When downgrading, first disable transactions in function configurations before downgrading the clusters.

Alternatives

Explicit Transaction API

Instead of automatic transaction wrapping, we could expose explicit transaction APIs in the Context interface:

java
Transaction newTransaction();
void commitTransaction(Transaction txn);
void abortTransaction(Transaction txn);

This approach gives function authors more control but requires code changes to use transactions. We rejected this approach to keep the programming model simpler and avoid breaking the Function interface.

General Notes

The implementation of transaction support in Pulsar Functions should be considered a stepping stone toward a more comprehensive exactly-once processing model for Pulsar's stream processing capabilities. Future work may include Transaction support in Pulsar IO connectors.

Links