The Pipeline Framework

Appearance

Orchestrator Runtime ​

The orchestrator runtime is the generated part of a TPF application that starts the pipeline, calls each step in order, tracks execution state when configured, and exposes generated REST, gRPC, LOCAL, or FUNCTION-platform entry points.

Current applications host their own generated orchestrator. For implementation notes on separating orchestration from worker execution, see Durable Coordinator.

Guide Pages ​

  1. Overview covers runtime modes, generated entry points, and generated structure.
  2. Queue-Async Runtime covers durable execution, crash behaviour, idempotency, providers, and HA baseline configuration.
  3. Checkpoint Handoff covers reliable cross-pipeline publication and admission.
  4. Await Runtime Setup covers durable external suspend/resume adapters and runtime configuration.
  5. Command Steps covers replay-safe external effects with command ids, effect logging, duplicate policy, and connector execution.

Runtime Modes ​

TPF supports two orchestrator runtime modes:

  1. SYNC (default): in-process request/response execution.
  2. QUEUE_ASYNC: background async execution with stored execution state, dispatch, retry, recovery, and dead-letter handling.

Set mode with:

properties
pipeline.orchestrator.mode=SYNC
# or
pipeline.orchestrator.mode=QUEUE_ASYNC

Use Queue-Async Runtime before relying on background execution in production-style environments.

For latency and topology tradeoffs between SYNC, compute-first QUEUE_ASYNC, current FUNCTION, and future all-serverless HA, see Runtime Boundaries And Performance.

kind: await and kind: command are framework-owned I/O shells on top of QUEUE_ASYNC. Await owns suspend/resume around a deferred external result. Command owns idempotent external effects that should be recorded, retried, replayed, or dead-lettered as part of the pipeline lifecycle.

Generated Transport Entry Points ​

Generated orchestrator entry points use the selected transport. Transport modes are REST, gRPC, and LOCAL; platform modes are COMPUTE and FUNCTION.

  1. REST:
    • POST /pipeline/run
    • POST /pipeline/run-async
    • GET /pipeline/executions/{executionId}
    • GET /pipeline/executions/{executionId}/result
    • POST /pipeline/ingest
    • GET /pipeline/subscribe
  2. gRPC:
    • Run
    • RunAsync
    • GetExecutionStatus
    • GetExecutionResult
    • Ingest
    • Subscribe
  3. LOCAL:
    • in-process client calls for same-JVM deployments

In FUNCTION platform mode, TPF also generates provider handler artifacts. These are not a transport mode:

  • PipelineRunFunctionHandler
  • PipelineRunAsyncFunctionHandler
  • PipelineExecutionStatusFunctionHandler
  • PipelineExecutionResultFunctionHandler

Generated Structure ​

text
orchestrator-svc/
├── src/main/java/<base>/orchestrator/service/
│   ├── PipelineRunResource.java
│   ├── OrchestratorGrpcService.java
│   ├── PipelineRunFunctionHandler.java
│   ├── PipelineRunAsyncFunctionHandler.java
│   ├── PipelineExecutionStatusFunctionHandler.java
│   └── PipelineExecutionResultFunctionHandler.java
└── src/main/resources/application.properties