阅读中文版本

Build a Columnar Query Engine

Columnar Query Engine — a complete walkthrough of analytical query processing, covering column reader, predicate pushdown, vectorized aggregate, and testing.

columnar_query_engine/
├── src/index.ts
├── src/column-reader.ts
├── src/predicate-pushdown.ts
├── tests/columnar_query_engine.test.ts
└── package.json

// Column reader: analytical query processing core
export function columnreader(input: ColumnreaderInput): ColumnreaderResult {
  // Build the column reader that anchors analytical query processing and defines the core data shape.
  return processColumnreader(input);
}

// Predicate pushdown — handles analytical query processing Add the predicate pushdown so
export class Predicatepushdown {
  constructor(private opts: PredicatepushdownOptions) {}
  async run(ctx: Context): Promise<PredicatepushdownOutput> {
    return this.execute(ctx);
  }
}

// Vectorized aggregate: analytical query processing stage
export const vectorizedaggregate = (next: Next) => async (req: Request) => {
  const out = await transform(req);
  return next(out);
};

// Spill to disk for analytical query processing
export async function spilltodisk(stream: AsyncIterable<Chunk>): Promise<Aggregate> {
  let acc = init();
  for await (const c of stream) acc = merge(acc, c);
  return acc;
}

// Query planner: analytical query processing finalization
export function queryplanner(state: State): Result {
  validate(state);
  return finalize(state);
}

Project overview

This project builds a analytical query processing system whose reason for existing is scanning billions of rows for aggregates without deserializing irrelevant columns. The temptation in analytical query 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 wide select that forces the engine to materialize many columns and exhausts the row-group cache as the first thing designed, not the last.

You will implement Column reader, Predicate pushdown, Vectorized aggregate, Spill to disk, and Query planner, each as a module with one responsibility and a test that proves it in isolation. The analytical query processing pipeline is then the wiring that composes them, and the integration test is where the composition is shown to survive a wide select that forces the engine to materialize many columns and exhausts the row-group cache.

The walkthrough is written for a developer who has used a analytical query 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 column reader that returns the right shape, a predicate pushdown that rejects bad input, an integration test that goes green — so you can stop at any point and still have a analytical query processing system that works for the cases covered so far.

What you will build

The finished analytical query processing system exposes five capabilities, each with a focused test:

• Column reader
• Predicate pushdown
• Vectorized aggregate
• Spill to disk
• Query planner

Column reader is the foundation that every later stage assumes; Predicate pushdown is where analytical query processing correctness lives; Vectorized aggregate handles a wide select that forces the engine to materialize many columns and exhausts the row-group cache; Spill to disk makes the system observable; and Query planner is the graceful-degradation layer that keeps overload from becoming corruption.

The build keeps the analytical query 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 Column reader, Predicate pushdown, and Vectorized aggregate, with Spill to disk tying them together. Data flows as columns stored as typed, compressed arrays with min/max statistics per chunk to skip blocks during scans.

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 Vectorized aggregate does not need to stand up Column reader and Predicate pushdown to exercise the interesting branch.

The data model is deliberately explicit. Columns stored as typed, compressed arrays with min/max statistics per chunk to skip blocks during scans 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 analytical query 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 Vectorized aggregate is designed before the happy path is finished, because a wide select that forces the engine to materialize many columns and exhausts the row-group cache is the case that distinguishes a analytical query 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 analytical query processing setup is not blocked by a missing tool:

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

The runtime dependencies are chosen to keep the analytical query processing datapath legible:

• apache-arrow (^17.0.0) — Columnar in-memory format
• duckdb-async (^1.0.0) — Embedded analytical runtime

# setup for columnar-query-engine
pnpm init
pnpm add apache-arrow duckdb-async
pnpm add -D typescript tsx vitest

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

Directory structure

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

columnar_query_engine/
├── src/index.ts
├── src/column-reader.ts
├── src/predicate-pushdown.ts
├── tests/columnar_query_engine.test.ts
└── package.json

Phase 1: Column reader

Build the column reader that anchors analytical query processing and defines the core data shape.

// Column reader: analytical query processing core
export function columnreader(input: ColumnreaderInput): ColumnreaderResult {
  // Build the column reader that anchors analytical query processing and defines the core data shape.
  return processColumnreader(input);
}

After this phase the column reader works in isolation and is covered by a unit test. The column reader 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 column reader as the single source of truth for the analytical query processing data shape. Validate inputs at the boundary and reject early, so a malformed value never reaches Predicate pushdown or Vectorized aggregate where its effect would be harder to localize.

Phase 2: Predicate pushdown

Add the predicate pushdown so analytical query processing can scanning billions of rows correctly.

// Predicate pushdown — handles analytical query processing Add the predicate pushdown so
export class Predicatepushdown {
  constructor(private opts: PredicatepushdownOptions) {}
  async run(ctx: Context): Promise<PredicatepushdownOutput> {
    return this.execute(ctx);
  }
}

After this phase the predicate pushdown works in isolation and is covered by a unit test. This is where analytical query processing correctness lives: the predicate pushdown is what makes scanning billions of rows for aggregates without deserializing irrelevant columns true, and its test is the proof.

Keep the predicate pushdown 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 predicate pushdown about the transformation.

Phase 3: Vectorized aggregate

Wire the vectorized aggregate to handle a wide select that without losing data.

// Vectorized aggregate: analytical query processing stage
export const vectorizedaggregate = (next: Next) => async (req: Request) => {
  const out = await transform(req);
  return next(out);
};

After this phase the vectorized aggregate works in isolation and is covered by a unit test. The vectorized aggregate is the difference between a analytical query processing system that works on the happy path and one that survives a wide select that forces the engine to materialize many columns and exhausts the row-group cache.

Design the failure before the success. Decide what vectorized aggregate does when a wide select that forces the engine to materialize many columns and exhausts the row-group cache: reject, retry, shed, or degrade — and write that test first. A analytical query processing system whose failure behavior is unspecified will pick one at random under pressure, and it is usually the wrong one.

Phase 4: Spill to disk

Integrate the spill to disk to make analytical query processing observable and testable end to end.

// Spill to disk for analytical query processing
export async function spilltodisk(stream: AsyncIterable<Chunk>): Promise<Aggregate> {
  let acc = init();
  for await (const c of stream) acc = merge(acc, c);
  return acc;
}

After this phase the spill to disk works in isolation and is covered by a unit test. Observability is not a finishing touch; it is how you know the analytical query 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 spill to disk is what turns “something is slow” into “the predicate pushdown is slow for inputs matching this shape,” which is the difference between an alert and an investigation.

Phase 5: Query planner

Finalize the query planner so the analytical query processing system degrades gracefully under overload.

// Query planner: analytical query processing finalization
export function queryplanner(state: State): Result {
  validate(state);
  return finalize(state);
}

After this phase the query planner works in isolation and is covered by a unit test. The query planner is what keeps overload from becoming corruption: when the analytical query 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 query planner — shed, queue, or backpressure — and make it explicit in code and in a test. Leaving the behavior to chance is how a analytical query processing traffic spike becomes a data loss incident.

Testing

Each phase has a focused test: column reader, predicate pushdown, and vectorized aggregate are exercised in isolation, then an integration test drives the full analytical query processing pipeline and asserts the a wide select that case is handled.

Start with unit tests that pin each component’s contract: the column reader returns the documented shape, the predicate pushdown transforms correctly, and the vectorized aggregate 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 analytical query processing scenario end to end. Assert the happy path first, then deliberately trigger a wide select that forces the engine to materialize many columns and exhausts the row-group cache and assert the system degrades rather than corrupts. Because the components are pure where practical, the integration test stays deterministic rather than flaky.

A analytical query 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 analytical query 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 vectorized aggregate failure-mode assertions live as tests rather than as a manual checklist that drifts out of date.

# run the analytical query processing test suite
pnpm vitest run

Deployment and operations

Run at least two analytical query processing instances behind a load balancer and share state in Redis. Set resource limits so analytical query 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 analytical query processing behavior is global rather than per instance. Size limits so the system sheds load before it runs out of memory, because a analytical query 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 analytical query 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 analytical query 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 analytical query processing inputs at the boundary, reject oversized payloads before buffering them, and keep secrets in the environment rather than the repository.

The analytical query processing system is a choke point, so it is the right place to reject malformed input before it reaches predicate pushdown or vectorized aggregate. 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 analytical query 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 analytical query processing system, and the lockfile is what makes the audit meaningful.

Observability

Emit one structured event per analytical query 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 analytical query 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 vectorized aggregate failure rate separately from the overall error rate. A jump in vectorized aggregate failures usually means a wide select that forces the engine to materialize many columns and exhausts the row-group cache, not a bug in predicate pushdown, and the two have different owners and different fixes.

Chart the share of analytical query 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 wide select that forces the engine to materialize many columns and exhausts the row-group cache. Reproduce with a minimal fixture, then add a guard test so the analytical query processing system rejects the condition loudly.

Performance regresses under analytical query processing load. Profile with the sampling analyzer, locate the hot path in predicate pushdown, and add a targeted benchmark to lock the regression.

Keep the analytical query processing dashboards separate from dependency dashboards. When an operation fails, the correlation id lets you decide in seconds whether the fault is column reader, predicate pushdown, or vectorized aggregate, instead of guessing from a timeout.

Limitations and next steps

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

The analytical query processing system is single-region by design; extending to multi-region means making column reader shareable across regions, which is a real change to the consistency model rather than a config flag. The vectorized aggregate 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 analytical query processing correctness that is the point.

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