Begin -> consume -> correlate/respond -> endEndpoint handlers
Source and sink endpoint contracts for Python. The exact generated signature follows connector type, endpoint direction, protocol mode, input type, result type, and error type.
When to use
Use this page while implementing a generated business function or endpoint handler.
Behavior
- A source endpoint converts transport input into graph values. A sink endpoint converts graph values into transport requests and optionally emits responses back into the graph.
- Begin creates state, consume performs the protocol conversion, result correlation handles asynchronous graph responses, and end releases resources.
- When an Input and Sink share a request/response endpoint, generated soft associations coordinate the protocol boundary but do not replace persisted stream links.
Result correlation
A source can emit work into the graph and wait for a later result. Register the callback under the same message ID returned by the handler correlation method. The framework uses the stream context to route the result back to the pending request.
Workflow and activity endpoints
Where Temporal is supported, the endpoint configuration determines whether generated code binds the function as a workflow or an activity. This is endpoint behavior; it does not introduce an additional stream type.
Source lifecycle
Common lifecycle phases; names below are normalized for readability.
concurrencyKafka sourceLimits concurrently processed records. 0 means no handler-level limit; Kafka partitioning still bounds work.
begin requestSource lifecycleCreates request-local state and may return an updated message context before any input is emitted.
consume messageSource lifecycleDecodes the transport request or record, emits values through the source stream context, and optionally registers result correlation.
get message IDSource correlationReturns the stable key used to route a downstream result back to the request that emitted the input. Return an empty value when no result routing is used.
end requestSource lifecycleRuns after completion or failure and releases request-local resources. It receives the first lifecycle error.
EOFStreaming gRPC sourceSignals that the client has finished sending request messages while the RPC may still be waiting for graph results.
Python signatures
Concrete methods and helpers exposed by this runtime.
concurrency(sc) -> intPython Kafka source0 leaves concurrency unrestricted by the handler.
await begin_request(sc) -> StatePython sourceCreates per-request state before decoding input.
await consume_message(sc, state, transport_value, result_ctx)Python sourceDecodes transport data, registers result callbacks, and awaits sc.collect.
get_message_id(sc, state, result) -> strPython sourceReturns the correlation key for a downstream result.
await end_request(sc, error, state)Python lifecycleFinalizes request resources after completion or failure.
result_ctx.set_result_callback / donePython source helperRegisters a callback by message ID and marks synchronous production complete.
begin_request / consume_message / handle_response / end_requestPython sinkImplements outbound transport mapping and result handling.
Sink lifecycle
Outbound lifecycle phases used when supported by the selected connector.
begin requestSink lifecycleCreates state for one outbound interaction and may enrich the message context.
consume messageSink lifecycleMaps a graph value to the transport request, sends it, or writes a Kafka record.
handle responseHTTP or gRPC sinkConverts a transport response to the sink result stream and can emit a typed error.
get stream IDKafka sink correlationSelects the message context or correlation identifier attached to the outbound record when the generated signature requires it.
end requestSink lifecycleFinalizes the interaction after success or failure and releases request-local resources.
Shared endpoint types
Roles repeated across transport-specific generated signatures.
Handler statePer request or recordState returned by the begin method and passed to every later lifecycle method. Use it instead of storing request-specific data on the singleton handler.
Source stream contextGraph boundaryEmits decoded input values and typed endpoint errors into the graph.
Sink stream contextGraph boundaryEmits decoded transport responses and typed endpoint errors from an outbound endpoint.
Result contextAsync correlationRegisters callbacks by message ID and marks synchronous request production complete.
SenderStreaming transportSends protocol responses or requests without exposing the connector implementation.
Transport-specific values
Connector adapters currently exposed by this framework.
HTTPaiohttp request/result typesRead inbound requests or build outbound responses through generated endpoint adapters.
gRPCgrpc.aio sender/result typesSupports unary and streaming RPC lifecycle methods.
Kafka`ConsumerMessage`, `ResultContext`Read bytes and metadata, then mark or commit after the desired result callback.
Temporalworkflow and activity endpoint contractsThe generated endpoint role determines whether the function participates in a Temporal workflow or activity.
CronAPScheduler integrationInvokes the configured input endpoint on its schedule.
Local source/sinkCustom endpoint ProtocolsConnect application-owned async producers or consumers.
Generated extension pattern
result_ctx.set_result_callback(message_id, on_result)
await sc.collect(decoded_value)
# on_result calls result_ctx.done() when the request can finish