阅读中文版本

Build an ETL Pipeline Orchestrator

ETL Pipeline Orchestrator — a complete walkthrough of batch data processing, covering stage graph, task scheduler, retry policy, and testing.

etl_pipeline_orchestrator/
├── src/index.ts
├── src/stage-graph.ts
├── src/task-scheduler.ts
├── tests/etl_pipeline_orchestrator.test.ts
└── package.json

// Stage graph: batch data processing core
export function stagegraph(input: StagegraphInput): StagegraphResult {
  // Build the stage graph that anchors batch data processing and defines the core data shape.
  return processStagegraph(input);
}

// Task scheduler — handles batch data processing Add the task scheduler so
export class Taskscheduler {
  constructor(private opts: TaskschedulerOptions) {}
  async run(ctx: Context): Promise<TaskschedulerOutput> {
    return this.execute(ctx);
  }
}

// Retry policy: batch data processing stage
export const retrypolicy = (next: Next) => async (req: Request) => {
  const out = await transform(req);
  return next(out);
};

// Checkpoint store for batch data processing
export async function checkpointstore(stream: AsyncIterable<Chunk>): Promise<Aggregate> {
  let acc = init();
  for await (const c of stream) acc = merge(acc, c);
  return acc;
}

// Progress reporter: batch data processing finalization
export function progressreporter(state: State): Result {
  validate(state);
  return finalize(state);
}

Project overview

This project builds a batch data processing system whose reason for existing is running extract-transform-load stages with retries, ordering, and observability across large datasets. The temptation in batch data processing is to bolt the interesting part onto a working prototype and discover the hard cases in production; this build inverts that order and treats a stage that partially writes its output and then fails, leaving a dirty dataset that a retry doubles as the first thing designed, not the last.

You will implement Stage graph, Task scheduler, Retry policy, Checkpoint store, and Progress reporter, each as a module with one responsibility and a test that proves it in isolation. The batch data processing pipeline is then the wiring that composes them, and the integration test is where the composition is shown to survive a stage that partially writes its output and then fails, leaving a dirty dataset that a retry doubles.

The walkthrough is written for a developer who has used a batch data processing system but has not had to make one honest. It favors explicit data shapes over clever abstractions, and it names the failure each design choice prevents rather than presenting the choice as a style preference.

Each phase ends with something observable — a stage graph that returns the right shape, a task scheduler that rejects bad input, an integration test that goes green — so you can stop at any point and still have a batch data processing system that works for the cases covered so far.

What you will build

The finished batch data processing system exposes five capabilities, each with a focused test:

• Stage graph
• Task scheduler
• Retry policy
• Checkpoint store
• Progress reporter

Stage graph is the foundation that every later stage assumes; Task scheduler is where batch data processing correctness lives; Retry policy handles a stage that partially writes its output and then fails, leaving a dirty dataset that a retry doubles; Checkpoint store makes the system observable; and Progress reporter is the graceful-degradation layer that keeps overload from becoming corruption.

The build keeps the batch data processing logic in ordinary modules you can read in one sitting, and pulls in dependencies only for the parts that are genuinely tedious to hand-roll. That keeps the diff between the tutorial and a real codebase small, and keeps the failure modes visible in code rather than hidden behind a framework.

Architecture and data model

The system is organized around Stage graph, Task scheduler, and Retry policy, with Checkpoint store tying them together. Data flows as a directed acyclic graph of stages each producing a dataset, scheduled with dependencies and checkpointed for resume.

Each component is a module with a pure interface where practical: the same input produces the same output, and side effects are pushed to the edges. That discipline is what makes the phases independently testable, because a test for Retry policy does not need to stand up Stage graph and Task scheduler to exercise the interesting branch.

The data model is deliberately explicit. A directed acyclic graph of stages each producing a dataset, scheduled with dependencies and checkpointed for resume Keeping the shape written down — in a schema, a type, or a fixture — means a change to the shape is a visible diff rather than a silent runtime shift, and it means the integration test can assert against a concrete shape instead of guessing.

Configuration and code are kept apart: the configuration says what batch data processing should do, the modules say how. A reviewer can change a threshold or a policy without reading the implementation, which keeps the risky changes small and lets the diff stand as the review.

The failure path through Retry policy is designed before the happy path is finished, because a stage that partially writes its output and then fails, leaving a dirty dataset that a retry doubles is the case that distinguishes a batch data processing system from a demo. Every component declares what it does when its inputs are wrong, and the integration test asserts each of those declarations.

Prerequisites and dependencies

Have these ready before you install anything, so the batch data processing setup is not blocked by a missing tool:

• Node.js 20 LTS or later
• Familiarity with batch data processing fundamentals
• A local environment to run the test suite

The runtime dependencies are chosen to keep the batch data processing datapath legible:

• fastify (^5.0.0) — Orchestration API
• redis (^4.7.0) — Checkpoint store

# setup for etl-pipeline-orchestrator
pnpm init
pnpm add fastify redis
pnpm add -D typescript tsx vitest

Versions are pinned to a major range so the batch data processing install is reproducible; commit the lockfile. The stage graph reads its operational parameters from the environment and validates them at startup, so a missing value fails the boot rather than the first batch data processing request.

Directory structure

Keep each batch data processing component in its own file so it can be tested without starting the others:

etl_pipeline_orchestrator/
├── src/index.ts
├── src/stage-graph.ts
├── src/task-scheduler.ts
├── tests/etl_pipeline_orchestrator.test.ts
└── package.json

Phase 1: Stage graph

Build the stage graph that anchors batch data processing and defines the core data shape.

// Stage graph: batch data processing core
export function stagegraph(input: StagegraphInput): StagegraphResult {
  // Build the stage graph that anchors batch data processing and defines the core data shape.
  return processStagegraph(input);
}

After this phase the stage graph works in isolation and is covered by a unit test. The stage graph defines the data shape that every later stage assumes, so getting it right first means the later phases compile against a stable contract.

Treat the stage graph as the single source of truth for the batch data processing data shape. Validate inputs at the boundary and reject early, so a malformed value never reaches Task scheduler or Retry policy where its effect would be harder to localize.

Phase 2: Task scheduler

Add the task scheduler so batch data processing can running extract-transform-load stages with correctly.

// Task scheduler — handles batch data processing Add the task scheduler so
export class Taskscheduler {
  constructor(private opts: TaskschedulerOptions) {}
  async run(ctx: Context): Promise<TaskschedulerOutput> {
    return this.execute(ctx);
  }
}

After this phase the task scheduler works in isolation and is covered by a unit test. This is where batch data processing correctness lives: the task scheduler is what makes running extract-transform-load stages with retries, ordering, and observability across large datasets true, and its test is the proof.

Keep the task scheduler pure where you can. A function that takes an input and returns an output is trivial to test and impossible to mock wrong; a class with hidden state is the opposite. Push side effects to the caller and keep the task scheduler about the transformation.

Phase 3: Retry policy

Wire the retry policy to handle a stage that partially without losing data.

// Retry policy: batch data processing stage
export const retrypolicy = (next: Next) => async (req: Request) => {
  const out = await transform(req);
  return next(out);
};

After this phase the retry policy works in isolation and is covered by a unit test. The retry policy is the difference between a batch data processing system that works on the happy path and one that survives a stage that partially writes its output and then fails, leaving a dirty dataset that a retry doubles.

Design the failure before the success. Decide what retry policy does when a stage that partially writes its output and then fails, leaving a dirty dataset that a retry doubles: reject, retry, shed, or degrade — and write that test first. A batch data processing system whose failure behavior is unspecified will pick one at random under pressure, and it is usually the wrong one.

Phase 4: Checkpoint store

Integrate the checkpoint store to make batch data processing observable and testable end to end.

// Checkpoint store for batch data processing
export async function checkpointstore(stream: AsyncIterable<Chunk>): Promise<Aggregate> {
  let acc = init();
  for await (const c of stream) acc = merge(acc, c);
  return acc;
}

After this phase the checkpoint store works in isolation and is covered by a unit test. Observability is not a finishing touch; it is how you know the batch data processing system is doing what it claims in production.

Emit one structured event per meaningful action, with enough context to localize a failure without a second round-trip. The checkpoint store is what turns “something is slow” into “the task scheduler is slow for inputs matching this shape,” which is the difference between an alert and an investigation.

Phase 5: Progress reporter

Finalize the progress reporter so the batch data processing system degrades gracefully under overload.

// Progress reporter: batch data processing finalization
export function progressreporter(state: State): Result {
  validate(state);
  return finalize(state);
}

After this phase the progress reporter works in isolation and is covered by a unit test. The progress reporter is what keeps overload from becoming corruption: when the batch data processing system is asked for more than it can do, it degrades in a defined way instead of failing in an undefined one.

Pick a degradation strategy for progress reporter — shed, queue, or backpressure — and make it explicit in code and in a test. Leaving the behavior to chance is how a batch data processing traffic spike becomes a data loss incident.

Testing

Each phase has a focused test: stage graph, task scheduler, and retry policy are exercised in isolation, then an integration test drives the full batch data processing pipeline and asserts the a stage that partially case is handled.

Start with unit tests that pin each component’s contract: the stage graph returns the documented shape, the task scheduler transforms correctly, and the retry policy fails in the documented way. These tests are fast and they localize regressions to the component that broke.

Then add an integration test that composes the components and drives a realistic batch data processing scenario end to end. Assert the happy path first, then deliberately trigger a stage that partially writes its output and then fails, leaving a dirty dataset that a retry doubles and assert the system degrades rather than corrupts. Because the components are pure where practical, the integration test stays deterministic rather than flaky.

A batch data processing suite must also prove the negative: bad input is rejected at the boundary, overload is shed rather than queued indefinitely, and a partial failure does not leave the system in an inconsistent state. These are the tests that catch the regressions a happy-path suite never would.

The entire batch data processing suite runs with no external service; every dependency is either in-process or stubbed, so every case is reproducible in CI. That is what lets the retry policy failure-mode assertions live as tests rather than as a manual checklist that drifts out of date.

# run the batch data processing test suite
pnpm vitest run

Deployment and operations

Run at least two batch data processing instances behind a load balancer and share state in Redis. Set resource limits so batch data processing overload surfaces as backpressure rather than failure.

Run at least two stateless instances behind a load balancer, and move any shared state into a shared store so the batch data processing behavior is global rather than per instance. Size limits so the system sheds load before it runs out of memory, because a batch data processing system that OOMs under load fails in a way that is hard to reason about.

Drain on deploy: stop accepting new work, let in-flight batch data processing operations finish, then exit. Because the system holds no state that cannot be reconstructed, a rollout is a matter of cycling instances one at a time while watching the error rate.

When a dependency is unavailable, decide explicitly whether the batch data processing system fails closed or degrades. Leaving that choice to chance is how a dependency outage becomes a silent data loss, so make the choice visible in code and in a test.

Security considerations

Validate all batch data processing inputs at the boundary, reject oversized payloads before buffering them, and keep secrets in the environment rather than the repository.

The batch data processing system is a choke point, so it is the right place to reject malformed input before it reaches task scheduler or retry policy. Validate the shape and the size at the edge, and let the inner components assume what they receive is well-formed.

Never log secrets or full payloads; log enough to localize a failure without recording data that should not be on disk. Treat any external call as untrusted and validate its response before using it, because a batch data processing system that trusts its inputs blindly is one malformed response away from corruption.

Keep dependencies pinned and run an audit in CI. A transitive dependency with a known vulnerability is still a vulnerability in your batch data processing system, and the lockfile is what makes the audit meaningful.

Observability

Emit one structured event per batch data processing operation with the inputs’ shape, the outcome, and a correlation id. Track error rate and latency per component and alert on a step change rather than a single spike.

The correlation id is the thread that ties a user-visible symptom to the batch data processing decision and to the component log entry for the same operation. Without it, an investigation starts from a timestamp and a guess.

Watch the retry policy failure rate separately from the overall error rate. A jump in retry policy failures usually means a stage that partially writes its output and then fails, leaving a dirty dataset that a retry doubles, not a bug in task scheduler, and the two have different owners and different fixes.

Chart the share of batch data processing operations that shed, queue, or succeed. A change in the mix points at the cause faster than an aggregate latency can, because each class has a different root.

Troubleshooting

A stage that partially writes its output and then fails, leaving a dirty dataset that a retry doubles. Reproduce with a minimal fixture, then add a guard test so the batch data processing system rejects the condition loudly.

Performance regresses under batch data processing load. Profile with the sampling analyzer, locate the hot path in task scheduler, and add a targeted benchmark to lock the regression.

Keep the batch data processing dashboards separate from dependency dashboards. When an operation fails, the correlation id lets you decide in seconds whether the fault is stage graph, task scheduler, or retry policy, instead of guessing from a timeout.

Limitations and next steps

• Single-region; multi-region replication is out of scope.
• The stage graph is in-process and not horizontally shared by default.
• No auth layer; assume deployment behind an edge that enforces identity.

The batch data processing system is single-region by design; extending to multi-region means making stage graph shareable across regions, which is a real change to the consistency model rather than a config flag. The retry policy is in-process, which is correct for one instance but must move to a shared store before scaling out.

There is no authentication or authorization layer; the system assumes it is deployed behind an edge that enforces identity. Adding one is straightforward but out of scope for this build, and doing it here would distract from the batch data processing correctness that is the point.

The natural next step, once the batch data processing pipeline is trusted in production, is to add a short-TTL cache for idempotent reads and a richer checkpoint store that exports histograms rather than counters. Both are additive and neither changes the contracts this build establishes.