Data Pipeline

Data Pipeline Design

Core Principles

Data pipelines are the arteries of modern data platforms. A well-designed pipeline is reliable, observable, idempotent, and cost-efficient. The following principles guide every decision:

1. Idempotency First — Running a pipeline twice should produce the same result. This enables safe retries and backfills without data duplication.
2. Observability by Default — Every stage must emit metrics, logs, and lineage metadata. If you can't see it, you can't fix it.
3. Fail Gracefully — Assume failures will happen. Design dead letter queues, retry logic with exponential backoff, and alerting on anomalies.
4. Incremental Processing — Process only what's changed. Full refreshes are for schema migrations and backfills only.
5. Data Contracts — Define and enforce schemas at every boundary. Catch drift before it reaches downstream consumers.
6. Separation of Concerns — Extract, transform, load are distinct phases. Each should be independently testable and debuggable.
7. Cost Awareness — Every byte processed costs money. Partition, compress, and prune aggressively.

Pipeline Maturity Model

Target: At minimum Level 3 for production pipelines. Level 4 for critical business data.

Pipeline Architecture Patterns

Batch vs Streaming

When to choose batch: Business reports don't need sub-minute freshness. Batch is simpler, cheaper, and easier to debug.

When to choose streaming: You need real-time decisions (fraud, pricing, monitoring). Be prepared for the operational complexity.

Lambda vs Kappa Architecture

Lambda Architecture — Run batch and streaming paths in parallel, merge results at query time.

Streaming path: Source → Stream processor → Speed layer → Serving layer
Batch path:    Source → Batch processor → Batch view     ↗

• Pros: Handles both real-time and historical accuracy
• Cons: Code duplication (batch + streaming logic), complex reconciliation

Kappa Architecture — Everything is a stream. Batch is just replaying a stream from the beginning.

Source → Stream processor → Serving layer (with replay capability)

• Pros: Single codebase, simpler operational model
• Cons: Requires robust stream processing infrastructure (Kafka + Flink/Kafka Streams)

Recommendation: Start with Kappa unless you have existing batch infrastructure. The unified model reduces maintenance burden significantly.

Medallion Architecture (Bronze/Silver/Gold)

This is the de facto standard for modern data lakes and lakehouses (Databricks, Iceberg, Delta Lake).

Bronze (Raw): Landing zone — raw data as-is from sources. Schema-on-read. Immutable.

• Append-only, no transformations
• Preserves original data for reprocessing
• Partitioned by ingestion date

Silver (Cleaned): Validated, deduplicated, enriched data.

• Schema enforced, quality checks applied
• Joins, type casting, null handling
• Suitable for data science exploration

Gold (Aggregated): Business-level aggregates, metrics, and reporting tables.

• Denormalized for query performance
• Aggregated at business grain (daily, monthly)
• Powers dashboards, ML features, and APIs

Bronze ──► Silver ──► Gold ──► Consumers
  │          │          │
  │          │          └── BI dashboards
  │          │          └── Feature store
  │          │          └── Reporting
  │          │
  │          └── Data science
  │          └── Ad-hoc queries
  │
  └── Reprocessing / backfills

Key benefit: Each layer acts as a checkpoint. If gold is corrupted, replay from silver. If silver has issues, replay from bronze.

ETL vs ELT

ETL (Extract, Transform, Load)

Extract → Transform (in staging area) → Load

• When to use: Transform logic is complex, source system is slow/stressed, target system can't handle complex transformations
• Pros: Less load on target, cleaner data at load time
• Cons: Requires a transformation engine, more pipeline code, harder to debug

ELT (Extract, Load, Transform)

Extract → Load (raw) → Transform (in target)

• When to use: Target is a powerful warehouse (Snowflake, BigQuery, Redshift, Databricks), raw data needs to be preserved
• Pros: Simpler pipeline, leverages warehouse compute power, raw data always available
• Cons: More expensive (compute on transformed data may be wasteful), raw data takes storage

Transformation Strategies

Recommendation: Prefer ELT + dbt for 80% of pipelines. Use ETL with Spark/Beam only when transformations are too complex for SQL (ML feature engineering, graph processing, custom aggregations).

Data Quality

Schema Validation

Validate schemas at every pipeline boundary. Use schema registries (Confluent Schema Registry, JSON Schema, Avro, Protobuf).

# Example: Schema validation with Great Expectations
import great_expectations as ge

df = ge.read_csv("raw_orders.csv")
df.expect_column_values_to_not_be_null("order_id")
df.expect_column_values_to_be_between("amount", 0, 100000)
df.expect_column_values_to_be_in_set("status", ["pending", "shipped", "delivered"])
validation_result = df.validate()
assert validation_result["success"], "Schema validation failed!"

Data Contracts

Define a contract between producers and consumers:

# data_contracts/orders.yaml
version: 1
table: orders
columns:
  order_id: { type: string, nullable: false, unique: true }
  user_id: { type: string, nullable: false }
  amount: { type: decimal(10,2), nullable: false, min: 0 }
  status: { type: string, nullable: false, enum: ["pending", "shipped", "delivered"] }
  created_at: { type: timestamp, nullable: false }
freshness: { sla: 1h, check_on: created_at }
volume: { min_rows: 100, max_rows: 1_000_000 }

Freshness Checks

Alert when data stops arriving:

-- Freshness check (runs every 5 minutes)
SELECT
  CURRENT_TIMESTAMP AS check_time,
  MAX(created_at) AS latest_record,
  DATEDIFF('minute', MAX(created_at), CURRENT_TIMESTAMP) AS staleness_minutes
FROM orders
HAVING staleness_minutes > 60;  -- SLA is 1 hour

Anomaly Detection

Detect unexpected changes in volume, schema, or values:

# Volume anomaly detection
expected_row_count = 10000  # from historical baseline
actual_count = spark.sql("SELECT COUNT(*) FROM orders").collect()[0][0]

threshold = 0.3  # 30% deviation
if abs(actual_count - expected_row_count) / expected_row_count > threshold:
    alert(f"Volume anomaly: expected {expected_row_count}, got {actual_count}")

dbt Testing

# dbt/schema.yml
version: 2

models:
  - name: orders
    description: "Cleaned orders table in Silver layer"
    columns:
      - name: order_id
        tests:
          - unique
          - not_null
      - name: amount
        tests:
          - not_null
          - dbt_expectations.expect_column_values_to_be_between:
              min_value: 0
              max_value: 100000
    tests:
      - dbt_utils.recency:
          datepart: hour
          field: created_at
          interval: 1

Monitoring & Observability

Data Observability

Monitor the five pillars: freshness, volume, schema, distribution, lineage.

Tools: Monte Carlo, Sifflet, Datadog, OpenLineage, Marquez, DQOps.

Lineage Tracking

Every transformation should log its inputs and outputs:

# OpenLineage example
from openlineage.client import OpenLineageClient
from openlineage.client.run import RunEvent, RunState

client = OpenLineageClient(url="http://localhost:8080")
client.emit(RunEvent(
    eventType=RunState.COMPLETE,
    eventTime=datetime.now().isoformat(),
    run=Run(runId=str(uuid4())),
    job=Job(namespace="my_namespace", name="transform_orders"),
    inputs=[Dataset(namespace="my_database", name="raw_orders")],
    outputs=[Dataset(namespace="my_database", name="silver_orders")]
))

Freshness SLAs

Define SLAs and alert when breached:

-- Alert if any critical table hasn't been updated in the expected window
WITH table_freshness AS (
  SELECT
    'orders' AS table_name,
    MAX(created_at) AS last_update,
    DATEDIFF('hour', MAX(created_at), CURRENT_TIMESTAMP) AS hours_since_update
  FROM orders
  UNION ALL
  SELECT
    'inventory',
    MAX(updated_at),
    DATEDIFF('hour', MAX(updated_at), CURRENT_TIMESTAMP)
  FROM inventory
)
SELECT * FROM table_freshness
WHERE hours_since_update > (
  CASE table_name
    WHEN 'orders' THEN 1     -- 1 hour SLA
    WHEN 'inventory' THEN 4  -- 4 hour SLA
    ELSE 24                  -- default 24 hour SLA
  END
);

Alerting Rules

• P0 (Critical): Pipeline down, data missing for > SLA, schema corruption → Page on-call immediately
• P1 (High): Anomalous volume (>30% deviation), freshness breach warning → Notify Slack, create ticket
• P2 (Medium): Minor schema drift (nullable → non-nullable), performance degradation → Daily digest
• P3 (Low): Deprecation warnings, non-critical schema changes → Logged in weekly report

Orchestration

Airflow Patterns

# Airflow DAG with idempotent tasks
from airflow import DAG
from airflow.operators.python import PythonOperator
from datetime import datetime, timedelta

default_args = {
    "owner": "data-team",
    "retries": 3,
    "retry_delay": timedelta(minutes=5),
    "retry_exponential_backoff": True,
    "max_retry_delay": timedelta(hours=1),
}

with DAG(
    dag_id="orders_pipeline",
    start_date=datetime(2024, 1, 1),
    schedule="0 6 * * *",  # Daily at 6 AM
    catchup=False,  # Avoid automatic backfill
    tags=["production", "etl"],
    default_args=default_args,
) as dag:

    extract = PythonOperator(
        task_id="extract_orders",
        python_callable=lambda: print("Extracting..."),
    )

    validate = PythonOperator(
        task_id="validate_schema",
        python_callable=lambda: print("Validating..."),
    )

    load = PythonOperator(
        task_id="load_to_silver",
        python_callable=lambda: print("Loading..."),
    )

    # DAG structure
    extract >> validate >> load

Dagster Patterns

# Dagster with asset-based approach
from dagster import asset, AssetExecutionContext, materialize, Definitions

@asset
def raw_orders():
    """Extract raw orders from source."""
    return extract_from_api()

@asset
def silver_orders(context: AssetExecutionContext, raw_orders):
    """Clean and validate raw orders."""
    cleaned = clean_data(raw_orders)
    validate_schema(cleaned)
    context.log.info(f"Processed {len(cleaned)} orders")
    return cleaned

@asset
def gold_daily_orders(silver_orders):
    """Aggregate orders to daily grain."""
    return silver_orders.groupby("date").agg({"amount": "sum"}).reset_index()

defs = Definitions(assets=[raw_orders, silver_orders, gold_daily_orders])

Prefect Patterns

# Prefect flow with caching and retries
from prefect import flow, task
from prefect.tasks import task_input_hash
from datetime import timedelta

@task(
    retries=3,
    retry_delay_seconds=30,
    cache_key_fn=task_input_hash,
    cache_expiration=timedelta(hours=1),
)
def fetch_data(date: str) -> list:
    response = requests.get(f"https://api.example.com/orders?date={date}")
    response.raise_for_status()
    return response.json()

@task
def transform(data: list) -> list:
    return [{"order_id": d["id"], "amount": float(d["total"])} for d in data]

@flow
def orders_pipeline(date: str):
    raw = fetch_data(date)
    transformed = transform(raw)
    return transformed

DAG Design Principles

1. Single Responsibility — Each task does one thing and does it well
2. Idempotent Tasks — Re-running a task produces identical results
3. Deterministic Ordering — Dependencies are explicit, not implicit
4. Minimal Fan-out — Too many parallel tasks overwhelm resources. Batch where possible
5. Task Granularity — Too fine-grained = overhead. Too coarse = long-running, hard to restart
6. Retry from Point of Failure — Don't restart the entire DAG on a single task failure

Task Idempotency

A task is idempotent if running it N times produces the same result as running it once.

How to achieve:

• Use MERGE or INSERT OVERWRITE instead of INSERT INTO
• Include a dedup step: ROW_NUMBER() OVER (PARTITION BY id ORDER BY updated_at DESC) = 1
• Use deterministic partition overwrites: INSERT OVERWRITE TABLE orders PARTITION(ds='2024-01-01')
• For streaming: use exactly-once semantics with Kafka offsets

Backfilling

Backfilling reprocesses data for a historical time window.

Safe backfill strategy:

1. Ensure all tasks are idempotent
2. Use partition-aware processing (process only affected partitions)
3. Run backfill in a sandbox environment first
4. Validate row counts before swapping into production
5. Use Airflow's backfill feature: airflow dags backfill orders_pipeline -s 2024-01-01 -e 2024-01-07

# Safe backfill: reprocess a date range
from datetime import date, timedelta

def backfill_range(start_date: date, end_date: date):
    current = start_date
    while current <= end_date:
        # Process single partition — safe and restartable
        process_partition(current.isoformat())
        current += timedelta(days=1)

Error Handling

Retry Logic

import time
from functools import wraps

def retry_with_exponential_backoff(max_retries=5, base_delay=5):
    def decorator(func):
        @wraps(func)
        def wrapper(*args, **kwargs):
            for attempt in range(max_retries):
                try:
                    return func(*args, **kwargs)
                except Exception as e:
                    if attempt == max_retries - 1:
                        raise
                    delay = base_delay * (2 ** attempt)  # 5, 10, 20, 40, 80 seconds
                    time.sleep(delay)
            return None
        return wrapper
    return decorator

Dead Letter Queues (DLQ)

When a record can't be processed, don't drop it — route it to a DLQ:

def process_with_dlq(records: list, dlq_topic: str = "pipeline.errors"):
    success_count = 0
    error_count = 0

    for record in records:
        try:
            process_record(record)
            success_count += 1
        except Exception as e:
            # Route failed record to DLQ with error metadata
            dlq_publish({
                "original_record": record,
                "error": str(e),
                "timestamp": datetime.now().isoformat(),
                "pipeline": "orders_etl"
            })
            error_count += 1

    # Emit metrics
    print(f"Processed: {success_count}, Failed (routed to DLQ): {error_count}")

Alerting Triggers

def alert_on_failure(context):
    """Send alert when task fails after all retries."""
    dag_id = context["dag"].dag_id
    task_id = context["task"].task_id
    execution_date = context["execution_date"]

    message = f"""
    🚨 Pipeline FAILURE
    DAG: {dag_id}
    Task: {task_id}
    Execution: {execution_date}
    """

    # Send to multiple channels
    send_slack(message, channel="#data-alerts")
    send_pagerduty(message, severity="critical")

Cost Optimization

Partitioning

Partition by date — the most common and effective strategy:

-- Partition by ingestion date
CREATE TABLE orders (
    order_id STRING,
    amount DECIMAL(10,2),
    created_at TIMESTAMP
)
PARTITIONED BY (ds STRING)  -- 'yyyy-mm-dd'
STORED AS PARQUET;

Query only needed partitions:

SELECT * FROM orders WHERE ds = '2024-01-15';  -- Scans 1 partition = 1/365 of data
SELECT * FROM orders WHERE ds >= '2024-01-01' AND ds < '2024-02-01';  -- Scans 31 partitions

Incremental Processing

Never reprocess the full dataset. Track watermark and process only new/changed records.

from datetime import datetime, timedelta

def incremental_load():
    # Read watermark from last successful run
    last_run = get_watermark("orders_pipeline")  # e.g., 2024-01-15 06:00:00

    # Process only records after watermark
    new_records = fetch_orders_since(last_run)

    if new_records:
        process(new_records)
        # Update watermark to now
        set_watermark("orders_pipeline", datetime.now())
    else:
        print("No new records to process.")

Compression

Use columnar formats with compression:

Rule of thumb: Use Parquet + ZSTD for storage, Parquet + Snappy for performance-critical paths.

Additional Cost Tips

1. Predicate pushdown — Push filters into the storage layer (Parquet row group pruning)
2. Cluster by high-cardinality columns — Optimizes large joins and aggregations
3. Use materialized views — Pre-compute expensive aggregations, refresh incrementally
4. Auto-scaling — Rightsize compute per pipeline stage (Spark dynamic allocation, Airflow worker pool sizing)
5. Data lifecycle — Archive or delete data older than 90 days from hot storage; move to cold/glacier

Common Mistakes

❌ Not Handling Schema Drift

Problem: Source adds a column, pipeline silently drops it or crashes.

Solution: Implement schema-on-read with evolution strategies:

-- Delta Lake / Iceberg: allow schema evolution
ALTER TABLE bronze_orders ADD COLUMN discount DECIMAL(5,2);

Use schema registries to detect and alert on drift.

❌ No Data Quality Checks

Problem: Bad data flows silently to dashboards. Decisions are made on garbage.

Solution: Add quality gates at every stage:

• Bronze → Silver: Schema validation, null checks
• Silver → Gold: Business rule validation, uniqueness checks
• Gold → Dashboard: Freshness SLA, volume anomaly detection

❌ Fragile Dependencies

Problem: Pipelines depend on implicit upstream completion (e.g., "wait 2 hours after midnight").

Solution: Use explicit dependency tracking:

# Bad: Implicit wait
def wait_for_upstream():
    time.sleep(7200)  # Pray it's done in 2 hours

# Good: Sensor checks for upstream completion
from airflow.sensors.time_delta import TimeDeltaSensor
wait_for_upstream = ExternalTaskSensor(
    task_id="wait_for_upstream",
    external_dag_id="source_ingestion",
    external_task_id="complete",
    timeout=3600,
)

❌ Monolithic DAGs

Problem: One massive DAG with 100+ tasks. Hard to debug, impossible to maintain.

Solution: Break into focused DAGs with clear boundaries:

• ingestion_dag — Source → Bronze
• cleaning_dag — Bronze → Silver
• aggregation_dag — Silver → Gold
• export_dag — Gold → BI tool

Use ExternalTaskSensor or dataset-driven scheduling for cross-DAG dependencies.

❌ Ignoring Backfill Strategy

Problem: Need to reprocess 3 months of data but pipeline only supports incremental loads.

Solution: Design for both from day one. Include a mode parameter:

def run_pipeline(mode: str = "incremental", start_date: str = None, end_date: str = None):
    if mode == "full_refresh":
        clear_partitions(start_date, end_date)
        process_full_range(start_date, end_date)
    else:
        incremental_load()

❌ No Monitoring or Alerting

Problem: Pipeline silently fails at 2 AM, nobody notices until 9 AM standup.

Solution: Invest in observability before you need it. Set up at minimum:

1. Failure alerts (Slack + PagerDuty)
2. Freshness SLA checks
3. Volume anomaly detection
4. Pipeline duration tracking (alert on slow runs)

❌ Over-Engineering

Problem: Adding streaming infrastructure for a daily batch report.

Solution: Match architecture to actual requirements. Start batch, move to streaming only when latency demands it. "We might need real-time someday" is not a reason to build a streaming pipeline today.

Scoring & Evaluation

Use this rubric to evaluate pipeline quality:

Score targets:

• 7-10: Development/QA pipeline
• 11-15: Production pipeline (acceptable)
• 16-21: Enterprise-grade pipeline (excellent)